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>
116 #include <linux/compat.h>
118 #include <linux/uaccess.h>
120 #include <linux/netdevice.h>
121 #include <net/protocol.h>
122 #include <linux/skbuff.h>
123 #include <net/net_namespace.h>
124 #include <net/request_sock.h>
125 #include <net/sock.h>
126 #include <linux/net_tstamp.h>
127 #include <net/xfrm.h>
128 #include <linux/ipsec.h>
129 #include <net/cls_cgroup.h>
130 #include <net/netprio_cgroup.h>
131 #include <linux/sock_diag.h>
133 #include <linux/filter.h>
134 #include <net/sock_reuseport.h>
135 #include <net/bpf_sk_storage.h>
137 #include <trace/events/sock.h>
140 #include <net/busy_poll.h>
142 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
145 static void sock_inuse_add(struct net *net, int val);
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
157 bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
163 EXPORT_SYMBOL(sk_ns_capable);
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
174 bool sk_capable(const struct sock *sk, int cap)
176 return sk_ns_capable(sk, &init_user_ns, cap);
178 EXPORT_SYMBOL(sk_capable);
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
189 bool sk_net_capable(const struct sock *sk, int cap)
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 EXPORT_SYMBOL(sk_net_capable);
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
200 static struct lock_class_key af_family_keys[AF_MAX];
201 static struct lock_class_key af_family_kern_keys[AF_MAX];
202 static struct lock_class_key af_family_slock_keys[AF_MAX];
203 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
211 #define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
229 static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
232 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
235 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
239 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
242 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
245 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
248 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 _sock_locks("rlock-")
251 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252 _sock_locks("wlock-")
254 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255 _sock_locks("elock-")
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
262 static struct lock_class_key af_callback_keys[AF_MAX];
263 static struct lock_class_key af_rlock_keys[AF_MAX];
264 static struct lock_class_key af_wlock_keys[AF_MAX];
265 static struct lock_class_key af_elock_keys[AF_MAX];
266 static struct lock_class_key af_kern_callback_keys[AF_MAX];
268 /* Run time adjustable parameters. */
269 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270 EXPORT_SYMBOL(sysctl_wmem_max);
271 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272 EXPORT_SYMBOL(sysctl_rmem_max);
273 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
276 /* Maximal space eaten by iovec or ancillary data plus some space */
277 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278 EXPORT_SYMBOL(sysctl_optmem_max);
280 int sysctl_tstamp_allow_data __read_mostly = 1;
282 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283 EXPORT_SYMBOL_GPL(memalloc_socks_key);
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
293 void sk_set_memalloc(struct sock *sk)
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_branch_inc(&memalloc_socks_key);
299 EXPORT_SYMBOL_GPL(sk_set_memalloc);
301 void sk_clear_memalloc(struct sock *sk)
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_branch_dec(&memalloc_socks_key);
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
321 unsigned int noreclaim_flag;
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
326 noreclaim_flag = memalloc_noreclaim_save();
327 ret = sk->sk_backlog_rcv(sk, skb);
328 memalloc_noreclaim_restore(noreclaim_flag);
332 EXPORT_SYMBOL(__sk_backlog_rcv);
334 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
336 struct __kernel_sock_timeval tv;
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 return sizeof(old_tv);
360 *(struct __kernel_sock_timeval *)optval = tv;
364 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
367 struct __kernel_sock_timeval tv;
369 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
370 struct old_timeval32 tv32;
372 if (optlen < sizeof(tv32))
375 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
377 tv.tv_sec = tv32.tv_sec;
378 tv.tv_usec = tv32.tv_usec;
379 } else if (old_timeval) {
380 struct __kernel_old_timeval old_tv;
382 if (optlen < sizeof(old_tv))
384 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
386 tv.tv_sec = old_tv.tv_sec;
387 tv.tv_usec = old_tv.tv_usec;
389 if (optlen < sizeof(tv))
391 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
394 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
398 static int warned __read_mostly;
401 if (warned < 10 && net_ratelimit()) {
403 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 __func__, current->comm, task_pid_nr(current));
408 *timeo_p = MAX_SCHEDULE_TIMEOUT;
409 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
412 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
416 static void sock_warn_obsolete_bsdism(const char *name)
419 static char warncomm[TASK_COMM_LEN];
420 if (strcmp(warncomm, current->comm) && warned < 5) {
421 strcpy(warncomm, current->comm);
422 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
428 static bool sock_needs_netstamp(const struct sock *sk)
430 switch (sk->sk_family) {
439 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
441 if (sk->sk_flags & flags) {
442 sk->sk_flags &= ~flags;
443 if (sock_needs_netstamp(sk) &&
444 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
445 net_disable_timestamp();
450 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
453 struct sk_buff_head *list = &sk->sk_receive_queue;
455 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
456 atomic_inc(&sk->sk_drops);
457 trace_sock_rcvqueue_full(sk, skb);
461 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
462 atomic_inc(&sk->sk_drops);
467 skb_set_owner_r(skb, sk);
469 /* we escape from rcu protected region, make sure we dont leak
474 spin_lock_irqsave(&list->lock, flags);
475 sock_skb_set_dropcount(sk, skb);
476 __skb_queue_tail(list, skb);
477 spin_unlock_irqrestore(&list->lock, flags);
479 if (!sock_flag(sk, SOCK_DEAD))
480 sk->sk_data_ready(sk);
483 EXPORT_SYMBOL(__sock_queue_rcv_skb);
485 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
489 err = sk_filter(sk, skb);
493 return __sock_queue_rcv_skb(sk, skb);
495 EXPORT_SYMBOL(sock_queue_rcv_skb);
497 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
498 const int nested, unsigned int trim_cap, bool refcounted)
500 int rc = NET_RX_SUCCESS;
502 if (sk_filter_trim_cap(sk, skb, trim_cap))
503 goto discard_and_relse;
507 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
508 atomic_inc(&sk->sk_drops);
509 goto discard_and_relse;
512 bh_lock_sock_nested(sk);
515 if (!sock_owned_by_user(sk)) {
517 * trylock + unlock semantics:
519 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
521 rc = sk_backlog_rcv(sk, skb);
523 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
524 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
526 atomic_inc(&sk->sk_drops);
527 goto discard_and_relse;
539 EXPORT_SYMBOL(__sk_receive_skb);
541 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
543 struct dst_entry *dst = __sk_dst_get(sk);
545 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
546 sk_tx_queue_clear(sk);
547 sk->sk_dst_pending_confirm = 0;
548 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
555 EXPORT_SYMBOL(__sk_dst_check);
557 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
559 struct dst_entry *dst = sk_dst_get(sk);
561 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
569 EXPORT_SYMBOL(sk_dst_check);
571 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
573 int ret = -ENOPROTOOPT;
574 #ifdef CONFIG_NETDEVICES
575 struct net *net = sock_net(sk);
579 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
586 sk->sk_bound_dev_if = ifindex;
587 if (sk->sk_prot->rehash)
588 sk->sk_prot->rehash(sk);
599 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
605 ret = sock_bindtoindex_locked(sk, ifindex);
611 EXPORT_SYMBOL(sock_bindtoindex);
613 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
615 int ret = -ENOPROTOOPT;
616 #ifdef CONFIG_NETDEVICES
617 struct net *net = sock_net(sk);
618 char devname[IFNAMSIZ];
625 /* Bind this socket to a particular device like "eth0",
626 * as specified in the passed interface name. If the
627 * name is "" or the option length is zero the socket
630 if (optlen > IFNAMSIZ - 1)
631 optlen = IFNAMSIZ - 1;
632 memset(devname, 0, sizeof(devname));
635 if (copy_from_sockptr(devname, optval, optlen))
639 if (devname[0] != '\0') {
640 struct net_device *dev;
643 dev = dev_get_by_name_rcu(net, devname);
645 index = dev->ifindex;
652 return sock_bindtoindex(sk, index, true);
659 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
660 int __user *optlen, int len)
662 int ret = -ENOPROTOOPT;
663 #ifdef CONFIG_NETDEVICES
664 struct net *net = sock_net(sk);
665 char devname[IFNAMSIZ];
667 if (sk->sk_bound_dev_if == 0) {
676 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
680 len = strlen(devname) + 1;
683 if (copy_to_user(optval, devname, len))
688 if (put_user(len, optlen))
699 bool sk_mc_loop(struct sock *sk)
701 if (dev_recursion_level())
705 switch (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);
718 void sock_set_reuseaddr(struct sock *sk)
721 sk->sk_reuse = SK_CAN_REUSE;
724 EXPORT_SYMBOL(sock_set_reuseaddr);
726 void sock_set_reuseport(struct sock *sk)
729 sk->sk_reuseport = true;
732 EXPORT_SYMBOL(sock_set_reuseport);
734 void sock_no_linger(struct sock *sk)
737 sk->sk_lingertime = 0;
738 sock_set_flag(sk, SOCK_LINGER);
741 EXPORT_SYMBOL(sock_no_linger);
743 void sock_set_priority(struct sock *sk, u32 priority)
746 sk->sk_priority = priority;
749 EXPORT_SYMBOL(sock_set_priority);
751 void sock_set_sndtimeo(struct sock *sk, s64 secs)
754 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
755 sk->sk_sndtimeo = secs * HZ;
757 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
760 EXPORT_SYMBOL(sock_set_sndtimeo);
762 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
765 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
766 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
767 sock_set_flag(sk, SOCK_RCVTSTAMP);
768 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
770 sock_reset_flag(sk, SOCK_RCVTSTAMP);
771 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
772 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
776 void sock_enable_timestamps(struct sock *sk)
779 __sock_set_timestamps(sk, true, false, true);
782 EXPORT_SYMBOL(sock_enable_timestamps);
784 void sock_set_keepalive(struct sock *sk)
787 if (sk->sk_prot->keepalive)
788 sk->sk_prot->keepalive(sk, true);
789 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
792 EXPORT_SYMBOL(sock_set_keepalive);
794 static void __sock_set_rcvbuf(struct sock *sk, int val)
796 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
797 * as a negative value.
