1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
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;
337 if (timeo == MAX_SCHEDULE_TIMEOUT) {
341 tv.tv_sec = timeo / HZ;
342 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
345 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
346 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
347 *(struct old_timeval32 *)optval = tv32;
352 struct __kernel_old_timeval old_tv;
353 old_tv.tv_sec = tv.tv_sec;
354 old_tv.tv_usec = tv.tv_usec;
355 *(struct __kernel_old_timeval *)optval = old_tv;
356 return sizeof(old_tv);
359 *(struct __kernel_sock_timeval *)optval = tv;
363 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
365 struct __kernel_sock_timeval tv;
367 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
368 struct old_timeval32 tv32;
370 if (optlen < sizeof(tv32))
373 if (copy_from_user(&tv32, optval, sizeof(tv32)))
375 tv.tv_sec = tv32.tv_sec;
376 tv.tv_usec = tv32.tv_usec;
377 } else if (old_timeval) {
378 struct __kernel_old_timeval old_tv;
380 if (optlen < sizeof(old_tv))
382 if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
384 tv.tv_sec = old_tv.tv_sec;
385 tv.tv_usec = old_tv.tv_usec;
387 if (optlen < sizeof(tv))
389 if (copy_from_user(&tv, optval, sizeof(tv)))
392 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
396 static int warned __read_mostly;
399 if (warned < 10 && net_ratelimit()) {
401 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
402 __func__, current->comm, task_pid_nr(current));
406 *timeo_p = MAX_SCHEDULE_TIMEOUT;
407 if (tv.tv_sec == 0 && tv.tv_usec == 0)
409 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
410 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
414 static void sock_warn_obsolete_bsdism(const char *name)
417 static char warncomm[TASK_COMM_LEN];
418 if (strcmp(warncomm, current->comm) && warned < 5) {
419 strcpy(warncomm, current->comm);
420 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
426 static bool sock_needs_netstamp(const struct sock *sk)
428 switch (sk->sk_family) {
437 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
439 if (sk->sk_flags & flags) {
440 sk->sk_flags &= ~flags;
441 if (sock_needs_netstamp(sk) &&
442 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
443 net_disable_timestamp();
448 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
451 struct sk_buff_head *list = &sk->sk_receive_queue;
453 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
454 atomic_inc(&sk->sk_drops);
455 trace_sock_rcvqueue_full(sk, skb);
459 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
460 atomic_inc(&sk->sk_drops);
465 skb_set_owner_r(skb, sk);
467 /* we escape from rcu protected region, make sure we dont leak
472 spin_lock_irqsave(&list->lock, flags);
473 sock_skb_set_dropcount(sk, skb);
474 __skb_queue_tail(list, skb);
475 spin_unlock_irqrestore(&list->lock, flags);
477 if (!sock_flag(sk, SOCK_DEAD))
478 sk->sk_data_ready(sk);
481 EXPORT_SYMBOL(__sock_queue_rcv_skb);
483 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
487 err = sk_filter(sk, skb);
491 return __sock_queue_rcv_skb(sk, skb);
493 EXPORT_SYMBOL(sock_queue_rcv_skb);
495 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
496 const int nested, unsigned int trim_cap, bool refcounted)
498 int rc = NET_RX_SUCCESS;
500 if (sk_filter_trim_cap(sk, skb, trim_cap))
501 goto discard_and_relse;
505 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
506 atomic_inc(&sk->sk_drops);
507 goto discard_and_relse;
510 bh_lock_sock_nested(sk);
513 if (!sock_owned_by_user(sk)) {
515 * trylock + unlock semantics:
517 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
519 rc = sk_backlog_rcv(sk, skb);
521 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
522 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
524 atomic_inc(&sk->sk_drops);
525 goto discard_and_relse;
537 EXPORT_SYMBOL(__sk_receive_skb);
539 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
541 struct dst_entry *dst = __sk_dst_get(sk);
543 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
544 sk_tx_queue_clear(sk);
545 sk->sk_dst_pending_confirm = 0;
546 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
553 EXPORT_SYMBOL(__sk_dst_check);
555 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
557 struct dst_entry *dst = sk_dst_get(sk);
559 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
567 EXPORT_SYMBOL(sk_dst_check);
569 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
571 int ret = -ENOPROTOOPT;
572 #ifdef CONFIG_NETDEVICES
573 struct net *net = sock_net(sk);
577 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
584 sk->sk_bound_dev_if = ifindex;
585 if (sk->sk_prot->rehash)
586 sk->sk_prot->rehash(sk);
597 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
603 ret = sock_bindtoindex_locked(sk, ifindex);
609 EXPORT_SYMBOL(sock_bindtoindex);
611 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
614 int ret = -ENOPROTOOPT;
615 #ifdef CONFIG_NETDEVICES
616 struct net *net = sock_net(sk);
617 char devname[IFNAMSIZ];
624 /* Bind this socket to a particular device like "eth0",
625 * as specified in the passed interface name. If the
626 * name is "" or the option length is zero the socket
629 if (optlen > IFNAMSIZ - 1)
630 optlen = IFNAMSIZ - 1;
631 memset(devname, 0, sizeof(devname));
634 if (copy_from_user(devname, optval, optlen))
638 if (devname[0] != '\0') {
639 struct net_device *dev;
642 dev = dev_get_by_name_rcu(net, devname);
644 index = dev->ifindex;
651 return sock_bindtoindex(sk, index, true);
658 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
659 int __user *optlen, int len)
661 int ret = -ENOPROTOOPT;
662 #ifdef CONFIG_NETDEVICES
663 struct net *net = sock_net(sk);
664 char devname[IFNAMSIZ];
666 if (sk->sk_bound_dev_if == 0) {
675 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
679 len = strlen(devname) + 1;
682 if (copy_to_user(optval, devname, len))
687 if (put_user(len, optlen))
698 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
702 sock_set_flag(sk, bit);
704 sock_reset_flag(sk, bit);
707 bool sk_mc_loop(struct sock *sk)
709 if (dev_recursion_level())
713 switch (sk->sk_family) {
715 return inet_sk(sk)->mc_loop;
716 #if IS_ENABLED(CONFIG_IPV6)
718 return inet6_sk(sk)->mc_loop;
724 EXPORT_SYMBOL(sk_mc_loop);
726 void sock_set_reuseaddr(struct sock *sk)
729 sk->sk_reuse = SK_CAN_REUSE;
732 EXPORT_SYMBOL(sock_set_reuseaddr);
734 void sock_set_reuseport(struct sock *sk)
737 sk->sk_reuseport = true;
740 EXPORT_SYMBOL(sock_set_reuseport);
742 void sock_no_linger(struct sock *sk)
745 sk->sk_lingertime = 0;
746 sock_set_flag(sk, SOCK_LINGER);
749 EXPORT_SYMBOL(sock_no_linger);
751 void sock_set_priority(struct sock *sk, u32 priority)
754 sk->sk_priority = priority;
757 EXPORT_SYMBOL(sock_set_priority);
759 void sock_set_sndtimeo(struct sock *sk, s64 secs)
762 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
763 sk->sk_sndtimeo = secs * HZ;
765 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
768 EXPORT_SYMBOL(sock_set_sndtimeo);
770 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
773 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
774 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
775 sock_set_flag(sk, SOCK_RCVTSTAMP);
776 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
778 sock_reset_flag(sk, SOCK_RCVTSTAMP);
779 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
780 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
784 void sock_enable_timestamps(struct sock *sk)
787 __sock_set_timestamps(sk, true, false, true);
790 EXPORT_SYMBOL(sock_enable_timestamps);
792 void sock_set_keepalive(struct sock *sk)
795 if (sk->sk_prot->keepalive)
796 sk->sk_prot->keepalive(sk, true);
797 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
800 EXPORT_SYMBOL(sock_set_keepalive);
802 static void __sock_set_rcvbuf(struct sock *sk, int val)
804 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
805 * as a negative value.
807 val = min_t(int, val, INT_MAX / 2);
808 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
810 /* We double it on the way in to account for "struct sk_buff" etc.
811 * overhead. Applications assume that the SO_RCVBUF setting they make
812 * will allow that much actual data to be received on that socket.
814 * Applications are unaware that "struct sk_buff" and other overheads
815 * allocate from the receive buffer during socket buffer allocation.
817 * And after considering the possible alternatives, returning the value
818 * we actually used in getsockopt is the most desirable behavior.
820 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
823 void sock_set_rcvbuf(struct sock *sk, int val)
826 __sock_set_rcvbuf(sk, val);
829 EXPORT_SYMBOL(sock_set_rcvbuf);
832 * This is meant for all protocols to use and covers goings on
833 * at the socket level. Everything here is generic.
