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_REUSEPORT_BPF:
1043 ret = reuseport_detach_prog(sk);
1046 case SO_DETACH_FILTER:
1047 ret = sk_detach_filter(sk);
1050 case SO_LOCK_FILTER:
1051 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1054 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1059 set_bit(SOCK_PASSSEC, &sock->flags);
1061 clear_bit(SOCK_PASSSEC, &sock->flags);
1064 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1066 } else if (val != sk->sk_mark) {
1073 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1076 case SO_WIFI_STATUS:
1077 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1081 if (sock->ops->set_peek_off)
1082 ret = sock->ops->set_peek_off(sk, val);
1088 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1091 case SO_SELECT_ERR_QUEUE:
1092 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1095 #ifdef CONFIG_NET_RX_BUSY_POLL
1097 /* allow unprivileged users to decrease the value */
1098 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1104 sk->sk_ll_usec = val;
1109 case SO_MAX_PACING_RATE:
1111 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1113 if (sizeof(ulval) != sizeof(val) &&
1114 optlen >= sizeof(ulval) &&
1115 get_user(ulval, (unsigned long __user *)optval)) {
1120 cmpxchg(&sk->sk_pacing_status,
1123 sk->sk_max_pacing_rate = ulval;
1124 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1127 case SO_INCOMING_CPU:
1128 sk->sk_incoming_cpu = val;
1133 dst_negative_advice(sk);
1137 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1138 if (!((sk->sk_type == SOCK_STREAM &&
1139 sk->sk_protocol == IPPROTO_TCP) ||
1140 (sk->sk_type == SOCK_DGRAM &&
1141 sk->sk_protocol == IPPROTO_UDP)))
1143 } else if (sk->sk_family != PF_RDS) {
1147 if (val < 0 || val > 1)
1150 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1155 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1157 } else if (optlen != sizeof(struct sock_txtime)) {
1159 } else if (copy_from_user(&sk_txtime, optval,
1160 sizeof(struct sock_txtime))) {
1162 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1165 sock_valbool_flag(sk, SOCK_TXTIME, true);
1166 sk->sk_clockid = sk_txtime.clockid;
1167 sk->sk_txtime_deadline_mode =
1168 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1169 sk->sk_txtime_report_errors =
1170 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1174 case SO_BINDTOIFINDEX:
1175 ret = sock_setbindtodevice_locked(sk, val);
1185 EXPORT_SYMBOL(sock_setsockopt);
1188 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1189 struct ucred *ucred)
1191 ucred->pid = pid_vnr(pid);
1192 ucred->uid = ucred->gid = -1;
1194 struct user_namespace *current_ns = current_user_ns();
1196 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1197 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1201 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1203 struct user_namespace *user_ns = current_user_ns();
1206 for (i = 0; i < src->ngroups; i++)
1207 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1213 int sock_getsockopt(struct socket *sock, int level, int optname,
1214 char __user *optval, int __user *optlen)
1216 struct sock *sk = sock->sk;
1221 unsigned long ulval;
1223 struct old_timeval32 tm32;
1224 struct __kernel_old_timeval tm;
1225 struct __kernel_sock_timeval stm;
1226 struct sock_txtime txtime;
1229 int lv = sizeof(int);
1232 if (get_user(len, optlen))
1237 memset(&v, 0, sizeof(v));
1241 v.val = sock_flag(sk, SOCK_DBG);
1245 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1249 v.val = sock_flag(sk, SOCK_BROADCAST);
1253 v.val = sk->sk_sndbuf;
1257 v.val = sk->sk_rcvbuf;
1261 v.val = sk->sk_reuse;
1265 v.val = sk->sk_reuseport;
1269 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1273 v.val = sk->sk_type;
1277 v.val = sk->sk_protocol;
1281 v.val = sk->sk_family;
1285 v.val = -sock_error(sk);
1287 v.val = xchg(&sk->sk_err_soft, 0);
1291 v.val = sock_flag(sk, SOCK_URGINLINE);
1295 v.val = sk->sk_no_check_tx;
1299 v.val = sk->sk_priority;
1303 lv = sizeof(v.ling);
1304 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1305 v.ling.l_linger = sk->sk_lingertime / HZ;
1309 sock_warn_obsolete_bsdism("getsockopt");
1312 case SO_TIMESTAMP_OLD:
1313 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1314 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1315 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1318 case SO_TIMESTAMPNS_OLD:
1319 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1322 case SO_TIMESTAMP_NEW:
1323 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1326 case SO_TIMESTAMPNS_NEW:
1327 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1330 case SO_TIMESTAMPING_OLD:
1331 v.val = sk->sk_tsflags;
1334 case SO_RCVTIMEO_OLD:
1335 case SO_RCVTIMEO_NEW:
1336 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1339 case SO_SNDTIMEO_OLD:
1340 case SO_SNDTIMEO_NEW:
1341 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1345 v.val = sk->sk_rcvlowat;
1353 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1358 struct ucred peercred;
1359 if (len > sizeof(peercred))
1360 len = sizeof(peercred);
1361 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1362 if (copy_to_user(optval, &peercred, len))
1371 if (!sk->sk_peer_cred)
1374 n = sk->sk_peer_cred->group_info->ngroups;
1375 if (len < n * sizeof(gid_t)) {
1376 len = n * sizeof(gid_t);
1377 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1379 len = n * sizeof(gid_t);
1381 ret = groups_to_user((gid_t __user *)optval,
1382 sk->sk_peer_cred->group_info);
1392 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1397 if (copy_to_user(optval, address, len))
1402 /* Dubious BSD thing... Probably nobody even uses it, but
1403 * the UNIX standard wants it for whatever reason... -DaveM
1406 v.val = sk->sk_state == TCP_LISTEN;
1410 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1414 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1417 v.val = sk->sk_mark;
1421 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1424 case SO_WIFI_STATUS:
1425 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1429 if (!sock->ops->set_peek_off)
1432 v.val = sk->sk_peek_off;
1435 v.val = sock_flag(sk, SOCK_NOFCS);
1438 case SO_BINDTODEVICE:
1439 return sock_getbindtodevice(sk, optval, optlen, len);
1442 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1448 case SO_LOCK_FILTER:
1449 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1452 case SO_BPF_EXTENSIONS:
1453 v.val = bpf_tell_extensions();
1456 case SO_SELECT_ERR_QUEUE:
1457 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1460 #ifdef CONFIG_NET_RX_BUSY_POLL
1462 v.val = sk->sk_ll_usec;
1466 case SO_MAX_PACING_RATE:
1467 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1468 lv = sizeof(v.ulval);
1469 v.ulval = sk->sk_max_pacing_rate;
1472 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1476 case SO_INCOMING_CPU:
1477 v.val = sk->sk_incoming_cpu;
1482 u32 meminfo[SK_MEMINFO_VARS];
1484 sk_get_meminfo(sk, meminfo);
1486 len = min_t(unsigned int, len, sizeof(meminfo));
1487 if (copy_to_user(optval, &meminfo, len))
1493 #ifdef CONFIG_NET_RX_BUSY_POLL
1494 case SO_INCOMING_NAPI_ID:
1495 v.val = READ_ONCE(sk->sk_napi_id);
1497 /* aggregate non-NAPI IDs down to 0 */
1498 if (v.val < MIN_NAPI_ID)
1508 v.val64 = sock_gen_cookie(sk);
1512 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1516 lv = sizeof(v.txtime);
1517 v.txtime.clockid = sk->sk_clockid;
1518 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1519 SOF_TXTIME_DEADLINE_MODE : 0;
1520 v.txtime.flags |= sk->sk_txtime_report_errors ?
