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, READ_ONCE(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 /* IPV6_ADDRFORM can change sk->sk_family under us. */
705 switch (READ_ONCE(sk->sk_family)) {
707 return inet_sk(sk)->mc_loop;
708 #if IS_ENABLED(CONFIG_IPV6)
710 return inet6_sk(sk)->mc_loop;
716 EXPORT_SYMBOL(sk_mc_loop);
719 * This is meant for all protocols to use and covers goings on
720 * at the socket level. Everything here is generic.
723 int sock_setsockopt(struct socket *sock, int level, int optname,
724 char __user *optval, unsigned int optlen)
726 struct sock_txtime sk_txtime;
727 struct sock *sk = sock->sk;
734 * Options without arguments
737 if (optname == SO_BINDTODEVICE)
738 return sock_setbindtodevice(sk, optval, optlen);
740 if (optlen < sizeof(int))
743 if (get_user(val, (int __user *)optval))
746 valbool = val ? 1 : 0;
752 if (val && !capable(CAP_NET_ADMIN))
755 sock_valbool_flag(sk, SOCK_DBG, valbool);
758 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
761 sk->sk_reuseport = valbool;
770 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
774 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
777 /* Don't error on this BSD doesn't and if you think
778 * about it this is right. Otherwise apps have to
779 * play 'guess the biggest size' games. RCVBUF/SNDBUF
780 * are treated in BSD as hints
782 val = min_t(u32, val, sysctl_wmem_max);
784 /* Ensure val * 2 fits into an int, to prevent max_t()
785 * from treating it as a negative value.
787 val = min_t(int, val, INT_MAX / 2);
788 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
789 WRITE_ONCE(sk->sk_sndbuf,
790 max_t(int, val * 2, SOCK_MIN_SNDBUF));
791 /* Wake up sending tasks if we upped the value. */
792 sk->sk_write_space(sk);
796 if (!capable(CAP_NET_ADMIN)) {
801 /* No negative values (to prevent underflow, as val will be
809 /* Don't error on this BSD doesn't and if you think
810 * about it this is right. Otherwise apps have to
811 * play 'guess the biggest size' games. RCVBUF/SNDBUF
812 * are treated in BSD as hints
814 val = min_t(u32, val, sysctl_rmem_max);
816 /* Ensure val * 2 fits into an int, to prevent max_t()
817 * from treating it as a negative value.
819 val = min_t(int, val, INT_MAX / 2);
820 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
822 * We double it on the way in to account for
823 * "struct sk_buff" etc. overhead. Applications
824 * assume that the SO_RCVBUF setting they make will
825 * allow that much actual data to be received on that
828 * Applications are unaware that "struct sk_buff" and
829 * other overheads allocate from the receive buffer
830 * during socket buffer allocation.
832 * And after considering the possible alternatives,
833 * returning the value we actually used in getsockopt
834 * is the most desirable behavior.
836 WRITE_ONCE(sk->sk_rcvbuf,
837 max_t(int, val * 2, SOCK_MIN_RCVBUF));
841 if (!capable(CAP_NET_ADMIN)) {
846 /* No negative values (to prevent underflow, as val will be
854 if (sk->sk_prot->keepalive)
855 sk->sk_prot->keepalive(sk, valbool);
856 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
860 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
864 sk->sk_no_check_tx = valbool;
868 if ((val >= 0 && val <= 6) ||
869 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
870 sk->sk_priority = val;
876 if (optlen < sizeof(ling)) {
877 ret = -EINVAL; /* 1003.1g */
880 if (copy_from_user(&ling, optval, sizeof(ling))) {
885 sock_reset_flag(sk, SOCK_LINGER);
887 #if (BITS_PER_LONG == 32)
888 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
889 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
892 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
893 sock_set_flag(sk, SOCK_LINGER);
898 sock_warn_obsolete_bsdism("setsockopt");
903 set_bit(SOCK_PASSCRED, &sock->flags);
905 clear_bit(SOCK_PASSCRED, &sock->flags);
908 case SO_TIMESTAMP_OLD:
909 case SO_TIMESTAMP_NEW:
910 case SO_TIMESTAMPNS_OLD:
911 case SO_TIMESTAMPNS_NEW:
913 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
914 sock_set_flag(sk, SOCK_TSTAMP_NEW);
916 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
918 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
919 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
921 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
922 sock_set_flag(sk, SOCK_RCVTSTAMP);
923 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
925 sock_reset_flag(sk, SOCK_RCVTSTAMP);
926 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
930 case SO_TIMESTAMPING_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 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
961 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
962 sock_enable_timestamp(sk,
963 SOCK_TIMESTAMPING_RX_SOFTWARE);
965 sock_disable_timestamp(sk,
966 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
972 if (sock->ops->set_rcvlowat)
973 ret = sock->ops->set_rcvlowat(sk, val);
975 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
978 case SO_RCVTIMEO_OLD:
979 case SO_RCVTIMEO_NEW:
980 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
983 case SO_SNDTIMEO_OLD:
984 case SO_SNDTIMEO_NEW:
985 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
988 case SO_ATTACH_FILTER:
990 if (optlen == sizeof(struct sock_fprog)) {
991 struct sock_fprog fprog;
994 if (copy_from_user(&fprog, optval, sizeof(fprog)))
997 ret = sk_attach_filter(&fprog, sk);
1003 if (optlen == sizeof(u32)) {
1007 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1010 ret = sk_attach_bpf(ufd, sk);
1014 case SO_ATTACH_REUSEPORT_CBPF:
1016 if (optlen == sizeof(struct sock_fprog)) {
1017 struct sock_fprog fprog;
1020 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1023 ret = sk_reuseport_attach_filter(&fprog, sk);
1027 case SO_ATTACH_REUSEPORT_EBPF:
1029 if (optlen == sizeof(u32)) {
1033 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1036 ret = sk_reuseport_attach_bpf(ufd, sk);
1040 case SO_DETACH_REUSEPORT_BPF:
1041 ret = reuseport_detach_prog(sk);
1044 case SO_DETACH_FILTER:
1045 ret = sk_detach_filter(sk);
1048 case SO_LOCK_FILTER:
1049 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1052 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1057 set_bit(SOCK_PASSSEC, &sock->flags);
1059 clear_bit(SOCK_PASSSEC, &sock->flags);
1062 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1064 } else if (val != sk->sk_mark) {
1071 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1074 case SO_WIFI_STATUS:
1075 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1079 if (sock->ops->set_peek_off)
1080 ret = sock->ops->set_peek_off(sk, val);
1086 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1089 case SO_SELECT_ERR_QUEUE:
1090 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1093 #ifdef CONFIG_NET_RX_BUSY_POLL
1095 /* allow unprivileged users to decrease the value */
1096 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1102 WRITE_ONCE(sk->sk_ll_usec, val);
1107 case SO_MAX_PACING_RATE:
1109 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1111 if (sizeof(ulval) != sizeof(val) &&
1112 optlen >= sizeof(ulval) &&
1113 get_user(ulval, (unsigned long __user *)optval)) {