836 int sock_setsockopt(struct socket *sock, int level, int optname,
837 char __user *optval, unsigned int optlen)
839 struct sock_txtime sk_txtime;
840 struct sock *sk = sock->sk;
847 * Options without arguments
850 if (optname == SO_BINDTODEVICE)
851 return sock_setbindtodevice(sk, optval, optlen);
853 if (optlen < sizeof(int))
856 if (get_user(val, (int __user *)optval))
859 valbool = val ? 1 : 0;
865 if (val && !capable(CAP_NET_ADMIN))
868 sock_valbool_flag(sk, SOCK_DBG, valbool);
871 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
874 sk->sk_reuseport = valbool;
883 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
887 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
890 /* Don't error on this BSD doesn't and if you think
891 * about it this is right. Otherwise apps have to
892 * play 'guess the biggest size' games. RCVBUF/SNDBUF
893 * are treated in BSD as hints
895 val = min_t(u32, val, sysctl_wmem_max);
897 /* Ensure val * 2 fits into an int, to prevent max_t()
898 * from treating it as a negative value.
900 val = min_t(int, val, INT_MAX / 2);
901 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
902 WRITE_ONCE(sk->sk_sndbuf,
903 max_t(int, val * 2, SOCK_MIN_SNDBUF));
904 /* Wake up sending tasks if we upped the value. */
905 sk->sk_write_space(sk);
909 if (!capable(CAP_NET_ADMIN)) {
914 /* No negative values (to prevent underflow, as val will be
922 /* Don't error on this BSD doesn't and if you think
923 * about it this is right. Otherwise apps have to
924 * play 'guess the biggest size' games. RCVBUF/SNDBUF
925 * are treated in BSD as hints
927 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
931 if (!capable(CAP_NET_ADMIN)) {
936 /* No negative values (to prevent underflow, as val will be
939 __sock_set_rcvbuf(sk, max(val, 0));
943 if (sk->sk_prot->keepalive)
944 sk->sk_prot->keepalive(sk, valbool);
945 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
949 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
953 sk->sk_no_check_tx = valbool;
957 if ((val >= 0 && val <= 6) ||
958 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
959 sk->sk_priority = val;
965 if (optlen < sizeof(ling)) {
966 ret = -EINVAL; /* 1003.1g */
969 if (copy_from_user(&ling, optval, sizeof(ling))) {
974 sock_reset_flag(sk, SOCK_LINGER);
976 #if (BITS_PER_LONG == 32)
977 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
978 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
981 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
982 sock_set_flag(sk, SOCK_LINGER);
987 sock_warn_obsolete_bsdism("setsockopt");
992 set_bit(SOCK_PASSCRED, &sock->flags);
994 clear_bit(SOCK_PASSCRED, &sock->flags);
997 case SO_TIMESTAMP_OLD:
998 __sock_set_timestamps(sk, valbool, false, false);
1000 case SO_TIMESTAMP_NEW:
1001 __sock_set_timestamps(sk, valbool, true, false);
1003 case SO_TIMESTAMPNS_OLD:
1004 __sock_set_timestamps(sk, valbool, false, true);
1006 case SO_TIMESTAMPNS_NEW:
1007 __sock_set_timestamps(sk, valbool, true, true);
1009 case SO_TIMESTAMPING_NEW:
1010 sock_set_flag(sk, SOCK_TSTAMP_NEW);
1012 case SO_TIMESTAMPING_OLD:
1013 if (val & ~SOF_TIMESTAMPING_MASK) {
1018 if (val & SOF_TIMESTAMPING_OPT_ID &&
1019 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1020 if (sk->sk_protocol == IPPROTO_TCP &&
1021 sk->sk_type == SOCK_STREAM) {
1022 if ((1 << sk->sk_state) &
1023 (TCPF_CLOSE | TCPF_LISTEN)) {
1027 sk->sk_tskey = tcp_sk(sk)->snd_una;
1033 if (val & SOF_TIMESTAMPING_OPT_STATS &&
1034 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1039 sk->sk_tsflags = val;
1040 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1041 sock_enable_timestamp(sk,
1042 SOCK_TIMESTAMPING_RX_SOFTWARE);
1044 if (optname == SO_TIMESTAMPING_NEW)
1045 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
1047 sock_disable_timestamp(sk,
1048 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1055 if (sock->ops->set_rcvlowat)
1056 ret = sock->ops->set_rcvlowat(sk, val);
1058 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1061 case SO_RCVTIMEO_OLD:
1062 case SO_RCVTIMEO_NEW:
1063 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
1066 case SO_SNDTIMEO_OLD:
1067 case SO_SNDTIMEO_NEW:
1068 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
1071 case SO_ATTACH_FILTER:
1073 if (optlen == sizeof(struct sock_fprog)) {
1074 struct sock_fprog fprog;
1077 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1080 ret = sk_attach_filter(&fprog, sk);
1086 if (optlen == sizeof(u32)) {
1090 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1093 ret = sk_attach_bpf(ufd, sk);
1097 case SO_ATTACH_REUSEPORT_CBPF:
1099 if (optlen == sizeof(struct sock_fprog)) {
1100 struct sock_fprog fprog;
1103 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1106 ret = sk_reuseport_attach_filter(&fprog, sk);
1110 case SO_ATTACH_REUSEPORT_EBPF:
1112 if (optlen == sizeof(u32)) {
1116 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1119 ret = sk_reuseport_attach_bpf(ufd, sk);
1123 case SO_DETACH_REUSEPORT_BPF:
1124 ret = reuseport_detach_prog(sk);
1127 case SO_DETACH_FILTER:
1128 ret = sk_detach_filter(sk);
1131 case SO_LOCK_FILTER:
1132 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1135 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1140 set_bit(SOCK_PASSSEC, &sock->flags);
1142 clear_bit(SOCK_PASSSEC, &sock->flags);
1145 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1147 } else if (val != sk->sk_mark) {
1154 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1157 case SO_WIFI_STATUS:
1158 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1162 if (sock->ops->set_peek_off)
1163 ret = sock->ops->set_peek_off(sk, val);
1169 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1172 case SO_SELECT_ERR_QUEUE:
1173 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1176 #ifdef CONFIG_NET_RX_BUSY_POLL
1178 /* allow unprivileged users to decrease the value */
1179 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1185 sk->sk_ll_usec = val;
1190 case SO_MAX_PACING_RATE:
1192 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1194 if (sizeof(ulval) != sizeof(val) &&
1195 optlen >= sizeof(ulval) &&
1196 get_user(ulval, (unsigned long __user *)optval)) {
1201 cmpxchg(&sk->sk_pacing_status,
1204 sk->sk_max_pacing_rate = ulval;
1205 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1208 case SO_INCOMING_CPU:
1209 WRITE_ONCE(sk->sk_incoming_cpu, val);
1214 dst_negative_advice(sk);
1218 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1219 if (!((sk->sk_type == SOCK_STREAM &&
1220 sk->sk_protocol == IPPROTO_TCP) ||
1221 (sk->sk_type == SOCK_DGRAM &&
1222 sk->sk_protocol == IPPROTO_UDP)))
1224 } else if (sk->sk_family != PF_RDS) {
1228 if (val < 0 || val > 1)
1231 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1236 if (optlen != sizeof(struct sock_txtime)) {
1239 } else if (copy_from_user(&sk_txtime, optval,
1240 sizeof(struct sock_txtime))) {
1243 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1247 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1248 * scheduler has enough safe guards.