1521 SOF_TXTIME_REPORT_ERRORS : 0;
1524 case SO_BINDTOIFINDEX:
1525 v.val = sk->sk_bound_dev_if;
1529 /* We implement the SO_SNDLOWAT etc to not be settable
1532 return -ENOPROTOOPT;
1537 if (copy_to_user(optval, &v, len))
1540 if (put_user(len, optlen))
1546 * Initialize an sk_lock.
1548 * (We also register the sk_lock with the lock validator.)
1550 static inline void sock_lock_init(struct sock *sk)
1552 if (sk->sk_kern_sock)
1553 sock_lock_init_class_and_name(
1555 af_family_kern_slock_key_strings[sk->sk_family],
1556 af_family_kern_slock_keys + sk->sk_family,
1557 af_family_kern_key_strings[sk->sk_family],
1558 af_family_kern_keys + sk->sk_family);
1560 sock_lock_init_class_and_name(
1562 af_family_slock_key_strings[sk->sk_family],
1563 af_family_slock_keys + sk->sk_family,
1564 af_family_key_strings[sk->sk_family],
1565 af_family_keys + sk->sk_family);
1569 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1570 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1571 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1573 static void sock_copy(struct sock *nsk, const struct sock *osk)
1575 #ifdef CONFIG_SECURITY_NETWORK
1576 void *sptr = nsk->sk_security;
1578 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1580 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1581 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1583 #ifdef CONFIG_SECURITY_NETWORK
1584 nsk->sk_security = sptr;
1585 security_sk_clone(osk, nsk);
1589 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1593 struct kmem_cache *slab;
1597 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1600 if (want_init_on_alloc(priority))
1601 sk_prot_clear_nulls(sk, prot->obj_size);
1603 sk = kmalloc(prot->obj_size, priority);
1606 if (security_sk_alloc(sk, family, priority))
1609 if (!try_module_get(prot->owner))
1611 sk_tx_queue_clear(sk);
1617 security_sk_free(sk);
1620 kmem_cache_free(slab, sk);
1626 static void sk_prot_free(struct proto *prot, struct sock *sk)
1628 struct kmem_cache *slab;
1629 struct module *owner;
1631 owner = prot->owner;
1634 cgroup_sk_free(&sk->sk_cgrp_data);
1635 mem_cgroup_sk_free(sk);
1636 security_sk_free(sk);
1638 kmem_cache_free(slab, sk);
1645 * sk_alloc - All socket objects are allocated here
1646 * @net: the applicable net namespace
1647 * @family: protocol family
1648 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1649 * @prot: struct proto associated with this new sock instance
1650 * @kern: is this to be a kernel socket?
1652 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1653 struct proto *prot, int kern)
1657 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1659 sk->sk_family = family;
1661 * See comment in struct sock definition to understand
1662 * why we need sk_prot_creator -acme
1664 sk->sk_prot = sk->sk_prot_creator = prot;
1665 sk->sk_kern_sock = kern;
1667 sk->sk_net_refcnt = kern ? 0 : 1;
1668 if (likely(sk->sk_net_refcnt)) {
1670 sock_inuse_add(net, 1);
1673 sock_net_set(sk, net);
1674 refcount_set(&sk->sk_wmem_alloc, 1);
1676 mem_cgroup_sk_alloc(sk);
1677 cgroup_sk_alloc(&sk->sk_cgrp_data);
1678 sock_update_classid(&sk->sk_cgrp_data);
1679 sock_update_netprioidx(&sk->sk_cgrp_data);
1684 EXPORT_SYMBOL(sk_alloc);
1686 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1687 * grace period. This is the case for UDP sockets and TCP listeners.
1689 static void __sk_destruct(struct rcu_head *head)
1691 struct sock *sk = container_of(head, struct sock, sk_rcu);
1692 struct sk_filter *filter;
1694 if (sk->sk_destruct)
1695 sk->sk_destruct(sk);
1697 filter = rcu_dereference_check(sk->sk_filter,
1698 refcount_read(&sk->sk_wmem_alloc) == 0);
1700 sk_filter_uncharge(sk, filter);
1701 RCU_INIT_POINTER(sk->sk_filter, NULL);
1703 if (rcu_access_pointer(sk->sk_reuseport_cb))
1704 reuseport_detach_sock(sk);
1706 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1708 #ifdef CONFIG_BPF_SYSCALL
1709 bpf_sk_storage_free(sk);
1712 if (atomic_read(&sk->sk_omem_alloc))
1713 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1714 __func__, atomic_read(&sk->sk_omem_alloc));
1716 if (sk->sk_frag.page) {
1717 put_page(sk->sk_frag.page);
1718 sk->sk_frag.page = NULL;
1721 if (sk->sk_peer_cred)
1722 put_cred(sk->sk_peer_cred);
1723 put_pid(sk->sk_peer_pid);
1724 if (likely(sk->sk_net_refcnt))
1725 put_net(sock_net(sk));
1726 sk_prot_free(sk->sk_prot_creator, sk);
1729 void sk_destruct(struct sock *sk)
1731 if (sock_flag(sk, SOCK_RCU_FREE))
1732 call_rcu(&sk->sk_rcu, __sk_destruct);
1734 __sk_destruct(&sk->sk_rcu);
1737 static void __sk_free(struct sock *sk)
1739 if (likely(sk->sk_net_refcnt))
1740 sock_inuse_add(sock_net(sk), -1);
1742 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1743 sock_diag_broadcast_destroy(sk);
1748 void sk_free(struct sock *sk)
1751 * We subtract one from sk_wmem_alloc and can know if
1752 * some packets are still in some tx queue.
