Linux Audio

Check our new training course

Loading...
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   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.
   6 *
   7 *		Generic socket support routines. Memory allocators, socket lock/release
   8 *		handler for protocols to use and generic option handler.
   9 *
  10 * Authors:	Ross Biro
  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>
  14 *
  15 * Fixes:
  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
  36 *					TCP layer surgery.
  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
  64 *					(compatibility fix)
  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
  82 *
  83 * To Fix:
  84 */
  85
  86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  87
  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>
  94#include <linux/in.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>
 117
 118#include <linux/uaccess.h>
 119
 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>
 132
 133#include <linux/filter.h>
 134#include <net/sock_reuseport.h>
 135#include <net/bpf_sk_storage.h>
 136
 137#include <trace/events/sock.h>
 138
 139#include <net/tcp.h>
 140#include <net/busy_poll.h>
 141
 142#include <linux/ethtool.h>
 143
 144static DEFINE_MUTEX(proto_list_mutex);
 145static LIST_HEAD(proto_list);
 146
 147static void sock_inuse_add(struct net *net, int val);
 148
 149/**
 150 * sk_ns_capable - General socket capability test
 151 * @sk: Socket to use a capability on or through
 152 * @user_ns: The user namespace of the capability to use
 153 * @cap: The capability to use
 154 *
 155 * Test to see if the opener of the socket had when the socket was
 156 * created and the current process has the capability @cap in the user
 157 * namespace @user_ns.
 158 */
 159bool sk_ns_capable(const struct sock *sk,
 160		   struct user_namespace *user_ns, int cap)
 161{
 162	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
 163		ns_capable(user_ns, cap);
 164}
 165EXPORT_SYMBOL(sk_ns_capable);
 166
 167/**
 168 * sk_capable - Socket global capability test
 169 * @sk: Socket to use a capability on or through
 170 * @cap: The global capability to use
 171 *
 172 * Test to see if the opener of the socket had when the socket was
 173 * created and the current process has the capability @cap in all user
 174 * namespaces.
 175 */
 176bool sk_capable(const struct sock *sk, int cap)
 177{
 178	return sk_ns_capable(sk, &init_user_ns, cap);
 179}
 180EXPORT_SYMBOL(sk_capable);
 181
 182/**
 183 * sk_net_capable - Network namespace socket capability test
 184 * @sk: Socket to use a capability on or through
 185 * @cap: The capability to use
 186 *
 187 * Test to see if the opener of the socket had when the socket was created
 188 * and the current process has the capability @cap over the network namespace
 189 * the socket is a member of.
 190 */
 191bool sk_net_capable(const struct sock *sk, int cap)
 192{
 193	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
 194}
 195EXPORT_SYMBOL(sk_net_capable);
 196
 197/*
 198 * Each address family might have different locking rules, so we have
 199 * one slock key per address family and separate keys for internal and
 200 * userspace sockets.
 201 */
 202static struct lock_class_key af_family_keys[AF_MAX];
 203static struct lock_class_key af_family_kern_keys[AF_MAX];
 204static struct lock_class_key af_family_slock_keys[AF_MAX];
 205static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
 206
 207/*
 208 * Make lock validator output more readable. (we pre-construct these
 209 * strings build-time, so that runtime initialization of socket
 210 * locks is fast):
 211 */
 212
 213#define _sock_locks(x)						  \
 214  x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
 215  x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
 216  x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
 217  x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
 218  x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
 219  x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
 220  x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
 221  x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
 222  x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
 223  x "27"       ,	x "28"          ,	x "AF_CAN"      , \
 224  x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
 225  x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
 226  x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
 227  x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
 228  x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
 229  x "AF_MAX"
 230
 231static const char *const af_family_key_strings[AF_MAX+1] = {
 232	_sock_locks("sk_lock-")
 233};
 234static const char *const af_family_slock_key_strings[AF_MAX+1] = {
 235	_sock_locks("slock-")
 236};
 237static const char *const af_family_clock_key_strings[AF_MAX+1] = {
 238	_sock_locks("clock-")
 239};
 240
 241static const char *const af_family_kern_key_strings[AF_MAX+1] = {
 242	_sock_locks("k-sk_lock-")
 243};
 244static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
 245	_sock_locks("k-slock-")
 246};
 247static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
 248	_sock_locks("k-clock-")
 249};
 250static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
 251	_sock_locks("rlock-")
 252};
 253static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
 254	_sock_locks("wlock-")
 255};
 256static const char *const af_family_elock_key_strings[AF_MAX+1] = {
 257	_sock_locks("elock-")
 258};
 259
 260/*
 261 * sk_callback_lock and sk queues locking rules are per-address-family,
 262 * so split the lock classes by using a per-AF key:
 263 */
 264static struct lock_class_key af_callback_keys[AF_MAX];
 265static struct lock_class_key af_rlock_keys[AF_MAX];
 266static struct lock_class_key af_wlock_keys[AF_MAX];
 267static struct lock_class_key af_elock_keys[AF_MAX];
 268static struct lock_class_key af_kern_callback_keys[AF_MAX];
 269
 270/* Run time adjustable parameters. */
 271__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 272EXPORT_SYMBOL(sysctl_wmem_max);
 273__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 274EXPORT_SYMBOL(sysctl_rmem_max);
 275__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 276__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 277
 278/* Maximal space eaten by iovec or ancillary data plus some space */
 279int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 280EXPORT_SYMBOL(sysctl_optmem_max);
 281
 282int sysctl_tstamp_allow_data __read_mostly = 1;
 283
 284DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
 285EXPORT_SYMBOL_GPL(memalloc_socks_key);
 286
 287/**
 288 * sk_set_memalloc - sets %SOCK_MEMALLOC
 289 * @sk: socket to set it on
 290 *
 291 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
 292 * It's the responsibility of the admin to adjust min_free_kbytes
 293 * to meet the requirements
 294 */
 295void sk_set_memalloc(struct sock *sk)
 296{
 297	sock_set_flag(sk, SOCK_MEMALLOC);
 298	sk->sk_allocation |= __GFP_MEMALLOC;
 299	static_branch_inc(&memalloc_socks_key);
 300}
 301EXPORT_SYMBOL_GPL(sk_set_memalloc);
 302
 303void sk_clear_memalloc(struct sock *sk)
 304{
 305	sock_reset_flag(sk, SOCK_MEMALLOC);
 306	sk->sk_allocation &= ~__GFP_MEMALLOC;
 307	static_branch_dec(&memalloc_socks_key);
 308
 309	/*
 310	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
 311	 * progress of swapping. SOCK_MEMALLOC may be cleared while
 312	 * it has rmem allocations due to the last swapfile being deactivated
 313	 * but there is a risk that the socket is unusable due to exceeding
 314	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
 315	 */
 316	sk_mem_reclaim(sk);
 317}
 318EXPORT_SYMBOL_GPL(sk_clear_memalloc);
 319
 320int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 321{
 322	int ret;
 323	unsigned int noreclaim_flag;
 324
 325	/* these should have been dropped before queueing */
 326	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
 327
 328	noreclaim_flag = memalloc_noreclaim_save();
 329	ret = sk->sk_backlog_rcv(sk, skb);
 330	memalloc_noreclaim_restore(noreclaim_flag);
 331
 332	return ret;
 333}
 334EXPORT_SYMBOL(__sk_backlog_rcv);
 335
 336void sk_error_report(struct sock *sk)
 337{
 338	sk->sk_error_report(sk);
 339
 340	switch (sk->sk_family) {
 341	case AF_INET:
 342		fallthrough;
 343	case AF_INET6:
 344		trace_inet_sk_error_report(sk);
 345		break;
 346	default:
 347		break;
 348	}
 349}
 350EXPORT_SYMBOL(sk_error_report);
 351
 352static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
 353{
 354	struct __kernel_sock_timeval tv;
 355
 356	if (timeo == MAX_SCHEDULE_TIMEOUT) {
 357		tv.tv_sec = 0;
 358		tv.tv_usec = 0;
 359	} else {
 360		tv.tv_sec = timeo / HZ;
 361		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
 362	}
 363
 364	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
 365		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
 366		*(struct old_timeval32 *)optval = tv32;
 367		return sizeof(tv32);
 368	}
 369
 370	if (old_timeval) {
 371		struct __kernel_old_timeval old_tv;
 372		old_tv.tv_sec = tv.tv_sec;
 373		old_tv.tv_usec = tv.tv_usec;
 374		*(struct __kernel_old_timeval *)optval = old_tv;
 375		return sizeof(old_tv);
 376	}
 377
 378	*(struct __kernel_sock_timeval *)optval = tv;
 379	return sizeof(tv);
 380}
 381
 382static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
 383			    bool old_timeval)
 384{
 385	struct __kernel_sock_timeval tv;
 386
 387	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
 388		struct old_timeval32 tv32;
 389
 390		if (optlen < sizeof(tv32))
 391			return -EINVAL;
 392
 393		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
 394			return -EFAULT;
 395		tv.tv_sec = tv32.tv_sec;
 396		tv.tv_usec = tv32.tv_usec;
 397	} else if (old_timeval) {
 398		struct __kernel_old_timeval old_tv;
 399
 400		if (optlen < sizeof(old_tv))
 401			return -EINVAL;
 402		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
 403			return -EFAULT;
 404		tv.tv_sec = old_tv.tv_sec;
 405		tv.tv_usec = old_tv.tv_usec;
 406	} else {
 407		if (optlen < sizeof(tv))
 408			return -EINVAL;
 409		if (copy_from_sockptr(&tv, optval, sizeof(tv)))
 410			return -EFAULT;
 411	}
 412	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
 413		return -EDOM;
 414
 415	if (tv.tv_sec < 0) {
 416		static int warned __read_mostly;
 417
 418		*timeo_p = 0;
 419		if (warned < 10 && net_ratelimit()) {
 420			warned++;
 421			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
 422				__func__, current->comm, task_pid_nr(current));
 423		}
 424		return 0;
 425	}
 426	*timeo_p = MAX_SCHEDULE_TIMEOUT;
 427	if (tv.tv_sec == 0 && tv.tv_usec == 0)
 428		return 0;
 429	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
 430		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
 431	return 0;
 432}
 433
 
 
 
 
 
 
 
 
 
 
 