1250 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1251 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1255 sock_valbool_flag(sk, SOCK_TXTIME, true);
1256 sk->sk_clockid = sk_txtime.clockid;
1257 sk->sk_txtime_deadline_mode =
1258 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1259 sk->sk_txtime_report_errors =
1260 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1263 case SO_BINDTOIFINDEX:
1264 ret = sock_bindtoindex_locked(sk, val);
1274 EXPORT_SYMBOL(sock_setsockopt);
1277 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1278 struct ucred *ucred)
1280 ucred->pid = pid_vnr(pid);
1281 ucred->uid = ucred->gid = -1;
1283 struct user_namespace *current_ns = current_user_ns();
1285 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1286 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1290 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1292 struct user_namespace *user_ns = current_user_ns();
1295 for (i = 0; i < src->ngroups; i++)
1296 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1302 int sock_getsockopt(struct socket *sock, int level, int optname,
1303 char __user *optval, int __user *optlen)
1305 struct sock *sk = sock->sk;
1310 unsigned long ulval;
1312 struct old_timeval32 tm32;
1313 struct __kernel_old_timeval tm;
1314 struct __kernel_sock_timeval stm;
1315 struct sock_txtime txtime;
1318 int lv = sizeof(int);
1321 if (get_user(len, optlen))
1326 memset(&v, 0, sizeof(v));
1330 v.val = sock_flag(sk, SOCK_DBG);
1334 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1338 v.val = sock_flag(sk, SOCK_BROADCAST);
1342 v.val = sk->sk_sndbuf;
1346 v.val = sk->sk_rcvbuf;
1350 v.val = sk->sk_reuse;
1354 v.val = sk->sk_reuseport;
1358 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1362 v.val = sk->sk_type;
1366 v.val = sk->sk_protocol;
1370 v.val = sk->sk_family;
1374 v.val = -sock_error(sk);
1376 v.val = xchg(&sk->sk_err_soft, 0);
1380 v.val = sock_flag(sk, SOCK_URGINLINE);
1384 v.val = sk->sk_no_check_tx;
1388 v.val = sk->sk_priority;
1392 lv = sizeof(v.ling);
1393 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1394 v.ling.l_linger = sk->sk_lingertime / HZ;
1398 sock_warn_obsolete_bsdism("getsockopt");
1401 case SO_TIMESTAMP_OLD:
1402 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1403 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1404 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1407 case SO_TIMESTAMPNS_OLD:
1408 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1411 case SO_TIMESTAMP_NEW:
1412 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1415 case SO_TIMESTAMPNS_NEW:
1416 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1419 case SO_TIMESTAMPING_OLD:
1420 v.val = sk->sk_tsflags;
1423 case SO_RCVTIMEO_OLD:
1424 case SO_RCVTIMEO_NEW:
1425 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1428 case SO_SNDTIMEO_OLD:
1429 case SO_SNDTIMEO_NEW:
1430 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1434 v.val = sk->sk_rcvlowat;
1442 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1447 struct ucred peercred;
1448 if (len > sizeof(peercred))
1449 len = sizeof(peercred);
1450 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1451 if (copy_to_user(optval, &peercred, len))
1460 if (!sk->sk_peer_cred)
1463 n = sk->sk_peer_cred->group_info->ngroups;
1464 if (len < n * sizeof(gid_t)) {
1465 len = n * sizeof(gid_t);
1466 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1468 len = n * sizeof(gid_t);
1470 ret = groups_to_user((gid_t __user *)optval,
1471 sk->sk_peer_cred->group_info);
1481 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1486 if (copy_to_user(optval, address, len))
1491 /* Dubious BSD thing... Probably nobody even uses it, but
1492 * the UNIX standard wants it for whatever reason... -DaveM
1495 v.val = sk->sk_state == TCP_LISTEN;
1499 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1503 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1506 v.val = sk->sk_mark;
1510 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1513 case SO_WIFI_STATUS:
1514 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1518 if (!sock->ops->set_peek_off)
1521 v.val = sk->sk_peek_off;
1524 v.val = sock_flag(sk, SOCK_NOFCS);
1527 case SO_BINDTODEVICE:
1528 return sock_getbindtodevice(sk, optval, optlen, len);
1531 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1537 case SO_LOCK_FILTER:
1538 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1541 case SO_BPF_EXTENSIONS:
1542 v.val = bpf_tell_extensions();
1545 case SO_SELECT_ERR_QUEUE:
1546 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1549 #ifdef CONFIG_NET_RX_BUSY_POLL
1551 v.val = sk->sk_ll_usec;
1555 case SO_MAX_PACING_RATE:
1556 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1557 lv = sizeof(v.ulval);
1558 v.ulval = sk->sk_max_pacing_rate;
1561 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1565 case SO_INCOMING_CPU:
1566 v.val = READ_ONCE(sk->sk_incoming_cpu);
1571 u32 meminfo[SK_MEMINFO_VARS];
1573 sk_get_meminfo(sk, meminfo);
1575 len = min_t(unsigned int, len, sizeof(meminfo));
1576 if (copy_to_user(optval, &meminfo, len))
1582 #ifdef CONFIG_NET_RX_BUSY_POLL
1583 case SO_INCOMING_NAPI_ID:
1584 v.val = READ_ONCE(sk->sk_napi_id);
1586 /* aggregate non-NAPI IDs down to 0 */
1587 if (v.val < MIN_NAPI_ID)
1597 v.val64 = sock_gen_cookie(sk);
1601 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1605 lv = sizeof(v.txtime);
1606 v.txtime.clockid = sk->sk_clockid;
1607 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1608 SOF_TXTIME_DEADLINE_MODE : 0;
1609 v.txtime.flags |= sk->sk_txtime_report_errors ?
1610 SOF_TXTIME_REPORT_ERRORS : 0;
1613 case SO_BINDTOIFINDEX:
1614 v.val = sk->sk_bound_dev_if;
1618 /* We implement the SO_SNDLOWAT etc to not be settable
1621 return -ENOPROTOOPT;
1626 if (copy_to_user(optval, &v, len))
1629 if (put_user(len, optlen))
1635 * Initialize an sk_lock.
1637 * (We also register the sk_lock with the lock validator.)
1639 static inline void sock_lock_init(struct sock *sk)
1641 if (sk->sk_kern_sock)
1642 sock_lock_init_class_and_name(
1644 af_family_kern_slock_key_strings[sk->sk_family],
1645 af_family_kern_slock_keys + sk->sk_family,
1646 af_family_kern_key_strings[sk->sk_family],
1647 af_family_kern_keys + sk->sk_family);
1649 sock_lock_init_class_and_name(
1651 af_family_slock_key_strings[sk->sk_family],
1652 af_family_slock_keys + sk->sk_family,
1653 af_family_key_strings[sk->sk_family],
1654 af_family_keys + sk->sk_family);
1658 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1659 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1660 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1662 static void sock_copy(struct sock *nsk, const struct sock *osk)
1664 const struct proto *prot = READ_ONCE(osk->sk_prot);
1665 #ifdef CONFIG_SECURITY_NETWORK
1666 void *sptr = nsk->sk_security;
1668 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1670 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1671 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1673 #ifdef CONFIG_SECURITY_NETWORK
1674 nsk->sk_security = sptr;
1675 security_sk_clone(osk, nsk);
1679 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1683 struct kmem_cache *slab;
1687 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1690 if (want_init_on_alloc(priority))
1691 sk_prot_clear_nulls(sk, prot->obj_size);
1693 sk = kmalloc(prot->obj_size, priority);
1696 if (security_sk_alloc(sk, family, priority))
1699 if (!try_module_get(prot->owner))
1701 sk_tx_queue_clear(sk);
1707 security_sk_free(sk);
1710 kmem_cache_free(slab, sk);
1716 static void sk_prot_free(struct proto *prot, struct sock *sk)
1718 struct kmem_cache *slab;
1719 struct module *owner;
1721 owner = prot->owner;
1724 cgroup_sk_free(&sk->sk_cgrp_data);
1725 mem_cgroup_sk_free(sk);
1726 security_sk_free(sk);
1728 kmem_cache_free(slab, sk);
1735 * sk_alloc - All socket objects are allocated here
1736 * @net: the applicable net namespace
1737 * @family: protocol family
1738 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1739 * @prot: struct proto associated with this new sock instance
1740 * @kern: is this to be a kernel socket?
1742 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1743 struct proto *prot, int kern)
1747 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1749 sk->sk_family = family;
1751 * See comment in struct sock definition to understand
1752 * why we need sk_prot_creator -acme
1754 sk->sk_prot = sk->sk_prot_creator = prot;
1755 sk->sk_kern_sock = kern;
1757 sk->sk_net_refcnt = kern ? 0 : 1;
1758 if (likely(sk->sk_net_refcnt)) {
1760 sock_inuse_add(net, 1);
1763 sock_net_set(sk, net);
1764 refcount_set(&sk->sk_wmem_alloc, 1);
1766 mem_cgroup_sk_alloc(sk);
1767 cgroup_sk_alloc(&sk->sk_cgrp_data);
1768 sock_update_classid(&sk->sk_cgrp_data);
1769 sock_update_netprioidx(&sk->sk_cgrp_data);
1770 sk_tx_queue_clear(sk);
1775 EXPORT_SYMBOL(sk_alloc);
1777 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1778 * grace period. This is the case for UDP sockets and TCP listeners.