1753 * If not null, sock_wfree() will call __sk_free(sk) later
1755 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1758 EXPORT_SYMBOL(sk_free);
1760 static void sk_init_common(struct sock *sk)
1762 skb_queue_head_init(&sk->sk_receive_queue);
1763 skb_queue_head_init(&sk->sk_write_queue);
1764 skb_queue_head_init(&sk->sk_error_queue);
1766 rwlock_init(&sk->sk_callback_lock);
1767 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1768 af_rlock_keys + sk->sk_family,
1769 af_family_rlock_key_strings[sk->sk_family]);
1770 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1771 af_wlock_keys + sk->sk_family,
1772 af_family_wlock_key_strings[sk->sk_family]);
1773 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1774 af_elock_keys + sk->sk_family,
1775 af_family_elock_key_strings[sk->sk_family]);
1776 lockdep_set_class_and_name(&sk->sk_callback_lock,
1777 af_callback_keys + sk->sk_family,
1778 af_family_clock_key_strings[sk->sk_family]);
1782 * sk_clone_lock - clone a socket, and lock its clone
1783 * @sk: the socket to clone
1784 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1786 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1788 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1791 bool is_charged = true;
1793 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1794 if (newsk != NULL) {
1795 struct sk_filter *filter;
1797 sock_copy(newsk, sk);
1799 newsk->sk_prot_creator = sk->sk_prot;
1802 if (likely(newsk->sk_net_refcnt))
1803 get_net(sock_net(newsk));
1804 sk_node_init(&newsk->sk_node);
1805 sock_lock_init(newsk);
1806 bh_lock_sock(newsk);
1807 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1808 newsk->sk_backlog.len = 0;
1810 atomic_set(&newsk->sk_rmem_alloc, 0);
1812 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1814 refcount_set(&newsk->sk_wmem_alloc, 1);
1815 atomic_set(&newsk->sk_omem_alloc, 0);
1816 sk_init_common(newsk);
1818 newsk->sk_dst_cache = NULL;
1819 newsk->sk_dst_pending_confirm = 0;
1820 newsk->sk_wmem_queued = 0;
1821 newsk->sk_forward_alloc = 0;
1822 atomic_set(&newsk->sk_drops, 0);
1823 newsk->sk_send_head = NULL;
1824 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1825 atomic_set(&newsk->sk_zckey, 0);
1827 sock_reset_flag(newsk, SOCK_DONE);
1828 mem_cgroup_sk_alloc(newsk);
1829 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1832 filter = rcu_dereference(sk->sk_filter);
1834 /* though it's an empty new sock, the charging may fail
1835 * if sysctl_optmem_max was changed between creation of
1836 * original socket and cloning
1838 is_charged = sk_filter_charge(newsk, filter);
1839 RCU_INIT_POINTER(newsk->sk_filter, filter);
1842 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1843 /* We need to make sure that we don't uncharge the new
1844 * socket if we couldn't charge it in the first place
1845 * as otherwise we uncharge the parent's filter.
1848 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1849 sk_free_unlock_clone(newsk);
1853 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1855 if (bpf_sk_storage_clone(sk, newsk)) {
1856 sk_free_unlock_clone(newsk);
1862 newsk->sk_err_soft = 0;
1863 newsk->sk_priority = 0;
1864 newsk->sk_incoming_cpu = raw_smp_processor_id();
1865 if (likely(newsk->sk_net_refcnt))
1866 sock_inuse_add(sock_net(newsk), 1);
1869 * Before updating sk_refcnt, we must commit prior changes to memory
1870 * (Documentation/RCU/rculist_nulls.txt for details)
1873 refcount_set(&newsk->sk_refcnt, 2);
1876 * Increment the counter in the same struct proto as the master
1877 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1878 * is the same as sk->sk_prot->socks, as this field was copied
1881 * This _changes_ the previous behaviour, where
1882 * tcp_create_openreq_child always was incrementing the
1883 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1884 * to be taken into account in all callers. -acme
1886 sk_refcnt_debug_inc(newsk);
1887 sk_set_socket(newsk, NULL);
1888 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1890 if (newsk->sk_prot->sockets_allocated)
1891 sk_sockets_allocated_inc(newsk);
1893 if (sock_needs_netstamp(sk) &&
1894 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1895 net_enable_timestamp();
1900 EXPORT_SYMBOL_GPL(sk_clone_lock);
1902 void sk_free_unlock_clone(struct sock *sk)
1904 /* It is still raw copy of parent, so invalidate
1905 * destructor and make plain sk_free() */
1906 sk->sk_destruct = NULL;
1910 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1912 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1916 sk_dst_set(sk, dst);
1917 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1918 if (sk->sk_route_caps & NETIF_F_GSO)
1919 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1920 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1921 if (sk_can_gso(sk)) {
1922 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1923 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1925 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1926 sk->sk_gso_max_size = dst->dev->gso_max_size;
1927 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1930 sk->sk_gso_max_segs = max_segs;
1932 EXPORT_SYMBOL_GPL(sk_setup_caps);
1935 * Simple resource managers for sockets.
1940 * Write buffer destructor automatically called from kfree_skb.
1942 void sock_wfree(struct sk_buff *skb)
1944 struct sock *sk = skb->sk;
1945 unsigned int len = skb->truesize;
1947 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1949 * Keep a reference on sk_wmem_alloc, this will be released
1950 * after sk_write_space() call
1952 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1953 sk->sk_write_space(sk);
1957 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1958 * could not do because of in-flight packets
1960 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1963 EXPORT_SYMBOL(sock_wfree);
1965 /* This variant of sock_wfree() is used by TCP,
1966 * since it sets SOCK_USE_WRITE_QUEUE.
1968 void __sock_wfree(struct sk_buff *skb)
1970 struct sock *sk = skb->sk;
1972 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1976 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1981 if (unlikely(!sk_fullsock(sk))) {
1982 skb->destructor = sock_edemux;
1987 skb->destructor = sock_wfree;
1988 skb_set_hash_from_sk(skb, sk);
1990 * We used to take a refcount on sk, but following operation
1991 * is enough to guarantee sk_free() wont free this sock until
1992 * all in-flight packets are completed
1994 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1996 EXPORT_SYMBOL(skb_set_owner_w);
1998 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2000 #ifdef CONFIG_TLS_DEVICE
2001 /* Drivers depend on in-order delivery for crypto offload,
2002 * partial orphan breaks out-of-order-OK logic.
2007 return (skb->destructor == sock_wfree ||
2008 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2011 /* This helper is used by netem, as it can hold packets in its
2012 * delay queue. We want to allow the owner socket to send more
2013 * packets, as if they were already TX completed by a typical driver.
2014 * But we also want to keep skb->sk set because some packet schedulers
2015 * rely on it (sch_fq for example).