799 val = min_t(int, val, INT_MAX / 2);
800 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
802 /* We double it on the way in to account for "struct sk_buff" etc.
803 * overhead. Applications assume that the SO_RCVBUF setting they make
804 * will allow that much actual data to be received on that socket.
806 * Applications are unaware that "struct sk_buff" and other overheads
807 * allocate from the receive buffer during socket buffer allocation.
809 * And after considering the possible alternatives, returning the value
810 * we actually used in getsockopt is the most desirable behavior.
812 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
815 void sock_set_rcvbuf(struct sock *sk, int val)
818 __sock_set_rcvbuf(sk, val);
821 EXPORT_SYMBOL(sock_set_rcvbuf);
824 * This is meant for all protocols to use and covers goings on
825 * at the socket level. Everything here is generic.
828 int sock_setsockopt(struct socket *sock, int level, int optname,
829 sockptr_t optval, unsigned int optlen)
831 struct sock_txtime sk_txtime;
832 struct sock *sk = sock->sk;
839 * Options without arguments
842 if (optname == SO_BINDTODEVICE)
843 return sock_setbindtodevice(sk, optval, optlen);
845 if (optlen < sizeof(int))
848 if (copy_from_sockptr(&val, optval, sizeof(val)))
851 valbool = val ? 1 : 0;
857 if (val && !capable(CAP_NET_ADMIN))
860 sock_valbool_flag(sk, SOCK_DBG, valbool);
863 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
866 sk->sk_reuseport = valbool;
875 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
879 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
882 /* Don't error on this BSD doesn't and if you think
883 * about it this is right. Otherwise apps have to
884 * play 'guess the biggest size' games. RCVBUF/SNDBUF
885 * are treated in BSD as hints
887 val = min_t(u32, val, sysctl_wmem_max);
889 /* Ensure val * 2 fits into an int, to prevent max_t()
890 * from treating it as a negative value.
892 val = min_t(int, val, INT_MAX / 2);
893 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
894 WRITE_ONCE(sk->sk_sndbuf,
895 max_t(int, val * 2, SOCK_MIN_SNDBUF));
896 /* Wake up sending tasks if we upped the value. */
897 sk->sk_write_space(sk);
901 if (!capable(CAP_NET_ADMIN)) {
906 /* No negative values (to prevent underflow, as val will be
914 /* Don't error on this BSD doesn't and if you think
915 * about it this is right. Otherwise apps have to
916 * play 'guess the biggest size' games. RCVBUF/SNDBUF
917 * are treated in BSD as hints
919 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
923 if (!capable(CAP_NET_ADMIN)) {
928 /* No negative values (to prevent underflow, as val will be
931 __sock_set_rcvbuf(sk, max(val, 0));
935 if (sk->sk_prot->keepalive)
936 sk->sk_prot->keepalive(sk, valbool);
937 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
941 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
945 sk->sk_no_check_tx = valbool;
949 if ((val >= 0 && val <= 6) ||
950 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
951 sk->sk_priority = val;
957 if (optlen < sizeof(ling)) {
958 ret = -EINVAL; /* 1003.1g */
961 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
966 sock_reset_flag(sk, SOCK_LINGER);
968 #if (BITS_PER_LONG == 32)
969 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
970 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
973 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
974 sock_set_flag(sk, SOCK_LINGER);
979 sock_warn_obsolete_bsdism("setsockopt");
984 set_bit(SOCK_PASSCRED, &sock->flags);
986 clear_bit(SOCK_PASSCRED, &sock->flags);
989 case SO_TIMESTAMP_OLD:
990 __sock_set_timestamps(sk, valbool, false, false);
992 case SO_TIMESTAMP_NEW:
993 __sock_set_timestamps(sk, valbool, true, false);
995 case SO_TIMESTAMPNS_OLD:
996 __sock_set_timestamps(sk, valbool, false, true);
998 case SO_TIMESTAMPNS_NEW:
999 __sock_set_timestamps(sk, valbool, true, true);
1001 case SO_TIMESTAMPING_NEW:
1002 sock_set_flag(sk, SOCK_TSTAMP_NEW);
1004 case SO_TIMESTAMPING_OLD:
1005 if (val & ~SOF_TIMESTAMPING_MASK) {
1010 if (val & SOF_TIMESTAMPING_OPT_ID &&
1011 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1012 if (sk->sk_protocol == IPPROTO_TCP &&
1013 sk->sk_type == SOCK_STREAM) {
1014 if ((1 << sk->sk_state) &
1015 (TCPF_CLOSE | TCPF_LISTEN)) {
1019 sk->sk_tskey = tcp_sk(sk)->snd_una;
1025 if (val & SOF_TIMESTAMPING_OPT_STATS &&
1026 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1031 sk->sk_tsflags = val;
1032 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1033 sock_enable_timestamp(sk,
1034 SOCK_TIMESTAMPING_RX_SOFTWARE);
1036 if (optname == SO_TIMESTAMPING_NEW)
1037 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
1039 sock_disable_timestamp(sk,
1040 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1047 if (sock->ops->set_rcvlowat)
1048 ret = sock->ops->set_rcvlowat(sk, val);
1050 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1053 case SO_RCVTIMEO_OLD:
1054 case SO_RCVTIMEO_NEW:
1055 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1056 optlen, optname == SO_RCVTIMEO_OLD);
1059 case SO_SNDTIMEO_OLD:
1060 case SO_SNDTIMEO_NEW:
1061 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1062 optlen, optname == SO_SNDTIMEO_OLD);
1065 case SO_ATTACH_FILTER: {
1066 struct sock_fprog fprog;
1068 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1070 ret = sk_attach_filter(&fprog, sk);
1075 if (optlen == sizeof(u32)) {
1079 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1082 ret = sk_attach_bpf(ufd, sk);
1086 case SO_ATTACH_REUSEPORT_CBPF: {
1087 struct sock_fprog fprog;
1089 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1091 ret = sk_reuseport_attach_filter(&fprog, sk);
1094 case SO_ATTACH_REUSEPORT_EBPF:
1096 if (optlen == sizeof(u32)) {
1100 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1103 ret = sk_reuseport_attach_bpf(ufd, sk);
1107 case SO_DETACH_REUSEPORT_BPF:
1108 ret = reuseport_detach_prog(sk);
1111 case SO_DETACH_FILTER:
1112 ret = sk_detach_filter(sk);
1115 case SO_LOCK_FILTER:
1116 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1119 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1124 set_bit(SOCK_PASSSEC, &sock->flags);
1126 clear_bit(SOCK_PASSSEC, &sock->flags);
1129 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1131 } else if (val != sk->sk_mark) {
1138 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1141 case SO_WIFI_STATUS:
1142 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1146 if (sock->ops->set_peek_off)
1147 ret = sock->ops->set_peek_off(sk, val);
1153 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1156 case SO_SELECT_ERR_QUEUE:
1157 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1160 #ifdef CONFIG_NET_RX_BUSY_POLL
1162 /* allow unprivileged users to decrease the value */
1163 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1169 sk->sk_ll_usec = val;
1174 case SO_MAX_PACING_RATE:
1176 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1178 if (sizeof(ulval) != sizeof(val) &&
1179 optlen >= sizeof(ulval) &&
1180 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1185 cmpxchg(&sk->sk_pacing_status,
1188 sk->sk_max_pacing_rate = ulval;
1189 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1192 case SO_INCOMING_CPU:
1193 WRITE_ONCE(sk->sk_incoming_cpu, val);
1198 dst_negative_advice(sk);
1202 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1203 if (!((sk->sk_type == SOCK_STREAM &&
1204 sk->sk_protocol == IPPROTO_TCP) ||
1205 (sk->sk_type == SOCK_DGRAM &&
1206 sk->sk_protocol == IPPROTO_UDP)))
1208 } else if (sk->sk_family != PF_RDS) {
1212 if (val < 0 || val > 1)
1215 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1220 if (optlen != sizeof(struct sock_txtime)) {
1223 } else if (copy_from_sockptr(&sk_txtime, optval,
1224 sizeof(struct sock_txtime))) {
1227 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1231 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1232 * scheduler has enough safe guards.