1780 static void __sk_destruct(struct rcu_head *head)
1782 struct sock *sk = container_of(head, struct sock, sk_rcu);
1783 struct sk_filter *filter;
1785 if (sk->sk_destruct)
1786 sk->sk_destruct(sk);
1788 filter = rcu_dereference_check(sk->sk_filter,
1789 refcount_read(&sk->sk_wmem_alloc) == 0);
1791 sk_filter_uncharge(sk, filter);
1792 RCU_INIT_POINTER(sk->sk_filter, NULL);
1795 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1797 #ifdef CONFIG_BPF_SYSCALL
1798 bpf_sk_storage_free(sk);
1801 if (atomic_read(&sk->sk_omem_alloc))
1802 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1803 __func__, atomic_read(&sk->sk_omem_alloc));
1805 if (sk->sk_frag.page) {
1806 put_page(sk->sk_frag.page);
1807 sk->sk_frag.page = NULL;
1810 if (sk->sk_peer_cred)
1811 put_cred(sk->sk_peer_cred);
1812 put_pid(sk->sk_peer_pid);
1813 if (likely(sk->sk_net_refcnt))
1814 put_net(sock_net(sk));
1815 sk_prot_free(sk->sk_prot_creator, sk);
1818 void sk_destruct(struct sock *sk)
1820 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1822 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1823 reuseport_detach_sock(sk);
1824 use_call_rcu = true;
1828 call_rcu(&sk->sk_rcu, __sk_destruct);
1830 __sk_destruct(&sk->sk_rcu);
1833 static void __sk_free(struct sock *sk)
1835 if (likely(sk->sk_net_refcnt))
1836 sock_inuse_add(sock_net(sk), -1);
1838 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1839 sock_diag_broadcast_destroy(sk);
1844 void sk_free(struct sock *sk)
1847 * We subtract one from sk_wmem_alloc and can know if
1848 * some packets are still in some tx queue.
1849 * If not null, sock_wfree() will call __sk_free(sk) later
1851 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1854 EXPORT_SYMBOL(sk_free);
1856 static void sk_init_common(struct sock *sk)
1858 skb_queue_head_init(&sk->sk_receive_queue);
1859 skb_queue_head_init(&sk->sk_write_queue);
1860 skb_queue_head_init(&sk->sk_error_queue);
1862 rwlock_init(&sk->sk_callback_lock);
1863 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1864 af_rlock_keys + sk->sk_family,
1865 af_family_rlock_key_strings[sk->sk_family]);
1866 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1867 af_wlock_keys + sk->sk_family,
1868 af_family_wlock_key_strings[sk->sk_family]);
1869 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1870 af_elock_keys + sk->sk_family,
1871 af_family_elock_key_strings[sk->sk_family]);
1872 lockdep_set_class_and_name(&sk->sk_callback_lock,
1873 af_callback_keys + sk->sk_family,
1874 af_family_clock_key_strings[sk->sk_family]);
1878 * sk_clone_lock - clone a socket, and lock its clone
1879 * @sk: the socket to clone
1880 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1882 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1884 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1886 struct proto *prot = READ_ONCE(sk->sk_prot);
1888 bool is_charged = true;
1890 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1891 if (newsk != NULL) {
1892 struct sk_filter *filter;
1894 sock_copy(newsk, sk);
1896 newsk->sk_prot_creator = prot;
1899 if (likely(newsk->sk_net_refcnt))
1900 get_net(sock_net(newsk));
1901 sk_node_init(&newsk->sk_node);
1902 sock_lock_init(newsk);
1903 bh_lock_sock(newsk);
1904 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1905 newsk->sk_backlog.len = 0;
1907 atomic_set(&newsk->sk_rmem_alloc, 0);
1909 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1911 refcount_set(&newsk->sk_wmem_alloc, 1);
1912 atomic_set(&newsk->sk_omem_alloc, 0);
1913 sk_init_common(newsk);
1915 newsk->sk_dst_cache = NULL;
1916 newsk->sk_dst_pending_confirm = 0;
1917 newsk->sk_wmem_queued = 0;
1918 newsk->sk_forward_alloc = 0;
1919 atomic_set(&newsk->sk_drops, 0);
1920 newsk->sk_send_head = NULL;
1921 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1922 atomic_set(&newsk->sk_zckey, 0);
1924 sock_reset_flag(newsk, SOCK_DONE);
1926 /* sk->sk_memcg will be populated at accept() time */
1927 newsk->sk_memcg = NULL;
1929 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1932 filter = rcu_dereference(sk->sk_filter);
1934 /* though it's an empty new sock, the charging may fail
1935 * if sysctl_optmem_max was changed between creation of
1936 * original socket and cloning
1938 is_charged = sk_filter_charge(newsk, filter);
1939 RCU_INIT_POINTER(newsk->sk_filter, filter);
1942 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1943 /* We need to make sure that we don't uncharge the new
1944 * socket if we couldn't charge it in the first place
1945 * as otherwise we uncharge the parent's filter.
1948 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1949 sk_free_unlock_clone(newsk);
1953 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1955 if (bpf_sk_storage_clone(sk, newsk)) {
1956 sk_free_unlock_clone(newsk);
1961 /* Clear sk_user_data if parent had the pointer tagged
1962 * as not suitable for copying when cloning.
1964 if (sk_user_data_is_nocopy(newsk))
1965 newsk->sk_user_data = NULL;
1968 newsk->sk_err_soft = 0;
1969 newsk->sk_priority = 0;
1970 newsk->sk_incoming_cpu = raw_smp_processor_id();
1971 if (likely(newsk->sk_net_refcnt))
1972 sock_inuse_add(sock_net(newsk), 1);
1975 * Before updating sk_refcnt, we must commit prior changes to memory
1976 * (Documentation/RCU/rculist_nulls.txt for details)
1979 refcount_set(&newsk->sk_refcnt, 2);
1982 * Increment the counter in the same struct proto as the master
1983 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1984 * is the same as sk->sk_prot->socks, as this field was copied
1987 * This _changes_ the previous behaviour, where
1988 * tcp_create_openreq_child always was incrementing the
1989 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1990 * to be taken into account in all callers. -acme
1992 sk_refcnt_debug_inc(newsk);
1993 sk_set_socket(newsk, NULL);
1994 sk_tx_queue_clear(newsk);
1995 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1997 if (newsk->sk_prot->sockets_allocated)
1998 sk_sockets_allocated_inc(newsk);
2000 if (sock_needs_netstamp(sk) &&
2001 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2002 net_enable_timestamp();
2007 EXPORT_SYMBOL_GPL(sk_clone_lock);
2009 void sk_free_unlock_clone(struct sock *sk)
2011 /* It is still raw copy of parent, so invalidate
2012 * destructor and make plain sk_free() */
2013 sk->sk_destruct = NULL;
2017 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2019 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2023 sk_dst_set(sk, dst);
2024 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2025 if (sk->sk_route_caps & NETIF_F_GSO)
2026 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2027 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2028 if (sk_can_gso(sk)) {
2029 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2030 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2032 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2033 sk->sk_gso_max_size = dst->dev->gso_max_size;
2034 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2037 sk->sk_gso_max_segs = max_segs;
2039 EXPORT_SYMBOL_GPL(sk_setup_caps);
2042 * Simple resource managers for sockets.
2047 * Write buffer destructor automatically called from kfree_skb.
2049 void sock_wfree(struct sk_buff *skb)
2051 struct sock *sk = skb->sk;
2052 unsigned int len = skb->truesize;
2054 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2056 * Keep a reference on sk_wmem_alloc, this will be released
2057 * after sk_write_space() call
2059 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2060 sk->sk_write_space(sk);
2064 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2065 * could not do because of in-flight packets
2067 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2070 EXPORT_SYMBOL(sock_wfree);
2072 /* This variant of sock_wfree() is used by TCP,
2073 * since it sets SOCK_USE_WRITE_QUEUE.
2075 void __sock_wfree(struct sk_buff *skb)
2077 struct sock *sk = skb->sk;
2079 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2083 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2088 if (unlikely(!sk_fullsock(sk))) {
2089 skb->destructor = sock_edemux;
2094 skb->destructor = sock_wfree;
2095 skb_set_hash_from_sk(skb, sk);
2097 * We used to take a refcount on sk, but following operation
2098 * is enough to guarantee sk_free() wont free this sock until
2099 * all in-flight packets are completed
2101 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2103 EXPORT_SYMBOL(skb_set_owner_w);
2105 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2107 #ifdef CONFIG_TLS_DEVICE
2108 /* Drivers depend on in-order delivery for crypto offload,
2109 * partial orphan breaks out-of-order-OK logic.