2017 void skb_orphan_partial(struct sk_buff *skb)
2019 if (skb_is_tcp_pure_ack(skb))
2022 if (can_skb_orphan_partial(skb)) {
2023 struct sock *sk = skb->sk;
2025 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2026 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2027 skb->destructor = sock_efree;
2033 EXPORT_SYMBOL(skb_orphan_partial);
2036 * Read buffer destructor automatically called from kfree_skb.
2038 void sock_rfree(struct sk_buff *skb)
2040 struct sock *sk = skb->sk;
2041 unsigned int len = skb->truesize;
2043 atomic_sub(len, &sk->sk_rmem_alloc);
2044 sk_mem_uncharge(sk, len);
2046 EXPORT_SYMBOL(sock_rfree);
2049 * Buffer destructor for skbs that are not used directly in read or write
2050 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2052 void sock_efree(struct sk_buff *skb)
2056 EXPORT_SYMBOL(sock_efree);
2058 kuid_t sock_i_uid(struct sock *sk)
2062 read_lock_bh(&sk->sk_callback_lock);
2063 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2064 read_unlock_bh(&sk->sk_callback_lock);
2067 EXPORT_SYMBOL(sock_i_uid);
2069 unsigned long sock_i_ino(struct sock *sk)
2073 read_lock_bh(&sk->sk_callback_lock);
2074 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2075 read_unlock_bh(&sk->sk_callback_lock);
2078 EXPORT_SYMBOL(sock_i_ino);
2081 * Allocate a skb from the socket's send buffer.
2083 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2086 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2087 struct sk_buff *skb = alloc_skb(size, priority);
2089 skb_set_owner_w(skb, sk);
2095 EXPORT_SYMBOL(sock_wmalloc);
2097 static void sock_ofree(struct sk_buff *skb)
2099 struct sock *sk = skb->sk;
2101 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2104 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2107 struct sk_buff *skb;
2109 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2110 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2114 skb = alloc_skb(size, priority);
2118 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2120 skb->destructor = sock_ofree;
2125 * Allocate a memory block from the socket's option memory buffer.
2127 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2129 if ((unsigned int)size <= sysctl_optmem_max &&
2130 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2132 /* First do the add, to avoid the race if kmalloc
2135 atomic_add(size, &sk->sk_omem_alloc);
2136 mem = kmalloc(size, priority);
2139 atomic_sub(size, &sk->sk_omem_alloc);
2143 EXPORT_SYMBOL(sock_kmalloc);
2145 /* Free an option memory block. Note, we actually want the inline
2146 * here as this allows gcc to detect the nullify and fold away the
2147 * condition entirely.
2149 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2152 if (WARN_ON_ONCE(!mem))
2158 atomic_sub(size, &sk->sk_omem_alloc);
2161 void sock_kfree_s(struct sock *sk, void *mem, int size)
2163 __sock_kfree_s(sk, mem, size, false);
2165 EXPORT_SYMBOL(sock_kfree_s);
2167 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2169 __sock_kfree_s(sk, mem, size, true);
2171 EXPORT_SYMBOL(sock_kzfree_s);
2173 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2174 I think, these locks should be removed for datagram sockets.
2176 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2180 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2184 if (signal_pending(current))
2186 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2187 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2188 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2190 if (sk->sk_shutdown & SEND_SHUTDOWN)
2194 timeo = schedule_timeout(timeo);
2196 finish_wait(sk_sleep(sk), &wait);
2202 * Generic send/receive buffer handlers
2205 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2206 unsigned long data_len, int noblock,
2207 int *errcode, int max_page_order)
2209 struct sk_buff *skb;
2213 timeo = sock_sndtimeo(sk, noblock);
2215 err = sock_error(sk);
2220 if (sk->sk_shutdown & SEND_SHUTDOWN)
2223 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2226 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2227 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2231 if (signal_pending(current))
2233 timeo = sock_wait_for_wmem(sk, timeo);
2235 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2236 errcode, sk->sk_allocation);
2238 skb_set_owner_w(skb, sk);
2242 err = sock_intr_errno(timeo);
2247 EXPORT_SYMBOL(sock_alloc_send_pskb);
2249 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2250 int noblock, int *errcode)
2252 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2254 EXPORT_SYMBOL(sock_alloc_send_skb);
2256 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2257 struct sockcm_cookie *sockc)
2261 switch (cmsg->cmsg_type) {
2263 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2265 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2267 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2269 case SO_TIMESTAMPING_OLD:
2270 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2273 tsflags = *(u32 *)CMSG_DATA(cmsg);
2274 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2277 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2278 sockc->tsflags |= tsflags;
2281 if (!sock_flag(sk, SOCK_TXTIME))
2283 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2285 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2287 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2289 case SCM_CREDENTIALS:
2296 EXPORT_SYMBOL(__sock_cmsg_send);
2298 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2299 struct sockcm_cookie *sockc)
2301 struct cmsghdr *cmsg;
2304 for_each_cmsghdr(cmsg, msg) {
2305 if (!CMSG_OK(msg, cmsg))
2307 if (cmsg->cmsg_level != SOL_SOCKET)
2309 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2315 EXPORT_SYMBOL(sock_cmsg_send);
2317 static void sk_enter_memory_pressure(struct sock *sk)
2319 if (!sk->sk_prot->enter_memory_pressure)
2322 sk->sk_prot->enter_memory_pressure(sk);
2325 static void sk_leave_memory_pressure(struct sock *sk)
2327 if (sk->sk_prot->leave_memory_pressure) {
2328 sk->sk_prot->leave_memory_pressure(sk);
2330 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2332 if (memory_pressure && *memory_pressure)
2333 *memory_pressure = 0;
2337 /* On 32bit arches, an skb frag is limited to 2^15 */
2338 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2339 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2342 * skb_page_frag_refill - check that a page_frag contains enough room
2343 * @sz: minimum size of the fragment we want to get
2344 * @pfrag: pointer to page_frag
2345 * @gfp: priority for memory allocation
2347 * Note: While this allocator tries to use high order pages, there is
2348 * no guarantee that allocations succeed. Therefore, @sz MUST be
2349 * less or equal than PAGE_SIZE.