1234 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1235 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1239 sock_valbool_flag(sk, SOCK_TXTIME, true);
1240 sk->sk_clockid = sk_txtime.clockid;
1241 sk->sk_txtime_deadline_mode =
1242 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1243 sk->sk_txtime_report_errors =
1244 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1247 case SO_BINDTOIFINDEX:
1248 ret = sock_bindtoindex_locked(sk, val);
1258 EXPORT_SYMBOL(sock_setsockopt);
1261 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1262 struct ucred *ucred)
1264 ucred->pid = pid_vnr(pid);
1265 ucred->uid = ucred->gid = -1;
1267 struct user_namespace *current_ns = current_user_ns();
1269 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1270 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1274 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1276 struct user_namespace *user_ns = current_user_ns();
1279 for (i = 0; i < src->ngroups; i++)
1280 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1286 int sock_getsockopt(struct socket *sock, int level, int optname,
1287 char __user *optval, int __user *optlen)
1289 struct sock *sk = sock->sk;
1294 unsigned long ulval;
1296 struct old_timeval32 tm32;
1297 struct __kernel_old_timeval tm;
1298 struct __kernel_sock_timeval stm;
1299 struct sock_txtime txtime;
1302 int lv = sizeof(int);
1305 if (get_user(len, optlen))
1310 memset(&v, 0, sizeof(v));
1314 v.val = sock_flag(sk, SOCK_DBG);
1318 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1322 v.val = sock_flag(sk, SOCK_BROADCAST);
1326 v.val = sk->sk_sndbuf;
1330 v.val = sk->sk_rcvbuf;
1334 v.val = sk->sk_reuse;
1338 v.val = sk->sk_reuseport;
1342 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1346 v.val = sk->sk_type;
1350 v.val = sk->sk_protocol;
1354 v.val = sk->sk_family;
1358 v.val = -sock_error(sk);
1360 v.val = xchg(&sk->sk_err_soft, 0);
1364 v.val = sock_flag(sk, SOCK_URGINLINE);
1368 v.val = sk->sk_no_check_tx;
1372 v.val = sk->sk_priority;
1376 lv = sizeof(v.ling);
1377 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1378 v.ling.l_linger = sk->sk_lingertime / HZ;
1382 sock_warn_obsolete_bsdism("getsockopt");
1385 case SO_TIMESTAMP_OLD:
1386 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1387 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1388 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1391 case SO_TIMESTAMPNS_OLD:
1392 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1395 case SO_TIMESTAMP_NEW:
1396 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1399 case SO_TIMESTAMPNS_NEW:
1400 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1403 case SO_TIMESTAMPING_OLD:
1404 v.val = sk->sk_tsflags;
1407 case SO_RCVTIMEO_OLD:
1408 case SO_RCVTIMEO_NEW:
1409 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1412 case SO_SNDTIMEO_OLD:
1413 case SO_SNDTIMEO_NEW:
1414 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1418 v.val = sk->sk_rcvlowat;
1426 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1431 struct ucred peercred;
1432 if (len > sizeof(peercred))
1433 len = sizeof(peercred);
1434 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1435 if (copy_to_user(optval, &peercred, len))
1444 if (!sk->sk_peer_cred)
1447 n = sk->sk_peer_cred->group_info->ngroups;
1448 if (len < n * sizeof(gid_t)) {
1449 len = n * sizeof(gid_t);
1450 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1452 len = n * sizeof(gid_t);
1454 ret = groups_to_user((gid_t __user *)optval,
1455 sk->sk_peer_cred->group_info);
1465 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1470 if (copy_to_user(optval, address, len))
1475 /* Dubious BSD thing... Probably nobody even uses it, but
1476 * the UNIX standard wants it for whatever reason... -DaveM
1479 v.val = sk->sk_state == TCP_LISTEN;
1483 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1487 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1490 v.val = sk->sk_mark;
1494 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1497 case SO_WIFI_STATUS:
1498 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1502 if (!sock->ops->set_peek_off)
1505 v.val = sk->sk_peek_off;
1508 v.val = sock_flag(sk, SOCK_NOFCS);
1511 case SO_BINDTODEVICE:
1512 return sock_getbindtodevice(sk, optval, optlen, len);
1515 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1521 case SO_LOCK_FILTER:
1522 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1525 case SO_BPF_EXTENSIONS:
1526 v.val = bpf_tell_extensions();
1529 case SO_SELECT_ERR_QUEUE:
1530 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1533 #ifdef CONFIG_NET_RX_BUSY_POLL
1535 v.val = sk->sk_ll_usec;
1539 case SO_MAX_PACING_RATE:
1540 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1541 lv = sizeof(v.ulval);
1542 v.ulval = sk->sk_max_pacing_rate;
1545 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1549 case SO_INCOMING_CPU:
1550 v.val = READ_ONCE(sk->sk_incoming_cpu);
1555 u32 meminfo[SK_MEMINFO_VARS];
1557 sk_get_meminfo(sk, meminfo);
1559 len = min_t(unsigned int, len, sizeof(meminfo));
1560 if (copy_to_user(optval, &meminfo, len))
1566 #ifdef CONFIG_NET_RX_BUSY_POLL
1567 case SO_INCOMING_NAPI_ID:
1568 v.val = READ_ONCE(sk->sk_napi_id);
1570 /* aggregate non-NAPI IDs down to 0 */
1571 if (v.val < MIN_NAPI_ID)
1581 v.val64 = sock_gen_cookie(sk);
1585 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1589 lv = sizeof(v.txtime);
1590 v.txtime.clockid = sk->sk_clockid;
1591 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1592 SOF_TXTIME_DEADLINE_MODE : 0;
1593 v.txtime.flags |= sk->sk_txtime_report_errors ?
1594 SOF_TXTIME_REPORT_ERRORS : 0;
1597 case SO_BINDTOIFINDEX:
1598 v.val = sk->sk_bound_dev_if;
1602 /* We implement the SO_SNDLOWAT etc to not be settable
1605 return -ENOPROTOOPT;
1610 if (copy_to_user(optval, &v, len))
1613 if (put_user(len, optlen))
1619 * Initialize an sk_lock.
1621 * (We also register the sk_lock with the lock validator.)
1623 static inline void sock_lock_init(struct sock *sk)
1625 if (sk->sk_kern_sock)
1626 sock_lock_init_class_and_name(
1628 af_family_kern_slock_key_strings[sk->sk_family],
1629 af_family_kern_slock_keys + sk->sk_family,
1630 af_family_kern_key_strings[sk->sk_family],
1631 af_family_kern_keys + sk->sk_family);
1633 sock_lock_init_class_and_name(
1635 af_family_slock_key_strings[sk->sk_family],
1636 af_family_slock_keys + sk->sk_family,
1637 af_family_key_strings[sk->sk_family],
1638 af_family_keys + sk->sk_family);
1642 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1643 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1644 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1646 static void sock_copy(struct sock *nsk, const struct sock *osk)
1648 const struct proto *prot = READ_ONCE(osk->sk_prot);
1649 #ifdef CONFIG_SECURITY_NETWORK
1650 void *sptr = nsk->sk_security;
1652 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1654 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1655 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1657 #ifdef CONFIG_SECURITY_NETWORK
1658 nsk->sk_security = sptr;
1659 security_sk_clone(osk, nsk);
1663 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1667 struct kmem_cache *slab;
1671 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1674 if (want_init_on_alloc(priority))
1675 sk_prot_clear_nulls(sk, prot->obj_size);
1677 sk = kmalloc(prot->obj_size, priority);
1680 if (security_sk_alloc(sk, family, priority))
1683 if (!try_module_get(prot->owner))
1685 sk_tx_queue_clear(sk);
1691 security_sk_free(sk);
1694 kmem_cache_free(slab, sk);
1700 static void sk_prot_free(struct proto *prot, struct sock *sk)
1702 struct kmem_cache *slab;
1703 struct module *owner;
1705 owner = prot->owner;
1708 cgroup_sk_free(&sk->sk_cgrp_data);
1709 mem_cgroup_sk_free(sk);
1710 security_sk_free(sk);
1712 kmem_cache_free(slab, sk);
1719 * sk_alloc - All socket objects are allocated here
1720 * @net: the applicable net namespace
1721 * @family: protocol family
1722 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1723 * @prot: struct proto associated with this new sock instance
1724 * @kern: is this to be a kernel socket?