 
 434static bool sock_needs_netstamp(const struct sock *sk)
 435{
 436	switch (sk->sk_family) {
 437	case AF_UNSPEC:
 438	case AF_UNIX:
 439		return false;
 440	default:
 441		return true;
 442	}
 443}
 444
 445static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
 446{
 447	if (sk->sk_flags & flags) {
 448		sk->sk_flags &= ~flags;
 449		if (sock_needs_netstamp(sk) &&
 450		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
 451			net_disable_timestamp();
 452	}
 453}
 454
 455
 456int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 457{
 458	unsigned long flags;
 459	struct sk_buff_head *list = &sk->sk_receive_queue;
 460
 461	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
 462		atomic_inc(&sk->sk_drops);
 463		trace_sock_rcvqueue_full(sk, skb);
 464		return -ENOMEM;
 465	}
 466
 467	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
 468		atomic_inc(&sk->sk_drops);
 469		return -ENOBUFS;
 470	}
 471
 472	skb->dev = NULL;
 473	skb_set_owner_r(skb, sk);
 474
 475	/* we escape from rcu protected region, make sure we dont leak
 476	 * a norefcounted dst
 477	 */
 478	skb_dst_force(skb);
 479
 480	spin_lock_irqsave(&list->lock, flags);
 481	sock_skb_set_dropcount(sk, skb);
 482	__skb_queue_tail(list, skb);
 483	spin_unlock_irqrestore(&list->lock, flags);
 484
 485	if (!sock_flag(sk, SOCK_DEAD))
 486		sk->sk_data_ready(sk);
 487	return 0;
 488}
 489EXPORT_SYMBOL(__sock_queue_rcv_skb);
 490
 491int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 492{
 493	int err;
 494
 495	err = sk_filter(sk, skb);
 496	if (err)
 497		return err;
 498
 499	return __sock_queue_rcv_skb(sk, skb);
 500}
 501EXPORT_SYMBOL(sock_queue_rcv_skb);
 502
 503int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
 504		     const int nested, unsigned int trim_cap, bool refcounted)
 505{
 506	int rc = NET_RX_SUCCESS;
 507
 508	if (sk_filter_trim_cap(sk, skb, trim_cap))
 509		goto discard_and_relse;
 510
 511	skb->dev = NULL;
 512
 513	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
 514		atomic_inc(&sk->sk_drops);
 515		goto discard_and_relse;
 516	}
 517	if (nested)
 518		bh_lock_sock_nested(sk);
 519	else
 520		bh_lock_sock(sk);
 521	if (!sock_owned_by_user(sk)) {
 522		/*
 523		 * trylock + unlock semantics:
 524		 */
 525		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 526
 527		rc = sk_backlog_rcv(sk, skb);
 528
 529		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
 530	} else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
 531		bh_unlock_sock(sk);
 532		atomic_inc(&sk->sk_drops);
 533		goto discard_and_relse;
 534	}
 535
 536	bh_unlock_sock(sk);
 537out:
 538	if (refcounted)
 539		sock_put(sk);
 540	return rc;
 541discard_and_relse:
 542	kfree_skb(skb);
 543	goto out;
 544}
 545EXPORT_SYMBOL(__sk_receive_skb);
 546
 547INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
 548							  u32));
 549INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
 550							   u32));
 551struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 552{
 553	struct dst_entry *dst = __sk_dst_get(sk);
 554
 555	if (dst && dst->obsolete &&
 556	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
 557			       dst, cookie) == NULL) {
 558		sk_tx_queue_clear(sk);
 559		sk->sk_dst_pending_confirm = 0;
 560		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
 561		dst_release(dst);
 562		return NULL;
 563	}
 564
 565	return dst;
 566}
 567EXPORT_SYMBOL(__sk_dst_check);
 568
 569struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 570{
 571	struct dst_entry *dst = sk_dst_get(sk);
 572
 573	if (dst && dst->obsolete &&
 574	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
 575			       dst, cookie) == NULL) {
 576		sk_dst_reset(sk);
 577		dst_release(dst);
 578		return NULL;
 579	}
 580
 581	return dst;
 582}
 583EXPORT_SYMBOL(sk_dst_check);
 584
 585static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
 586{
 587	int ret = -ENOPROTOOPT;
 588#ifdef CONFIG_NETDEVICES
 589	struct net *net = sock_net(sk);
 590
 591	/* Sorry... */
 592	ret = -EPERM;
 593	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
 594		goto out;
 595
 596	ret = -EINVAL;
 597	if (ifindex < 0)
 598		goto out;
 599
 600	sk->sk_bound_dev_if = ifindex;
 601	if (sk->sk_prot->rehash)
 602		sk->sk_prot->rehash(sk);
 603	sk_dst_reset(sk);
 604
 605	ret = 0;
 606
 607out:
 608#endif
 609
 610	return ret;
 611}
 612
 613int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
 614{
 615	int ret;
 616
 617	if (lock_sk)
 618		lock_sock(sk);
 619	ret = sock_bindtoindex_locked(sk, ifindex);
 620	if (lock_sk)
 621		release_sock(sk);
 622
 623	return ret;
 624}
 625EXPORT_SYMBOL(sock_bindtoindex);
 626
 627static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
 628{
 629	int ret = -ENOPROTOOPT;
 630#ifdef CONFIG_NETDEVICES
 631	struct net *net = sock_net(sk);
 632	char devname[IFNAMSIZ];
 633	int index;
 634
 635	ret = -EINVAL;
 636	if (optlen < 0)
 637		goto out;
 638
 639	/* Bind this socket to a particular device like "eth0",
 640	 * as specified in the passed interface name. If the
 641	 * name is "" or the option length is zero the socket
 642	 * is not bound.
 643	 */
 644	if (optlen > IFNAMSIZ - 1)
 645		optlen = IFNAMSIZ - 1;
 646	memset(devname, 0, sizeof(devname));
 647
 648	ret = -EFAULT;
 649	if (copy_from_sockptr(devname, optval, optlen))
 650		goto out;
 651
 652	index = 0;
 653	if (devname[0] != '\0') {
 654		struct net_device *dev;
 655
 656		rcu_read_lock();
 657		dev = dev_get_by_name_rcu(net, devname);
 658		if (dev)
 659			index = dev->ifindex;
 660		rcu_read_unlock();
 661		ret = -ENODEV;
 662		if (!dev)
 663			goto out;
 664	}
 665
 666	return sock_bindtoindex(sk, index, true);
 667out:
 668#endif
 669
 670	return ret;
 671}
 672
 673static int sock_getbindtodevice(struct sock *sk, char __user *optval,
 674				int __user *optlen, int len)
 675{
 676	int ret = -ENOPROTOOPT;
 677#ifdef CONFIG_NETDEVICES
 678	struct net *net = sock_net(sk);
 679	char devname[IFNAMSIZ];
 680
 681	if (sk->sk_bound_dev_if == 0) {
 682		len = 0;
 683		goto zero;
 684	}
 685
 686	ret = -EINVAL;
 687	if (len < IFNAMSIZ)
 688		goto out;
 689
 690	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
 691	if (ret)
 692		goto out;
 693
 694	len = strlen(devname) + 1;
 695
 696	ret = -EFAULT;
 697	if (copy_to_user(optval, devname, len))
 698		goto out;
 699
 700zero:
 701	ret = -EFAULT;
 702	if (put_user(len, optlen))
 703		goto out;
 704
 705	ret = 0;
 706
 707out:
 708#endif
 709
 710	return ret;
 711}
 712
 713bool sk_mc_loop(struct sock *sk)
 714{
 715	if (dev_recursion_level())
 716		return false;
 717	if (!sk)
 718		return true;
 719	switch (sk->sk_family) {
 720	case AF_INET:
 721		return inet_sk(sk)->mc_loop;
 722#if IS_ENABLED(CONFIG_IPV6)
 723	case AF_INET6:
 724		return inet6_sk(sk)->mc_loop;
 725#endif
 726	}
 727	WARN_ON_ONCE(1);
 728	return true;
 729}
 730EXPORT_SYMBOL(sk_mc_loop);
 731
 732void sock_set_reuseaddr(struct sock *sk)
 733{
 734	lock_sock(sk);
 735	sk->sk_reuse = SK_CAN_REUSE;
 736	release_sock(sk);
 737}
 738EXPORT_SYMBOL(sock_set_reuseaddr);
 739
 740void sock_set_reuseport(struct sock *sk)
 741{
 742	lock_sock(sk);
 743	sk->sk_reuseport = true;
 744	release_sock(sk);
 745}
 746EXPORT_SYMBOL(sock_set_reuseport);
 747
 748void sock_no_linger(struct sock *sk)
 749{
 750	lock_sock(sk);
 751	sk->sk_lingertime = 0;
 752	sock_set_flag(sk, SOCK_LINGER);
 753	release_sock(sk);
 754}
 755EXPORT_SYMBOL(sock_no_linger);
 756
 757void sock_set_priority(struct sock *sk, u32 priority)
 758{
 759	lock_sock(sk);
 760	sk->sk_priority = priority;
 761	release_sock(sk);
 762}
 763EXPORT_SYMBOL(sock_set_priority);
 764
 765void sock_set_sndtimeo(struct sock *sk, s64 secs)
 766{
 767	lock_sock(sk);
 768	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
 769		sk->sk_sndtimeo = secs * HZ;
 770	else
 771		sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
 772	release_sock(sk);
 773}
 774EXPORT_SYMBOL(sock_set_sndtimeo);
 775
 776static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
 777{
 778	if (val)  {
 779		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
 780		sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
 781		sock_set_flag(sk, SOCK_RCVTSTAMP);
 782		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
 783	} else {
 784		sock_reset_flag(sk, SOCK_RCVTSTAMP);
 785		sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 
 786	}
 787}
 788
 789void sock_enable_timestamps(struct sock *sk)
 790{
 791	lock_sock(sk);
 792	__sock_set_timestamps(sk, true, false, true);
 793	release_sock(sk);
 794}
 795EXPORT_SYMBOL(sock_enable_timestamps);
 796
 797void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
 798{
 799	switch (optname) {
 800	case SO_TIMESTAMP_OLD:
 801		__sock_set_timestamps(sk, valbool, false, false);
 802		break;
 803	case SO_TIMESTAMP_NEW:
 804		__sock_set_timestamps(sk, valbool, true, false);
 805		break;
 806	case SO_TIMESTAMPNS_OLD:
 807		__sock_set_timestamps(sk, valbool, false, true);
 808		break;
 809	case SO_TIMESTAMPNS_NEW:
 810		__sock_set_timestamps(sk, valbool, true, true);
 811		break;
 812	}
 813}
 814
 815static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
 816{
 817	struct net *net = sock_net(sk);
 818	struct net_device *dev = NULL;
 819	bool match = false;
 820	int *vclock_index;
 821	int i, num;
 822
 823	if (sk->sk_bound_dev_if)
 824		dev = dev_get_by_index(net, sk->sk_bound_dev_if);
 825
 826	if (!dev) {
 827		pr_err("%s: sock not bind to device\n", __func__);
 828		return -EOPNOTSUPP;
 829	}
 830
 831	num = ethtool_get_phc_vclocks(dev, &vclock_index);
 832	for (i = 0; i < num; i++) {
 833		if (*(vclock_index + i) == phc_index) {
 834			match = true;
 835			break;
 836		}
 837	}
 838
 839	if (num > 0)
 840		kfree(vclock_index);
 841
 842	if (!match)
 843		return -EINVAL;
 844
 845	sk->sk_bind_phc = phc_index;
 846
 847	return 0;
 848}
 849
 850int sock_set_timestamping(struct sock *sk, int optname,
 851			  struct so_timestamping timestamping)
 852{
 853	int val = timestamping.flags;
 854	int ret;
 855
 856	if (val & ~SOF_TIMESTAMPING_MASK)
 857		return -EINVAL;
 858
 859	if (val & SOF_TIMESTAMPING_OPT_ID &&
 860	    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
 861		if (sk->sk_protocol == IPPROTO_TCP &&
 862		    sk->sk_type == SOCK_STREAM) {
 863			if ((1 << sk->sk_state) &
 864			    (TCPF_CLOSE | TCPF_LISTEN))
 865				return -EINVAL;
 866			sk->sk_tskey = tcp_sk(sk)->snd_una;
 867		} else {
 868			sk->sk_tskey = 0;
 869		}
 870	}
 871
 872	if (val & SOF_TIMESTAMPING_OPT_STATS &&
 873	    !(val & SOF_TIMESTAMPING_OPT_TSONLY))
 874		return -EINVAL;
 875
 876	if (val & SOF_TIMESTAMPING_BIND_PHC) {
 877		ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
 878		if (ret)
 879			return ret;
 880	}
 881
 882	sk->sk_tsflags = val;
 883	sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
 884
 885	if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 886		sock_enable_timestamp(sk,
 887				      SOCK_TIMESTAMPING_RX_SOFTWARE);
 888	else
 889		sock_disable_timestamp(sk,
 890				       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
 891	return 0;
 892}
 893
 894void sock_set_keepalive(struct sock *sk)
 895{
 896	lock_sock(sk);
 897	if (sk->sk_prot->keepalive)
 898		sk->sk_prot->keepalive(sk, true);
 899	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
 900	release_sock(sk);
 901}
 902EXPORT_SYMBOL(sock_set_keepalive);
 903
 904static void __sock_set_rcvbuf(struct sock *sk, int val)
 905{
 906	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
 907	 * as a negative value.
 908	 */
 909	val = min_t(int, val, INT_MAX / 2);
 910	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 911
 912	/* We double it on the way in to account for "struct sk_buff" etc.
 913	 * overhead.   Applications assume that the SO_RCVBUF setting they make
 914	 * will allow that much actual data to be received on that socket.
 915	 *
 916	 * Applications are unaware that "struct sk_buff" and other overheads
 917	 * allocate from the receive buffer during socket buffer allocation.
 918	 *
 919	 * And after considering the possible alternatives, returning the value
 920	 * we actually used in getsockopt is the most desirable behavior.
 921	 */
 922	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
 923}
 924
 925void sock_set_rcvbuf(struct sock *sk, int val)
 926{
 927	lock_sock(sk);
 928	__sock_set_rcvbuf(sk, val);
 929	release_sock(sk);
 930}
 931EXPORT_SYMBOL(sock_set_rcvbuf);
 932
 933static void __sock_set_mark(struct sock *sk, u32 val)
 934{
 935	if (val != sk->sk_mark) {
 936		sk->sk_mark = val;
 937		sk_dst_reset(sk);
 938	}
 939}
 940
 941void sock_set_mark(struct sock *sk, u32 val)
 942{
 943	lock_sock(sk);
 944	__sock_set_mark(sk, val);
 945	release_sock(sk);
 946}
 947EXPORT_SYMBOL(sock_set_mark);
 948
 949/*
 950 *	This is meant for all protocols to use and covers goings on
 951 *	at the socket level. Everything here is generic.
 952 */
 953
 954int sock_setsockopt(struct socket *sock, int level, int optname,
 955		    sockptr_t optval, unsigned int optlen)
 956{
 957	struct so_timestamping timestamping;
 958	struct sock_txtime sk_txtime;
 959	struct sock *sk = sock->sk;
 960	int val;
 961	int valbool;
 962	struct linger ling;
 963	int ret = 0;
 964
 965	/*
 966	 *	Options without arguments
 967	 */
 968
 969	if (optname == SO_BINDTODEVICE)
 970		return sock_setbindtodevice(sk, optval, optlen);
 971
 972	if (optlen < sizeof(int))
 973		return -EINVAL;
 974
 975	if (copy_from_sockptr(&val, optval, sizeof(val)))
 976		return -EFAULT;
 977
 978	valbool = val ? 1 : 0;
 979
 980	lock_sock(sk);
 981
 982	switch (optname) {
 983	case SO_DEBUG:
 984		if (val && !capable(CAP_NET_ADMIN))
 985			ret = -EACCES;
 986		else
 987			sock_valbool_flag(sk, SOCK_DBG, valbool);
 988		break;
 989	case SO_REUSEADDR:
 990		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
 991		break;
 992	case SO_REUSEPORT:
 993		sk->sk_reuseport = valbool;
 994		break;
 995	case SO_TYPE:
 996	case SO_PROTOCOL:
 997	case SO_DOMAIN:
 998	case SO_ERROR:
 999		ret = -ENOPROTOOPT;
1000		break;
1001	case SO_DONTROUTE:
1002		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1003		sk_dst_reset(sk);
1004		break;
1005	case SO_BROADCAST:
1006		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1007		break;
1008	case SO_SNDBUF:
1009		/* Don't error on this BSD doesn't and if you think
1010		 * about it this is right. Otherwise apps have to
1011		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1012		 * are treated in BSD as hints
1013		 */
1014		val = min_t(u32, val, sysctl_wmem_max);
1015set_sndbuf:
1016		/* Ensure val * 2 fits into an int, to prevent max_t()
1017		 * from treating it as a negative value.
1018		 */
1019		val = min_t(int, val, INT_MAX / 2);
1020		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1021		WRITE_ONCE(sk->sk_sndbuf,
1022			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
1023		/* Wake up sending tasks if we upped the value. */
1024		sk->sk_write_space(sk);
1025		break;
1026
1027	case SO_SNDBUFFORCE:
1028		if (!capable(CAP_NET_ADMIN)) {
1029			ret = -EPERM;
1030			break;
1031		}
1032
1033		/* No negative values (to prevent underflow, as val will be
1034		 * multiplied by 2).
1035		 */
1036		if (val < 0)
1037			val = 0;
1038		goto set_sndbuf;
1039
1040	case SO_RCVBUF:
1041		/* Don't error on this BSD doesn't and if you think
1042		 * about it this is right. Otherwise apps have to
1043		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1044		 * are treated in BSD as hints
1045		 */
1046		__sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
1047		break;
1048
1049	case SO_RCVBUFFORCE:
1050		if (!capable(CAP_NET_ADMIN)) {
1051			ret = -EPERM;
1052			break;
1053		}
1054
1055		/* No negative values (to prevent underflow, as val will be
1056		 * multiplied by 2).
1057		 */
1058		__sock_set_rcvbuf(sk, max(val, 0));
1059		break;
1060
1061	case SO_KEEPALIVE:
1062		if (sk->sk_prot->keepalive)
1063			sk->sk_prot->keepalive(sk, valbool);
1064		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1065		break;
1066
1067	case SO_OOBINLINE:
1068		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1069		break;
1070
1071	case SO_NO_CHECK:
1072		sk->sk_no_check_tx = valbool;
1073		break;
1074
1075	case SO_PRIORITY:
1076		if ((val >= 0 && val <= 6) ||
1077		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1078			sk->sk_priority = val;
1079		else
1080			ret = -EPERM;
1081		break;
1082
1083	case SO_LINGER:
1084		if (optlen < sizeof(ling)) {
1085			ret = -EINVAL;	/* 1003.1g */
1086			break;
1087		}
1088		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1089			ret = -EFAULT;
1090			break;
1091		}
1092		if (!ling.l_onoff)
1093			sock_reset_flag(sk, SOCK_LINGER);
1094		else {
1095#if (BITS_PER_LONG == 32)
1096			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1097				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1098			else
1099#endif
1100				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1101			sock_set_flag(sk, SOCK_LINGER);
1102		}
1103		break;
1104
1105	case SO_BSDCOMPAT:
 