2351 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2354 if (page_ref_count(pfrag->page) == 1) {
2358 if (pfrag->offset + sz <= pfrag->size)
2360 put_page(pfrag->page);
2364 if (SKB_FRAG_PAGE_ORDER &&
2365 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2366 /* Avoid direct reclaim but allow kswapd to wake */
2367 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2368 __GFP_COMP | __GFP_NOWARN |
2370 SKB_FRAG_PAGE_ORDER);
2371 if (likely(pfrag->page)) {
2372 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2376 pfrag->page = alloc_page(gfp);
2377 if (likely(pfrag->page)) {
2378 pfrag->size = PAGE_SIZE;
2383 EXPORT_SYMBOL(skb_page_frag_refill);
2385 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2387 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2390 sk_enter_memory_pressure(sk);
2391 sk_stream_moderate_sndbuf(sk);
2394 EXPORT_SYMBOL(sk_page_frag_refill);
2396 static void __lock_sock(struct sock *sk)
2397 __releases(&sk->sk_lock.slock)
2398 __acquires(&sk->sk_lock.slock)
2403 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2404 TASK_UNINTERRUPTIBLE);
2405 spin_unlock_bh(&sk->sk_lock.slock);
2407 spin_lock_bh(&sk->sk_lock.slock);
2408 if (!sock_owned_by_user(sk))
2411 finish_wait(&sk->sk_lock.wq, &wait);
2414 void __release_sock(struct sock *sk)
2415 __releases(&sk->sk_lock.slock)
2416 __acquires(&sk->sk_lock.slock)
2418 struct sk_buff *skb, *next;
2420 while ((skb = sk->sk_backlog.head) != NULL) {
2421 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2423 spin_unlock_bh(&sk->sk_lock.slock);
2428 WARN_ON_ONCE(skb_dst_is_noref(skb));
2429 skb_mark_not_on_list(skb);
2430 sk_backlog_rcv(sk, skb);
2435 } while (skb != NULL);
2437 spin_lock_bh(&sk->sk_lock.slock);
2441 * Doing the zeroing here guarantee we can not loop forever
2442 * while a wild producer attempts to flood us.
2444 sk->sk_backlog.len = 0;
2447 void __sk_flush_backlog(struct sock *sk)
2449 spin_lock_bh(&sk->sk_lock.slock);
2451 spin_unlock_bh(&sk->sk_lock.slock);
2455 * sk_wait_data - wait for data to arrive at sk_receive_queue
2456 * @sk: sock to wait on
2457 * @timeo: for how long
2458 * @skb: last skb seen on sk_receive_queue
2460 * Now socket state including sk->sk_err is changed only under lock,
2461 * hence we may omit checks after joining wait queue.
2462 * We check receive queue before schedule() only as optimization;
2463 * it is very likely that release_sock() added new data.
2465 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2467 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2470 add_wait_queue(sk_sleep(sk), &wait);
2471 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2472 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2473 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2474 remove_wait_queue(sk_sleep(sk), &wait);
2477 EXPORT_SYMBOL(sk_wait_data);
2480 * __sk_mem_raise_allocated - increase memory_allocated
2482 * @size: memory size to allocate
2483 * @amt: pages to allocate
2484 * @kind: allocation type
2486 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2488 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2490 struct proto *prot = sk->sk_prot;
2491 long allocated = sk_memory_allocated_add(sk, amt);
2492 bool charged = true;
2494 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2495 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2496 goto suppress_allocation;
2499 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2500 sk_leave_memory_pressure(sk);
2504 /* Under pressure. */
2505 if (allocated > sk_prot_mem_limits(sk, 1))
2506 sk_enter_memory_pressure(sk);
2508 /* Over hard limit. */
2509 if (allocated > sk_prot_mem_limits(sk, 2))
2510 goto suppress_allocation;
2512 /* guarantee minimum buffer size under pressure */
2513 if (kind == SK_MEM_RECV) {
2514 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2517 } else { /* SK_MEM_SEND */
2518 int wmem0 = sk_get_wmem0(sk, prot);
2520 if (sk->sk_type == SOCK_STREAM) {
2521 if (sk->sk_wmem_queued < wmem0)
2523 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2528 if (sk_has_memory_pressure(sk)) {
2531 if (!sk_under_memory_pressure(sk))
2533 alloc = sk_sockets_allocated_read_positive(sk);
2534 if (sk_prot_mem_limits(sk, 2) > alloc *
2535 sk_mem_pages(sk->sk_wmem_queued +
2536 atomic_read(&sk->sk_rmem_alloc) +
2537 sk->sk_forward_alloc))
2541 suppress_allocation:
2543 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2544 sk_stream_moderate_sndbuf(sk);
2546 /* Fail only if socket is _under_ its sndbuf.
2547 * In this case we cannot block, so that we have to fail.
2549 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2553 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2554 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2556 sk_memory_allocated_sub(sk, amt);
2558 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2559 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2563 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2566 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2568 * @size: memory size to allocate
2569 * @kind: allocation type
2571 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2572 * rmem allocation. This function assumes that protocols which have
2573 * memory_pressure use sk_wmem_queued as write buffer accounting.
2575 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2577 int ret, amt = sk_mem_pages(size);
2579 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2580 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2582 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2585 EXPORT_SYMBOL(__sk_mem_schedule);
2588 * __sk_mem_reduce_allocated - reclaim memory_allocated
2590 * @amount: number of quanta
2592 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2594 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2596 sk_memory_allocated_sub(sk, amount);
2598 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2599 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2601 if (sk_under_memory_pressure(sk) &&
2602 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2603 sk_leave_memory_pressure(sk);
2605 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2608 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2610 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2612 void __sk_mem_reclaim(struct sock *sk, int amount)
2614 amount >>= SK_MEM_QUANTUM_SHIFT;
2615 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2616 __sk_mem_reduce_allocated(sk, amount);
2618 EXPORT_SYMBOL(__sk_mem_reclaim);
2620 int sk_set_peek_off(struct sock *sk, int val)
2622 sk->sk_peek_off = val;
2625 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2628 * Set of default routines for initialising struct proto_ops when
2629 * the protocol does not support a particular function. In certain
2630 * cases where it makes no sense for a protocol to have a "do nothing"
2631 * function, some default processing is provided.