1118 cmpxchg(&sk->sk_pacing_status,
1121 /* Pairs with READ_ONCE() from sk_getsockopt() */
1122 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1123 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1126 case SO_INCOMING_CPU:
1127 WRITE_ONCE(sk->sk_incoming_cpu, val);
1132 dst_negative_advice(sk);
1136 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1137 if (!((sk->sk_type == SOCK_STREAM &&
1138 sk->sk_protocol == IPPROTO_TCP) ||
1139 (sk->sk_type == SOCK_DGRAM &&
1140 sk->sk_protocol == IPPROTO_UDP)))
1142 } else if (sk->sk_family != PF_RDS) {
1146 if (val < 0 || val > 1)
1149 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1154 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1156 } else if (optlen != sizeof(struct sock_txtime)) {
1158 } else if (copy_from_user(&sk_txtime, optval,
1159 sizeof(struct sock_txtime))) {
1161 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1164 sock_valbool_flag(sk, SOCK_TXTIME, true);
1165 sk->sk_clockid = sk_txtime.clockid;
1166 sk->sk_txtime_deadline_mode =
1167 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1168 sk->sk_txtime_report_errors =
1169 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1173 case SO_BINDTOIFINDEX:
1174 ret = sock_setbindtodevice_locked(sk, val);
1184 EXPORT_SYMBOL(sock_setsockopt);
1186 static const struct cred *sk_get_peer_cred(struct sock *sk)
1188 const struct cred *cred;
1190 spin_lock(&sk->sk_peer_lock);
1191 cred = get_cred(sk->sk_peer_cred);
1192 spin_unlock(&sk->sk_peer_lock);
1197 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1198 struct ucred *ucred)
1200 ucred->pid = pid_vnr(pid);
1201 ucred->uid = ucred->gid = -1;
1203 struct user_namespace *current_ns = current_user_ns();
1205 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1206 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1210 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1212 struct user_namespace *user_ns = current_user_ns();
1215 for (i = 0; i < src->ngroups; i++)
1216 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1222 int sock_getsockopt(struct socket *sock, int level, int optname,
1223 char __user *optval, int __user *optlen)
1225 struct sock *sk = sock->sk;
1230 unsigned long ulval;
1232 struct old_timeval32 tm32;
1233 struct __kernel_old_timeval tm;
1234 struct __kernel_sock_timeval stm;
1235 struct sock_txtime txtime;
1238 int lv = sizeof(int);
1241 if (get_user(len, optlen))
1246 memset(&v, 0, sizeof(v));
1250 v.val = sock_flag(sk, SOCK_DBG);
1254 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1258 v.val = sock_flag(sk, SOCK_BROADCAST);
1262 v.val = READ_ONCE(sk->sk_sndbuf);
1266 v.val = READ_ONCE(sk->sk_rcvbuf);
1270 v.val = sk->sk_reuse;
1274 v.val = sk->sk_reuseport;
1278 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1282 v.val = sk->sk_type;
1286 v.val = sk->sk_protocol;
1290 v.val = sk->sk_family;
1294 v.val = -sock_error(sk);
1296 v.val = xchg(&sk->sk_err_soft, 0);
1300 v.val = sock_flag(sk, SOCK_URGINLINE);
1304 v.val = sk->sk_no_check_tx;
1308 v.val = sk->sk_priority;
1312 lv = sizeof(v.ling);
1313 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1314 v.ling.l_linger = sk->sk_lingertime / HZ;
1318 sock_warn_obsolete_bsdism("getsockopt");
1321 case SO_TIMESTAMP_OLD:
1322 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1323 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1324 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1327 case SO_TIMESTAMPNS_OLD:
1328 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1331 case SO_TIMESTAMP_NEW:
1332 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1335 case SO_TIMESTAMPNS_NEW:
1336 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1339 case SO_TIMESTAMPING_OLD:
1340 v.val = sk->sk_tsflags;
1343 case SO_RCVTIMEO_OLD:
1344 case SO_RCVTIMEO_NEW:
1345 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1348 case SO_SNDTIMEO_OLD:
1349 case SO_SNDTIMEO_NEW:
1350 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1354 v.val = READ_ONCE(sk->sk_rcvlowat);
1362 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1367 struct ucred peercred;
1368 if (len > sizeof(peercred))
1369 len = sizeof(peercred);
1371 spin_lock(&sk->sk_peer_lock);
1372 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1373 spin_unlock(&sk->sk_peer_lock);
1375 if (copy_to_user(optval, &peercred, len))
1382 const struct cred *cred;
1385 cred = sk_get_peer_cred(sk);
1389 n = cred->group_info->ngroups;
1390 if (len < n * sizeof(gid_t)) {
1391 len = n * sizeof(gid_t);
1393 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1395 len = n * sizeof(gid_t);
1397 ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1408 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1413 if (copy_to_user(optval, address, len))
1418 /* Dubious BSD thing... Probably nobody even uses it, but
1419 * the UNIX standard wants it for whatever reason... -DaveM
1422 v.val = sk->sk_state == TCP_LISTEN;
1426 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1430 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1433 v.val = sk->sk_mark;
1437 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1440 case SO_WIFI_STATUS:
1441 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1445 if (!sock->ops->set_peek_off)
1448 v.val = READ_ONCE(sk->sk_peek_off);
1451 v.val = sock_flag(sk, SOCK_NOFCS);
1454 case SO_BINDTODEVICE:
1455 return sock_getbindtodevice(sk, optval, optlen, len);
1458 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1464 case SO_LOCK_FILTER:
1465 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1468 case SO_BPF_EXTENSIONS:
1469 v.val = bpf_tell_extensions();
1472 case SO_SELECT_ERR_QUEUE:
1473 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1476 #ifdef CONFIG_NET_RX_BUSY_POLL
1478 v.val = READ_ONCE(sk->sk_ll_usec);
1482 case SO_MAX_PACING_RATE:
1483 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1484 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1485 lv = sizeof(v.ulval);
1486 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1489 v.val = min_t(unsigned long, ~0U,
1490 READ_ONCE(sk->sk_max_pacing_rate));
1494 case SO_INCOMING_CPU:
1495 v.val = READ_ONCE(sk->sk_incoming_cpu);
1500 u32 meminfo[SK_MEMINFO_VARS];
1502 sk_get_meminfo(sk, meminfo);
1504 len = min_t(unsigned int, len, sizeof(meminfo));
1505 if (copy_to_user(optval, &meminfo, len))
1511 #ifdef CONFIG_NET_RX_BUSY_POLL
1512 case SO_INCOMING_NAPI_ID:
1513 v.val = READ_ONCE(sk->sk_napi_id);
1515 /* aggregate non-NAPI IDs down to 0 */
1516 if (v.val < MIN_NAPI_ID)
1526 v.val64 = sock_gen_cookie(sk);
1530 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1534 lv = sizeof(v.txtime);
1535 v.txtime.clockid = sk->sk_clockid;
1536 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1537 SOF_TXTIME_DEADLINE_MODE : 0;
1538 v.txtime.flags |= sk->sk_txtime_report_errors ?
1539 SOF_TXTIME_REPORT_ERRORS : 0;
1542 case SO_BINDTOIFINDEX:
1543 v.val = sk->sk_bound_dev_if;
1547 /* We implement the SO_SNDLOWAT etc to not be settable
1550 return -ENOPROTOOPT;
1555 if (copy_to_user(optval, &v, len))
1558 if (put_user(len, optlen))
1564 * Initialize an sk_lock.
1566 * (We also register the sk_lock with the lock validator.)