1106		break;
1107
1108	case SO_PASSCRED:
1109		if (valbool)
1110			set_bit(SOCK_PASSCRED, &sock->flags);
1111		else
1112			clear_bit(SOCK_PASSCRED, &sock->flags);
1113		break;
1114
1115	case SO_TIMESTAMP_OLD:
 
 
1116	case SO_TIMESTAMP_NEW:
 
 
1117	case SO_TIMESTAMPNS_OLD:
 
 
1118	case SO_TIMESTAMPNS_NEW:
1119		sock_set_timestamp(sk, optname, valbool);
1120		break;
1121
1122	case SO_TIMESTAMPING_NEW:
 
 
1123	case SO_TIMESTAMPING_OLD:
1124		if (optlen == sizeof(timestamping)) {
1125			if (copy_from_sockptr(&timestamping, optval,
1126					      sizeof(timestamping))) {
1127				ret = -EFAULT;
1128				break;
 
 
 
 
 
 
 
 
 
 
 
 
1129			}
1130		} else {
1131			memset(&timestamping, 0, sizeof(timestamping));
1132			timestamping.flags = val;
1133		}
1134		ret = sock_set_timestamping(sk, optname, timestamping);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1135		break;
1136
1137	case SO_RCVLOWAT:
1138		if (val < 0)
1139			val = INT_MAX;
1140		if (sock->ops->set_rcvlowat)
1141			ret = sock->ops->set_rcvlowat(sk, val);
1142		else
1143			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1144		break;
1145
1146	case SO_RCVTIMEO_OLD:
1147	case SO_RCVTIMEO_NEW:
1148		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1149				       optlen, optname == SO_RCVTIMEO_OLD);
1150		break;
1151
1152	case SO_SNDTIMEO_OLD:
1153	case SO_SNDTIMEO_NEW:
1154		ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1155				       optlen, optname == SO_SNDTIMEO_OLD);
1156		break;
1157
1158	case SO_ATTACH_FILTER: {
1159		struct sock_fprog fprog;
1160
1161		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1162		if (!ret)
1163			ret = sk_attach_filter(&fprog, sk);
1164		break;
1165	}
1166	case SO_ATTACH_BPF:
1167		ret = -EINVAL;
1168		if (optlen == sizeof(u32)) {
1169			u32 ufd;
1170
1171			ret = -EFAULT;
1172			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1173				break;
1174
1175			ret = sk_attach_bpf(ufd, sk);
1176		}
1177		break;
1178
1179	case SO_ATTACH_REUSEPORT_CBPF: {
1180		struct sock_fprog fprog;
1181
1182		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1183		if (!ret)
1184			ret = sk_reuseport_attach_filter(&fprog, sk);
1185		break;
1186	}
1187	case SO_ATTACH_REUSEPORT_EBPF:
1188		ret = -EINVAL;
1189		if (optlen == sizeof(u32)) {
1190			u32 ufd;
1191
1192			ret = -EFAULT;
1193			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1194				break;
1195
1196			ret = sk_reuseport_attach_bpf(ufd, sk);
1197		}
1198		break;
1199
1200	case SO_DETACH_REUSEPORT_BPF:
1201		ret = reuseport_detach_prog(sk);
1202		break;
1203
1204	case SO_DETACH_FILTER:
1205		ret = sk_detach_filter(sk);
1206		break;
1207
1208	case SO_LOCK_FILTER:
1209		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1210			ret = -EPERM;
1211		else
1212			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1213		break;
1214
1215	case SO_PASSSEC:
1216		if (valbool)
1217			set_bit(SOCK_PASSSEC, &sock->flags);
1218		else
1219			clear_bit(SOCK_PASSSEC, &sock->flags);
1220		break;
1221	case SO_MARK:
1222		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1223			ret = -EPERM;
1224			break;
 
 
1225		}
1226
1227		__sock_set_mark(sk, val);
1228		break;
1229
1230	case SO_RXQ_OVFL:
1231		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1232		break;
1233
1234	case SO_WIFI_STATUS:
1235		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1236		break;
1237
1238	case SO_PEEK_OFF:
1239		if (sock->ops->set_peek_off)
1240			ret = sock->ops->set_peek_off(sk, val);
1241		else
1242			ret = -EOPNOTSUPP;
1243		break;
1244
1245	case SO_NOFCS:
1246		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1247		break;
1248
1249	case SO_SELECT_ERR_QUEUE:
1250		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1251		break;
1252
1253#ifdef CONFIG_NET_RX_BUSY_POLL
1254	case SO_BUSY_POLL:
1255		/* allow unprivileged users to decrease the value */
1256		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1257			ret = -EPERM;
1258		else {
1259			if (val < 0)
1260				ret = -EINVAL;
1261			else
1262				WRITE_ONCE(sk->sk_ll_usec, val);
1263		}
1264		break;
1265	case SO_PREFER_BUSY_POLL:
1266		if (valbool && !capable(CAP_NET_ADMIN))
1267			ret = -EPERM;
1268		else
1269			WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1270		break;
1271	case SO_BUSY_POLL_BUDGET:
1272		if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) {
1273			ret = -EPERM;
1274		} else {
1275			if (val < 0 || val > U16_MAX)
1276				ret = -EINVAL;
1277			else
1278				WRITE_ONCE(sk->sk_busy_poll_budget, val);
1279		}
1280		break;
1281#endif
1282
1283	case SO_MAX_PACING_RATE:
1284		{
1285		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1286
1287		if (sizeof(ulval) != sizeof(val) &&
1288		    optlen >= sizeof(ulval) &&
1289		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1290			ret = -EFAULT;
1291			break;
1292		}
1293		if (ulval != ~0UL)
1294			cmpxchg(&sk->sk_pacing_status,
1295				SK_PACING_NONE,
1296				SK_PACING_NEEDED);
1297		sk->sk_max_pacing_rate = ulval;
1298		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1299		break;
1300		}
1301	case SO_INCOMING_CPU:
1302		WRITE_ONCE(sk->sk_incoming_cpu, val);
1303		break;
1304
1305	case SO_CNX_ADVICE:
1306		if (val == 1)
1307			dst_negative_advice(sk);
1308		break;
1309
1310	case SO_ZEROCOPY:
1311		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1312			if (!((sk->sk_type == SOCK_STREAM &&
1313			       sk->sk_protocol == IPPROTO_TCP) ||
1314			      (sk->sk_type == SOCK_DGRAM &&
1315			       sk->sk_protocol == IPPROTO_UDP)))
1316				ret = -ENOTSUPP;
1317		} else if (sk->sk_family != PF_RDS) {
1318			ret = -ENOTSUPP;
1319		}
1320		if (!ret) {
1321			if (val < 0 || val > 1)
1322				ret = -EINVAL;
1323			else
1324				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1325		}
1326		break;
1327
1328	case SO_TXTIME:
1329		if (optlen != sizeof(struct sock_txtime)) {
1330			ret = -EINVAL;
1331			break;
1332		} else if (copy_from_sockptr(&sk_txtime, optval,
1333			   sizeof(struct sock_txtime))) {
1334			ret = -EFAULT;
1335			break;
1336		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1337			ret = -EINVAL;
1338			break;
1339		}
1340		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1341		 * scheduler has enough safe guards.
1342		 */
1343		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1344		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1345			ret = -EPERM;
1346			break;
1347		}
1348		sock_valbool_flag(sk, SOCK_TXTIME, true);
1349		sk->sk_clockid = sk_txtime.clockid;
1350		sk->sk_txtime_deadline_mode =
1351			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1352		sk->sk_txtime_report_errors =
1353			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1354		break;
1355
1356	case SO_BINDTOIFINDEX:
1357		ret = sock_bindtoindex_locked(sk, val);
1358		break;
1359
1360	default:
1361		ret = -ENOPROTOOPT;
1362		break;
1363	}
1364	release_sock(sk);
1365	return ret;
1366}
1367EXPORT_SYMBOL(sock_setsockopt);
1368
1369static const struct cred *sk_get_peer_cred(struct sock *sk)
1370{
1371	const struct cred *cred;
1372
1373	spin_lock(&sk->sk_peer_lock);
1374	cred = get_cred(sk->sk_peer_cred);
1375	spin_unlock(&sk->sk_peer_lock);
1376
1377	return cred;
1378}
1379
1380static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1381			  struct ucred *ucred)
1382{
1383	ucred->pid = pid_vnr(pid);
1384	ucred->uid = ucred->gid = -1;
1385	if (cred) {
1386		struct user_namespace *current_ns = current_user_ns();
1387
1388		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1389		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1390	}
1391}
1392
1393static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1394{
1395	struct user_namespace *user_ns = current_user_ns();
1396	int i;
1397
1398	for (i = 0; i < src->ngroups; i++)
1399		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1400			return -EFAULT;
1401
1402	return 0;
1403}
1404
1405int sock_getsockopt(struct socket *sock, int level, int optname,
1406		    char __user *optval, int __user *optlen)
1407{
1408	struct sock *sk = sock->sk;
1409
1410	union {
1411		int val;
1412		u64 val64;
1413		unsigned long ulval;
1414		struct linger ling;
1415		struct old_timeval32 tm32;
1416		struct __kernel_old_timeval tm;
1417		struct  __kernel_sock_timeval stm;
1418		struct sock_txtime txtime;
1419		struct so_timestamping timestamping;
1420	} v;
1421
1422	int lv = sizeof(int);
1423	int len;
1424
1425	if (get_user(len, optlen))
1426		return -EFAULT;
1427	if (len < 0)
1428		return -EINVAL;
1429
1430	memset(&v, 0, sizeof(v));
1431
1432	switch (optname) {
1433	case SO_DEBUG:
1434		v.val = sock_flag(sk, SOCK_DBG);
1435		break;
1436
1437	case SO_DONTROUTE:
1438		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1439		break;
1440
1441	case SO_BROADCAST:
1442		v.val = sock_flag(sk, SOCK_BROADCAST);
1443		break;
1444
1445	case SO_SNDBUF:
1446		v.val = sk->sk_sndbuf;
1447		break;
1448
1449	case SO_RCVBUF:
1450		v.val = sk->sk_rcvbuf;
1451		break;
1452
1453	case SO_REUSEADDR:
1454		v.val = sk->sk_reuse;
1455		break;
1456
1457	case SO_REUSEPORT:
1458		v.val = sk->sk_reuseport;
1459		break;
1460
1461	case SO_KEEPALIVE:
1462		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1463		break;
1464
1465	case SO_TYPE:
1466		v.val = sk->sk_type;
1467		break;
1468
1469	case SO_PROTOCOL:
1470		v.val = sk->sk_protocol;
1471		break;
1472
1473	case SO_DOMAIN:
1474		v.val = sk->sk_family;
1475		break;
1476
1477	case SO_ERROR:
1478		v.val = -sock_error(sk);
1479		if (v.val == 0)
1480			v.val = xchg(&sk->sk_err_soft, 0);
1481		break;
1482
1483	case SO_OOBINLINE:
1484		v.val = sock_flag(sk, SOCK_URGINLINE);
1485		break;
1486
1487	case SO_NO_CHECK:
1488		v.val = sk->sk_no_check_tx;
1489		break;
1490
1491	case SO_PRIORITY:
1492		v.val = sk->sk_priority;
1493		break;
1494
1495	case SO_LINGER:
1496		lv		= sizeof(v.ling);
1497		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1498		v.ling.l_linger	= sk->sk_lingertime / HZ;
1499		break;
1500
1501	case SO_BSDCOMPAT:
 