2114 return (skb->destructor == sock_wfree ||
2115 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2118 /* This helper is used by netem, as it can hold packets in its
2119 * delay queue. We want to allow the owner socket to send more
2120 * packets, as if they were already TX completed by a typical driver.
2121 * But we also want to keep skb->sk set because some packet schedulers
2122 * rely on it (sch_fq for example).
2124 void skb_orphan_partial(struct sk_buff *skb)
2126 if (skb_is_tcp_pure_ack(skb))
2129 if (can_skb_orphan_partial(skb)) {
2130 struct sock *sk = skb->sk;
2132 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2133 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2134 skb->destructor = sock_efree;
2140 EXPORT_SYMBOL(skb_orphan_partial);
2143 * Read buffer destructor automatically called from kfree_skb.
2145 void sock_rfree(struct sk_buff *skb)
2147 struct sock *sk = skb->sk;
2148 unsigned int len = skb->truesize;
2150 atomic_sub(len, &sk->sk_rmem_alloc);
2151 sk_mem_uncharge(sk, len);
2153 EXPORT_SYMBOL(sock_rfree);
2156 * Buffer destructor for skbs that are not used directly in read or write
2157 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2159 void sock_efree(struct sk_buff *skb)
2163 EXPORT_SYMBOL(sock_efree);
2165 /* Buffer destructor for prefetch/receive path where reference count may
2166 * not be held, e.g. for listen sockets.
2169 void sock_pfree(struct sk_buff *skb)
2171 if (sk_is_refcounted(skb->sk))
2172 sock_gen_put(skb->sk);
2174 EXPORT_SYMBOL(sock_pfree);
2175 #endif /* CONFIG_INET */
2177 kuid_t sock_i_uid(struct sock *sk)
2181 read_lock_bh(&sk->sk_callback_lock);
2182 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2183 read_unlock_bh(&sk->sk_callback_lock);
2186 EXPORT_SYMBOL(sock_i_uid);
2188 unsigned long sock_i_ino(struct sock *sk)
2192 read_lock_bh(&sk->sk_callback_lock);
2193 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2194 read_unlock_bh(&sk->sk_callback_lock);
2197 EXPORT_SYMBOL(sock_i_ino);
2200 * Allocate a skb from the socket's send buffer.
2202 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2206 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2207 struct sk_buff *skb = alloc_skb(size, priority);
2210 skb_set_owner_w(skb, sk);
2216 EXPORT_SYMBOL(sock_wmalloc);
2218 static void sock_ofree(struct sk_buff *skb)
2220 struct sock *sk = skb->sk;
2222 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2225 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2228 struct sk_buff *skb;
2230 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2231 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2235 skb = alloc_skb(size, priority);
2239 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2241 skb->destructor = sock_ofree;
2246 * Allocate a memory block from the socket's option memory buffer.
2248 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2250 if ((unsigned int)size <= sysctl_optmem_max &&
2251 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2253 /* First do the add, to avoid the race if kmalloc
2256 atomic_add(size, &sk->sk_omem_alloc);
2257 mem = kmalloc(size, priority);
2260 atomic_sub(size, &sk->sk_omem_alloc);
2264 EXPORT_SYMBOL(sock_kmalloc);
2266 /* Free an option memory block. Note, we actually want the inline
2267 * here as this allows gcc to detect the nullify and fold away the
2268 * condition entirely.
2270 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2273 if (WARN_ON_ONCE(!mem))
2279 atomic_sub(size, &sk->sk_omem_alloc);
2282 void sock_kfree_s(struct sock *sk, void *mem, int size)
2284 __sock_kfree_s(sk, mem, size, false);
2286 EXPORT_SYMBOL(sock_kfree_s);
2288 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2290 __sock_kfree_s(sk, mem, size, true);
2292 EXPORT_SYMBOL(sock_kzfree_s);
2294 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2295 I think, these locks should be removed for datagram sockets.
2297 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2301 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2305 if (signal_pending(current))
2307 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2308 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2309 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2311 if (sk->sk_shutdown & SEND_SHUTDOWN)
2315 timeo = schedule_timeout(timeo);
2317 finish_wait(sk_sleep(sk), &wait);
2323 * Generic send/receive buffer handlers
2326 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2327 unsigned long data_len, int noblock,
2328 int *errcode, int max_page_order)
2330 struct sk_buff *skb;
2334 timeo = sock_sndtimeo(sk, noblock);
2336 err = sock_error(sk);
2341 if (sk->sk_shutdown & SEND_SHUTDOWN)
2344 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2347 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2348 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2352 if (signal_pending(current))
2354 timeo = sock_wait_for_wmem(sk, timeo);
2356 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2357 errcode, sk->sk_allocation);
2359 skb_set_owner_w(skb, sk);
2363 err = sock_intr_errno(timeo);
2368 EXPORT_SYMBOL(sock_alloc_send_pskb);
2370 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2371 int noblock, int *errcode)
2373 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2375 EXPORT_SYMBOL(sock_alloc_send_skb);
2377 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2378 struct sockcm_cookie *sockc)
2382 switch (cmsg->cmsg_type) {
2384 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2386 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2388 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2390 case SO_TIMESTAMPING_OLD:
2391 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2394 tsflags = *(u32 *)CMSG_DATA(cmsg);
2395 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2398 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2399 sockc->tsflags |= tsflags;
2402 if (!sock_flag(sk, SOCK_TXTIME))
2404 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2406 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2408 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2410 case SCM_CREDENTIALS:
2417 EXPORT_SYMBOL(__sock_cmsg_send);
2419 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2420 struct sockcm_cookie *sockc)
2422 struct cmsghdr *cmsg;
2425 for_each_cmsghdr(cmsg, msg) {
2426 if (!CMSG_OK(msg, cmsg))
2428 if (cmsg->cmsg_level != SOL_SOCKET)
2430 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2436 EXPORT_SYMBOL(sock_cmsg_send);
2438 static void sk_enter_memory_pressure(struct sock *sk)
2440 if (!sk->sk_prot->enter_memory_pressure)
2443 sk->sk_prot->enter_memory_pressure(sk);
2446 static void sk_leave_memory_pressure(struct sock *sk)
2448 if (sk->sk_prot->leave_memory_pressure) {
2449 sk->sk_prot->leave_memory_pressure(sk);
2451 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2453 if (memory_pressure && READ_ONCE(*memory_pressure))
2454 WRITE_ONCE(*memory_pressure, 0);
2458 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2459 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2462 * skb_page_frag_refill - check that a page_frag contains enough room
2463 * @sz: minimum size of the fragment we want to get
2464 * @pfrag: pointer to page_frag
2465 * @gfp: priority for memory allocation
2467 * Note: While this allocator tries to use high order pages, there is
2468 * no guarantee that allocations succeed. Therefore, @sz MUST be
2469 * less or equal than PAGE_SIZE.
2471 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2474 if (page_ref_count(pfrag->page) == 1) {
2478 if (pfrag->offset + sz <= pfrag->size)
2480 put_page(pfrag->page);
2484 if (SKB_FRAG_PAGE_ORDER &&
2485 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2486 /* Avoid direct reclaim but allow kswapd to wake */
2487 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2488 __GFP_COMP | __GFP_NOWARN |
2490 SKB_FRAG_PAGE_ORDER);
2491 if (likely(pfrag->page)) {
2492 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2496 pfrag->page = alloc_page(gfp);
2497 if (likely(pfrag->page)) {
2498 pfrag->size = PAGE_SIZE;
2503 EXPORT_SYMBOL(skb_page_frag_refill);
2505 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2507 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2510 sk_enter_memory_pressure(sk);
2511 sk_stream_moderate_sndbuf(sk);
2514 EXPORT_SYMBOL(sk_page_frag_refill);
2516 static void __lock_sock(struct sock *sk)
2517 __releases(&sk->sk_lock.slock)
2518 __acquires(&sk->sk_lock.slock)
2523 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2524 TASK_UNINTERRUPTIBLE);
2525 spin_unlock_bh(&sk->sk_lock.slock);
2527 spin_lock_bh(&sk->sk_lock.slock);
2528 if (!sock_owned_by_user(sk))
2531 finish_wait(&sk->sk_lock.wq, &wait);
2534 void __release_sock(struct sock *sk)
2535 __releases(&sk->sk_lock.slock)
2536 __acquires(&sk->sk_lock.slock)
2538 struct sk_buff *skb, *next;
2540 while ((skb = sk->sk_backlog.head) != NULL) {
2541 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2543 spin_unlock_bh(&sk->sk_lock.slock);
2548 WARN_ON_ONCE(skb_dst_is_noref(skb));
2549 skb_mark_not_on_list(skb);
2550 sk_backlog_rcv(sk, skb);
2555 } while (skb != NULL);
2557 spin_lock_bh(&sk->sk_lock.slock);
2561 * Doing the zeroing here guarantee we can not loop forever
2562 * while a wild producer attempts to flood us.