1726 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1727 struct proto *prot, int kern)
1731 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1733 sk->sk_family = family;
1735 * See comment in struct sock definition to understand
1736 * why we need sk_prot_creator -acme
1738 sk->sk_prot = sk->sk_prot_creator = prot;
1739 sk->sk_kern_sock = kern;
1741 sk->sk_net_refcnt = kern ? 0 : 1;
1742 if (likely(sk->sk_net_refcnt)) {
1744 sock_inuse_add(net, 1);
1747 sock_net_set(sk, net);
1748 refcount_set(&sk->sk_wmem_alloc, 1);
1750 mem_cgroup_sk_alloc(sk);
1751 cgroup_sk_alloc(&sk->sk_cgrp_data);
1752 sock_update_classid(&sk->sk_cgrp_data);
1753 sock_update_netprioidx(&sk->sk_cgrp_data);
1754 sk_tx_queue_clear(sk);
1759 EXPORT_SYMBOL(sk_alloc);
1761 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1762 * grace period. This is the case for UDP sockets and TCP listeners.
1764 static void __sk_destruct(struct rcu_head *head)
1766 struct sock *sk = container_of(head, struct sock, sk_rcu);
1767 struct sk_filter *filter;
1769 if (sk->sk_destruct)
1770 sk->sk_destruct(sk);
1772 filter = rcu_dereference_check(sk->sk_filter,
1773 refcount_read(&sk->sk_wmem_alloc) == 0);
1775 sk_filter_uncharge(sk, filter);
1776 RCU_INIT_POINTER(sk->sk_filter, NULL);
1779 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1781 #ifdef CONFIG_BPF_SYSCALL
1782 bpf_sk_storage_free(sk);
1785 if (atomic_read(&sk->sk_omem_alloc))
1786 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1787 __func__, atomic_read(&sk->sk_omem_alloc));
1789 if (sk->sk_frag.page) {
1790 put_page(sk->sk_frag.page);
1791 sk->sk_frag.page = NULL;
1794 if (sk->sk_peer_cred)
1795 put_cred(sk->sk_peer_cred);
1796 put_pid(sk->sk_peer_pid);
1797 if (likely(sk->sk_net_refcnt))
1798 put_net(sock_net(sk));
1799 sk_prot_free(sk->sk_prot_creator, sk);
1802 void sk_destruct(struct sock *sk)
1804 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1806 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1807 reuseport_detach_sock(sk);
1808 use_call_rcu = true;
1812 call_rcu(&sk->sk_rcu, __sk_destruct);
1814 __sk_destruct(&sk->sk_rcu);
1817 static void __sk_free(struct sock *sk)
1819 if (likely(sk->sk_net_refcnt))
1820 sock_inuse_add(sock_net(sk), -1);
1822 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1823 sock_diag_broadcast_destroy(sk);
1828 void sk_free(struct sock *sk)
1831 * We subtract one from sk_wmem_alloc and can know if
1832 * some packets are still in some tx queue.
1833 * If not null, sock_wfree() will call __sk_free(sk) later
1835 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1838 EXPORT_SYMBOL(sk_free);
1840 static void sk_init_common(struct sock *sk)
1842 skb_queue_head_init(&sk->sk_receive_queue);
1843 skb_queue_head_init(&sk->sk_write_queue);
1844 skb_queue_head_init(&sk->sk_error_queue);
1846 rwlock_init(&sk->sk_callback_lock);
1847 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1848 af_rlock_keys + sk->sk_family,
1849 af_family_rlock_key_strings[sk->sk_family]);
1850 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1851 af_wlock_keys + sk->sk_family,
1852 af_family_wlock_key_strings[sk->sk_family]);
1853 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1854 af_elock_keys + sk->sk_family,
1855 af_family_elock_key_strings[sk->sk_family]);
1856 lockdep_set_class_and_name(&sk->sk_callback_lock,
1857 af_callback_keys + sk->sk_family,
1858 af_family_clock_key_strings[sk->sk_family]);
1862 * sk_clone_lock - clone a socket, and lock its clone
1863 * @sk: the socket to clone
1864 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1866 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1868 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1870 struct proto *prot = READ_ONCE(sk->sk_prot);
1872 bool is_charged = true;
1874 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1875 if (newsk != NULL) {
1876 struct sk_filter *filter;
1878 sock_copy(newsk, sk);
1880 newsk->sk_prot_creator = prot;
1883 if (likely(newsk->sk_net_refcnt))
1884 get_net(sock_net(newsk));
1885 sk_node_init(&newsk->sk_node);
1886 sock_lock_init(newsk);
1887 bh_lock_sock(newsk);
1888 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1889 newsk->sk_backlog.len = 0;
1891 atomic_set(&newsk->sk_rmem_alloc, 0);
1893 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1895 refcount_set(&newsk->sk_wmem_alloc, 1);
1896 atomic_set(&newsk->sk_omem_alloc, 0);
1897 sk_init_common(newsk);
1899 newsk->sk_dst_cache = NULL;
1900 newsk->sk_dst_pending_confirm = 0;
1901 newsk->sk_wmem_queued = 0;
1902 newsk->sk_forward_alloc = 0;
1903 atomic_set(&newsk->sk_drops, 0);
1904 newsk->sk_send_head = NULL;
1905 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1906 atomic_set(&newsk->sk_zckey, 0);
1908 sock_reset_flag(newsk, SOCK_DONE);
1910 /* sk->sk_memcg will be populated at accept() time */
1911 newsk->sk_memcg = NULL;
1913 cgroup_sk_clone(&newsk->sk_cgrp_data);
1916 filter = rcu_dereference(sk->sk_filter);
1918 /* though it's an empty new sock, the charging may fail
1919 * if sysctl_optmem_max was changed between creation of
1920 * original socket and cloning
1922 is_charged = sk_filter_charge(newsk, filter);
1923 RCU_INIT_POINTER(newsk->sk_filter, filter);
1926 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1927 /* We need to make sure that we don't uncharge the new
1928 * socket if we couldn't charge it in the first place
1929 * as otherwise we uncharge the parent's filter.
1932 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1933 sk_free_unlock_clone(newsk);
1937 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1939 if (bpf_sk_storage_clone(sk, newsk)) {
1940 sk_free_unlock_clone(newsk);
1945 /* Clear sk_user_data if parent had the pointer tagged
1946 * as not suitable for copying when cloning.
1948 if (sk_user_data_is_nocopy(newsk))
1949 newsk->sk_user_data = NULL;
1952 newsk->sk_err_soft = 0;
1953 newsk->sk_priority = 0;
1954 newsk->sk_incoming_cpu = raw_smp_processor_id();
1955 if (likely(newsk->sk_net_refcnt))
1956 sock_inuse_add(sock_net(newsk), 1);
1959 * Before updating sk_refcnt, we must commit prior changes to memory
1960 * (Documentation/RCU/rculist_nulls.rst for details)
1963 refcount_set(&newsk->sk_refcnt, 2);
1966 * Increment the counter in the same struct proto as the master
1967 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1968 * is the same as sk->sk_prot->socks, as this field was copied
1971 * This _changes_ the previous behaviour, where
1972 * tcp_create_openreq_child always was incrementing the
1973 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1974 * to be taken into account in all callers. -acme
1976 sk_refcnt_debug_inc(newsk);
1977 sk_set_socket(newsk, NULL);
1978 sk_tx_queue_clear(newsk);
1979 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1981 if (newsk->sk_prot->sockets_allocated)
1982 sk_sockets_allocated_inc(newsk);
1984 if (sock_needs_netstamp(sk) &&
1985 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1986 net_enable_timestamp();
1991 EXPORT_SYMBOL_GPL(sk_clone_lock);
1993 void sk_free_unlock_clone(struct sock *sk)
1995 /* It is still raw copy of parent, so invalidate
1996 * destructor and make plain sk_free() */
1997 sk->sk_destruct = NULL;
2001 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2003 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2007 sk_dst_set(sk, dst);
2008 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2009 if (sk->sk_route_caps & NETIF_F_GSO)
2010 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2011 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2012 if (sk_can_gso(sk)) {
2013 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2014 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2016 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2017 sk->sk_gso_max_size = dst->dev->gso_max_size;
2018 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2021 sk->sk_gso_max_segs = max_segs;
2023 EXPORT_SYMBOL_GPL(sk_setup_caps);
2026 * Simple resource managers for sockets.