1502		break;
1503
1504	case SO_TIMESTAMP_OLD:
1505		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1506				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1507				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1508		break;
1509
1510	case SO_TIMESTAMPNS_OLD:
1511		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1512		break;
1513
1514	case SO_TIMESTAMP_NEW:
1515		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1516		break;
1517
1518	case SO_TIMESTAMPNS_NEW:
1519		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1520		break;
1521
1522	case SO_TIMESTAMPING_OLD:
1523		lv = sizeof(v.timestamping);
1524		v.timestamping.flags = sk->sk_tsflags;
1525		v.timestamping.bind_phc = sk->sk_bind_phc;
1526		break;
1527
1528	case SO_RCVTIMEO_OLD:
1529	case SO_RCVTIMEO_NEW:
1530		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1531		break;
1532
1533	case SO_SNDTIMEO_OLD:
1534	case SO_SNDTIMEO_NEW:
1535		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1536		break;
1537
1538	case SO_RCVLOWAT:
1539		v.val = sk->sk_rcvlowat;
1540		break;
1541
1542	case SO_SNDLOWAT:
1543		v.val = 1;
1544		break;
1545
1546	case SO_PASSCRED:
1547		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1548		break;
1549
1550	case SO_PEERCRED:
1551	{
1552		struct ucred peercred;
1553		if (len > sizeof(peercred))
1554			len = sizeof(peercred);
1555
1556		spin_lock(&sk->sk_peer_lock);
1557		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1558		spin_unlock(&sk->sk_peer_lock);
1559
1560		if (copy_to_user(optval, &peercred, len))
1561			return -EFAULT;
1562		goto lenout;
1563	}
1564
1565	case SO_PEERGROUPS:
1566	{
1567		const struct cred *cred;
1568		int ret, n;
1569
1570		cred = sk_get_peer_cred(sk);
1571		if (!cred)
1572			return -ENODATA;
1573
1574		n = cred->group_info->ngroups;
1575		if (len < n * sizeof(gid_t)) {
1576			len = n * sizeof(gid_t);
1577			put_cred(cred);
1578			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1579		}
1580		len = n * sizeof(gid_t);
1581
1582		ret = groups_to_user((gid_t __user *)optval, cred->group_info);
1583		put_cred(cred);
1584		if (ret)
1585			return ret;
1586		goto lenout;
1587	}
1588
1589	case SO_PEERNAME:
1590	{
1591		char address[128];
1592
1593		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1594		if (lv < 0)
1595			return -ENOTCONN;
1596		if (lv < len)
1597			return -EINVAL;
1598		if (copy_to_user(optval, address, len))
1599			return -EFAULT;
1600		goto lenout;
1601	}
1602
1603	/* Dubious BSD thing... Probably nobody even uses it, but
1604	 * the UNIX standard wants it for whatever reason... -DaveM
1605	 */
1606	case SO_ACCEPTCONN:
1607		v.val = sk->sk_state == TCP_LISTEN;
1608		break;
1609
1610	case SO_PASSSEC:
1611		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1612		break;
1613
1614	case SO_PEERSEC:
1615		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1616
1617	case SO_MARK:
1618		v.val = sk->sk_mark;
1619		break;
1620
1621	case SO_RXQ_OVFL:
1622		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1623		break;
1624
1625	case SO_WIFI_STATUS:
1626		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1627		break;
1628
1629	case SO_PEEK_OFF:
1630		if (!sock->ops->set_peek_off)
1631			return -EOPNOTSUPP;
1632
1633		v.val = sk->sk_peek_off;
1634		break;
1635	case SO_NOFCS:
1636		v.val = sock_flag(sk, SOCK_NOFCS);
1637		break;
1638
1639	case SO_BINDTODEVICE:
1640		return sock_getbindtodevice(sk, optval, optlen, len);
1641
1642	case SO_GET_FILTER:
1643		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1644		if (len < 0)
1645			return len;
1646
1647		goto lenout;
1648
1649	case SO_LOCK_FILTER:
1650		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1651		break;
1652
1653	case SO_BPF_EXTENSIONS:
1654		v.val = bpf_tell_extensions();
1655		break;
1656
1657	case SO_SELECT_ERR_QUEUE:
1658		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1659		break;
1660
1661#ifdef CONFIG_NET_RX_BUSY_POLL
1662	case SO_BUSY_POLL:
1663		v.val = sk->sk_ll_usec;
1664		break;
1665	case SO_PREFER_BUSY_POLL:
1666		v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1667		break;
1668#endif
1669
1670	case SO_MAX_PACING_RATE:
1671		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1672			lv = sizeof(v.ulval);
1673			v.ulval = sk->sk_max_pacing_rate;
1674		} else {
1675			/* 32bit version */
1676			v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1677		}
1678		break;
1679
1680	case SO_INCOMING_CPU:
1681		v.val = READ_ONCE(sk->sk_incoming_cpu);
1682		break;
1683
1684	case SO_MEMINFO:
1685	{
1686		u32 meminfo[SK_MEMINFO_VARS];
1687
1688		sk_get_meminfo(sk, meminfo);
1689
1690		len = min_t(unsigned int, len, sizeof(meminfo));
1691		if (copy_to_user(optval, &meminfo, len))
1692			return -EFAULT;
1693
1694		goto lenout;
1695	}
1696
1697#ifdef CONFIG_NET_RX_BUSY_POLL
1698	case SO_INCOMING_NAPI_ID:
1699		v.val = READ_ONCE(sk->sk_napi_id);
1700
1701		/* aggregate non-NAPI IDs down to 0 */
1702		if (v.val < MIN_NAPI_ID)
1703			v.val = 0;
1704
1705		break;
1706#endif
1707
1708	case SO_COOKIE:
1709		lv = sizeof(u64);
1710		if (len < lv)
1711			return -EINVAL;
1712		v.val64 = sock_gen_cookie(sk);
1713		break;
1714
1715	case SO_ZEROCOPY:
1716		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1717		break;
1718
1719	case SO_TXTIME:
1720		lv = sizeof(v.txtime);
1721		v.txtime.clockid = sk->sk_clockid;
1722		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1723				  SOF_TXTIME_DEADLINE_MODE : 0;
1724		v.txtime.flags |= sk->sk_txtime_report_errors ?
1725				  SOF_TXTIME_REPORT_ERRORS : 0;
1726		break;
1727
1728	case SO_BINDTOIFINDEX:
1729		v.val = sk->sk_bound_dev_if;
1730		break;
1731
1732	case SO_NETNS_COOKIE:
1733		lv = sizeof(u64);
1734		if (len != lv)
1735			return -EINVAL;
1736		v.val64 = sock_net(sk)->net_cookie;
1737		break;
1738
1739	default:
1740		/* We implement the SO_SNDLOWAT etc to not be settable
1741		 * (1003.1g 7).
1742		 */
1743		return -ENOPROTOOPT;
1744	}
1745
1746	if (len > lv)
1747		len = lv;
1748	if (copy_to_user(optval, &v, len))
1749		return -EFAULT;
1750lenout:
1751	if (put_user(len, optlen))
1752		return -EFAULT;
1753	return 0;
1754}
1755
1756/*
1757 * Initialize an sk_lock.
1758 *
1759 * (We also register the sk_lock with the lock validator.)
1760 */
1761static inline void sock_lock_init(struct sock *sk)
1762{
1763	if (sk->sk_kern_sock)
1764		sock_lock_init_class_and_name(
1765			sk,
1766			af_family_kern_slock_key_strings[sk->sk_family],
1767			af_family_kern_slock_keys + sk->sk_family,
1768			af_family_kern_key_strings[sk->sk_family],
1769			af_family_kern_keys + sk->sk_family);
1770	else
1771		sock_lock_init_class_and_name(
1772			sk,
1773			af_family_slock_key_strings[sk->sk_family],
1774			af_family_slock_keys + sk->sk_family,
1775			af_family_key_strings[sk->sk_family],
1776			af_family_keys + sk->sk_family);
1777}
1778
1779/*
1780 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1781 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1782 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1783 */
1784static void sock_copy(struct sock *nsk, const struct sock *osk)
1785{
1786	const struct proto *prot = READ_ONCE(osk->sk_prot);
1787#ifdef CONFIG_SECURITY_NETWORK
1788	void *sptr = nsk->sk_security;
1789#endif
1790
1791	/* If we move sk_tx_queue_mapping out of the private section,
1792	 * we must check if sk_tx_queue_clear() is called after
1793	 * sock_copy() in sk_clone_lock().
1794	 */
1795	BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
1796		     offsetof(struct sock, sk_dontcopy_begin) ||
1797		     offsetof(struct sock, sk_tx_queue_mapping) >=
1798		     offsetof(struct sock, sk_dontcopy_end));
1799
1800	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1801
1802	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1803	       prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1804
1805#ifdef CONFIG_SECURITY_NETWORK
1806	nsk->sk_security = sptr;
1807	security_sk_clone(osk, nsk);
1808#endif
1809}
1810
1811static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1812		int family)
1813{
1814	struct sock *sk;
1815	struct kmem_cache *slab;
1816
1817	slab = prot->slab;
1818	if (slab != NULL) {
1819		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1820		if (!sk)
1821			return sk;
1822		if (want_init_on_alloc(priority))
1823			sk_prot_clear_nulls(sk, prot->obj_size);
1824	} else
1825		sk = kmalloc(prot->obj_size, priority);
1826
1827	if (sk != NULL) {
1828		if (security_sk_alloc(sk, family, priority))
1829			goto out_free;
1830
1831		if (!try_module_get(prot->owner))
1832			goto out_free_sec;
 