2634 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2638 EXPORT_SYMBOL(sock_no_bind);
2640 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2645 EXPORT_SYMBOL(sock_no_connect);
2647 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2651 EXPORT_SYMBOL(sock_no_socketpair);
2653 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2658 EXPORT_SYMBOL(sock_no_accept);
2660 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2665 EXPORT_SYMBOL(sock_no_getname);
2667 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2671 EXPORT_SYMBOL(sock_no_ioctl);
2673 int sock_no_listen(struct socket *sock, int backlog)
2677 EXPORT_SYMBOL(sock_no_listen);
2679 int sock_no_shutdown(struct socket *sock, int how)
2683 EXPORT_SYMBOL(sock_no_shutdown);
2685 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2686 char __user *optval, unsigned int optlen)
2690 EXPORT_SYMBOL(sock_no_setsockopt);
2692 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2693 char __user *optval, int __user *optlen)
2697 EXPORT_SYMBOL(sock_no_getsockopt);
2699 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2703 EXPORT_SYMBOL(sock_no_sendmsg);
2705 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2709 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2711 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2716 EXPORT_SYMBOL(sock_no_recvmsg);
2718 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2720 /* Mirror missing mmap method error code */
2723 EXPORT_SYMBOL(sock_no_mmap);
2725 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2728 struct msghdr msg = {.msg_flags = flags};
2730 char *kaddr = kmap(page);
2731 iov.iov_base = kaddr + offset;
2733 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2737 EXPORT_SYMBOL(sock_no_sendpage);
2739 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2740 int offset, size_t size, int flags)
2743 struct msghdr msg = {.msg_flags = flags};
2745 char *kaddr = kmap(page);
2747 iov.iov_base = kaddr + offset;
2749 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2753 EXPORT_SYMBOL(sock_no_sendpage_locked);
2756 * Default Socket Callbacks
2759 static void sock_def_wakeup(struct sock *sk)
2761 struct socket_wq *wq;
2764 wq = rcu_dereference(sk->sk_wq);
2765 if (skwq_has_sleeper(wq))
2766 wake_up_interruptible_all(&wq->wait);
2770 static void sock_def_error_report(struct sock *sk)
2772 struct socket_wq *wq;
2775 wq = rcu_dereference(sk->sk_wq);
2776 if (skwq_has_sleeper(wq))
2777 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2778 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2782 static void sock_def_readable(struct sock *sk)
2784 struct socket_wq *wq;
2787 wq = rcu_dereference(sk->sk_wq);
2788 if (skwq_has_sleeper(wq))
2789 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2790 EPOLLRDNORM | EPOLLRDBAND);
2791 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2795 static void sock_def_write_space(struct sock *sk)
2797 struct socket_wq *wq;
2801 /* Do not wake up a writer until he can make "significant"
2804 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2805 wq = rcu_dereference(sk->sk_wq);
2806 if (skwq_has_sleeper(wq))
2807 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2808 EPOLLWRNORM | EPOLLWRBAND);
2810 /* Should agree with poll, otherwise some programs break */
2811 if (sock_writeable(sk))
2812 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2818 static void sock_def_destruct(struct sock *sk)
2822 void sk_send_sigurg(struct sock *sk)
2824 if (sk->sk_socket && sk->sk_socket->file)
2825 if (send_sigurg(&sk->sk_socket->file->f_owner))
2826 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2828 EXPORT_SYMBOL(sk_send_sigurg);
2830 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2831 unsigned long expires)
2833 if (!mod_timer(timer, expires))
2836 EXPORT_SYMBOL(sk_reset_timer);
2838 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2840 if (del_timer(timer))
2843 EXPORT_SYMBOL(sk_stop_timer);
2845 void sock_init_data(struct socket *sock, struct sock *sk)
2848 sk->sk_send_head = NULL;
2850 timer_setup(&sk->sk_timer, NULL, 0);
2852 sk->sk_allocation = GFP_KERNEL;
2853 sk->sk_rcvbuf = sysctl_rmem_default;
2854 sk->sk_sndbuf = sysctl_wmem_default;
2855 sk->sk_state = TCP_CLOSE;
2856 sk_set_socket(sk, sock);
2858 sock_set_flag(sk, SOCK_ZAPPED);
2861 sk->sk_type = sock->type;
2862 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2864 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2866 RCU_INIT_POINTER(sk->sk_wq, NULL);
2867 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2870 rwlock_init(&sk->sk_callback_lock);
2871 if (sk->sk_kern_sock)
2872 lockdep_set_class_and_name(
2873 &sk->sk_callback_lock,
2874 af_kern_callback_keys + sk->sk_family,
2875 af_family_kern_clock_key_strings[sk->sk_family]);
2877 lockdep_set_class_and_name(
2878 &sk->sk_callback_lock,
2879 af_callback_keys + sk->sk_family,
2880 af_family_clock_key_strings[sk->sk_family]);
2882 sk->sk_state_change = sock_def_wakeup;
2883 sk->sk_data_ready = sock_def_readable;
2884 sk->sk_write_space = sock_def_write_space;
2885 sk->sk_error_report = sock_def_error_report;
2886 sk->sk_destruct = sock_def_destruct;
2888 sk->sk_frag.page = NULL;
2889 sk->sk_frag.offset = 0;
2890 sk->sk_peek_off = -1;
2892 sk->sk_peer_pid = NULL;
2893 sk->sk_peer_cred = NULL;
2894 sk->sk_write_pending = 0;
2895 sk->sk_rcvlowat = 1;
2896 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2897 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2899 sk->sk_stamp = SK_DEFAULT_STAMP;
2900 #if BITS_PER_LONG==32
2901 seqlock_init(&sk->sk_stamp_seq);
2903 atomic_set(&sk->sk_zckey, 0);
2905 #ifdef CONFIG_NET_RX_BUSY_POLL
2907 sk->sk_ll_usec = sysctl_net_busy_read;
2910 sk->sk_max_pacing_rate = ~0UL;
2911 sk->sk_pacing_rate = ~0UL;
2912 sk->sk_pacing_shift = 10;
2913 sk->sk_incoming_cpu = -1;
2915 sk_rx_queue_clear(sk);
2917 * Before updating sk_refcnt, we must commit prior changes to memory
2918 * (Documentation/RCU/rculist_nulls.txt for details)
2921 refcount_set(&sk->sk_refcnt, 1);
2922 atomic_set(&sk->sk_drops, 0);
2924 EXPORT_SYMBOL(sock_init_data);
2926 void lock_sock_nested(struct sock *sk, int subclass)
2929 spin_lock_bh(&sk->sk_lock.slock);
2930 if (sk->sk_lock.owned)
2932 sk->sk_lock.owned = 1;
2933 spin_unlock(&sk->sk_lock.slock);
2935 * The sk_lock has mutex_lock() semantics here:
2937 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2940 EXPORT_SYMBOL(lock_sock_nested);
2942 void release_sock(struct sock *sk)
2944 spin_lock_bh(&sk->sk_lock.slock);
2945 if (sk->sk_backlog.tail)
2948 /* Warning : release_cb() might need to release sk ownership,
2949 * ie call sock_release_ownership(sk) before us.