2031 * Write buffer destructor automatically called from kfree_skb.
2033 void sock_wfree(struct sk_buff *skb)
2035 struct sock *sk = skb->sk;
2036 unsigned int len = skb->truesize;
2038 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2040 * Keep a reference on sk_wmem_alloc, this will be released
2041 * after sk_write_space() call
2043 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2044 sk->sk_write_space(sk);
2048 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2049 * could not do because of in-flight packets
2051 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2054 EXPORT_SYMBOL(sock_wfree);
2056 /* This variant of sock_wfree() is used by TCP,
2057 * since it sets SOCK_USE_WRITE_QUEUE.
2059 void __sock_wfree(struct sk_buff *skb)
2061 struct sock *sk = skb->sk;
2063 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2067 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2072 if (unlikely(!sk_fullsock(sk))) {
2073 skb->destructor = sock_edemux;
2078 skb->destructor = sock_wfree;
2079 skb_set_hash_from_sk(skb, sk);
2081 * We used to take a refcount on sk, but following operation
2082 * is enough to guarantee sk_free() wont free this sock until
2083 * all in-flight packets are completed
2085 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2087 EXPORT_SYMBOL(skb_set_owner_w);
2089 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2091 #ifdef CONFIG_TLS_DEVICE
2092 /* Drivers depend on in-order delivery for crypto offload,
2093 * partial orphan breaks out-of-order-OK logic.
2098 return (skb->destructor == sock_wfree ||
2099 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2102 /* This helper is used by netem, as it can hold packets in its
2103 * delay queue. We want to allow the owner socket to send more
2104 * packets, as if they were already TX completed by a typical driver.
2105 * But we also want to keep skb->sk set because some packet schedulers
2106 * rely on it (sch_fq for example).
2108 void skb_orphan_partial(struct sk_buff *skb)
2110 if (skb_is_tcp_pure_ack(skb))
2113 if (can_skb_orphan_partial(skb)) {
2114 struct sock *sk = skb->sk;
2116 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2117 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2118 skb->destructor = sock_efree;
2124 EXPORT_SYMBOL(skb_orphan_partial);
2127 * Read buffer destructor automatically called from kfree_skb.
2129 void sock_rfree(struct sk_buff *skb)
2131 struct sock *sk = skb->sk;
2132 unsigned int len = skb->truesize;
2134 atomic_sub(len, &sk->sk_rmem_alloc);
2135 sk_mem_uncharge(sk, len);
2137 EXPORT_SYMBOL(sock_rfree);
2140 * Buffer destructor for skbs that are not used directly in read or write
2141 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2143 void sock_efree(struct sk_buff *skb)
2147 EXPORT_SYMBOL(sock_efree);
2149 /* Buffer destructor for prefetch/receive path where reference count may
2150 * not be held, e.g. for listen sockets.
2153 void sock_pfree(struct sk_buff *skb)
2155 if (sk_is_refcounted(skb->sk))
2156 sock_gen_put(skb->sk);
2158 EXPORT_SYMBOL(sock_pfree);
2159 #endif /* CONFIG_INET */
2161 kuid_t sock_i_uid(struct sock *sk)
2165 read_lock_bh(&sk->sk_callback_lock);
2166 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2167 read_unlock_bh(&sk->sk_callback_lock);
2170 EXPORT_SYMBOL(sock_i_uid);
2172 unsigned long sock_i_ino(struct sock *sk)
2176 read_lock_bh(&sk->sk_callback_lock);
2177 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2178 read_unlock_bh(&sk->sk_callback_lock);
2181 EXPORT_SYMBOL(sock_i_ino);
2184 * Allocate a skb from the socket's send buffer.
2186 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2190 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2191 struct sk_buff *skb = alloc_skb(size, priority);
2194 skb_set_owner_w(skb, sk);
2200 EXPORT_SYMBOL(sock_wmalloc);
2202 static void sock_ofree(struct sk_buff *skb)
2204 struct sock *sk = skb->sk;
2206 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2209 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2212 struct sk_buff *skb;
2214 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2215 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2219 skb = alloc_skb(size, priority);
2223 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2225 skb->destructor = sock_ofree;
2230 * Allocate a memory block from the socket's option memory buffer.
2232 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2234 if ((unsigned int)size <= sysctl_optmem_max &&
2235 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2237 /* First do the add, to avoid the race if kmalloc
2240 atomic_add(size, &sk->sk_omem_alloc);
2241 mem = kmalloc(size, priority);
2244 atomic_sub(size, &sk->sk_omem_alloc);
2248 EXPORT_SYMBOL(sock_kmalloc);
2250 /* Free an option memory block. Note, we actually want the inline
2251 * here as this allows gcc to detect the nullify and fold away the
2252 * condition entirely.
2254 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2257 if (WARN_ON_ONCE(!mem))
2263 atomic_sub(size, &sk->sk_omem_alloc);
2266 void sock_kfree_s(struct sock *sk, void *mem, int size)
2268 __sock_kfree_s(sk, mem, size, false);
2270 EXPORT_SYMBOL(sock_kfree_s);
2272 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2274 __sock_kfree_s(sk, mem, size, true);
2276 EXPORT_SYMBOL(sock_kzfree_s);
2278 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2279 I think, these locks should be removed for datagram sockets.
2281 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2285 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2289 if (signal_pending(current))
2291 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2292 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2293 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2295 if (sk->sk_shutdown & SEND_SHUTDOWN)
2299 timeo = schedule_timeout(timeo);
2301 finish_wait(sk_sleep(sk), &wait);
2307 * Generic send/receive buffer handlers
2310 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2311 unsigned long data_len, int noblock,
2312 int *errcode, int max_page_order)
2314 struct sk_buff *skb;
2318 timeo = sock_sndtimeo(sk, noblock);
2320 err = sock_error(sk);
2325 if (sk->sk_shutdown & SEND_SHUTDOWN)
2328 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2331 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2332 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2336 if (signal_pending(current))
2338 timeo = sock_wait_for_wmem(sk, timeo);
2340 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2341 errcode, sk->sk_allocation);
2343 skb_set_owner_w(skb, sk);
2347 err = sock_intr_errno(timeo);
2352 EXPORT_SYMBOL(sock_alloc_send_pskb);
2354 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2355 int noblock, int *errcode)
2357 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2359 EXPORT_SYMBOL(sock_alloc_send_skb);
2361 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2362 struct sockcm_cookie *sockc)
2366 switch (cmsg->cmsg_type) {
2368 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2370 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2372 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2374 case SO_TIMESTAMPING_OLD:
2375 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2378 tsflags = *(u32 *)CMSG_DATA(cmsg);
2379 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2382 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2383 sockc->tsflags |= tsflags;
2386 if (!sock_flag(sk, SOCK_TXTIME))
2388 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2390 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2392 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2394 case SCM_CREDENTIALS:
2401 EXPORT_SYMBOL(__sock_cmsg_send);
2403 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2404 struct sockcm_cookie *sockc)
2406 struct cmsghdr *cmsg;
2409 for_each_cmsghdr(cmsg, msg) {
2410 if (!CMSG_OK(msg, cmsg))
2412 if (cmsg->cmsg_level != SOL_SOCKET)
2414 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2420 EXPORT_SYMBOL(sock_cmsg_send);
2422 static void sk_enter_memory_pressure(struct sock *sk)
2424 if (!sk->sk_prot->enter_memory_pressure)
2427 sk->sk_prot->enter_memory_pressure(sk);
2430 static void sk_leave_memory_pressure(struct sock *sk)
2432 if (sk->sk_prot->leave_memory_pressure) {
2433 sk->sk_prot->leave_memory_pressure(sk);
2435 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2437 if (memory_pressure && READ_ONCE(*memory_pressure))
2438 WRITE_ONCE(*memory_pressure, 0);
2442 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2443 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2446 * skb_page_frag_refill - check that a page_frag contains enough room
2447 * @sz: minimum size of the fragment we want to get
2448 * @pfrag: pointer to page_frag
2449 * @gfp: priority for memory allocation
2451 * Note: While this allocator tries to use high order pages, there is
2452 * no guarantee that allocations succeed. Therefore, @sz MUST be
2453 * less or equal than PAGE_SIZE.