2564 sk->sk_backlog.len = 0;
2567 void __sk_flush_backlog(struct sock *sk)
2569 spin_lock_bh(&sk->sk_lock.slock);
2571 spin_unlock_bh(&sk->sk_lock.slock);
2575 * sk_wait_data - wait for data to arrive at sk_receive_queue
2576 * @sk: sock to wait on
2577 * @timeo: for how long
2578 * @skb: last skb seen on sk_receive_queue
2580 * Now socket state including sk->sk_err is changed only under lock,
2581 * hence we may omit checks after joining wait queue.
2582 * We check receive queue before schedule() only as optimization;
2583 * it is very likely that release_sock() added new data.
2585 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2587 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2590 add_wait_queue(sk_sleep(sk), &wait);
2591 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2592 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2593 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2594 remove_wait_queue(sk_sleep(sk), &wait);
2597 EXPORT_SYMBOL(sk_wait_data);
2600 * __sk_mem_raise_allocated - increase memory_allocated
2602 * @size: memory size to allocate
2603 * @amt: pages to allocate
2604 * @kind: allocation type
2606 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2608 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2610 struct proto *prot = sk->sk_prot;
2611 long allocated = sk_memory_allocated_add(sk, amt);
2612 bool charged = true;
2614 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2615 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2616 goto suppress_allocation;
2619 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2620 sk_leave_memory_pressure(sk);
2624 /* Under pressure. */
2625 if (allocated > sk_prot_mem_limits(sk, 1))
2626 sk_enter_memory_pressure(sk);
2628 /* Over hard limit. */
2629 if (allocated > sk_prot_mem_limits(sk, 2))
2630 goto suppress_allocation;
2632 /* guarantee minimum buffer size under pressure */
2633 if (kind == SK_MEM_RECV) {
2634 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2637 } else { /* SK_MEM_SEND */
2638 int wmem0 = sk_get_wmem0(sk, prot);
2640 if (sk->sk_type == SOCK_STREAM) {
2641 if (sk->sk_wmem_queued < wmem0)
2643 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2648 if (sk_has_memory_pressure(sk)) {
2651 if (!sk_under_memory_pressure(sk))
2653 alloc = sk_sockets_allocated_read_positive(sk);
2654 if (sk_prot_mem_limits(sk, 2) > alloc *
2655 sk_mem_pages(sk->sk_wmem_queued +
2656 atomic_read(&sk->sk_rmem_alloc) +
2657 sk->sk_forward_alloc))
2661 suppress_allocation:
2663 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2664 sk_stream_moderate_sndbuf(sk);
2666 /* Fail only if socket is _under_ its sndbuf.
2667 * In this case we cannot block, so that we have to fail.
2669 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2673 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2674 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2676 sk_memory_allocated_sub(sk, amt);
2678 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2679 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2683 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2686 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2688 * @size: memory size to allocate
2689 * @kind: allocation type
2691 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2692 * rmem allocation. This function assumes that protocols which have
2693 * memory_pressure use sk_wmem_queued as write buffer accounting.
2695 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2697 int ret, amt = sk_mem_pages(size);
2699 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2700 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2702 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2705 EXPORT_SYMBOL(__sk_mem_schedule);
2708 * __sk_mem_reduce_allocated - reclaim memory_allocated
2710 * @amount: number of quanta
2712 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2714 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2716 sk_memory_allocated_sub(sk, amount);
2718 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2719 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2721 if (sk_under_memory_pressure(sk) &&
2722 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2723 sk_leave_memory_pressure(sk);
2725 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2728 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2730 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2732 void __sk_mem_reclaim(struct sock *sk, int amount)
2734 amount >>= SK_MEM_QUANTUM_SHIFT;
2735 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2736 __sk_mem_reduce_allocated(sk, amount);
2738 EXPORT_SYMBOL(__sk_mem_reclaim);
2740 int sk_set_peek_off(struct sock *sk, int val)
2742 sk->sk_peek_off = val;
2745 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2748 * Set of default routines for initialising struct proto_ops when
2749 * the protocol does not support a particular function. In certain
2750 * cases where it makes no sense for a protocol to have a "do nothing"
2751 * function, some default processing is provided.
2754 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2758 EXPORT_SYMBOL(sock_no_bind);
2760 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2765 EXPORT_SYMBOL(sock_no_connect);
2767 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2771 EXPORT_SYMBOL(sock_no_socketpair);
2773 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2778 EXPORT_SYMBOL(sock_no_accept);
2780 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2785 EXPORT_SYMBOL(sock_no_getname);
2787 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2791 EXPORT_SYMBOL(sock_no_ioctl);
2793 int sock_no_listen(struct socket *sock, int backlog)
2797 EXPORT_SYMBOL(sock_no_listen);
2799 int sock_no_shutdown(struct socket *sock, int how)
2803 EXPORT_SYMBOL(sock_no_shutdown);
2805 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2806 char __user *optval, unsigned int optlen)
2810 EXPORT_SYMBOL(sock_no_setsockopt);
2812 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2813 char __user *optval, int __user *optlen)
2817 EXPORT_SYMBOL(sock_no_getsockopt);
2819 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2823 EXPORT_SYMBOL(sock_no_sendmsg);
2825 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2829 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2831 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2836 EXPORT_SYMBOL(sock_no_recvmsg);
2838 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2840 /* Mirror missing mmap method error code */
2843 EXPORT_SYMBOL(sock_no_mmap);
2845 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2848 struct msghdr msg = {.msg_flags = flags};
2850 char *kaddr = kmap(page);
2851 iov.iov_base = kaddr + offset;
2853 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2857 EXPORT_SYMBOL(sock_no_sendpage);
2859 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2860 int offset, size_t size, int flags)
2863 struct msghdr msg = {.msg_flags = flags};
2865 char *kaddr = kmap(page);
2867 iov.iov_base = kaddr + offset;
2869 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2873 EXPORT_SYMBOL(sock_no_sendpage_locked);
2876 * Default Socket Callbacks
2879 static void sock_def_wakeup(struct sock *sk)
2881 struct socket_wq *wq;
2884 wq = rcu_dereference(sk->sk_wq);
2885 if (skwq_has_sleeper(wq))
2886 wake_up_interruptible_all(&wq->wait);
2890 static void sock_def_error_report(struct sock *sk)
2892 struct socket_wq *wq;
2895 wq = rcu_dereference(sk->sk_wq);
2896 if (skwq_has_sleeper(wq))
2897 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2898 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2902 void sock_def_readable(struct sock *sk)
2904 struct socket_wq *wq;
2907 wq = rcu_dereference(sk->sk_wq);
2908 if (skwq_has_sleeper(wq))
2909 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2910 EPOLLRDNORM | EPOLLRDBAND);
2911 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2915 static void sock_def_write_space(struct sock *sk)
2917 struct socket_wq *wq;
2921 /* Do not wake up a writer until he can make "significant"
2924 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2925 wq = rcu_dereference(sk->sk_wq);
2926 if (skwq_has_sleeper(wq))
2927 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2928 EPOLLWRNORM | EPOLLWRBAND);
2930 /* Should agree with poll, otherwise some programs break */
2931 if (sock_writeable(sk))
2932 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2938 static void sock_def_destruct(struct sock *sk)
2942 void sk_send_sigurg(struct sock *sk)
2944 if (sk->sk_socket && sk->sk_socket->file)