1568 static inline void sock_lock_init(struct sock *sk)
1570 if (sk->sk_kern_sock)
1571 sock_lock_init_class_and_name(
1573 af_family_kern_slock_key_strings[sk->sk_family],
1574 af_family_kern_slock_keys + sk->sk_family,
1575 af_family_kern_key_strings[sk->sk_family],
1576 af_family_kern_keys + sk->sk_family);
1578 sock_lock_init_class_and_name(
1580 af_family_slock_key_strings[sk->sk_family],
1581 af_family_slock_keys + sk->sk_family,
1582 af_family_key_strings[sk->sk_family],
1583 af_family_keys + sk->sk_family);
1587 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1588 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1589 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1591 static void sock_copy(struct sock *nsk, const struct sock *osk)
1593 #ifdef CONFIG_SECURITY_NETWORK
1594 void *sptr = nsk->sk_security;
1596 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1598 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1599 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1601 #ifdef CONFIG_SECURITY_NETWORK
1602 nsk->sk_security = sptr;
1603 security_sk_clone(osk, nsk);
1607 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1611 struct kmem_cache *slab;
1615 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1618 if (want_init_on_alloc(priority))
1619 sk_prot_clear_nulls(sk, prot->obj_size);
1621 sk = kmalloc(prot->obj_size, priority);
1624 if (security_sk_alloc(sk, family, priority))
1627 if (!try_module_get(prot->owner))
1629 sk_tx_queue_clear(sk);
1635 security_sk_free(sk);
1638 kmem_cache_free(slab, sk);
1644 static void sk_prot_free(struct proto *prot, struct sock *sk)
1646 struct kmem_cache *slab;
1647 struct module *owner;
1649 owner = prot->owner;
1652 cgroup_sk_free(&sk->sk_cgrp_data);
1653 mem_cgroup_sk_free(sk);
1654 security_sk_free(sk);
1656 kmem_cache_free(slab, sk);
1663 * sk_alloc - All socket objects are allocated here
1664 * @net: the applicable net namespace
1665 * @family: protocol family
1666 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1667 * @prot: struct proto associated with this new sock instance
1668 * @kern: is this to be a kernel socket?
1670 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1671 struct proto *prot, int kern)
1675 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1677 sk->sk_family = family;
1679 * See comment in struct sock definition to understand
1680 * why we need sk_prot_creator -acme
1682 sk->sk_prot = sk->sk_prot_creator = prot;
1683 sk->sk_kern_sock = kern;
1685 sk->sk_net_refcnt = kern ? 0 : 1;
1686 if (likely(sk->sk_net_refcnt)) {
1688 sock_inuse_add(net, 1);
1691 sock_net_set(sk, net);
1692 refcount_set(&sk->sk_wmem_alloc, 1);
1694 mem_cgroup_sk_alloc(sk);
1695 cgroup_sk_alloc(&sk->sk_cgrp_data);
1696 sock_update_classid(&sk->sk_cgrp_data);
1697 sock_update_netprioidx(&sk->sk_cgrp_data);
1698 sk_tx_queue_clear(sk);
1703 EXPORT_SYMBOL(sk_alloc);
1705 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1706 * grace period. This is the case for UDP sockets and TCP listeners.
1708 static void __sk_destruct(struct rcu_head *head)
1710 struct sock *sk = container_of(head, struct sock, sk_rcu);
1711 struct sk_filter *filter;
1713 if (sk->sk_destruct)
1714 sk->sk_destruct(sk);
1716 filter = rcu_dereference_check(sk->sk_filter,
1717 refcount_read(&sk->sk_wmem_alloc) == 0);
1719 sk_filter_uncharge(sk, filter);
1720 RCU_INIT_POINTER(sk->sk_filter, NULL);
1723 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1725 #ifdef CONFIG_BPF_SYSCALL
1726 bpf_sk_storage_free(sk);
1729 if (atomic_read(&sk->sk_omem_alloc))
1730 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1731 __func__, atomic_read(&sk->sk_omem_alloc));
1733 if (sk->sk_frag.page) {
1734 put_page(sk->sk_frag.page);
1735 sk->sk_frag.page = NULL;
1738 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
1739 put_cred(sk->sk_peer_cred);
1740 put_pid(sk->sk_peer_pid);
1742 if (likely(sk->sk_net_refcnt))
1743 put_net(sock_net(sk));
1744 sk_prot_free(sk->sk_prot_creator, sk);
1747 void sk_destruct(struct sock *sk)
1749 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1751 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1752 reuseport_detach_sock(sk);
1753 use_call_rcu = true;
1757 call_rcu(&sk->sk_rcu, __sk_destruct);
1759 __sk_destruct(&sk->sk_rcu);
1762 static void __sk_free(struct sock *sk)
1764 if (likely(sk->sk_net_refcnt))
1765 sock_inuse_add(sock_net(sk), -1);
1767 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1768 sock_diag_broadcast_destroy(sk);
1773 void sk_free(struct sock *sk)
1776 * We subtract one from sk_wmem_alloc and can know if
1777 * some packets are still in some tx queue.
1778 * If not null, sock_wfree() will call __sk_free(sk) later
1780 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1783 EXPORT_SYMBOL(sk_free);
1785 static void sk_init_common(struct sock *sk)
1787 skb_queue_head_init(&sk->sk_receive_queue);
1788 skb_queue_head_init(&sk->sk_write_queue);
1789 skb_queue_head_init(&sk->sk_error_queue);
1791 rwlock_init(&sk->sk_callback_lock);
1792 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1793 af_rlock_keys + sk->sk_family,
1794 af_family_rlock_key_strings[sk->sk_family]);
1795 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1796 af_wlock_keys + sk->sk_family,
1797 af_family_wlock_key_strings[sk->sk_family]);
1798 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1799 af_elock_keys + sk->sk_family,
1800 af_family_elock_key_strings[sk->sk_family]);
1801 lockdep_set_class_and_name(&sk->sk_callback_lock,
1802 af_callback_keys + sk->sk_family,
1803 af_family_clock_key_strings[sk->sk_family]);
1807 * sk_clone_lock - clone a socket, and lock its clone
1808 * @sk: the socket to clone
1809 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1811 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1813 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1816 bool is_charged = true;
1818 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1819 if (newsk != NULL) {
1820 struct sk_filter *filter;
1822 sock_copy(newsk, sk);
1824 newsk->sk_prot_creator = sk->sk_prot;
1827 if (likely(newsk->sk_net_refcnt))
1828 get_net(sock_net(newsk));
1829 sk_node_init(&newsk->sk_node);
1830 sock_lock_init(newsk);
1831 bh_lock_sock(newsk);
1832 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1833 newsk->sk_backlog.len = 0;
1835 atomic_set(&newsk->sk_rmem_alloc, 0);
1837 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1839 refcount_set(&newsk->sk_wmem_alloc, 1);
1840 atomic_set(&newsk->sk_omem_alloc, 0);
1841 sk_init_common(newsk);
1843 newsk->sk_dst_cache = NULL;
1844 newsk->sk_dst_pending_confirm = 0;
1845 newsk->sk_wmem_queued = 0;
1846 newsk->sk_forward_alloc = 0;
1847 atomic_set(&newsk->sk_drops, 0);
1848 newsk->sk_send_head = NULL;
1849 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1850 atomic_set(&newsk->sk_zckey, 0);
1852 sock_reset_flag(newsk, SOCK_DONE);
1854 /* sk->sk_memcg will be populated at accept() time */
1855 newsk->sk_memcg = NULL;
1857 cgroup_sk_clone(&newsk->sk_cgrp_data);
1860 filter = rcu_dereference(sk->sk_filter);
1862 /* though it's an empty new sock, the charging may fail
1863 * if sysctl_optmem_max was changed between creation of
1864 * original socket and cloning
1866 is_charged = sk_filter_charge(newsk, filter);
1867 RCU_INIT_POINTER(newsk->sk_filter, filter);
1870 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1871 /* We need to make sure that we don't uncharge the new
1872 * socket if we couldn't charge it in the first place
1873 * as otherwise we uncharge the parent's filter.