2455 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2458 if (page_ref_count(pfrag->page) == 1) {
2462 if (pfrag->offset + sz <= pfrag->size)
2464 put_page(pfrag->page);
2468 if (SKB_FRAG_PAGE_ORDER &&
2469 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2470 /* Avoid direct reclaim but allow kswapd to wake */
2471 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2472 __GFP_COMP | __GFP_NOWARN |
2474 SKB_FRAG_PAGE_ORDER);
2475 if (likely(pfrag->page)) {
2476 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2480 pfrag->page = alloc_page(gfp);
2481 if (likely(pfrag->page)) {
2482 pfrag->size = PAGE_SIZE;
2487 EXPORT_SYMBOL(skb_page_frag_refill);
2489 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2491 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2494 sk_enter_memory_pressure(sk);
2495 sk_stream_moderate_sndbuf(sk);
2498 EXPORT_SYMBOL(sk_page_frag_refill);
2500 static void __lock_sock(struct sock *sk)
2501 __releases(&sk->sk_lock.slock)
2502 __acquires(&sk->sk_lock.slock)
2507 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2508 TASK_UNINTERRUPTIBLE);
2509 spin_unlock_bh(&sk->sk_lock.slock);
2511 spin_lock_bh(&sk->sk_lock.slock);
2512 if (!sock_owned_by_user(sk))
2515 finish_wait(&sk->sk_lock.wq, &wait);
2518 void __release_sock(struct sock *sk)
2519 __releases(&sk->sk_lock.slock)
2520 __acquires(&sk->sk_lock.slock)
2522 struct sk_buff *skb, *next;
2524 while ((skb = sk->sk_backlog.head) != NULL) {
2525 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2527 spin_unlock_bh(&sk->sk_lock.slock);
2532 WARN_ON_ONCE(skb_dst_is_noref(skb));
2533 skb_mark_not_on_list(skb);
2534 sk_backlog_rcv(sk, skb);
2539 } while (skb != NULL);
2541 spin_lock_bh(&sk->sk_lock.slock);
2545 * Doing the zeroing here guarantee we can not loop forever
2546 * while a wild producer attempts to flood us.
2548 sk->sk_backlog.len = 0;
2551 void __sk_flush_backlog(struct sock *sk)
2553 spin_lock_bh(&sk->sk_lock.slock);
2555 spin_unlock_bh(&sk->sk_lock.slock);
2559 * sk_wait_data - wait for data to arrive at sk_receive_queue
2560 * @sk: sock to wait on
2561 * @timeo: for how long
2562 * @skb: last skb seen on sk_receive_queue
2564 * Now socket state including sk->sk_err is changed only under lock,
2565 * hence we may omit checks after joining wait queue.
2566 * We check receive queue before schedule() only as optimization;
2567 * it is very likely that release_sock() added new data.
2569 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2571 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2574 add_wait_queue(sk_sleep(sk), &wait);
2575 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2576 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2577 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2578 remove_wait_queue(sk_sleep(sk), &wait);
2581 EXPORT_SYMBOL(sk_wait_data);
2584 * __sk_mem_raise_allocated - increase memory_allocated
2586 * @size: memory size to allocate
2587 * @amt: pages to allocate
2588 * @kind: allocation type
2590 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2592 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2594 struct proto *prot = sk->sk_prot;
2595 long allocated = sk_memory_allocated_add(sk, amt);
2596 bool charged = true;
2598 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2599 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2600 goto suppress_allocation;
2603 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2604 sk_leave_memory_pressure(sk);
2608 /* Under pressure. */
2609 if (allocated > sk_prot_mem_limits(sk, 1))
2610 sk_enter_memory_pressure(sk);
2612 /* Over hard limit. */
2613 if (allocated > sk_prot_mem_limits(sk, 2))
2614 goto suppress_allocation;
2616 /* guarantee minimum buffer size under pressure */
2617 if (kind == SK_MEM_RECV) {
2618 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2621 } else { /* SK_MEM_SEND */
2622 int wmem0 = sk_get_wmem0(sk, prot);
2624 if (sk->sk_type == SOCK_STREAM) {
2625 if (sk->sk_wmem_queued < wmem0)
2627 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2632 if (sk_has_memory_pressure(sk)) {
2635 if (!sk_under_memory_pressure(sk))
2637 alloc = sk_sockets_allocated_read_positive(sk);
2638 if (sk_prot_mem_limits(sk, 2) > alloc *
2639 sk_mem_pages(sk->sk_wmem_queued +
2640 atomic_read(&sk->sk_rmem_alloc) +
2641 sk->sk_forward_alloc))
2645 suppress_allocation:
2647 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2648 sk_stream_moderate_sndbuf(sk);
2650 /* Fail only if socket is _under_ its sndbuf.
2651 * In this case we cannot block, so that we have to fail.
2653 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2657 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2658 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2660 sk_memory_allocated_sub(sk, amt);
2662 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2663 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2667 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2670 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2672 * @size: memory size to allocate
2673 * @kind: allocation type
2675 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2676 * rmem allocation. This function assumes that protocols which have
2677 * memory_pressure use sk_wmem_queued as write buffer accounting.
2679 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2681 int ret, amt = sk_mem_pages(size);
2683 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2684 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2686 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2689 EXPORT_SYMBOL(__sk_mem_schedule);
2692 * __sk_mem_reduce_allocated - reclaim memory_allocated
2694 * @amount: number of quanta
2696 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2698 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2700 sk_memory_allocated_sub(sk, amount);
2702 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2703 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2705 if (sk_under_memory_pressure(sk) &&
2706 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2707 sk_leave_memory_pressure(sk);
2709 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2712 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2714 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2716 void __sk_mem_reclaim(struct sock *sk, int amount)
2718 amount >>= SK_MEM_QUANTUM_SHIFT;
2719 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2720 __sk_mem_reduce_allocated(sk, amount);
2722 EXPORT_SYMBOL(__sk_mem_reclaim);
2724 int sk_set_peek_off(struct sock *sk, int val)
2726 sk->sk_peek_off = val;
2729 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2732 * Set of default routines for initialising struct proto_ops when
2733 * the protocol does not support a particular function. In certain
2734 * cases where it makes no sense for a protocol to have a "do nothing"
2735 * function, some default processing is provided.
2738 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2742 EXPORT_SYMBOL(sock_no_bind);
2744 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2749 EXPORT_SYMBOL(sock_no_connect);
2751 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2755 EXPORT_SYMBOL(sock_no_socketpair);
2757 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2762 EXPORT_SYMBOL(sock_no_accept);
2764 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2769 EXPORT_SYMBOL(sock_no_getname);
2771 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2775 EXPORT_SYMBOL(sock_no_ioctl);
2777 int sock_no_listen(struct socket *sock, int backlog)
2781 EXPORT_SYMBOL(sock_no_listen);
2783 int sock_no_shutdown(struct socket *sock, int how)
2787 EXPORT_SYMBOL(sock_no_shutdown);
2789 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2793 EXPORT_SYMBOL(sock_no_sendmsg);
2795 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2799 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2801 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2806 EXPORT_SYMBOL(sock_no_recvmsg);
2808 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2810 /* Mirror missing mmap method error code */
2813 EXPORT_SYMBOL(sock_no_mmap);
2816 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2817 * various sock-based usage counts.
2819 void __receive_sock(struct file *file)
2821 struct socket *sock;
2825 * The resulting value of "error" is ignored here since we only
2826 * need to take action when the file is a socket and testing
2827 * "sock" for NULL is sufficient.