2945 if (send_sigurg(&sk->sk_socket->file->f_owner))
2946 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2948 EXPORT_SYMBOL(sk_send_sigurg);
2950 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2951 unsigned long expires)
2953 if (!mod_timer(timer, expires))
2956 EXPORT_SYMBOL(sk_reset_timer);
2958 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2960 if (del_timer(timer))
2963 EXPORT_SYMBOL(sk_stop_timer);
2965 void sock_init_data(struct socket *sock, struct sock *sk)
2968 sk->sk_send_head = NULL;
2970 timer_setup(&sk->sk_timer, NULL, 0);
2972 sk->sk_allocation = GFP_KERNEL;
2973 sk->sk_rcvbuf = sysctl_rmem_default;
2974 sk->sk_sndbuf = sysctl_wmem_default;
2975 sk->sk_state = TCP_CLOSE;
2976 sk_set_socket(sk, sock);
2978 sock_set_flag(sk, SOCK_ZAPPED);
2981 sk->sk_type = sock->type;
2982 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2984 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2986 RCU_INIT_POINTER(sk->sk_wq, NULL);
2987 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2990 rwlock_init(&sk->sk_callback_lock);
2991 if (sk->sk_kern_sock)
2992 lockdep_set_class_and_name(
2993 &sk->sk_callback_lock,
2994 af_kern_callback_keys + sk->sk_family,
2995 af_family_kern_clock_key_strings[sk->sk_family]);
2997 lockdep_set_class_and_name(
2998 &sk->sk_callback_lock,
2999 af_callback_keys + sk->sk_family,
3000 af_family_clock_key_strings[sk->sk_family]);
3002 sk->sk_state_change = sock_def_wakeup;
3003 sk->sk_data_ready = sock_def_readable;
3004 sk->sk_write_space = sock_def_write_space;
3005 sk->sk_error_report = sock_def_error_report;
3006 sk->sk_destruct = sock_def_destruct;
3008 sk->sk_frag.page = NULL;
3009 sk->sk_frag.offset = 0;
3010 sk->sk_peek_off = -1;
3012 sk->sk_peer_pid = NULL;
3013 sk->sk_peer_cred = NULL;
3014 sk->sk_write_pending = 0;
3015 sk->sk_rcvlowat = 1;
3016 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3017 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3019 sk->sk_stamp = SK_DEFAULT_STAMP;
3020 #if BITS_PER_LONG==32
3021 seqlock_init(&sk->sk_stamp_seq);
3023 atomic_set(&sk->sk_zckey, 0);
3025 #ifdef CONFIG_NET_RX_BUSY_POLL
3027 sk->sk_ll_usec = sysctl_net_busy_read;
3030 sk->sk_max_pacing_rate = ~0UL;
3031 sk->sk_pacing_rate = ~0UL;
3032 WRITE_ONCE(sk->sk_pacing_shift, 10);
3033 sk->sk_incoming_cpu = -1;
3035 sk_rx_queue_clear(sk);
3037 * Before updating sk_refcnt, we must commit prior changes to memory
3038 * (Documentation/RCU/rculist_nulls.txt for details)
3041 refcount_set(&sk->sk_refcnt, 1);
3042 atomic_set(&sk->sk_drops, 0);
3044 EXPORT_SYMBOL(sock_init_data);
3046 void lock_sock_nested(struct sock *sk, int subclass)
3049 spin_lock_bh(&sk->sk_lock.slock);
3050 if (sk->sk_lock.owned)
3052 sk->sk_lock.owned = 1;
3053 spin_unlock(&sk->sk_lock.slock);
3055 * The sk_lock has mutex_lock() semantics here:
3057 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3060 EXPORT_SYMBOL(lock_sock_nested);
3062 void release_sock(struct sock *sk)
3064 spin_lock_bh(&sk->sk_lock.slock);
3065 if (sk->sk_backlog.tail)
3068 /* Warning : release_cb() might need to release sk ownership,
3069 * ie call sock_release_ownership(sk) before us.
3071 if (sk->sk_prot->release_cb)
3072 sk->sk_prot->release_cb(sk);
3074 sock_release_ownership(sk);
3075 if (waitqueue_active(&sk->sk_lock.wq))
3076 wake_up(&sk->sk_lock.wq);
3077 spin_unlock_bh(&sk->sk_lock.slock);
3079 EXPORT_SYMBOL(release_sock);
3082 * lock_sock_fast - fast version of lock_sock
3085 * This version should be used for very small section, where process wont block
3086 * return false if fast path is taken:
3088 * sk_lock.slock locked, owned = 0, BH disabled
3090 * return true if slow path is taken:
3092 * sk_lock.slock unlocked, owned = 1, BH enabled
3094 bool lock_sock_fast(struct sock *sk)
3097 spin_lock_bh(&sk->sk_lock.slock);
3099 if (!sk->sk_lock.owned)
3101 * Note : We must disable BH
3106 sk->sk_lock.owned = 1;
3107 spin_unlock(&sk->sk_lock.slock);
3109 * The sk_lock has mutex_lock() semantics here:
3111 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3115 EXPORT_SYMBOL(lock_sock_fast);
3117 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3118 bool timeval, bool time32)
3120 struct sock *sk = sock->sk;
3121 struct timespec64 ts;
3123 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3124 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3125 if (ts.tv_sec == -1)
3127 if (ts.tv_sec == 0) {
3128 ktime_t kt = ktime_get_real();
3129 sock_write_timestamp(sk, kt);
3130 ts = ktime_to_timespec64(kt);
3136 #ifdef CONFIG_COMPAT_32BIT_TIME
3138 return put_old_timespec32(&ts, userstamp);
3140 #ifdef CONFIG_SPARC64
3141 /* beware of padding in sparc64 timeval */
3142 if (timeval && !in_compat_syscall()) {
3143 struct __kernel_old_timeval __user tv = {
3144 .tv_sec = ts.tv_sec,
3145 .tv_usec = ts.tv_nsec,
3147 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3152 return put_timespec64(&ts, userstamp);
3154 EXPORT_SYMBOL(sock_gettstamp);
3156 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3158 if (!sock_flag(sk, flag)) {
3159 unsigned long previous_flags = sk->sk_flags;
3161 sock_set_flag(sk, flag);
3163 * we just set one of the two flags which require net
3164 * time stamping, but time stamping might have been on
3165 * already because of the other one
3167 if (sock_needs_netstamp(sk) &&
3168 !(previous_flags & SK_FLAGS_TIMESTAMP))
3169 net_enable_timestamp();
3173 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3174 int level, int type)
3176 struct sock_exterr_skb *serr;
3177 struct sk_buff *skb;
3181 skb = sock_dequeue_err_skb(sk);
3187 msg->msg_flags |= MSG_TRUNC;
3190 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3194 sock_recv_timestamp(msg, sk, skb);
3196 serr = SKB_EXT_ERR(skb);
3197 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3199 msg->msg_flags |= MSG_ERRQUEUE;
3207 EXPORT_SYMBOL(sock_recv_errqueue);
3210 * Get a socket option on an socket.
3212 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3213 * asynchronous errors should be reported by getsockopt. We assume
3214 * this means if you specify SO_ERROR (otherwise whats the point of it).
3216 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3217 char __user *optval, int __user *optlen)
3219 struct sock *sk = sock->sk;
3221 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3223 EXPORT_SYMBOL(sock_common_getsockopt);
3225 #ifdef CONFIG_COMPAT
3226 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3227 char __user *optval, int __user *optlen)
3229 struct sock *sk = sock->sk;
3231 if (sk->sk_prot->compat_getsockopt != NULL)
3232 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3234 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3236 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3239 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3242 struct sock *sk = sock->sk;
3246 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3247 flags & ~MSG_DONTWAIT, &addr_len);
3249 msg->msg_namelen = addr_len;
3252 EXPORT_SYMBOL(sock_common_recvmsg);
3255 * Set socket options on an inet socket.
3257 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3258 char __user *optval, unsigned int optlen)
3260 struct sock *sk = sock->sk;
3262 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3264 EXPORT_SYMBOL(sock_common_setsockopt);
3266 #ifdef CONFIG_COMPAT
3267 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3268 char __user *optval, unsigned int optlen)
3270 struct sock *sk = sock->sk;
3272 if (sk->sk_prot->compat_setsockopt != NULL)
3273 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3275 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3277 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3280 void sk_common_release(struct sock *sk)
3282 if (sk->sk_prot->destroy)
3283 sk->sk_prot->destroy(sk);
3286 * Observation: when sock_common_release is called, processes have
3287 * no access to socket. But net still has.
3288 * Step one, detach it from networking:
3290 * A. Remove from hash tables.
3293 sk->sk_prot->unhash(sk);
3296 * In this point socket cannot receive new packets, but it is possible
3297 * that some packets are in flight because some CPU runs receiver and
3298 * did hash table lookup before we unhashed socket. They will achieve
3299 * receive queue and will be purged by socket destructor.
3301 * Also we still have packets pending on receive queue and probably,
3302 * our own packets waiting in device queues. sock_destroy will drain
3303 * receive queue, but transmitted packets will delay socket destruction
3304 * until the last reference will be released.