1833	}
1834
1835	return sk;
1836
1837out_free_sec:
1838	security_sk_free(sk);
1839out_free:
1840	if (slab != NULL)
1841		kmem_cache_free(slab, sk);
1842	else
1843		kfree(sk);
1844	return NULL;
1845}
1846
1847static void sk_prot_free(struct proto *prot, struct sock *sk)
1848{
1849	struct kmem_cache *slab;
1850	struct module *owner;
1851
1852	owner = prot->owner;
1853	slab = prot->slab;
1854
1855	cgroup_sk_free(&sk->sk_cgrp_data);
1856	mem_cgroup_sk_free(sk);
1857	security_sk_free(sk);
1858	if (slab != NULL)
1859		kmem_cache_free(slab, sk);
1860	else
1861		kfree(sk);
1862	module_put(owner);
1863}
1864
1865/**
1866 *	sk_alloc - All socket objects are allocated here
1867 *	@net: the applicable net namespace
1868 *	@family: protocol family
1869 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1870 *	@prot: struct proto associated with this new sock instance
1871 *	@kern: is this to be a kernel socket?
1872 */
1873struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1874		      struct proto *prot, int kern)
1875{
1876	struct sock *sk;
1877
1878	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1879	if (sk) {
1880		sk->sk_family = family;
1881		/*
1882		 * See comment in struct sock definition to understand
1883		 * why we need sk_prot_creator -acme
1884		 */
1885		sk->sk_prot = sk->sk_prot_creator = prot;
1886		sk->sk_kern_sock = kern;
1887		sock_lock_init(sk);
1888		sk->sk_net_refcnt = kern ? 0 : 1;
1889		if (likely(sk->sk_net_refcnt)) {
1890			get_net(net);
1891			sock_inuse_add(net, 1);
1892		}
1893
1894		sock_net_set(sk, net);
1895		refcount_set(&sk->sk_wmem_alloc, 1);
1896
1897		mem_cgroup_sk_alloc(sk);
1898		cgroup_sk_alloc(&sk->sk_cgrp_data);
1899		sock_update_classid(&sk->sk_cgrp_data);
1900		sock_update_netprioidx(&sk->sk_cgrp_data);
1901		sk_tx_queue_clear(sk);
1902	}
1903
1904	return sk;
1905}
1906EXPORT_SYMBOL(sk_alloc);
1907
1908/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1909 * grace period. This is the case for UDP sockets and TCP listeners.
1910 */
1911static void __sk_destruct(struct rcu_head *head)
1912{
1913	struct sock *sk = container_of(head, struct sock, sk_rcu);
1914	struct sk_filter *filter;
1915
1916	if (sk->sk_destruct)
1917		sk->sk_destruct(sk);
1918
1919	filter = rcu_dereference_check(sk->sk_filter,
1920				       refcount_read(&sk->sk_wmem_alloc) == 0);
1921	if (filter) {
1922		sk_filter_uncharge(sk, filter);
1923		RCU_INIT_POINTER(sk->sk_filter, NULL);
1924	}
1925
1926	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1927
1928#ifdef CONFIG_BPF_SYSCALL
1929	bpf_sk_storage_free(sk);
1930#endif
1931
1932	if (atomic_read(&sk->sk_omem_alloc))
1933		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1934			 __func__, atomic_read(&sk->sk_omem_alloc));
1935
1936	if (sk->sk_frag.page) {
1937		put_page(sk->sk_frag.page);
1938		sk->sk_frag.page = NULL;
1939	}
1940
1941	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
1942	put_cred(sk->sk_peer_cred);
1943	put_pid(sk->sk_peer_pid);
1944
1945	if (likely(sk->sk_net_refcnt))
1946		put_net(sock_net(sk));
1947	sk_prot_free(sk->sk_prot_creator, sk);
1948}
1949
1950void sk_destruct(struct sock *sk)
1951{
1952	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1953
1954	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1955		reuseport_detach_sock(sk);
1956		use_call_rcu = true;
1957	}
1958
1959	if (use_call_rcu)
1960		call_rcu(&sk->sk_rcu, __sk_destruct);
1961	else
1962		__sk_destruct(&sk->sk_rcu);
1963}
1964
1965static void __sk_free(struct sock *sk)
1966{
1967	if (likely(sk->sk_net_refcnt))
1968		sock_inuse_add(sock_net(sk), -1);
1969
1970	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1971		sock_diag_broadcast_destroy(sk);
1972	else
1973		sk_destruct(sk);
1974}
1975
1976void sk_free(struct sock *sk)
1977{
1978	/*
1979	 * We subtract one from sk_wmem_alloc and can know if
1980	 * some packets are still in some tx queue.
1981	 * If not null, sock_wfree() will call __sk_free(sk) later
1982	 */
1983	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1984		__sk_free(sk);
1985}
1986EXPORT_SYMBOL(sk_free);
1987
1988static void sk_init_common(struct sock *sk)
1989{
1990	skb_queue_head_init(&sk->sk_receive_queue);
1991	skb_queue_head_init(&sk->sk_write_queue);
1992	skb_queue_head_init(&sk->sk_error_queue);
1993
1994	rwlock_init(&sk->sk_callback_lock);
1995	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1996			af_rlock_keys + sk->sk_family,
1997			af_family_rlock_key_strings[sk->sk_family]);
1998	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1999			af_wlock_keys + sk->sk_family,
2000			af_family_wlock_key_strings[sk->sk_family]);
2001	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2002			af_elock_keys + sk->sk_family,
2003			af_family_elock_key_strings[sk->sk_family]);
2004	lockdep_set_class_and_name(&sk->sk_callback_lock,
2005			af_callback_keys + sk->sk_family,
2006			af_family_clock_key_strings[sk->sk_family]);
2007}
2008
2009/**
2010 *	sk_clone_lock - clone a socket, and lock its clone
2011 *	@sk: the socket to clone
2012 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2013 *
2014 *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2015 */
2016struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2017{
2018	struct proto *prot = READ_ONCE(sk->sk_prot);
2019	struct sk_filter *filter;
2020	bool is_charged = true;
2021	struct sock *newsk;
 
2022
2023	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2024	if (!newsk)
2025		goto out;
2026
2027	sock_copy(newsk, sk);
2028
2029	newsk->sk_prot_creator = prot;
2030
2031	/* SANITY */
2032	if (likely(newsk->sk_net_refcnt))
2033		get_net(sock_net(newsk));
2034	sk_node_init(&newsk->sk_node);
2035	sock_lock_init(newsk);
2036	bh_lock_sock(newsk);
2037	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
2038	newsk->sk_backlog.len = 0;
2039
2040	atomic_set(&newsk->sk_rmem_alloc, 0);
2041
2042	/* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2043	refcount_set(&newsk->sk_wmem_alloc, 1);
2044
2045	atomic_set(&newsk->sk_omem_alloc, 0);
2046	sk_init_common(newsk);
2047
2048	newsk->sk_dst_cache	= NULL;
2049	newsk->sk_dst_pending_confirm = 0;
2050	newsk->sk_wmem_queued	= 0;
2051	newsk->sk_forward_alloc = 0;
2052	atomic_set(&newsk->sk_drops, 0);
2053	newsk->sk_send_head	= NULL;
2054	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2055	atomic_set(&newsk->sk_zckey, 0);
2056
2057	sock_reset_flag(newsk, SOCK_DONE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2058
2059	/* sk->sk_memcg will be populated at accept() time */
2060	newsk->sk_memcg = NULL;
2061
2062	cgroup_sk_clone(&newsk->sk_cgrp_data);
 
2063
2064	rcu_read_lock();
2065	filter = rcu_dereference(sk->sk_filter);
2066	if (filter != NULL)
2067		/* though it's an empty new sock, the charging may fail
2068		 * if sysctl_optmem_max was changed between creation of
2069		 * original socket and cloning
2070		 */
2071		is_charged = sk_filter_charge(newsk, filter);
2072	RCU_INIT_POINTER(newsk->sk_filter, filter);
2073	rcu_read_unlock();
2074
2075	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2076		/* We need to make sure that we don't uncharge the new
2077		 * socket if we couldn't charge it in the first place
2078		 * as otherwise we uncharge the parent's filter.
2079		 */
2080		if (!is_charged)
2081			RCU_INIT_POINTER(newsk->sk_filter, NULL);
2082		sk_free_unlock_clone(newsk);
2083		newsk = NULL;
2084		goto out;
2085	}
2086	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2087
2088	if (bpf_sk_storage_clone(sk, newsk)) {
2089		sk_free_unlock_clone(newsk);
2090		newsk = NULL;
2091		goto out;
2092	}
 
 
 
 
 
 
 
2093
2094	/* Clear sk_user_data if parent had the pointer tagged
2095	 * as not suitable for copying when cloning.
2096	 */
2097	if (sk_user_data_is_nocopy(newsk))
2098		newsk->sk_user_data = NULL;
2099
2100	newsk->sk_err	   = 0;
2101	newsk->sk_err_soft = 0;
2102	newsk->sk_priority = 0;
2103	newsk->sk_incoming_cpu = raw_smp_processor_id();
2104	if (likely(newsk->sk_net_refcnt))
2105		sock_inuse_add(sock_net(newsk), 1);
2106
2107	/* Before updating sk_refcnt, we must commit prior changes to memory
2108	 * (Documentation/RCU/rculist_nulls.rst for details)
2109	 */
2110	smp_wmb();
2111	refcount_set(&newsk->sk_refcnt, 2);
 
2112
2113	/* Increment the counter in the same struct proto as the master
2114	 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2115	 * is the same as sk->sk_prot->socks, as this field was copied
2116	 * with memcpy).
2117	 *
2118	 * This _changes_ the previous behaviour, where
2119	 * tcp_create_openreq_child always was incrementing the
2120	 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2121	 * to be taken into account in all callers. -acme
2122	 */
2123	sk_refcnt_debug_inc(newsk);
2124	sk_set_socket(newsk, NULL);
2125	sk_tx_queue_clear(newsk);
2126	RCU_INIT_POINTER(newsk->sk_wq, NULL);
2127
2128	if (newsk->sk_prot->sockets_allocated)
2129		sk_sockets_allocated_inc(newsk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2130
2131	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2132		net_enable_timestamp();
 
 
2133out:
2134	return newsk;
2135}
2136EXPORT_SYMBOL_GPL(sk_clone_lock);
2137
2138void sk_free_unlock_clone(struct sock *sk)
2139{
2140	/* It is still raw copy of parent, so invalidate
2141	 * destructor and make plain sk_free() */
2142	sk->sk_destruct = NULL;
2143	bh_unlock_sock(sk);
2144	sk_free(sk);
2145}
2146EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2147
2148void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2149{
2150	u32 max_segs = 1;
2151
2152	sk_dst_set(sk, dst);
2153	sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2154	if (sk->sk_route_caps & NETIF_F_GSO)
2155		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2156	sk->sk_route_caps &= ~sk->sk_route_nocaps;
2157	if (sk_can_gso(sk)) {
2158		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2159			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2160		} else {
2161			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2162			sk->sk_gso_max_size = dst->dev->gso_max_size;
2163			max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2164		}
2165	}
2166	sk->sk_gso_max_segs = max_segs;
2167}
2168EXPORT_SYMBOL_GPL(sk_setup_caps);
2169
2170/*
2171 *	Simple resource managers for sockets.
2172 */
2173
2174
2175/*
2176 * Write buffer destructor automatically called from kfree_skb.
2177 */
2178void sock_wfree(struct sk_buff *skb)
2179{
2180	struct sock *sk = skb->sk;
2181	unsigned int len = skb->truesize;
2182
2183	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2184		/*
2185		 * Keep a reference on sk_wmem_alloc, this will be released
2186		 * after sk_write_space() call
2187		 */
2188		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2189		sk->sk_write_space(sk);
2190		len = 1;
2191	}
2192	/*
2193	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2194	 * could not do because of in-flight packets
2195	 */
2196	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2197		__sk_free(sk);
2198}
2199EXPORT_SYMBOL(sock_wfree);
2200
2201/* This variant of sock_wfree() is used by TCP,
2202 * since it sets SOCK_USE_WRITE_QUEUE.
2203 */
2204void __sock_wfree(struct sk_buff *skb)
2205{
2206	struct sock *sk = skb->sk;
2207
2208	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2209		__sk_free(sk);
2210}
2211
2212void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2213{
2214	skb_orphan(skb);
2215	skb->sk = sk;
2216#ifdef CONFIG_INET
2217	if (unlikely(!sk_fullsock(sk))) {
2218		skb->destructor = sock_edemux;
2219		sock_hold(sk);
2220		return;
2221	}
2222#endif
2223	skb->destructor = sock_wfree;
2224	skb_set_hash_from_sk(skb, sk);
2225	/*
2226	 * We used to take a refcount on sk, but following operation
2227	 * is enough to guarantee sk_free() wont free this sock until
2228	 * all in-flight packets are completed
2229	 */
2230	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2231}
2232EXPORT_SYMBOL(skb_set_owner_w);
2233
2234static bool can_skb_orphan_partial(const struct sk_buff *skb)
2235{
2236#ifdef CONFIG_TLS_DEVICE
2237	/* Drivers depend on in-order delivery for crypto offload,
2238	 * partial orphan breaks out-of-order-OK logic.
2239	 */
2240	if (skb->decrypted)
2241		return false;
2242#endif
2243	return (skb->destructor == sock_wfree ||
2244		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2245}
2246
2247/* This helper is used by netem, as it can hold packets in its
2248 * delay queue. We want to allow the owner socket to send more
2249 * packets, as if they were already TX completed by a typical driver.
2250 * But we also want to keep skb->sk set because some packet schedulers
2251 * rely on it (sch_fq for example).
2252 */
2253void skb_orphan_partial(struct sk_buff *skb)
2254{
2255	if (skb_is_tcp_pure_ack(skb))
2256		return;
2257
2258	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2259		return;
2260
2261	skb_orphan(skb);
 