1876 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1877 sk_free_unlock_clone(newsk);
1881 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1883 if (bpf_sk_storage_clone(sk, newsk)) {
1884 sk_free_unlock_clone(newsk);
1890 newsk->sk_err_soft = 0;
1891 newsk->sk_priority = 0;
1892 newsk->sk_incoming_cpu = raw_smp_processor_id();
1893 if (likely(newsk->sk_net_refcnt))
1894 sock_inuse_add(sock_net(newsk), 1);
1897 * Before updating sk_refcnt, we must commit prior changes to memory
1898 * (Documentation/RCU/rculist_nulls.txt for details)
1901 refcount_set(&newsk->sk_refcnt, 2);
1904 * Increment the counter in the same struct proto as the master
1905 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1906 * is the same as sk->sk_prot->socks, as this field was copied
1909 * This _changes_ the previous behaviour, where
1910 * tcp_create_openreq_child always was incrementing the
1911 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1912 * to be taken into account in all callers. -acme
1914 sk_refcnt_debug_inc(newsk);
1915 sk_set_socket(newsk, NULL);
1916 sk_tx_queue_clear(newsk);
1917 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1919 if (newsk->sk_prot->sockets_allocated)
1920 sk_sockets_allocated_inc(newsk);
1922 if (sock_needs_netstamp(sk) &&
1923 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1924 net_enable_timestamp();
1929 EXPORT_SYMBOL_GPL(sk_clone_lock);
1931 void sk_free_unlock_clone(struct sock *sk)
1933 /* It is still raw copy of parent, so invalidate
1934 * destructor and make plain sk_free() */
1935 sk->sk_destruct = NULL;
1939 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1941 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1945 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1946 if (sk->sk_route_caps & NETIF_F_GSO)
1947 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1948 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1949 if (sk_can_gso(sk)) {
1950 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1951 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1953 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1954 sk->sk_gso_max_size = dst->dev->gso_max_size;
1955 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1958 sk->sk_gso_max_segs = max_segs;
1959 sk_dst_set(sk, dst);
1961 EXPORT_SYMBOL_GPL(sk_setup_caps);
1964 * Simple resource managers for sockets.
1969 * Write buffer destructor automatically called from kfree_skb.
1971 void sock_wfree(struct sk_buff *skb)
1973 struct sock *sk = skb->sk;
1974 unsigned int len = skb->truesize;
1976 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1978 * Keep a reference on sk_wmem_alloc, this will be released
1979 * after sk_write_space() call
1981 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1982 sk->sk_write_space(sk);
1986 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1987 * could not do because of in-flight packets
1989 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1992 EXPORT_SYMBOL(sock_wfree);
1994 /* This variant of sock_wfree() is used by TCP,
1995 * since it sets SOCK_USE_WRITE_QUEUE.
1997 void __sock_wfree(struct sk_buff *skb)
1999 struct sock *sk = skb->sk;
2001 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2005 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2010 if (unlikely(!sk_fullsock(sk))) {
2011 skb->destructor = sock_edemux;
2016 skb->destructor = sock_wfree;
2017 skb_set_hash_from_sk(skb, sk);
2019 * We used to take a refcount on sk, but following operation
2020 * is enough to guarantee sk_free() wont free this sock until
2021 * all in-flight packets are completed
2023 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2025 EXPORT_SYMBOL(skb_set_owner_w);
2027 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2029 #ifdef CONFIG_TLS_DEVICE
2030 /* Drivers depend on in-order delivery for crypto offload,
2031 * partial orphan breaks out-of-order-OK logic.
2036 return (skb->destructor == sock_wfree ||
2037 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2040 /* This helper is used by netem, as it can hold packets in its
2041 * delay queue. We want to allow the owner socket to send more
2042 * packets, as if they were already TX completed by a typical driver.
2043 * But we also want to keep skb->sk set because some packet schedulers
2044 * rely on it (sch_fq for example).
2046 void skb_orphan_partial(struct sk_buff *skb)
2048 if (skb_is_tcp_pure_ack(skb))
2051 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2056 EXPORT_SYMBOL(skb_orphan_partial);
2059 * Read buffer destructor automatically called from kfree_skb.
2061 void sock_rfree(struct sk_buff *skb)
2063 struct sock *sk = skb->sk;
2064 unsigned int len = skb->truesize;
2066 atomic_sub(len, &sk->sk_rmem_alloc);
2067 sk_mem_uncharge(sk, len);
2069 EXPORT_SYMBOL(sock_rfree);
2072 * Buffer destructor for skbs that are not used directly in read or write
2073 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2075 void sock_efree(struct sk_buff *skb)
2079 EXPORT_SYMBOL(sock_efree);
2081 kuid_t sock_i_uid(struct sock *sk)
2085 read_lock_bh(&sk->sk_callback_lock);
2086 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2087 read_unlock_bh(&sk->sk_callback_lock);
2090 EXPORT_SYMBOL(sock_i_uid);
2092 unsigned long __sock_i_ino(struct sock *sk)
2096 read_lock(&sk->sk_callback_lock);
2097 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2098 read_unlock(&sk->sk_callback_lock);
2101 EXPORT_SYMBOL(__sock_i_ino);
2103 unsigned long sock_i_ino(struct sock *sk)
2108 ino = __sock_i_ino(sk);
2112 EXPORT_SYMBOL(sock_i_ino);
2115 * Allocate a skb from the socket's send buffer.
2117 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2121 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2122 struct sk_buff *skb = alloc_skb(size, priority);
2125 skb_set_owner_w(skb, sk);
2131 EXPORT_SYMBOL(sock_wmalloc);
2133 static void sock_ofree(struct sk_buff *skb)
2135 struct sock *sk = skb->sk;
2137 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2140 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2143 struct sk_buff *skb;
2145 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2146 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2150 skb = alloc_skb(size, priority);
2154 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2156 skb->destructor = sock_ofree;
2161 * Allocate a memory block from the socket's option memory buffer.
2163 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2165 if ((unsigned int)size <= sysctl_optmem_max &&
2166 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2168 /* First do the add, to avoid the race if kmalloc
2171 atomic_add(size, &sk->sk_omem_alloc);
2172 mem = kmalloc(size, priority);
2175 atomic_sub(size, &sk->sk_omem_alloc);
2179 EXPORT_SYMBOL(sock_kmalloc);
2181 /* Free an option memory block. Note, we actually want the inline
2182 * here as this allows gcc to detect the nullify and fold away the
2183 * condition entirely.
2185 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2188 if (WARN_ON_ONCE(!mem))
2194 atomic_sub(size, &sk->sk_omem_alloc);
2197 void sock_kfree_s(struct sock *sk, void *mem, int size)
2199 __sock_kfree_s(sk, mem, size, false);
2201 EXPORT_SYMBOL(sock_kfree_s);
2203 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2205 __sock_kfree_s(sk, mem, size, true);
2207 EXPORT_SYMBOL(sock_kzfree_s);
2209 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2210 I think, these locks should be removed for datagram sockets.