3309 xfrm_sk_free_policy(sk);
3311 sk_refcnt_debug_release(sk);
3315 EXPORT_SYMBOL(sk_common_release);
3317 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3319 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3321 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3322 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3323 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3324 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3325 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3326 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3327 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3328 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3329 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3332 #ifdef CONFIG_PROC_FS
3333 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3335 int val[PROTO_INUSE_NR];
3338 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3340 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3342 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3344 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3346 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3348 int cpu, idx = prot->inuse_idx;
3351 for_each_possible_cpu(cpu)
3352 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3354 return res >= 0 ? res : 0;
3356 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3358 static void sock_inuse_add(struct net *net, int val)
3360 this_cpu_add(*net->core.sock_inuse, val);
3363 int sock_inuse_get(struct net *net)
3367 for_each_possible_cpu(cpu)
3368 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3373 EXPORT_SYMBOL_GPL(sock_inuse_get);
3375 static int __net_init sock_inuse_init_net(struct net *net)
3377 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3378 if (net->core.prot_inuse == NULL)
3381 net->core.sock_inuse = alloc_percpu(int);
3382 if (net->core.sock_inuse == NULL)
3388 free_percpu(net->core.prot_inuse);
3392 static void __net_exit sock_inuse_exit_net(struct net *net)
3394 free_percpu(net->core.prot_inuse);
3395 free_percpu(net->core.sock_inuse);
3398 static struct pernet_operations net_inuse_ops = {
3399 .init = sock_inuse_init_net,
3400 .exit = sock_inuse_exit_net,
3403 static __init int net_inuse_init(void)
3405 if (register_pernet_subsys(&net_inuse_ops))
3406 panic("Cannot initialize net inuse counters");
3411 core_initcall(net_inuse_init);
3413 static int assign_proto_idx(struct proto *prot)
3415 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3417 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3418 pr_err("PROTO_INUSE_NR exhausted\n");
3422 set_bit(prot->inuse_idx, proto_inuse_idx);
3426 static void release_proto_idx(struct proto *prot)
3428 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3429 clear_bit(prot->inuse_idx, proto_inuse_idx);
3432 static inline int assign_proto_idx(struct proto *prot)
3437 static inline void release_proto_idx(struct proto *prot)
3441 static void sock_inuse_add(struct net *net, int val)
3446 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3450 kfree(rsk_prot->slab_name);
3451 rsk_prot->slab_name = NULL;
3452 kmem_cache_destroy(rsk_prot->slab);
3453 rsk_prot->slab = NULL;
3456 static int req_prot_init(const struct proto *prot)
3458 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3463 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3465 if (!rsk_prot->slab_name)
3468 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3469 rsk_prot->obj_size, 0,
3470 SLAB_ACCOUNT | prot->slab_flags,
3473 if (!rsk_prot->slab) {
3474 pr_crit("%s: Can't create request sock SLAB cache!\n",
3481 int proto_register(struct proto *prot, int alloc_slab)
3486 prot->slab = kmem_cache_create_usercopy(prot->name,
3488 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3490 prot->useroffset, prot->usersize,
3493 if (prot->slab == NULL) {
3494 pr_crit("%s: Can't create sock SLAB cache!\n",
3499 if (req_prot_init(prot))
3500 goto out_free_request_sock_slab;
3502 if (prot->twsk_prot != NULL) {
3503 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3505 if (prot->twsk_prot->twsk_slab_name == NULL)
3506 goto out_free_request_sock_slab;
3508 prot->twsk_prot->twsk_slab =
3509 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3510 prot->twsk_prot->twsk_obj_size,
3515 if (prot->twsk_prot->twsk_slab == NULL)
3516 goto out_free_timewait_sock_slab_name;
3520 mutex_lock(&proto_list_mutex);
3521 ret = assign_proto_idx(prot);
3523 mutex_unlock(&proto_list_mutex);
3524 goto out_free_timewait_sock_slab_name;
3526 list_add(&prot->node, &proto_list);
3527 mutex_unlock(&proto_list_mutex);
3530 out_free_timewait_sock_slab_name:
3531 if (alloc_slab && prot->twsk_prot)
3532 kfree(prot->twsk_prot->twsk_slab_name);
3533 out_free_request_sock_slab:
3535 req_prot_cleanup(prot->rsk_prot);
3537 kmem_cache_destroy(prot->slab);
3543 EXPORT_SYMBOL(proto_register);
3545 void proto_unregister(struct proto *prot)
3547 mutex_lock(&proto_list_mutex);
3548 release_proto_idx(prot);
3549 list_del(&prot->node);
3550 mutex_unlock(&proto_list_mutex);
3552 kmem_cache_destroy(prot->slab);
3555 req_prot_cleanup(prot->rsk_prot);
3557 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3558 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3559 kfree(prot->twsk_prot->twsk_slab_name);
3560 prot->twsk_prot->twsk_slab = NULL;
3563 EXPORT_SYMBOL(proto_unregister);
3565 int sock_load_diag_module(int family, int protocol)
3568 if (!sock_is_registered(family))
3571 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3572 NETLINK_SOCK_DIAG, family);
3576 if (family == AF_INET &&
3577 protocol != IPPROTO_RAW &&
3578 !rcu_access_pointer(inet_protos[protocol]))
3582 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3583 NETLINK_SOCK_DIAG, family, protocol);
3585 EXPORT_SYMBOL(sock_load_diag_module);
3587 #ifdef CONFIG_PROC_FS
3588 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3589 __acquires(proto_list_mutex)
3591 mutex_lock(&proto_list_mutex);
3592 return seq_list_start_head(&proto_list, *pos);
3595 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3597 return seq_list_next(v, &proto_list, pos);
3600 static void proto_seq_stop(struct seq_file *seq, void *v)
3601 __releases(proto_list_mutex)
3603 mutex_unlock(&proto_list_mutex);
3606 static char proto_method_implemented(const void *method)
3608 return method == NULL ? 'n' : 'y';
3610 static long sock_prot_memory_allocated(struct proto *proto)
3612 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3615 static const char *sock_prot_memory_pressure(struct proto *proto)
3617 return proto->memory_pressure != NULL ?
3618 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3621 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3624 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3625 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3628 sock_prot_inuse_get(seq_file_net(seq), proto),
3629 sock_prot_memory_allocated(proto),
3630 sock_prot_memory_pressure(proto),
3632 proto->slab == NULL ? "no" : "yes",
3633 module_name(proto->owner),
3634 proto_method_implemented(proto->close),
3635 proto_method_implemented(proto->connect),
3636 proto_method_implemented(proto->disconnect),
3637 proto_method_implemented(proto->accept),
3638 proto_method_implemented(proto->ioctl),
3639 proto_method_implemented(proto->init),
3640 proto_method_implemented(proto->destroy),
3641 proto_method_implemented(proto->shutdown),
3642 proto_method_implemented(proto->setsockopt),
3643 proto_method_implemented(proto->getsockopt),
3644 proto_method_implemented(proto->sendmsg),
3645 proto_method_implemented(proto->recvmsg),
3646 proto_method_implemented(proto->sendpage),
3647 proto_method_implemented(proto->bind),
3648 proto_method_implemented(proto->backlog_rcv),
3649 proto_method_implemented(proto->hash),
3650 proto_method_implemented(proto->unhash),
3651 proto_method_implemented(proto->get_port),
3652 proto_method_implemented(proto->enter_memory_pressure));
3655 static int proto_seq_show(struct seq_file *seq, void *v)
3657 if (v == &proto_list)
3658 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3667 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3669 proto_seq_printf(seq, list_entry(v, struct proto, node));
3673 static const struct seq_operations proto_seq_ops = {
3674 .start = proto_seq_start,
3675 .next = proto_seq_next,
3676 .stop = proto_seq_stop,
3677 .show = proto_seq_show,
3680 static __net_init int proto_init_net(struct net *net)
3682 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3683 sizeof(struct seq_net_private)))
3689 static __net_exit void proto_exit_net(struct net *net)
3691 remove_proc_entry("protocols", net->proc_net);
3695 static __net_initdata struct pernet_operations proto_net_ops = {
3696 .init = proto_init_net,
3697 .exit = proto_exit_net,
3700 static int __init proto_init(void)
3702 return register_pernet_subsys(&proto_net_ops);
3705 subsys_initcall(proto_init);
3707 #endif /* PROC_FS */
3709 #ifdef CONFIG_NET_RX_BUSY_POLL
3710 bool sk_busy_loop_end(void *p, unsigned long start_time)
3712 struct sock *sk = p;
3714 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3715 sk_busy_loop_timeout(sk, start_time);
3717 EXPORT_SYMBOL(sk_busy_loop_end);
3718 #endif /* CONFIG_NET_RX_BUSY_POLL */
3720 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3722 if (!sk->sk_prot->bind_add)
3724 return sk->sk_prot->bind_add(sk, addr, addr_len);
3726 EXPORT_SYMBOL(sock_bind_add);