2829 sock = sock_from_file(file, &error);
2831 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2832 sock_update_classid(&sock->sk->sk_cgrp_data);
2836 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2839 struct msghdr msg = {.msg_flags = flags};
2841 char *kaddr = kmap(page);
2842 iov.iov_base = kaddr + offset;
2844 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2848 EXPORT_SYMBOL(sock_no_sendpage);
2850 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2851 int offset, size_t size, int flags)
2854 struct msghdr msg = {.msg_flags = flags};
2856 char *kaddr = kmap(page);
2858 iov.iov_base = kaddr + offset;
2860 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2864 EXPORT_SYMBOL(sock_no_sendpage_locked);
2867 * Default Socket Callbacks
2870 static void sock_def_wakeup(struct sock *sk)
2872 struct socket_wq *wq;
2875 wq = rcu_dereference(sk->sk_wq);
2876 if (skwq_has_sleeper(wq))
2877 wake_up_interruptible_all(&wq->wait);
2881 static void sock_def_error_report(struct sock *sk)
2883 struct socket_wq *wq;
2886 wq = rcu_dereference(sk->sk_wq);
2887 if (skwq_has_sleeper(wq))
2888 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2889 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2893 void sock_def_readable(struct sock *sk)
2895 struct socket_wq *wq;
2898 wq = rcu_dereference(sk->sk_wq);
2899 if (skwq_has_sleeper(wq))
2900 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2901 EPOLLRDNORM | EPOLLRDBAND);
2902 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2906 static void sock_def_write_space(struct sock *sk)
2908 struct socket_wq *wq;
2912 /* Do not wake up a writer until he can make "significant"
2915 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2916 wq = rcu_dereference(sk->sk_wq);
2917 if (skwq_has_sleeper(wq))
2918 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2919 EPOLLWRNORM | EPOLLWRBAND);
2921 /* Should agree with poll, otherwise some programs break */
2922 if (sock_writeable(sk))
2923 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2929 static void sock_def_destruct(struct sock *sk)
2933 void sk_send_sigurg(struct sock *sk)
2935 if (sk->sk_socket && sk->sk_socket->file)
2936 if (send_sigurg(&sk->sk_socket->file->f_owner))
2937 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2939 EXPORT_SYMBOL(sk_send_sigurg);
2941 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2942 unsigned long expires)
2944 if (!mod_timer(timer, expires))
2947 EXPORT_SYMBOL(sk_reset_timer);
2949 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2951 if (del_timer(timer))
2954 EXPORT_SYMBOL(sk_stop_timer);
2956 void sock_init_data(struct socket *sock, struct sock *sk)
2959 sk->sk_send_head = NULL;
2961 timer_setup(&sk->sk_timer, NULL, 0);
2963 sk->sk_allocation = GFP_KERNEL;
2964 sk->sk_rcvbuf = sysctl_rmem_default;
2965 sk->sk_sndbuf = sysctl_wmem_default;
2966 sk->sk_state = TCP_CLOSE;
2967 sk_set_socket(sk, sock);
2969 sock_set_flag(sk, SOCK_ZAPPED);
2972 sk->sk_type = sock->type;
2973 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2975 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2977 RCU_INIT_POINTER(sk->sk_wq, NULL);
2978 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2981 rwlock_init(&sk->sk_callback_lock);
2982 if (sk->sk_kern_sock)
2983 lockdep_set_class_and_name(
2984 &sk->sk_callback_lock,
2985 af_kern_callback_keys + sk->sk_family,
2986 af_family_kern_clock_key_strings[sk->sk_family]);
2988 lockdep_set_class_and_name(
2989 &sk->sk_callback_lock,
2990 af_callback_keys + sk->sk_family,
2991 af_family_clock_key_strings[sk->sk_family]);
2993 sk->sk_state_change = sock_def_wakeup;
2994 sk->sk_data_ready = sock_def_readable;
2995 sk->sk_write_space = sock_def_write_space;
2996 sk->sk_error_report = sock_def_error_report;
2997 sk->sk_destruct = sock_def_destruct;
2999 sk->sk_frag.page = NULL;
3000 sk->sk_frag.offset = 0;
3001 sk->sk_peek_off = -1;
3003 sk->sk_peer_pid = NULL;
3004 sk->sk_peer_cred = NULL;
3005 sk->sk_write_pending = 0;
3006 sk->sk_rcvlowat = 1;
3007 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3008 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3010 sk->sk_stamp = SK_DEFAULT_STAMP;
3011 #if BITS_PER_LONG==32
3012 seqlock_init(&sk->sk_stamp_seq);
3014 atomic_set(&sk->sk_zckey, 0);
3016 #ifdef CONFIG_NET_RX_BUSY_POLL
3018 sk->sk_ll_usec = sysctl_net_busy_read;
3021 sk->sk_max_pacing_rate = ~0UL;
3022 sk->sk_pacing_rate = ~0UL;
3023 WRITE_ONCE(sk->sk_pacing_shift, 10);
3024 sk->sk_incoming_cpu = -1;
3026 sk_rx_queue_clear(sk);
3028 * Before updating sk_refcnt, we must commit prior changes to memory
3029 * (Documentation/RCU/rculist_nulls.rst for details)
3032 refcount_set(&sk->sk_refcnt, 1);
3033 atomic_set(&sk->sk_drops, 0);
3035 EXPORT_SYMBOL(sock_init_data);
3037 void lock_sock_nested(struct sock *sk, int subclass)
3040 spin_lock_bh(&sk->sk_lock.slock);
3041 if (sk->sk_lock.owned)
3043 sk->sk_lock.owned = 1;
3044 spin_unlock(&sk->sk_lock.slock);
3046 * The sk_lock has mutex_lock() semantics here:
3048 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3051 EXPORT_SYMBOL(lock_sock_nested);
3053 void release_sock(struct sock *sk)
3055 spin_lock_bh(&sk->sk_lock.slock);
3056 if (sk->sk_backlog.tail)
3059 /* Warning : release_cb() might need to release sk ownership,
3060 * ie call sock_release_ownership(sk) before us.
3062 if (sk->sk_prot->release_cb)
3063 sk->sk_prot->release_cb(sk);
3065 sock_release_ownership(sk);
3066 if (waitqueue_active(&sk->sk_lock.wq))
3067 wake_up(&sk->sk_lock.wq);
3068 spin_unlock_bh(&sk->sk_lock.slock);
3070 EXPORT_SYMBOL(release_sock);
3073 * lock_sock_fast - fast version of lock_sock
3076 * This version should be used for very small section, where process wont block
3077 * return false if fast path is taken:
3079 * sk_lock.slock locked, owned = 0, BH disabled
3081 * return true if slow path is taken:
3083 * sk_lock.slock unlocked, owned = 1, BH enabled
3085 bool lock_sock_fast(struct sock *sk)
3088 spin_lock_bh(&sk->sk_lock.slock);
3090 if (!sk->sk_lock.owned)
3092 * Note : We must disable BH
3097 sk->sk_lock.owned = 1;
3098 spin_unlock(&sk->sk_lock.slock);
3100 * The sk_lock has mutex_lock() semantics here:
3102 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3106 EXPORT_SYMBOL(lock_sock_fast);
3108 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3109 bool timeval, bool time32)
3111 struct sock *sk = sock->sk;
3112 struct timespec64 ts;
3114 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3115 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3116 if (ts.tv_sec == -1)
3118 if (ts.tv_sec == 0) {
3119 ktime_t kt = ktime_get_real();
3120 sock_write_timestamp(sk, kt);
3121 ts = ktime_to_timespec64(kt);
3127 #ifdef CONFIG_COMPAT_32BIT_TIME
3129 return put_old_timespec32(&ts, userstamp);
3131 #ifdef CONFIG_SPARC64
3132 /* beware of padding in sparc64 timeval */
3133 if (timeval && !in_compat_syscall()) {
3134 struct __kernel_old_timeval __user tv = {
3135 .tv_sec = ts.tv_sec,
3136 .tv_usec = ts.tv_nsec,
3138 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3143 return put_timespec64(&ts, userstamp);
3145 EXPORT_SYMBOL(sock_gettstamp);
3147 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3149 if (!sock_flag(sk, flag)) {
3150 unsigned long previous_flags = sk->sk_flags;
3152 sock_set_flag(sk, flag);
3154 * we just set one of the two flags which require net
3155 * time stamping, but time stamping might have been on
3156 * already because of the other one
3158 if (sock_needs_netstamp(sk) &&
3159 !(previous_flags & SK_FLAGS_TIMESTAMP))
3160 net_enable_timestamp();
3164 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3165 int level, int type)
3167 struct sock_exterr_skb *serr;
3168 struct sk_buff *skb;
3172 skb = sock_dequeue_err_skb(sk);
3178 msg->msg_flags |= MSG_TRUNC;
3181 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3185 sock_recv_timestamp(msg, sk, skb);
3187 serr = SKB_EXT_ERR(skb);
3188 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3190 msg->msg_flags |= MSG_ERRQUEUE;
3198 EXPORT_SYMBOL(sock_recv_errqueue);
3201 * Get a socket option on an socket.