2212 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2216 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2220 if (signal_pending(current))
2222 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2223 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2224 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2226 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2228 if (READ_ONCE(sk->sk_err))
2230 timeo = schedule_timeout(timeo);
2232 finish_wait(sk_sleep(sk), &wait);
2238 * Generic send/receive buffer handlers
2241 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2242 unsigned long data_len, int noblock,
2243 int *errcode, int max_page_order)
2245 struct sk_buff *skb;
2249 timeo = sock_sndtimeo(sk, noblock);
2251 err = sock_error(sk);
2256 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2259 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2262 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2263 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2267 if (signal_pending(current))
2269 timeo = sock_wait_for_wmem(sk, timeo);
2271 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2272 errcode, sk->sk_allocation);
2274 skb_set_owner_w(skb, sk);
2278 err = sock_intr_errno(timeo);
2283 EXPORT_SYMBOL(sock_alloc_send_pskb);
2285 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2286 int noblock, int *errcode)
2288 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2290 EXPORT_SYMBOL(sock_alloc_send_skb);
2292 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2293 struct sockcm_cookie *sockc)
2297 switch (cmsg->cmsg_type) {
2299 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2301 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2303 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2305 case SO_TIMESTAMPING_OLD:
2306 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2309 tsflags = *(u32 *)CMSG_DATA(cmsg);
2310 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2313 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2314 sockc->tsflags |= tsflags;
2317 if (!sock_flag(sk, SOCK_TXTIME))
2319 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2321 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2323 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2325 case SCM_CREDENTIALS:
2332 EXPORT_SYMBOL(__sock_cmsg_send);
2334 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2335 struct sockcm_cookie *sockc)
2337 struct cmsghdr *cmsg;
2340 for_each_cmsghdr(cmsg, msg) {
2341 if (!CMSG_OK(msg, cmsg))
2343 if (cmsg->cmsg_level != SOL_SOCKET)
2345 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2351 EXPORT_SYMBOL(sock_cmsg_send);
2353 static void sk_enter_memory_pressure(struct sock *sk)
2355 if (!sk->sk_prot->enter_memory_pressure)
2358 sk->sk_prot->enter_memory_pressure(sk);
2361 static void sk_leave_memory_pressure(struct sock *sk)
2363 if (sk->sk_prot->leave_memory_pressure) {
2364 sk->sk_prot->leave_memory_pressure(sk);
2366 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2368 if (memory_pressure && READ_ONCE(*memory_pressure))
2369 WRITE_ONCE(*memory_pressure, 0);
2373 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2376 * skb_page_frag_refill - check that a page_frag contains enough room
2377 * @sz: minimum size of the fragment we want to get
2378 * @pfrag: pointer to page_frag
2379 * @gfp: priority for memory allocation
2381 * Note: While this allocator tries to use high order pages, there is
2382 * no guarantee that allocations succeed. Therefore, @sz MUST be
2383 * less or equal than PAGE_SIZE.
2385 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2388 if (page_ref_count(pfrag->page) == 1) {
2392 if (pfrag->offset + sz <= pfrag->size)
2394 put_page(pfrag->page);
2398 if (SKB_FRAG_PAGE_ORDER &&
2399 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2400 /* Avoid direct reclaim but allow kswapd to wake */
2401 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2402 __GFP_COMP | __GFP_NOWARN |
2404 SKB_FRAG_PAGE_ORDER);
2405 if (likely(pfrag->page)) {
2406 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2410 pfrag->page = alloc_page(gfp);
2411 if (likely(pfrag->page)) {
2412 pfrag->size = PAGE_SIZE;
2417 EXPORT_SYMBOL(skb_page_frag_refill);
2419 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2421 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2424 sk_enter_memory_pressure(sk);
2425 sk_stream_moderate_sndbuf(sk);
2428 EXPORT_SYMBOL(sk_page_frag_refill);
2430 static void __lock_sock(struct sock *sk)
2431 __releases(&sk->sk_lock.slock)
2432 __acquires(&sk->sk_lock.slock)
2437 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2438 TASK_UNINTERRUPTIBLE);
2439 spin_unlock_bh(&sk->sk_lock.slock);
2441 spin_lock_bh(&sk->sk_lock.slock);
2442 if (!sock_owned_by_user(sk))
2445 finish_wait(&sk->sk_lock.wq, &wait);
2448 void __release_sock(struct sock *sk)
2449 __releases(&sk->sk_lock.slock)
2450 __acquires(&sk->sk_lock.slock)
2452 struct sk_buff *skb, *next;
2454 while ((skb = sk->sk_backlog.head) != NULL) {
2455 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2457 spin_unlock_bh(&sk->sk_lock.slock);
2462 WARN_ON_ONCE(skb_dst_is_noref(skb));
2463 skb_mark_not_on_list(skb);
2464 sk_backlog_rcv(sk, skb);
2469 } while (skb != NULL);
2471 spin_lock_bh(&sk->sk_lock.slock);
2475 * Doing the zeroing here guarantee we can not loop forever
2476 * while a wild producer attempts to flood us.
2478 sk->sk_backlog.len = 0;
2481 void __sk_flush_backlog(struct sock *sk)
2483 spin_lock_bh(&sk->sk_lock.slock);
2485 spin_unlock_bh(&sk->sk_lock.slock);
2489 * sk_wait_data - wait for data to arrive at sk_receive_queue
2490 * @sk: sock to wait on
2491 * @timeo: for how long
2492 * @skb: last skb seen on sk_receive_queue
2494 * Now socket state including sk->sk_err is changed only under lock,
2495 * hence we may omit checks after joining wait queue.
2496 * We check receive queue before schedule() only as optimization;
2497 * it is very likely that release_sock() added new data.
2499 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2501 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2504 add_wait_queue(sk_sleep(sk), &wait);
2505 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2506 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2507 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2508 remove_wait_queue(sk_sleep(sk), &wait);
2511 EXPORT_SYMBOL(sk_wait_data);
2514 * __sk_mem_raise_allocated - increase memory_allocated
2516 * @size: memory size to allocate
2517 * @amt: pages to allocate
2518 * @kind: allocation type
2520 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2522 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2524 struct proto *prot = sk->sk_prot;
2525 long allocated = sk_memory_allocated_add(sk, amt);
2526 bool charged = true;
2528 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2529 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2530 goto suppress_allocation;
2533 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2534 sk_leave_memory_pressure(sk);
2538 /* Under pressure. */
2539 if (allocated > sk_prot_mem_limits(sk, 1))
2540 sk_enter_memory_pressure(sk);
2542 /* Over hard limit. */
2543 if (allocated > sk_prot_mem_limits(sk, 2))
2544 goto suppress_allocation;
2546 /* guarantee minimum buffer size under pressure */
2547 if (kind == SK_MEM_RECV) {
2548 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2551 } else { /* SK_MEM_SEND */
2552 int wmem0 = sk_get_wmem0(sk, prot);
2554 if (sk->sk_type == SOCK_STREAM) {
2555 if (sk->sk_wmem_queued < wmem0)
2557 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2562 if (sk_has_memory_pressure(sk)) {
2565 if (!sk_under_memory_pressure(sk))
2567 alloc = sk_sockets_allocated_read_positive(sk);
2568 if (sk_prot_mem_limits(sk, 2) > alloc *
2569 sk_mem_pages(sk->sk_wmem_queued +
2570 atomic_read(&sk->sk_rmem_alloc) +
2571 sk->sk_forward_alloc))
2575 suppress_allocation:
2577 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2578 sk_stream_moderate_sndbuf(sk);
2580 /* Fail only if socket is _under_ its sndbuf.
2581 * In this case we cannot block, so that we have to fail.
2583 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2587 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2588 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2590 sk_memory_allocated_sub(sk, amt);
2592 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2593 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2597 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2600 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2602 * @size: memory size to allocate
2603 * @kind: allocation type
2605 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2606 * rmem allocation. This function assumes that protocols which have
2607 * memory_pressure use sk_wmem_queued as write buffer accounting.