 
 
 
 
 
2262}
2263EXPORT_SYMBOL(skb_orphan_partial);
2264
2265/*
2266 * Read buffer destructor automatically called from kfree_skb.
2267 */
2268void sock_rfree(struct sk_buff *skb)
2269{
2270	struct sock *sk = skb->sk;
2271	unsigned int len = skb->truesize;
2272
2273	atomic_sub(len, &sk->sk_rmem_alloc);
2274	sk_mem_uncharge(sk, len);
2275}
2276EXPORT_SYMBOL(sock_rfree);
2277
2278/*
2279 * Buffer destructor for skbs that are not used directly in read or write
2280 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2281 */
2282void sock_efree(struct sk_buff *skb)
2283{
2284	sock_put(skb->sk);
2285}
2286EXPORT_SYMBOL(sock_efree);
2287
2288/* Buffer destructor for prefetch/receive path where reference count may
2289 * not be held, e.g. for listen sockets.
2290 */
2291#ifdef CONFIG_INET
2292void sock_pfree(struct sk_buff *skb)
2293{
2294	if (sk_is_refcounted(skb->sk))
2295		sock_gen_put(skb->sk);
2296}
2297EXPORT_SYMBOL(sock_pfree);
2298#endif /* CONFIG_INET */
2299
2300kuid_t sock_i_uid(struct sock *sk)
2301{
2302	kuid_t uid;
2303
2304	read_lock_bh(&sk->sk_callback_lock);
2305	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2306	read_unlock_bh(&sk->sk_callback_lock);
2307	return uid;
2308}
2309EXPORT_SYMBOL(sock_i_uid);
2310
2311unsigned long sock_i_ino(struct sock *sk)
2312{
2313	unsigned long ino;
2314
2315	read_lock_bh(&sk->sk_callback_lock);
2316	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2317	read_unlock_bh(&sk->sk_callback_lock);
2318	return ino;
2319}
2320EXPORT_SYMBOL(sock_i_ino);
2321
2322/*
2323 * Allocate a skb from the socket's send buffer.
2324 */
2325struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2326			     gfp_t priority)
2327{
2328	if (force ||
2329	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2330		struct sk_buff *skb = alloc_skb(size, priority);
2331
2332		if (skb) {
2333			skb_set_owner_w(skb, sk);
2334			return skb;
2335		}
2336	}
2337	return NULL;
2338}
2339EXPORT_SYMBOL(sock_wmalloc);
2340
2341static void sock_ofree(struct sk_buff *skb)
2342{
2343	struct sock *sk = skb->sk;
2344
2345	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2346}
2347
2348struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2349			     gfp_t priority)
2350{
2351	struct sk_buff *skb;
2352
2353	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2354	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2355	    sysctl_optmem_max)
2356		return NULL;
2357
2358	skb = alloc_skb(size, priority);
2359	if (!skb)
2360		return NULL;
2361
2362	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2363	skb->sk = sk;
2364	skb->destructor = sock_ofree;
2365	return skb;
2366}
2367
2368/*
2369 * Allocate a memory block from the socket's option memory buffer.
2370 */
2371void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2372{
2373	if ((unsigned int)size <= sysctl_optmem_max &&
2374	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2375		void *mem;
2376		/* First do the add, to avoid the race if kmalloc
2377		 * might sleep.
2378		 */
2379		atomic_add(size, &sk->sk_omem_alloc);
2380		mem = kmalloc(size, priority);
2381		if (mem)
2382			return mem;
2383		atomic_sub(size, &sk->sk_omem_alloc);
2384	}
2385	return NULL;
2386}
2387EXPORT_SYMBOL(sock_kmalloc);
2388
2389/* Free an option memory block. Note, we actually want the inline
2390 * here as this allows gcc to detect the nullify and fold away the
2391 * condition entirely.
2392 */
2393static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2394				  const bool nullify)
2395{
2396	if (WARN_ON_ONCE(!mem))
2397		return;
2398	if (nullify)
2399		kfree_sensitive(mem);
2400	else
2401		kfree(mem);
2402	atomic_sub(size, &sk->sk_omem_alloc);
2403}
2404
2405void sock_kfree_s(struct sock *sk, void *mem, int size)
2406{
2407	__sock_kfree_s(sk, mem, size, false);
2408}
2409EXPORT_SYMBOL(sock_kfree_s);
2410
2411void sock_kzfree_s(struct sock *sk, void *mem, int size)
2412{
2413	__sock_kfree_s(sk, mem, size, true);
2414}
2415EXPORT_SYMBOL(sock_kzfree_s);
2416
2417/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2418   I think, these locks should be removed for datagram sockets.
2419 */
2420static long sock_wait_for_wmem(struct sock *sk, long timeo)
2421{
2422	DEFINE_WAIT(wait);
2423
2424	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2425	for (;;) {
2426		if (!timeo)
2427			break;
2428		if (signal_pending(current))
2429			break;
2430		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2431		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2432		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2433			break;
2434		if (sk->sk_shutdown & SEND_SHUTDOWN)
2435			break;
2436		if (sk->sk_err)
2437			break;
2438		timeo = schedule_timeout(timeo);
2439	}
2440	finish_wait(sk_sleep(sk), &wait);
2441	return timeo;
2442}
2443
2444
2445/*
2446 *	Generic send/receive buffer handlers
2447 */
2448
2449struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2450				     unsigned long data_len, int noblock,
2451				     int *errcode, int max_page_order)
2452{
2453	struct sk_buff *skb;
2454	long timeo;
2455	int err;
2456
2457	timeo = sock_sndtimeo(sk, noblock);
2458	for (;;) {
2459		err = sock_error(sk);
2460		if (err != 0)
2461			goto failure;
2462
2463		err = -EPIPE;
2464		if (sk->sk_shutdown & SEND_SHUTDOWN)
2465			goto failure;
2466
2467		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2468			break;
2469
2470		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2471		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2472		err = -EAGAIN;
2473		if (!timeo)
2474			goto failure;
2475		if (signal_pending(current))
2476			goto interrupted;
2477		timeo = sock_wait_for_wmem(sk, timeo);
2478	}
2479	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2480				   errcode, sk->sk_allocation);
2481	if (skb)
2482		skb_set_owner_w(skb, sk);
2483	return skb;
2484
2485interrupted:
2486	err = sock_intr_errno(timeo);
2487failure:
2488	*errcode = err;
2489	return NULL;
2490}
2491EXPORT_SYMBOL(sock_alloc_send_pskb);
2492
2493struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2494				    int noblock, int *errcode)
2495{
2496	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2497}
2498EXPORT_SYMBOL(sock_alloc_send_skb);
2499
2500int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2501		     struct sockcm_cookie *sockc)
2502{
2503	u32 tsflags;
2504
2505	switch (cmsg->cmsg_type) {
2506	case SO_MARK:
2507		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2508			return -EPERM;
2509		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2510			return -EINVAL;
2511		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2512		break;
2513	case SO_TIMESTAMPING_OLD:
2514		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2515			return -EINVAL;
2516
2517		tsflags = *(u32 *)CMSG_DATA(cmsg);
2518		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2519			return -EINVAL;
2520
2521		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2522		sockc->tsflags |= tsflags;
2523		break;
2524	case SCM_TXTIME:
2525		if (!sock_flag(sk, SOCK_TXTIME))
2526			return -EINVAL;
2527		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2528			return -EINVAL;
2529		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2530		break;
2531	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2532	case SCM_RIGHTS:
2533	case SCM_CREDENTIALS:
2534		break;
2535	default:
2536		return -EINVAL;
2537	}
2538	return 0;
2539}
2540EXPORT_SYMBOL(__sock_cmsg_send);
2541
2542int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2543		   struct sockcm_cookie *sockc)
2544{
2545	struct cmsghdr *cmsg;
2546	int ret;
2547
2548	for_each_cmsghdr(cmsg, msg) {
2549		if (!CMSG_OK(msg, cmsg))
2550			return -EINVAL;
2551		if (cmsg->cmsg_level != SOL_SOCKET)
2552			continue;
2553		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2554		if (ret)
2555			return ret;
2556	}
2557	return 0;
2558}
2559EXPORT_SYMBOL(sock_cmsg_send);
2560
2561static void sk_enter_memory_pressure(struct sock *sk)
2562{
2563	if (!sk->sk_prot->enter_memory_pressure)
2564		return;
2565
2566	sk->sk_prot->enter_memory_pressure(sk);
2567}
2568
2569static void sk_leave_memory_pressure(struct sock *sk)
2570{
2571	if (sk->sk_prot->leave_memory_pressure) {
2572		sk->sk_prot->leave_memory_pressure(sk);
2573	} else {
2574		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2575
2576		if (memory_pressure && READ_ONCE(*memory_pressure))
2577			WRITE_ONCE(*memory_pressure, 0);
2578	}
2579}
2580
2581#define SKB_FRAG_PAGE_ORDER	get_order(32768)
2582DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2583
2584/**
2585 * skb_page_frag_refill - check that a page_frag contains enough room
2586 * @sz: minimum size of the fragment we want to get
2587 * @pfrag: pointer to page_frag
2588 * @gfp: priority for memory allocation
2589 *
2590 * Note: While this allocator tries to use high order pages, there is
2591 * no guarantee that allocations succeed. Therefore, @sz MUST be
2592 * less or equal than PAGE_SIZE.
2593 */
2594bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2595{
2596	if (pfrag->page) {
2597		if (page_ref_count(pfrag->page) == 1) {
2598			pfrag->offset = 0;
2599			return true;
2600		}
2601		if (pfrag->offset + sz <= pfrag->size)
2602			return true;
2603		put_page(pfrag->page);
2604	}
2605
2606	pfrag->offset = 0;
2607	if (SKB_FRAG_PAGE_ORDER &&
2608	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2609		/* Avoid direct reclaim but allow kswapd to wake */
2610		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2611					  __GFP_COMP | __GFP_NOWARN |
2612					  __GFP_NORETRY,
2613					  SKB_FRAG_PAGE_ORDER);
2614		if (likely(pfrag->page)) {