2951 if (sk->sk_prot->release_cb)
2952 sk->sk_prot->release_cb(sk);
2954 sock_release_ownership(sk);
2955 if (waitqueue_active(&sk->sk_lock.wq))
2956 wake_up(&sk->sk_lock.wq);
2957 spin_unlock_bh(&sk->sk_lock.slock);
2959 EXPORT_SYMBOL(release_sock);
2962 * lock_sock_fast - fast version of lock_sock
2965 * This version should be used for very small section, where process wont block
2966 * return false if fast path is taken:
2968 * sk_lock.slock locked, owned = 0, BH disabled
2970 * return true if slow path is taken:
2972 * sk_lock.slock unlocked, owned = 1, BH enabled
2974 bool lock_sock_fast(struct sock *sk)
2977 spin_lock_bh(&sk->sk_lock.slock);
2979 if (!sk->sk_lock.owned)
2981 * Note : We must disable BH
2986 sk->sk_lock.owned = 1;
2987 spin_unlock(&sk->sk_lock.slock);
2989 * The sk_lock has mutex_lock() semantics here:
2991 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2995 EXPORT_SYMBOL(lock_sock_fast);
2997 int sock_gettstamp(struct socket *sock, void __user *userstamp,
2998 bool timeval, bool time32)
3000 struct sock *sk = sock->sk;
3001 struct timespec64 ts;
3003 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3004 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3005 if (ts.tv_sec == -1)
3007 if (ts.tv_sec == 0) {
3008 ktime_t kt = ktime_get_real();
3009 sock_write_timestamp(sk, kt);;
3010 ts = ktime_to_timespec64(kt);
3016 #ifdef CONFIG_COMPAT_32BIT_TIME
3018 return put_old_timespec32(&ts, userstamp);
3020 #ifdef CONFIG_SPARC64
3021 /* beware of padding in sparc64 timeval */
3022 if (timeval && !in_compat_syscall()) {
3023 struct __kernel_old_timeval __user tv = {
3024 .tv_sec = ts.tv_sec,
3025 .tv_usec = ts.tv_nsec,
3027 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3032 return put_timespec64(&ts, userstamp);
3034 EXPORT_SYMBOL(sock_gettstamp);
3036 void sock_enable_timestamp(struct sock *sk, int flag)
3038 if (!sock_flag(sk, flag)) {
3039 unsigned long previous_flags = sk->sk_flags;
3041 sock_set_flag(sk, flag);
3043 * we just set one of the two flags which require net
3044 * time stamping, but time stamping might have been on
3045 * already because of the other one
3047 if (sock_needs_netstamp(sk) &&
3048 !(previous_flags & SK_FLAGS_TIMESTAMP))
3049 net_enable_timestamp();
3053 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3054 int level, int type)
3056 struct sock_exterr_skb *serr;
3057 struct sk_buff *skb;
3061 skb = sock_dequeue_err_skb(sk);
3067 msg->msg_flags |= MSG_TRUNC;
3070 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3074 sock_recv_timestamp(msg, sk, skb);
3076 serr = SKB_EXT_ERR(skb);
3077 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3079 msg->msg_flags |= MSG_ERRQUEUE;
3087 EXPORT_SYMBOL(sock_recv_errqueue);
3090 * Get a socket option on an socket.
3092 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3093 * asynchronous errors should be reported by getsockopt. We assume
3094 * this means if you specify SO_ERROR (otherwise whats the point of it).
3096 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3097 char __user *optval, int __user *optlen)
3099 struct sock *sk = sock->sk;
3101 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3103 EXPORT_SYMBOL(sock_common_getsockopt);
3105 #ifdef CONFIG_COMPAT
3106 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3107 char __user *optval, int __user *optlen)
3109 struct sock *sk = sock->sk;
3111 if (sk->sk_prot->compat_getsockopt != NULL)
3112 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3114 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3116 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3119 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3122 struct sock *sk = sock->sk;
3126 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3127 flags & ~MSG_DONTWAIT, &addr_len);
3129 msg->msg_namelen = addr_len;
3132 EXPORT_SYMBOL(sock_common_recvmsg);
3135 * Set socket options on an inet socket.
3137 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3138 char __user *optval, unsigned int optlen)
3140 struct sock *sk = sock->sk;
3142 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3144 EXPORT_SYMBOL(sock_common_setsockopt);
3146 #ifdef CONFIG_COMPAT
3147 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3148 char __user *optval, unsigned int optlen)
3150 struct sock *sk = sock->sk;
3152 if (sk->sk_prot->compat_setsockopt != NULL)
3153 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3155 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3157 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3160 void sk_common_release(struct sock *sk)
3162 if (sk->sk_prot->destroy)
3163 sk->sk_prot->destroy(sk);
3166 * Observation: when sock_common_release is called, processes have
3167 * no access to socket. But net still has.
3168 * Step one, detach it from networking:
3170 * A. Remove from hash tables.
3173 sk->sk_prot->unhash(sk);
3176 * In this point socket cannot receive new packets, but it is possible
3177 * that some packets are in flight because some CPU runs receiver and
3178 * did hash table lookup before we unhashed socket. They will achieve
3179 * receive queue and will be purged by socket destructor.
3181 * Also we still have packets pending on receive queue and probably,
3182 * our own packets waiting in device queues. sock_destroy will drain
3183 * receive queue, but transmitted packets will delay socket destruction
3184 * until the last reference will be released.
3189 xfrm_sk_free_policy(sk);
3191 sk_refcnt_debug_release(sk);
3195 EXPORT_SYMBOL(sk_common_release);
3197 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3199 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3201 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3202 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3203 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3204 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3205 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3206 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3207 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3208 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3209 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3212 #ifdef CONFIG_PROC_FS
3213 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3215 int val[PROTO_INUSE_NR];
3218 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3220 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3222 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3224 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3226 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3228 int cpu, idx = prot->inuse_idx;
3231 for_each_possible_cpu(cpu)
3232 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3234 return res >= 0 ? res : 0;
3236 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3238 static void sock_inuse_add(struct net *net, int val)
3240 this_cpu_add(*net->core.sock_inuse, val);
3243 int sock_inuse_get(struct net *net)
3247 for_each_possible_cpu(cpu)
3248 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3253 EXPORT_SYMBOL_GPL(sock_inuse_get);
3255 static int __net_init sock_inuse_init_net(struct net *net)
3257 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3258 if (net->core.prot_inuse == NULL)
3261 net->core.sock_inuse = alloc_percpu(int);
3262 if (net->core.sock_inuse == NULL)