2609 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2611 int ret, amt = sk_mem_pages(size);
2613 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2614 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2616 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2619 EXPORT_SYMBOL(__sk_mem_schedule);
2622 * __sk_mem_reduce_allocated - reclaim memory_allocated
2624 * @amount: number of quanta
2626 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2628 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2630 sk_memory_allocated_sub(sk, amount);
2632 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2633 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2635 if (sk_under_global_memory_pressure(sk) &&
2636 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2637 sk_leave_memory_pressure(sk);
2639 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2642 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2644 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2646 void __sk_mem_reclaim(struct sock *sk, int amount)
2648 amount >>= SK_MEM_QUANTUM_SHIFT;
2649 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2650 __sk_mem_reduce_allocated(sk, amount);
2652 EXPORT_SYMBOL(__sk_mem_reclaim);
2654 int sk_set_peek_off(struct sock *sk, int val)
2656 WRITE_ONCE(sk->sk_peek_off, val);
2659 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2662 * Set of default routines for initialising struct proto_ops when
2663 * the protocol does not support a particular function. In certain
2664 * cases where it makes no sense for a protocol to have a "do nothing"
2665 * function, some default processing is provided.
2668 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2672 EXPORT_SYMBOL(sock_no_bind);
2674 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2679 EXPORT_SYMBOL(sock_no_connect);
2681 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2685 EXPORT_SYMBOL(sock_no_socketpair);
2687 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2692 EXPORT_SYMBOL(sock_no_accept);
2694 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2699 EXPORT_SYMBOL(sock_no_getname);
2701 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2705 EXPORT_SYMBOL(sock_no_ioctl);
2707 int sock_no_listen(struct socket *sock, int backlog)
2711 EXPORT_SYMBOL(sock_no_listen);
2713 int sock_no_shutdown(struct socket *sock, int how)
2717 EXPORT_SYMBOL(sock_no_shutdown);
2719 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2720 char __user *optval, unsigned int optlen)
2724 EXPORT_SYMBOL(sock_no_setsockopt);
2726 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2727 char __user *optval, int __user *optlen)
2731 EXPORT_SYMBOL(sock_no_getsockopt);
2733 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2737 EXPORT_SYMBOL(sock_no_sendmsg);
2739 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2743 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2745 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2750 EXPORT_SYMBOL(sock_no_recvmsg);
2752 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2754 /* Mirror missing mmap method error code */
2757 EXPORT_SYMBOL(sock_no_mmap);
2760 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2761 * various sock-based usage counts.
2763 void __receive_sock(struct file *file)
2765 struct socket *sock;
2769 * The resulting value of "error" is ignored here since we only
2770 * need to take action when the file is a socket and testing
2771 * "sock" for NULL is sufficient.
2773 sock = sock_from_file(file, &error);
2775 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2776 sock_update_classid(&sock->sk->sk_cgrp_data);
2780 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2783 struct msghdr msg = {.msg_flags = flags};
2785 char *kaddr = kmap(page);
2786 iov.iov_base = kaddr + offset;
2788 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2792 EXPORT_SYMBOL(sock_no_sendpage);
2794 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2795 int offset, size_t size, int flags)
2798 struct msghdr msg = {.msg_flags = flags};
2800 char *kaddr = kmap(page);
2802 iov.iov_base = kaddr + offset;
2804 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2808 EXPORT_SYMBOL(sock_no_sendpage_locked);
2811 * Default Socket Callbacks
2814 static void sock_def_wakeup(struct sock *sk)
2816 struct socket_wq *wq;
2819 wq = rcu_dereference(sk->sk_wq);
2820 if (skwq_has_sleeper(wq))
2821 wake_up_interruptible_all(&wq->wait);
2825 static void sock_def_error_report(struct sock *sk)
2827 struct socket_wq *wq;
2830 wq = rcu_dereference(sk->sk_wq);
2831 if (skwq_has_sleeper(wq))
2832 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2833 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2837 static void sock_def_readable(struct sock *sk)
2839 struct socket_wq *wq;
2842 wq = rcu_dereference(sk->sk_wq);
2843 if (skwq_has_sleeper(wq))
2844 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2845 EPOLLRDNORM | EPOLLRDBAND);
2846 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2850 static void sock_def_write_space(struct sock *sk)
2852 struct socket_wq *wq;
2856 /* Do not wake up a writer until he can make "significant"
2859 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2860 wq = rcu_dereference(sk->sk_wq);
2861 if (skwq_has_sleeper(wq))
2862 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2863 EPOLLWRNORM | EPOLLWRBAND);
2865 /* Should agree with poll, otherwise some programs break */
2866 if (sock_writeable(sk))
2867 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2873 static void sock_def_destruct(struct sock *sk)
2877 void sk_send_sigurg(struct sock *sk)
2879 if (sk->sk_socket && sk->sk_socket->file)
2880 if (send_sigurg(&sk->sk_socket->file->f_owner))
2881 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2883 EXPORT_SYMBOL(sk_send_sigurg);
2885 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2886 unsigned long expires)
2888 if (!mod_timer(timer, expires))
2891 EXPORT_SYMBOL(sk_reset_timer);
2893 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2895 if (del_timer(timer))
2898 EXPORT_SYMBOL(sk_stop_timer);
2900 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
2902 if (del_timer_sync(timer))
2905 EXPORT_SYMBOL(sk_stop_timer_sync);
2907 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
2910 sk->sk_send_head = NULL;
2912 timer_setup(&sk->sk_timer, NULL, 0);
2914 sk->sk_allocation = GFP_KERNEL;
2915 sk->sk_rcvbuf = sysctl_rmem_default;
2916 sk->sk_sndbuf = sysctl_wmem_default;
2917 sk->sk_state = TCP_CLOSE;
2918 sk_set_socket(sk, sock);
2920 sock_set_flag(sk, SOCK_ZAPPED);
2923 sk->sk_type = sock->type;
2924 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2927 RCU_INIT_POINTER(sk->sk_wq, NULL);
2931 rwlock_init(&sk->sk_callback_lock);
2932 if (sk->sk_kern_sock)
2933 lockdep_set_class_and_name(
2934 &sk->sk_callback_lock,
2935 af_kern_callback_keys + sk->sk_family,
2936 af_family_kern_clock_key_strings[sk->sk_family]);
2938 lockdep_set_class_and_name(
2939 &sk->sk_callback_lock,
2940 af_callback_keys + sk->sk_family,
2941 af_family_clock_key_strings[sk->sk_family]);
2943 sk->sk_state_change = sock_def_wakeup;
2944 sk->sk_data_ready = sock_def_readable;
2945 sk->sk_write_space = sock_def_write_space;
2946 sk->sk_error_report = sock_def_error_report;
2947 sk->sk_destruct = sock_def_destruct;
2949 sk->sk_frag.page = NULL;
2950 sk->sk_frag.offset = 0;
2951 sk->sk_peek_off = -1;
2953 sk->sk_peer_pid = NULL;
2954 sk->sk_peer_cred = NULL;
2955 spin_lock_init(&sk->sk_peer_lock);
2957 sk->sk_write_pending = 0;
2958 sk->sk_rcvlowat = 1;
2959 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2960 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2962 sk->sk_stamp = SK_DEFAULT_STAMP;
2963 #if BITS_PER_LONG==32
2964 seqlock_init(&sk->sk_stamp_seq);
2966 atomic_set(&sk->sk_zckey, 0);
2968 #ifdef CONFIG_NET_RX_BUSY_POLL
2970 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
2973 sk->sk_max_pacing_rate = ~0UL;
2974 sk->sk_pacing_rate = ~0UL;
2975 WRITE_ONCE(sk->sk_pacing_shift, 10);
2976 sk->sk_incoming_cpu = -1;
2978 sk_rx_queue_clear(sk);
2980 * Before updating sk_refcnt, we must commit prior changes to memory
2981 * (Documentation/RCU/rculist_nulls.txt for details)
2984 refcount_set(&sk->sk_refcnt, 1);
2985 atomic_set(&sk->sk_drops, 0);
2987 EXPORT_SYMBOL(sock_init_data_uid);
2989 void sock_init_data(struct socket *sock, struct sock *sk)
2992 SOCK_INODE(sock)->i_uid :
2993 make_kuid(sock_net(sk)->user_ns, 0);
2995 sock_init_data_uid(sock, sk, uid);
2997 EXPORT_SYMBOL(sock_init_data);
2999 void lock_sock_nested(struct sock *sk, int subclass)
3002 spin_lock_bh(&sk->sk_lock.slock);
3003 if (sk->sk_lock.owned)
3005 sk->sk_lock.owned = 1;
3006 spin_unlock(&sk->sk_lock.slock);
3008 * The sk_lock has mutex_lock() semantics here:
3010 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3013 EXPORT_SYMBOL(lock_sock_nested);
3015 void release_sock(struct sock *sk)
3017 spin_lock_bh(&sk->sk_lock.slock);
3018 if (sk->sk_backlog.tail)
3021 /* Warning : release_cb() might need to release sk ownership,
3022 * ie call sock_release_ownership(sk) before us.