3203 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3204 * asynchronous errors should be reported by getsockopt. We assume
3205 * this means if you specify SO_ERROR (otherwise whats the point of it).
3207 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3208 char __user *optval, int __user *optlen)
3210 struct sock *sk = sock->sk;
3212 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3214 EXPORT_SYMBOL(sock_common_getsockopt);
3216 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3219 struct sock *sk = sock->sk;
3223 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3224 flags & ~MSG_DONTWAIT, &addr_len);
3226 msg->msg_namelen = addr_len;
3229 EXPORT_SYMBOL(sock_common_recvmsg);
3232 * Set socket options on an inet socket.
3234 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3235 sockptr_t optval, unsigned int optlen)
3237 struct sock *sk = sock->sk;
3239 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3241 EXPORT_SYMBOL(sock_common_setsockopt);
3243 void sk_common_release(struct sock *sk)
3245 if (sk->sk_prot->destroy)
3246 sk->sk_prot->destroy(sk);
3249 * Observation: when sock_common_release is called, processes have
3250 * no access to socket. But net still has.
3251 * Step one, detach it from networking:
3253 * A. Remove from hash tables.
3256 sk->sk_prot->unhash(sk);
3259 * In this point socket cannot receive new packets, but it is possible
3260 * that some packets are in flight because some CPU runs receiver and
3261 * did hash table lookup before we unhashed socket. They will achieve
3262 * receive queue and will be purged by socket destructor.
3264 * Also we still have packets pending on receive queue and probably,
3265 * our own packets waiting in device queues. sock_destroy will drain
3266 * receive queue, but transmitted packets will delay socket destruction
3267 * until the last reference will be released.
3272 xfrm_sk_free_policy(sk);
3274 sk_refcnt_debug_release(sk);
3278 EXPORT_SYMBOL(sk_common_release);
3280 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3282 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3284 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3285 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3286 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3287 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3288 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3289 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3290 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3291 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3292 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3295 #ifdef CONFIG_PROC_FS
3296 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3298 int val[PROTO_INUSE_NR];
3301 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3303 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3305 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3307 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3309 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3311 int cpu, idx = prot->inuse_idx;
3314 for_each_possible_cpu(cpu)
3315 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3317 return res >= 0 ? res : 0;
3319 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3321 static void sock_inuse_add(struct net *net, int val)
3323 this_cpu_add(*net->core.sock_inuse, val);
3326 int sock_inuse_get(struct net *net)
3330 for_each_possible_cpu(cpu)
3331 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3336 EXPORT_SYMBOL_GPL(sock_inuse_get);
3338 static int __net_init sock_inuse_init_net(struct net *net)
3340 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3341 if (net->core.prot_inuse == NULL)
3344 net->core.sock_inuse = alloc_percpu(int);
3345 if (net->core.sock_inuse == NULL)
3351 free_percpu(net->core.prot_inuse);
3355 static void __net_exit sock_inuse_exit_net(struct net *net)
3357 free_percpu(net->core.prot_inuse);
3358 free_percpu(net->core.sock_inuse);
3361 static struct pernet_operations net_inuse_ops = {
3362 .init = sock_inuse_init_net,
3363 .exit = sock_inuse_exit_net,
3366 static __init int net_inuse_init(void)
3368 if (register_pernet_subsys(&net_inuse_ops))
3369 panic("Cannot initialize net inuse counters");
3374 core_initcall(net_inuse_init);
3376 static int assign_proto_idx(struct proto *prot)
3378 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3380 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3381 pr_err("PROTO_INUSE_NR exhausted\n");
3385 set_bit(prot->inuse_idx, proto_inuse_idx);
3389 static void release_proto_idx(struct proto *prot)
3391 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3392 clear_bit(prot->inuse_idx, proto_inuse_idx);
3395 static inline int assign_proto_idx(struct proto *prot)
3400 static inline void release_proto_idx(struct proto *prot)
3404 static void sock_inuse_add(struct net *net, int val)
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_name;
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_name;
3489 list_add(&prot->node, &proto_list);
3490 mutex_unlock(&proto_list_mutex);
3493 out_free_timewait_sock_slab_name:
3494 if (alloc_slab && prot->twsk_prot)
3495 kfree(prot->twsk_prot->twsk_slab_name);
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);
3520 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3521 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3522 kfree(prot->twsk_prot->twsk_slab_name);
3523 prot->twsk_prot->twsk_slab = NULL;
3526 EXPORT_SYMBOL(proto_unregister);
3528 int sock_load_diag_module(int family, int protocol)
3531 if (!sock_is_registered(family))
3534 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3535 NETLINK_SOCK_DIAG, family);
3539 if (family == AF_INET &&
3540 protocol != IPPROTO_RAW &&
3541 protocol < MAX_INET_PROTOS &&
3542 !rcu_access_pointer(inet_protos[protocol]))
3546 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3547 NETLINK_SOCK_DIAG, family, protocol);
3549 EXPORT_SYMBOL(sock_load_diag_module);
3551 #ifdef CONFIG_PROC_FS
3552 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3553 __acquires(proto_list_mutex)
3555 mutex_lock(&proto_list_mutex);
3556 return seq_list_start_head(&proto_list, *pos);
3559 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3561 return seq_list_next(v, &proto_list, pos);
3564 static void proto_seq_stop(struct seq_file *seq, void *v)
3565 __releases(proto_list_mutex)
3567 mutex_unlock(&proto_list_mutex);
3570 static char proto_method_implemented(const void *method)
3572 return method == NULL ? 'n' : 'y';
3574 static long sock_prot_memory_allocated(struct proto *proto)
3576 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3579 static const char *sock_prot_memory_pressure(struct proto *proto)
3581 return proto->memory_pressure != NULL ?
3582 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3585 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3588 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3589 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3592 sock_prot_inuse_get(seq_file_net(seq), proto),
3593 sock_prot_memory_allocated(proto),
3594 sock_prot_memory_pressure(proto),
3596 proto->slab == NULL ? "no" : "yes",
3597 module_name(proto->owner),
3598 proto_method_implemented(proto->close),
3599 proto_method_implemented(proto->connect),
3600 proto_method_implemented(proto->disconnect),
3601 proto_method_implemented(proto->accept),
3602 proto_method_implemented(proto->ioctl),
3603 proto_method_implemented(proto->init),
3604 proto_method_implemented(proto->destroy),
3605 proto_method_implemented(proto->shutdown),
3606 proto_method_implemented(proto->setsockopt),
3607 proto_method_implemented(proto->getsockopt),
3608 proto_method_implemented(proto->sendmsg),
3609 proto_method_implemented(proto->recvmsg),
3610 proto_method_implemented(proto->sendpage),
3611 proto_method_implemented(proto->bind),
3612 proto_method_implemented(proto->backlog_rcv),
3613 proto_method_implemented(proto->hash),
3614 proto_method_implemented(proto->unhash),
3615 proto_method_implemented(proto->get_port),
3616 proto_method_implemented(proto->enter_memory_pressure));
3619 static int proto_seq_show(struct seq_file *seq, void *v)
3621 if (v == &proto_list)
3622 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3631 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3633 proto_seq_printf(seq, list_entry(v, struct proto, node));
3637 static const struct seq_operations proto_seq_ops = {
3638 .start = proto_seq_start,
3639 .next = proto_seq_next,
3640 .stop = proto_seq_stop,
3641 .show = proto_seq_show,
3644 static __net_init int proto_init_net(struct net *net)
3646 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3647 sizeof(struct seq_net_private)))
3653 static __net_exit void proto_exit_net(struct net *net)
3655 remove_proc_entry("protocols", net->proc_net);
3659 static __net_initdata struct pernet_operations proto_net_ops = {
3660 .init = proto_init_net,
3661 .exit = proto_exit_net,
3664 static int __init proto_init(void)
3666 return register_pernet_subsys(&proto_net_ops);
3669 subsys_initcall(proto_init);
3671 #endif /* PROC_FS */
3673 #ifdef CONFIG_NET_RX_BUSY_POLL
3674 bool sk_busy_loop_end(void *p, unsigned long start_time)
3676 struct sock *sk = p;
3678 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3679 sk_busy_loop_timeout(sk, start_time);
3681 EXPORT_SYMBOL(sk_busy_loop_end);
3682 #endif /* CONFIG_NET_RX_BUSY_POLL */
3684 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3686 if (!sk->sk_prot->bind_add)
3688 return sk->sk_prot->bind_add(sk, addr, addr_len);
3690 EXPORT_SYMBOL(sock_bind_add);