3268 free_percpu(net->core.prot_inuse);
3272 static void __net_exit sock_inuse_exit_net(struct net *net)
3274 free_percpu(net->core.prot_inuse);
3275 free_percpu(net->core.sock_inuse);
3278 static struct pernet_operations net_inuse_ops = {
3279 .init = sock_inuse_init_net,
3280 .exit = sock_inuse_exit_net,
3283 static __init int net_inuse_init(void)
3285 if (register_pernet_subsys(&net_inuse_ops))
3286 panic("Cannot initialize net inuse counters");
3291 core_initcall(net_inuse_init);
3293 static int assign_proto_idx(struct proto *prot)
3295 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3297 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3298 pr_err("PROTO_INUSE_NR exhausted\n");
3302 set_bit(prot->inuse_idx, proto_inuse_idx);
3306 static void release_proto_idx(struct proto *prot)
3308 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3309 clear_bit(prot->inuse_idx, proto_inuse_idx);
3312 static inline int assign_proto_idx(struct proto *prot)
3317 static inline void release_proto_idx(struct proto *prot)
3321 static void sock_inuse_add(struct net *net, int val)
3326 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3330 kfree(rsk_prot->slab_name);
3331 rsk_prot->slab_name = NULL;
3332 kmem_cache_destroy(rsk_prot->slab);
3333 rsk_prot->slab = NULL;
3336 static int req_prot_init(const struct proto *prot)
3338 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3343 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3345 if (!rsk_prot->slab_name)
3348 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3349 rsk_prot->obj_size, 0,
3350 SLAB_ACCOUNT | prot->slab_flags,
3353 if (!rsk_prot->slab) {
3354 pr_crit("%s: Can't create request sock SLAB cache!\n",
3361 int proto_register(struct proto *prot, int alloc_slab)
3366 prot->slab = kmem_cache_create_usercopy(prot->name,
3368 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3370 prot->useroffset, prot->usersize,
3373 if (prot->slab == NULL) {
3374 pr_crit("%s: Can't create sock SLAB cache!\n",
3379 if (req_prot_init(prot))
3380 goto out_free_request_sock_slab;
3382 if (prot->twsk_prot != NULL) {
3383 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3385 if (prot->twsk_prot->twsk_slab_name == NULL)
3386 goto out_free_request_sock_slab;
3388 prot->twsk_prot->twsk_slab =
3389 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3390 prot->twsk_prot->twsk_obj_size,
3395 if (prot->twsk_prot->twsk_slab == NULL)
3396 goto out_free_timewait_sock_slab_name;
3400 mutex_lock(&proto_list_mutex);
3401 ret = assign_proto_idx(prot);
3403 mutex_unlock(&proto_list_mutex);
3404 goto out_free_timewait_sock_slab_name;
3406 list_add(&prot->node, &proto_list);
3407 mutex_unlock(&proto_list_mutex);
3410 out_free_timewait_sock_slab_name:
3411 if (alloc_slab && prot->twsk_prot)
3412 kfree(prot->twsk_prot->twsk_slab_name);
3413 out_free_request_sock_slab:
3415 req_prot_cleanup(prot->rsk_prot);
3417 kmem_cache_destroy(prot->slab);
3423 EXPORT_SYMBOL(proto_register);
3425 void proto_unregister(struct proto *prot)
3427 mutex_lock(&proto_list_mutex);
3428 release_proto_idx(prot);
3429 list_del(&prot->node);
3430 mutex_unlock(&proto_list_mutex);
3432 kmem_cache_destroy(prot->slab);
3435 req_prot_cleanup(prot->rsk_prot);
3437 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3438 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3439 kfree(prot->twsk_prot->twsk_slab_name);
3440 prot->twsk_prot->twsk_slab = NULL;
3443 EXPORT_SYMBOL(proto_unregister);
3445 int sock_load_diag_module(int family, int protocol)
3448 if (!sock_is_registered(family))
3451 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3452 NETLINK_SOCK_DIAG, family);
3456 if (family == AF_INET &&
3457 protocol != IPPROTO_RAW &&
3458 !rcu_access_pointer(inet_protos[protocol]))
3462 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3463 NETLINK_SOCK_DIAG, family, protocol);
3465 EXPORT_SYMBOL(sock_load_diag_module);
3467 #ifdef CONFIG_PROC_FS
3468 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3469 __acquires(proto_list_mutex)
3471 mutex_lock(&proto_list_mutex);
3472 return seq_list_start_head(&proto_list, *pos);
3475 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3477 return seq_list_next(v, &proto_list, pos);
3480 static void proto_seq_stop(struct seq_file *seq, void *v)
3481 __releases(proto_list_mutex)
3483 mutex_unlock(&proto_list_mutex);
3486 static char proto_method_implemented(const void *method)
3488 return method == NULL ? 'n' : 'y';
3490 static long sock_prot_memory_allocated(struct proto *proto)
3492 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3495 static char *sock_prot_memory_pressure(struct proto *proto)
3497 return proto->memory_pressure != NULL ?
3498 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3501 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3504 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3505 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3508 sock_prot_inuse_get(seq_file_net(seq), proto),
3509 sock_prot_memory_allocated(proto),
3510 sock_prot_memory_pressure(proto),
3512 proto->slab == NULL ? "no" : "yes",
3513 module_name(proto->owner),
3514 proto_method_implemented(proto->close),
3515 proto_method_implemented(proto->connect),
3516 proto_method_implemented(proto->disconnect),
3517 proto_method_implemented(proto->accept),
3518 proto_method_implemented(proto->ioctl),
3519 proto_method_implemented(proto->init),
3520 proto_method_implemented(proto->destroy),
3521 proto_method_implemented(proto->shutdown),
3522 proto_method_implemented(proto->setsockopt),
3523 proto_method_implemented(proto->getsockopt),
3524 proto_method_implemented(proto->sendmsg),
3525 proto_method_implemented(proto->recvmsg),
3526 proto_method_implemented(proto->sendpage),
3527 proto_method_implemented(proto->bind),
3528 proto_method_implemented(proto->backlog_rcv),
3529 proto_method_implemented(proto->hash),
3530 proto_method_implemented(proto->unhash),
3531 proto_method_implemented(proto->get_port),
3532 proto_method_implemented(proto->enter_memory_pressure));
3535 static int proto_seq_show(struct seq_file *seq, void *v)
3537 if (v == &proto_list)
3538 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3547 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3549 proto_seq_printf(seq, list_entry(v, struct proto, node));
3553 static const struct seq_operations proto_seq_ops = {
3554 .start = proto_seq_start,
3555 .next = proto_seq_next,
3556 .stop = proto_seq_stop,
3557 .show = proto_seq_show,
3560 static __net_init int proto_init_net(struct net *net)
3562 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3563 sizeof(struct seq_net_private)))
3569 static __net_exit void proto_exit_net(struct net *net)
3571 remove_proc_entry("protocols", net->proc_net);
3575 static __net_initdata struct pernet_operations proto_net_ops = {
3576 .init = proto_init_net,
3577 .exit = proto_exit_net,
3580 static int __init proto_init(void)
3582 return register_pernet_subsys(&proto_net_ops);
3585 subsys_initcall(proto_init);
3587 #endif /* PROC_FS */
3589 #ifdef CONFIG_NET_RX_BUSY_POLL
3590 bool sk_busy_loop_end(void *p, unsigned long start_time)
3592 struct sock *sk = p;
3594 return !skb_queue_empty(&sk->sk_receive_queue) ||
3595 sk_busy_loop_timeout(sk, start_time);
3597 EXPORT_SYMBOL(sk_busy_loop_end);
3598 #endif /* CONFIG_NET_RX_BUSY_POLL */