3024 if (sk->sk_prot->release_cb)
3025 sk->sk_prot->release_cb(sk);
3027 sock_release_ownership(sk);
3028 if (waitqueue_active(&sk->sk_lock.wq))
3029 wake_up(&sk->sk_lock.wq);
3030 spin_unlock_bh(&sk->sk_lock.slock);
3032 EXPORT_SYMBOL(release_sock);
3035 * lock_sock_fast - fast version of lock_sock
3038 * This version should be used for very small section, where process wont block
3039 * return false if fast path is taken:
3041 * sk_lock.slock locked, owned = 0, BH disabled
3043 * return true if slow path is taken:
3045 * sk_lock.slock unlocked, owned = 1, BH enabled
3047 bool lock_sock_fast(struct sock *sk)
3050 spin_lock_bh(&sk->sk_lock.slock);
3052 if (!sk->sk_lock.owned)
3054 * Note : We must disable BH
3059 sk->sk_lock.owned = 1;
3060 spin_unlock(&sk->sk_lock.slock);
3062 * The sk_lock has mutex_lock() semantics here:
3064 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3068 EXPORT_SYMBOL(lock_sock_fast);
3070 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3071 bool timeval, bool time32)
3073 struct sock *sk = sock->sk;
3074 struct timespec64 ts;
3076 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3077 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3078 if (ts.tv_sec == -1)
3080 if (ts.tv_sec == 0) {
3081 ktime_t kt = ktime_get_real();
3082 sock_write_timestamp(sk, kt);;
3083 ts = ktime_to_timespec64(kt);
3089 #ifdef CONFIG_COMPAT_32BIT_TIME
3091 return put_old_timespec32(&ts, userstamp);
3093 #ifdef CONFIG_SPARC64
3094 /* beware of padding in sparc64 timeval */
3095 if (timeval && !in_compat_syscall()) {
3096 struct __kernel_old_timeval __user tv = {
3097 .tv_sec = ts.tv_sec,
3098 .tv_usec = ts.tv_nsec,
3100 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3105 return put_timespec64(&ts, userstamp);
3107 EXPORT_SYMBOL(sock_gettstamp);
3109 void sock_enable_timestamp(struct sock *sk, int flag)
3111 if (!sock_flag(sk, flag)) {
3112 unsigned long previous_flags = sk->sk_flags;
3114 sock_set_flag(sk, flag);
3116 * we just set one of the two flags which require net
3117 * time stamping, but time stamping might have been on
3118 * already because of the other one
3120 if (sock_needs_netstamp(sk) &&
3121 !(previous_flags & SK_FLAGS_TIMESTAMP))
3122 net_enable_timestamp();
3126 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3127 int level, int type)
3129 struct sock_exterr_skb *serr;
3130 struct sk_buff *skb;
3134 skb = sock_dequeue_err_skb(sk);
3140 msg->msg_flags |= MSG_TRUNC;
3143 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3147 sock_recv_timestamp(msg, sk, skb);
3149 serr = SKB_EXT_ERR(skb);
3150 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3152 msg->msg_flags |= MSG_ERRQUEUE;
3160 EXPORT_SYMBOL(sock_recv_errqueue);
3163 * Get a socket option on an socket.
3165 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3166 * asynchronous errors should be reported by getsockopt. We assume
3167 * this means if you specify SO_ERROR (otherwise whats the point of it).
3169 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3170 char __user *optval, int __user *optlen)
3172 struct sock *sk = sock->sk;
3174 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3176 EXPORT_SYMBOL(sock_common_getsockopt);
3178 #ifdef CONFIG_COMPAT
3179 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3180 char __user *optval, int __user *optlen)
3182 struct sock *sk = sock->sk;
3184 if (sk->sk_prot->compat_getsockopt != NULL)
3185 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3187 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3189 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3192 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3195 struct sock *sk = sock->sk;
3199 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3200 flags & ~MSG_DONTWAIT, &addr_len);
3202 msg->msg_namelen = addr_len;
3205 EXPORT_SYMBOL(sock_common_recvmsg);
3208 * Set socket options on an inet socket.
3210 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3211 char __user *optval, unsigned int optlen)
3213 struct sock *sk = sock->sk;
3215 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3217 EXPORT_SYMBOL(sock_common_setsockopt);
3219 #ifdef CONFIG_COMPAT
3220 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3221 char __user *optval, unsigned int optlen)
3223 struct sock *sk = sock->sk;
3225 if (sk->sk_prot->compat_setsockopt != NULL)
3226 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3228 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3230 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3233 void sk_common_release(struct sock *sk)
3235 if (sk->sk_prot->destroy)
3236 sk->sk_prot->destroy(sk);
3239 * Observation: when sock_common_release is called, processes have
3240 * no access to socket. But net still has.
3241 * Step one, detach it from networking:
3243 * A. Remove from hash tables.
3246 sk->sk_prot->unhash(sk);
3249 * In this point socket cannot receive new packets, but it is possible
3250 * that some packets are in flight because some CPU runs receiver and
3251 * did hash table lookup before we unhashed socket. They will achieve
3252 * receive queue and will be purged by socket destructor.
3254 * Also we still have packets pending on receive queue and probably,
3255 * our own packets waiting in device queues. sock_destroy will drain
3256 * receive queue, but transmitted packets will delay socket destruction
3257 * until the last reference will be released.