2615			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2616			return true;
2617		}
2618	}
2619	pfrag->page = alloc_page(gfp);
2620	if (likely(pfrag->page)) {
2621		pfrag->size = PAGE_SIZE;
2622		return true;
2623	}
2624	return false;
2625}
2626EXPORT_SYMBOL(skb_page_frag_refill);
2627
2628bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2629{
2630	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2631		return true;
2632
2633	sk_enter_memory_pressure(sk);
2634	sk_stream_moderate_sndbuf(sk);
2635	return false;
2636}
2637EXPORT_SYMBOL(sk_page_frag_refill);
2638
2639void __lock_sock(struct sock *sk)
2640	__releases(&sk->sk_lock.slock)
2641	__acquires(&sk->sk_lock.slock)
2642{
2643	DEFINE_WAIT(wait);
2644
2645	for (;;) {
2646		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2647					TASK_UNINTERRUPTIBLE);
2648		spin_unlock_bh(&sk->sk_lock.slock);
2649		schedule();
2650		spin_lock_bh(&sk->sk_lock.slock);
2651		if (!sock_owned_by_user(sk))
2652			break;
2653	}
2654	finish_wait(&sk->sk_lock.wq, &wait);
2655}
2656
2657void __release_sock(struct sock *sk)
2658	__releases(&sk->sk_lock.slock)
2659	__acquires(&sk->sk_lock.slock)
2660{
2661	struct sk_buff *skb, *next;
2662
2663	while ((skb = sk->sk_backlog.head) != NULL) {
2664		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2665
2666		spin_unlock_bh(&sk->sk_lock.slock);
2667
2668		do {
2669			next = skb->next;
2670			prefetch(next);
2671			WARN_ON_ONCE(skb_dst_is_noref(skb));
2672			skb_mark_not_on_list(skb);
2673			sk_backlog_rcv(sk, skb);
2674
2675			cond_resched();
2676
2677			skb = next;
2678		} while (skb != NULL);
2679
2680		spin_lock_bh(&sk->sk_lock.slock);
2681	}
2682
2683	/*
2684	 * Doing the zeroing here guarantee we can not loop forever
2685	 * while a wild producer attempts to flood us.
2686	 */
2687	sk->sk_backlog.len = 0;
2688}
2689
2690void __sk_flush_backlog(struct sock *sk)
2691{
2692	spin_lock_bh(&sk->sk_lock.slock);
2693	__release_sock(sk);
2694	spin_unlock_bh(&sk->sk_lock.slock);
2695}
2696
2697/**
2698 * sk_wait_data - wait for data to arrive at sk_receive_queue
2699 * @sk:    sock to wait on
2700 * @timeo: for how long
2701 * @skb:   last skb seen on sk_receive_queue
2702 *
2703 * Now socket state including sk->sk_err is changed only under lock,
2704 * hence we may omit checks after joining wait queue.
2705 * We check receive queue before schedule() only as optimization;
2706 * it is very likely that release_sock() added new data.
2707 */
2708int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2709{
2710	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2711	int rc;
2712
2713	add_wait_queue(sk_sleep(sk), &wait);
2714	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2715	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2716	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2717	remove_wait_queue(sk_sleep(sk), &wait);
2718	return rc;
2719}
2720EXPORT_SYMBOL(sk_wait_data);
2721
2722/**
2723 *	__sk_mem_raise_allocated - increase memory_allocated
2724 *	@sk: socket
2725 *	@size: memory size to allocate
2726 *	@amt: pages to allocate
2727 *	@kind: allocation type
2728 *
2729 *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2730 */
2731int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2732{
2733	struct proto *prot = sk->sk_prot;
2734	long allocated = sk_memory_allocated_add(sk, amt);
2735	bool charged = true;
2736
2737	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2738	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2739		goto suppress_allocation;
2740
2741	/* Under limit. */
2742	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2743		sk_leave_memory_pressure(sk);
2744		return 1;
2745	}
2746
2747	/* Under pressure. */
2748	if (allocated > sk_prot_mem_limits(sk, 1))
2749		sk_enter_memory_pressure(sk);
2750
2751	/* Over hard limit. */
2752	if (allocated > sk_prot_mem_limits(sk, 2))
2753		goto suppress_allocation;
2754
2755	/* guarantee minimum buffer size under pressure */
2756	if (kind == SK_MEM_RECV) {
2757		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2758			return 1;
2759
2760	} else { /* SK_MEM_SEND */
2761		int wmem0 = sk_get_wmem0(sk, prot);
2762
2763		if (sk->sk_type == SOCK_STREAM) {
2764			if (sk->sk_wmem_queued < wmem0)
2765				return 1;
2766		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2767				return 1;
2768		}
2769	}
2770
2771	if (sk_has_memory_pressure(sk)) {
2772		u64 alloc;
2773
2774		if (!sk_under_memory_pressure(sk))
2775			return 1;
2776		alloc = sk_sockets_allocated_read_positive(sk);
2777		if (sk_prot_mem_limits(sk, 2) > alloc *
2778		    sk_mem_pages(sk->sk_wmem_queued +
2779				 atomic_read(&sk->sk_rmem_alloc) +
2780				 sk->sk_forward_alloc))
2781			return 1;
2782	}
2783
2784suppress_allocation:
2785
2786	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2787		sk_stream_moderate_sndbuf(sk);
2788
2789		/* Fail only if socket is _under_ its sndbuf.
2790		 * In this case we cannot block, so that we have to fail.
2791		 */
2792		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2793			return 1;
2794	}
2795
2796	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2797		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2798
2799	sk_memory_allocated_sub(sk, amt);
2800
2801	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2802		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2803
2804	return 0;
2805}
2806EXPORT_SYMBOL(__sk_mem_raise_allocated);
2807
2808/**
2809 *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2810 *	@sk: socket
2811 *	@size: memory size to allocate
2812 *	@kind: allocation type
2813 *
2814 *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2815 *	rmem allocation. This function assumes that protocols which have
2816 *	memory_pressure use sk_wmem_queued as write buffer accounting.
2817 */
2818int __sk_mem_schedule(struct sock *sk, int size, int kind)
2819{
2820	int ret, amt = sk_mem_pages(size);
2821
2822	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2823	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2824	if (!ret)
2825		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2826	return ret;
2827}
2828EXPORT_SYMBOL(__sk_mem_schedule);
2829
2830/**
2831 *	__sk_mem_reduce_allocated - reclaim memory_allocated
2832 *	@sk: socket
2833 *	@amount: number of quanta
2834 *
2835 *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2836 */
2837void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2838{
2839	sk_memory_allocated_sub(sk, amount);
2840
2841	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2842		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2843
2844	if (sk_under_memory_pressure(sk) &&
2845	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2846		sk_leave_memory_pressure(sk);
2847}
2848EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2849
2850/**
2851 *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2852 *	@sk: socket
2853 *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2854 */
2855void __sk_mem_reclaim(struct sock *sk, int amount)
2856{
2857	amount >>= SK_MEM_QUANTUM_SHIFT;
2858	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2859	__sk_mem_reduce_allocated(sk, amount);
2860}
2861EXPORT_SYMBOL(__sk_mem_reclaim);
2862
2863int sk_set_peek_off(struct sock *sk, int val)
2864{
2865	sk->sk_peek_off = val;
2866	return 0;
2867}
2868EXPORT_SYMBOL_GPL(sk_set_peek_off);
2869
2870/*
2871 * Set of default routines for initialising struct proto_ops when
2872 * the protocol does not support a particular function. In certain
2873 * cases where it makes no sense for a protocol to have a "do nothing"
2874 * function, some default processing is provided.
2875 */
2876
2877int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2878{
2879	return -EOPNOTSUPP;
2880}
2881EXPORT_SYMBOL(sock_no_bind);
2882
2883int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2884		    int len, int flags)
2885{
2886	return -EOPNOTSUPP;
2887}
2888EXPORT_SYMBOL(sock_no_connect);
2889
2890int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2891{
2892	return -EOPNOTSUPP;
2893}
2894EXPORT_SYMBOL(sock_no_socketpair);
2895
2896int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2897		   bool kern)
2898{
2899	return -EOPNOTSUPP;
2900}
2901EXPORT_SYMBOL(sock_no_accept);
2902
2903int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2904		    int peer)
2905{
2906	return -EOPNOTSUPP;
2907}
2908EXPORT_SYMBOL(sock_no_getname);
2909
2910int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2911{
2912	return -EOPNOTSUPP;
2913}
2914EXPORT_SYMBOL(sock_no_ioctl);
2915
2916int sock_no_listen(struct socket *sock, int backlog)
2917{
2918	return -EOPNOTSUPP;
2919}
2920EXPORT_SYMBOL(sock_no_listen);
2921
2922int sock_no_shutdown(struct socket *sock, int how)
2923{
2924	return -EOPNOTSUPP;
2925}
2926EXPORT_SYMBOL(sock_no_shutdown);
2927
2928int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2929{
2930	return -EOPNOTSUPP;
2931}
2932EXPORT_SYMBOL(sock_no_sendmsg);
2933
2934int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2935{
2936	return -EOPNOTSUPP;
2937}
2938EXPORT_SYMBOL(sock_no_sendmsg_locked);
2939
2940int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2941		    int flags)
2942{
2943	return -EOPNOTSUPP;
2944}
2945EXPORT_SYMBOL(sock_no_recvmsg);
2946
2947int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2948{
2949	/* Mirror missing mmap method error code */
2950	return -ENODEV;
2951}
2952EXPORT_SYMBOL(sock_no_mmap);
2953
2954/*
2955 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2956 * various sock-based usage counts.
2957 */
2958void __receive_sock(struct file *file)
2959{
2960	struct socket *sock;
 