3262 xfrm_sk_free_policy(sk);
3264 sk_refcnt_debug_release(sk);
3268 EXPORT_SYMBOL(sk_common_release);
3270 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3272 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3274 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3275 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3276 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3277 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3278 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3279 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3280 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3281 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3282 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3285 #ifdef CONFIG_PROC_FS
3286 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3288 int val[PROTO_INUSE_NR];
3291 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3293 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3295 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3297 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3299 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3301 int cpu, idx = prot->inuse_idx;
3304 for_each_possible_cpu(cpu)
3305 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3307 return res >= 0 ? res : 0;
3309 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3311 static void sock_inuse_add(struct net *net, int val)
3313 this_cpu_add(*net->core.sock_inuse, val);
3316 int sock_inuse_get(struct net *net)
3320 for_each_possible_cpu(cpu)
3321 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3326 EXPORT_SYMBOL_GPL(sock_inuse_get);
3328 static int __net_init sock_inuse_init_net(struct net *net)
3330 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3331 if (net->core.prot_inuse == NULL)
3334 net->core.sock_inuse = alloc_percpu(int);
3335 if (net->core.sock_inuse == NULL)
3341 free_percpu(net->core.prot_inuse);
3345 static void __net_exit sock_inuse_exit_net(struct net *net)
3347 free_percpu(net->core.prot_inuse);
3348 free_percpu(net->core.sock_inuse);
3351 static struct pernet_operations net_inuse_ops = {
3352 .init = sock_inuse_init_net,
3353 .exit = sock_inuse_exit_net,
3356 static __init int net_inuse_init(void)
3358 if (register_pernet_subsys(&net_inuse_ops))
3359 panic("Cannot initialize net inuse counters");
3364 core_initcall(net_inuse_init);
3366 static int assign_proto_idx(struct proto *prot)
3368 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3370 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3371 pr_err("PROTO_INUSE_NR exhausted\n");
3375 set_bit(prot->inuse_idx, proto_inuse_idx);
3379 static void release_proto_idx(struct proto *prot)
3381 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3382 clear_bit(prot->inuse_idx, proto_inuse_idx);
3385 static inline int assign_proto_idx(struct proto *prot)
3390 static inline void release_proto_idx(struct proto *prot)
3394 static void sock_inuse_add(struct net *net, int val)
3399 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3403 kfree(twsk_prot->twsk_slab_name);
3404 twsk_prot->twsk_slab_name = NULL;
3405 kmem_cache_destroy(twsk_prot->twsk_slab);
3406 twsk_prot->twsk_slab = NULL;
3409 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3413 kfree(rsk_prot->slab_name);
3414 rsk_prot->slab_name = NULL;
3415 kmem_cache_destroy(rsk_prot->slab);
3416 rsk_prot->slab = NULL;
3419 static int req_prot_init(const struct proto *prot)
3421 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3426 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3428 if (!rsk_prot->slab_name)
3431 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3432 rsk_prot->obj_size, 0,
3433 SLAB_ACCOUNT | prot->slab_flags,
3436 if (!rsk_prot->slab) {
3437 pr_crit("%s: Can't create request sock SLAB cache!\n",
3444 int proto_register(struct proto *prot, int alloc_slab)
3449 prot->slab = kmem_cache_create_usercopy(prot->name,
3451 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3453 prot->useroffset, prot->usersize,
3456 if (prot->slab == NULL) {
3457 pr_crit("%s: Can't create sock SLAB cache!\n",
3462 if (req_prot_init(prot))
3463 goto out_free_request_sock_slab;
3465 if (prot->twsk_prot != NULL) {
3466 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3468 if (prot->twsk_prot->twsk_slab_name == NULL)
3469 goto out_free_request_sock_slab;
3471 prot->twsk_prot->twsk_slab =
3472 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3473 prot->twsk_prot->twsk_obj_size,
3478 if (prot->twsk_prot->twsk_slab == NULL)
3479 goto out_free_timewait_sock_slab;
3483 mutex_lock(&proto_list_mutex);
3484 ret = assign_proto_idx(prot);
3486 mutex_unlock(&proto_list_mutex);
3487 goto out_free_timewait_sock_slab;
3489 list_add(&prot->node, &proto_list);
3490 mutex_unlock(&proto_list_mutex);
3493 out_free_timewait_sock_slab:
3494 if (alloc_slab && prot->twsk_prot)
3495 tw_prot_cleanup(prot->twsk_prot);
3496 out_free_request_sock_slab:
3498 req_prot_cleanup(prot->rsk_prot);
3500 kmem_cache_destroy(prot->slab);
3506 EXPORT_SYMBOL(proto_register);
3508 void proto_unregister(struct proto *prot)
3510 mutex_lock(&proto_list_mutex);
3511 release_proto_idx(prot);
3512 list_del(&prot->node);
3513 mutex_unlock(&proto_list_mutex);
3515 kmem_cache_destroy(prot->slab);
3518 req_prot_cleanup(prot->rsk_prot);
3519 tw_prot_cleanup(prot->twsk_prot);
3521 EXPORT_SYMBOL(proto_unregister);
3523 int sock_load_diag_module(int family, int protocol)
3526 if (!sock_is_registered(family))
3529 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3530 NETLINK_SOCK_DIAG, family);
3534 if (family == AF_INET &&
3535 protocol != IPPROTO_RAW &&
3536 !rcu_access_pointer(inet_protos[protocol]))
3540 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3541 NETLINK_SOCK_DIAG, family, protocol);
3543 EXPORT_SYMBOL(sock_load_diag_module);
3545 #ifdef CONFIG_PROC_FS
3546 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3547 __acquires(proto_list_mutex)
3549 mutex_lock(&proto_list_mutex);
3550 return seq_list_start_head(&proto_list, *pos);
3553 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3555 return seq_list_next(v, &proto_list, pos);
3558 static void proto_seq_stop(struct seq_file *seq, void *v)
3559 __releases(proto_list_mutex)
3561 mutex_unlock(&proto_list_mutex);
3564 static char proto_method_implemented(const void *method)
3566 return method == NULL ? 'n' : 'y';
3568 static long sock_prot_memory_allocated(struct proto *proto)
3570 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3573 static const char *sock_prot_memory_pressure(struct proto *proto)
3575 return proto->memory_pressure != NULL ?
3576 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3579 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3582 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3583 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3586 sock_prot_inuse_get(seq_file_net(seq), proto),
3587 sock_prot_memory_allocated(proto),
3588 sock_prot_memory_pressure(proto),
3590 proto->slab == NULL ? "no" : "yes",
3591 module_name(proto->owner),
3592 proto_method_implemented(proto->close),
3593 proto_method_implemented(proto->connect),
3594 proto_method_implemented(proto->disconnect),
3595 proto_method_implemented(proto->accept),
3596 proto_method_implemented(proto->ioctl),
3597 proto_method_implemented(proto->init),
3598 proto_method_implemented(proto->destroy),
3599 proto_method_implemented(proto->shutdown),
3600 proto_method_implemented(proto->setsockopt),
3601 proto_method_implemented(proto->getsockopt),
3602 proto_method_implemented(proto->sendmsg),
3603 proto_method_implemented(proto->recvmsg),
3604 proto_method_implemented(proto->sendpage),
3605 proto_method_implemented(proto->bind),
3606 proto_method_implemented(proto->backlog_rcv),
3607 proto_method_implemented(proto->hash),
3608 proto_method_implemented(proto->unhash),
3609 proto_method_implemented(proto->get_port),
3610 proto_method_implemented(proto->enter_memory_pressure));
3613 static int proto_seq_show(struct seq_file *seq, void *v)
3615 if (v == &proto_list)
3616 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3625 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3627 proto_seq_printf(seq, list_entry(v, struct proto, node));
3631 static const struct seq_operations proto_seq_ops = {
3632 .start = proto_seq_start,
3633 .next = proto_seq_next,
3634 .stop = proto_seq_stop,
3635 .show = proto_seq_show,
3638 static __net_init int proto_init_net(struct net *net)
3640 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3641 sizeof(struct seq_net_private)))
3647 static __net_exit void proto_exit_net(struct net *net)
3649 remove_proc_entry("protocols", net->proc_net);
3653 static __net_initdata struct pernet_operations proto_net_ops = {
3654 .init = proto_init_net,
3655 .exit = proto_exit_net,
3658 static int __init proto_init(void)
3660 return register_pernet_subsys(&proto_net_ops);
3663 subsys_initcall(proto_init);
3665 #endif /* PROC_FS */
3667 #ifdef CONFIG_NET_RX_BUSY_POLL
3668 bool sk_busy_loop_end(void *p, unsigned long start_time)
3670 struct sock *sk = p;
3672 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3673 sk_busy_loop_timeout(sk, start_time);
3675 EXPORT_SYMBOL(sk_busy_loop_end);
3676 #endif /* CONFIG_NET_RX_BUSY_POLL */