2961
2962	sock = sock_from_file(file);
 
 
 
 
 
2963	if (sock) {
2964		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2965		sock_update_classid(&sock->sk->sk_cgrp_data);
2966	}
2967}
2968
2969ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2970{
2971	ssize_t res;
2972	struct msghdr msg = {.msg_flags = flags};
2973	struct kvec iov;
2974	char *kaddr = kmap(page);
2975	iov.iov_base = kaddr + offset;
2976	iov.iov_len = size;
2977	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2978	kunmap(page);
2979	return res;
2980}
2981EXPORT_SYMBOL(sock_no_sendpage);
2982
2983ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2984				int offset, size_t size, int flags)
2985{
2986	ssize_t res;
2987	struct msghdr msg = {.msg_flags = flags};
2988	struct kvec iov;
2989	char *kaddr = kmap(page);
2990
2991	iov.iov_base = kaddr + offset;
2992	iov.iov_len = size;
2993	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2994	kunmap(page);
2995	return res;
2996}
2997EXPORT_SYMBOL(sock_no_sendpage_locked);
2998
2999/*
3000 *	Default Socket Callbacks
3001 */
3002
3003static void sock_def_wakeup(struct sock *sk)
3004{
3005	struct socket_wq *wq;
3006
3007	rcu_read_lock();
3008	wq = rcu_dereference(sk->sk_wq);
3009	if (skwq_has_sleeper(wq))
3010		wake_up_interruptible_all(&wq->wait);
3011	rcu_read_unlock();
3012}
3013
3014static void sock_def_error_report(struct sock *sk)
3015{
3016	struct socket_wq *wq;
3017
3018	rcu_read_lock();
3019	wq = rcu_dereference(sk->sk_wq);
3020	if (skwq_has_sleeper(wq))
3021		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3022	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3023	rcu_read_unlock();
3024}
3025
3026void sock_def_readable(struct sock *sk)
3027{
3028	struct socket_wq *wq;
3029
3030	rcu_read_lock();
3031	wq = rcu_dereference(sk->sk_wq);
3032	if (skwq_has_sleeper(wq))
3033		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3034						EPOLLRDNORM | EPOLLRDBAND);
3035	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3036	rcu_read_unlock();
3037}
3038
3039static void sock_def_write_space(struct sock *sk)
3040{
3041	struct socket_wq *wq;
3042
3043	rcu_read_lock();
3044
3045	/* Do not wake up a writer until he can make "significant"
3046	 * progress.  --DaveM
3047	 */
3048	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
3049		wq = rcu_dereference(sk->sk_wq);
3050		if (skwq_has_sleeper(wq))
3051			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3052						EPOLLWRNORM | EPOLLWRBAND);
3053
3054		/* Should agree with poll, otherwise some programs break */
3055		if (sock_writeable(sk))
3056			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3057	}
3058
3059	rcu_read_unlock();
3060}
3061
3062static void sock_def_destruct(struct sock *sk)
3063{
3064}
3065
3066void sk_send_sigurg(struct sock *sk)
3067{
3068	if (sk->sk_socket && sk->sk_socket->file)
3069		if (send_sigurg(&sk->sk_socket->file->f_owner))
3070			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3071}
3072EXPORT_SYMBOL(sk_send_sigurg);
3073
3074void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3075		    unsigned long expires)
3076{
3077	if (!mod_timer(timer, expires))
3078		sock_hold(sk);
3079}
3080EXPORT_SYMBOL(sk_reset_timer);
3081
3082void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3083{
3084	if (del_timer(timer))
3085		__sock_put(sk);
3086}
3087EXPORT_SYMBOL(sk_stop_timer);
3088
3089void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3090{
3091	if (del_timer_sync(timer))
3092		__sock_put(sk);
3093}
3094EXPORT_SYMBOL(sk_stop_timer_sync);
3095
3096void sock_init_data(struct socket *sock, struct sock *sk)
3097{
3098	sk_init_common(sk);
3099	sk->sk_send_head	=	NULL;
3100
3101	timer_setup(&sk->sk_timer, NULL, 0);
3102
3103	sk->sk_allocation	=	GFP_KERNEL;
3104	sk->sk_rcvbuf		=	sysctl_rmem_default;
3105	sk->sk_sndbuf		=	sysctl_wmem_default;
3106	sk->sk_state		=	TCP_CLOSE;
3107	sk_set_socket(sk, sock);
3108
3109	sock_set_flag(sk, SOCK_ZAPPED);
3110
3111	if (sock) {
3112		sk->sk_type	=	sock->type;
3113		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3114		sock->sk	=	sk;
3115		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
3116	} else {
3117		RCU_INIT_POINTER(sk->sk_wq, NULL);
3118		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
3119	}
3120
3121	rwlock_init(&sk->sk_callback_lock);
3122	if (sk->sk_kern_sock)
3123		lockdep_set_class_and_name(
3124			&sk->sk_callback_lock,
3125			af_kern_callback_keys + sk->sk_family,
3126			af_family_kern_clock_key_strings[sk->sk_family]);
3127	else
3128		lockdep_set_class_and_name(
3129			&sk->sk_callback_lock,
3130			af_callback_keys + sk->sk_family,
3131			af_family_clock_key_strings[sk->sk_family]);
3132
3133	sk->sk_state_change	=	sock_def_wakeup;
3134	sk->sk_data_ready	=	sock_def_readable;
3135	sk->sk_write_space	=	sock_def_write_space;
3136	sk->sk_error_report	=	sock_def_error_report;
3137	sk->sk_destruct		=	sock_def_destruct;
3138
3139	sk->sk_frag.page	=	NULL;
3140	sk->sk_frag.offset	=	0;
3141	sk->sk_peek_off		=	-1;
3142
3143	sk->sk_peer_pid 	=	NULL;
3144	sk->sk_peer_cred	=	NULL;
3145	spin_lock_init(&sk->sk_peer_lock);
3146
3147	sk->sk_write_pending	=	0;
3148	sk->sk_rcvlowat		=	1;
3149	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3150	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3151
3152	sk->sk_stamp = SK_DEFAULT_STAMP;
3153#if BITS_PER_LONG==32
3154	seqlock_init(&sk->sk_stamp_seq);
3155#endif
3156	atomic_set(&sk->sk_zckey, 0);
3157
3158#ifdef CONFIG_NET_RX_BUSY_POLL
3159	sk->sk_napi_id		=	0;
3160	sk->sk_ll_usec		=	sysctl_net_busy_read;
3161#endif
3162
3163	sk->sk_max_pacing_rate = ~0UL;
3164	sk->sk_pacing_rate = ~0UL;
3165	WRITE_ONCE(sk->sk_pacing_shift, 10);
3166	sk->sk_incoming_cpu = -1;
3167
3168	sk_rx_queue_clear(sk);
3169	/*
3170	 * Before updating sk_refcnt, we must commit prior changes to memory
3171	 * (Documentation/RCU/rculist_nulls.rst for details)
3172	 */
3173	smp_wmb();
3174	refcount_set(&sk->sk_refcnt, 1);
3175	atomic_set(&sk->sk_drops, 0);
3176}
3177EXPORT_SYMBOL(sock_init_data);
3178
3179void lock_sock_nested(struct sock *sk, int subclass)
3180{
3181	might_sleep();
3182	spin_lock_bh(&sk->sk_lock.slock);
3183	if (sk->sk_lock.owned)
3184		__lock_sock(sk);
3185	sk->sk_lock.owned = 1;
3186	spin_unlock(&sk->sk_lock.slock);
3187	/*
3188	 * The sk_lock has mutex_lock() semantics here:
3189	 */
3190	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3191	local_bh_enable();
3192}
3193EXPORT_SYMBOL(lock_sock_nested);
3194
3195void release_sock(struct sock *sk)
3196{
3197	spin_lock_bh(&sk->sk_lock.slock);
3198	if (sk->sk_backlog.tail)
3199		__release_sock(sk);
3200
3201	/* Warning : release_cb() might need to release sk ownership,
3202	 * ie call sock_release_ownership(sk) before us.
3203	 */
3204	if (sk->sk_prot->release_cb)
3205		sk->sk_prot->release_cb(sk);
3206
3207	sock_release_ownership(sk);
3208	if (waitqueue_active(&sk->sk_lock.wq))
3209		wake_up(&sk->sk_lock.wq);
3210	spin_unlock_bh(&sk->sk_lock.slock);
3211}
3212EXPORT_SYMBOL(release_sock);
3213
3214/**
3215 * lock_sock_fast - fast version of lock_sock
3216 * @sk: socket
3217 *
3218 * This version should be used for very small section, where process wont block
3219 * return false if fast path is taken:
3220 *
3221 *   sk_lock.slock locked, owned = 0, BH disabled
3222 *
3223 * return true if slow path is taken:
3224 *
3225 *   sk_lock.slock unlocked, owned = 1, BH enabled
3226 */
3227bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3228{
3229	might_sleep();
3230	spin_lock_bh(&sk->sk_lock.slock);
3231
3232	if (!sk->sk_lock.owned)
3233		/*
3234		 * Note : We must disable BH
3235		 */
3236		return false;
3237
3238	__lock_sock(sk);
3239	sk->sk_lock.owned = 1;
3240	spin_unlock(&sk->sk_lock.slock);
3241	/*
3242	 * The sk_lock has mutex_lock() semantics here:
3243	 */
3244	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3245	__acquire(&sk->sk_lock.slock);
3246	local_bh_enable();
3247	return true;
3248}
3249EXPORT_SYMBOL(lock_sock_fast);
3250
3251int sock_gettstamp(struct socket *sock, void __user *userstamp,
3252		   bool timeval, bool time32)
3253{
3254	struct sock *sk = sock->sk;
3255	struct timespec64 ts;
3256
3257	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3258	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3259	if (ts.tv_sec == -1)
3260		return -ENOENT;
3261	if (ts.tv_sec == 0) {
3262		ktime_t kt = ktime_get_real();
3263		sock_write_timestamp(sk, kt);
3264		ts = ktime_to_timespec64(kt);
3265	}
3266
3267	if (timeval)
3268		ts.tv_nsec /= 1000;
3269
3270#ifdef CONFIG_COMPAT_32BIT_TIME
3271	if (time32)
3272		return put_old_timespec32(&ts, userstamp);
3273#endif
3274#ifdef CONFIG_SPARC64
3275	/* beware of padding in sparc64 timeval */
3276	if (timeval && !in_compat_syscall()) {
3277		struct __kernel_old_timeval __user tv = {
3278			.tv_sec = ts.tv_sec,
3279			.tv_usec = ts.tv_nsec,
3280		};
3281		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3282			return -EFAULT;
3283		return 0;
3284	}
3285#endif
3286	return put_timespec64(&ts, userstamp);
3287}
3288EXPORT_SYMBOL(sock_gettstamp);
3289
3290void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3291{
3292	if (!sock_flag(sk, flag)) {
3293		unsigned long previous_flags = sk->sk_flags;
3294
3295		sock_set_flag(sk, flag);
3296		/*
3297		 * we just set one of the two flags which require net
3298		 * time stamping, but time stamping might have been on
3299		 * already because of the other one
3300		 */
3301		if (sock_needs_netstamp(sk) &&
3302		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3303			net_enable_timestamp();
3304	}
3305}
3306
3307int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3308		       int level, int type)
3309{
3310	struct sock_exterr_skb *serr;
3311	struct sk_buff *skb;
3312	int copied, err;
3313
3314	err = -EAGAIN;
3315	skb = sock_dequeue_err_skb(sk);
3316	if (skb == NULL)
3317		goto out;
3318
3319	copied = skb->len;
3320	if (copied > len) {
3321		msg->msg_flags |= MSG_TRUNC;
3322		copied = len;
3323	}
3324	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3325	if (err)
3326		goto out_free_skb;
3327
3328	sock_recv_timestamp(msg, sk, skb);
3329
3330	serr = SKB_EXT_ERR(skb);
3331	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3332
3333	msg->msg_flags |= MSG_ERRQUEUE;
3334	err = copied;
3335
3336out_free_skb:
3337	kfree_skb(skb);
3338out:
3339	return err;
3340}
3341EXPORT_SYMBOL(sock_recv_errqueue);
3342
3343/*
3344 *	Get a socket option on an socket.
3345 *
3346 *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3347 *	asynchronous errors should be reported by getsockopt. We assume
3348 *	this means if you specify SO_ERROR (otherwise whats the point of it).
3349 */
3350int sock_common_getsockopt(struct socket *sock, int level, int optname,
3351			   char __user *optval, int __user *optlen)
3352{
3353	struct sock *sk = sock->sk;
3354
3355	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3356}
3357EXPORT_SYMBOL(sock_common_getsockopt);
3358
3359int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3360			int flags)
3361{
3362	struct sock *sk = sock->sk;
3363	int addr_len = 0;
3364	int err;
3365
3366	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3367				   flags & ~MSG_DONTWAIT, &addr_len);
3368	if (err >= 0)
3369		msg->msg_namelen = addr_len;
3370	return err;
3371}
3372EXPORT_SYMBOL(sock_common_recvmsg);
3373
3374/*
3375 *	Set socket options on an inet socket.
3376 */
3377int sock_common_setsockopt(struct socket *sock, int level, int optname,
3378			   sockptr_t optval, unsigned int optlen)
3379{
3380	struct sock *sk = sock->sk;
3381
3382	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3383}
3384EXPORT_SYMBOL(sock_common_setsockopt);
3385
3386void sk_common_release(struct sock *sk)
3387{
3388	if (sk->sk_prot->destroy)
3389		sk->sk_prot->destroy(sk);
3390
3391	/*
3392	 * Observation: when sk_common_release is called, processes have
3393	 * no access to socket. But net still has.
3394	 * Step one, detach it from networking:
3395	 *
3396	 * A. Remove from hash tables.
3397	 */
3398
3399	sk->sk_prot->unhash(sk);
3400
3401	/*
3402	 * In this point socket cannot receive new packets, but it is possible
3403	 * that some packets are in flight because some CPU runs receiver and
3404	 * did hash table lookup before we unhashed socket. They will achieve
3405	 * receive queue and will be purged by socket destructor.
3406	 *
3407	 * Also we still have packets pending on receive queue and probably,
3408	 * our own packets waiting in device queues. sock_destroy will drain
3409	 * receive queue, but transmitted packets will delay socket destruction
3410	 * until the last reference will be released.
3411	 */
3412
3413	sock_orphan(sk);
3414
3415	xfrm_sk_free_policy(sk);
3416
3417	sk_refcnt_debug_release(sk);
3418
3419	sock_put(sk);
3420}
3421EXPORT_SYMBOL(sk_common_release);
3422
3423void sk_get_meminfo(const struct sock *sk, u32 *mem)
3424{
3425	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3426
3427	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3428	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3429	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3430	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3431	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3432	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3433	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3434	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3435	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3436}
3437
3438#ifdef CONFIG_PROC_FS
3439#define PROTO_INUSE_NR	64	/* should be enough for the first time */
3440struct prot_inuse {
3441	int val[PROTO_INUSE_NR];
3442};
3443
3444static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3445
3446void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3447{
3448	__this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3449}
3450EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3451
3452int sock_prot_inuse_get(struct net *net, struct proto *prot)
3453{
3454	int cpu, idx = prot->inuse_idx;
3455	int res = 0;
3456
3457	for_each_possible_cpu(cpu)
3458		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3459
3460	return res >= 0 ? res : 0;
3461}
3462EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3463
3464static void sock_inuse_add(struct net *net, int val)
3465{
3466	this_cpu_add(*net->core.sock_inuse, val);
3467}
3468
3469int sock_inuse_get(struct net *net)
3470{
3471	int cpu, res = 0;
3472
3473	for_each_possible_cpu(cpu)
3474		res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3475
3476	return res;
3477}
3478
3479EXPORT_SYMBOL_GPL(sock_inuse_get);
3480
3481static int __net_init sock_inuse_init_net(struct net *net)
3482{
3483	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3484	if (net->core.prot_inuse == NULL)
3485		return -ENOMEM;
3486
3487	net->core.sock_inuse = alloc_percpu(int);
3488	if (net->core.sock_inuse == NULL)
3489		goto out;
3490
3491	return 0;
3492
3493out:
3494	free_percpu(net->core.prot_inuse);
3495	return -ENOMEM;
3496}
3497
3498static void __net_exit sock_inuse_exit_net(struct net *net)
3499{
3500	free_percpu(net->core.prot_inuse);
3501	free_percpu(net->core.sock_inuse);
3502}
3503
3504static struct pernet_operations net_inuse_ops = {
3505	.init = sock_inuse_init_net,
3506	.exit = sock_inuse_exit_net,
3507};
3508
3509static __init int net_inuse_init(void)
3510{
3511	if (register_pernet_subsys(&net_inuse_ops))
3512		panic("Cannot initialize net inuse counters");
3513
3514	return 0;
3515}
3516
3517core_initcall(net_inuse_init);
3518
3519static int assign_proto_idx(struct proto *prot)
3520{
3521	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3522
3523	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3524		pr_err("PROTO_INUSE_NR exhausted\n");
3525		return -ENOSPC;
3526	}
3527
3528	set_bit(prot->inuse_idx, proto_inuse_idx);
3529	return 0;
3530}
3531
3532static void release_proto_idx(struct proto *prot)
3533{
3534	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3535		clear_bit(prot->inuse_idx, proto_inuse_idx);
3536}
3537#else
3538static inline int assign_proto_idx(struct proto *prot)
3539{
3540	return 0;
3541}
3542
3543static inline void release_proto_idx(struct proto *prot)
3544{
3545}
3546
3547static void sock_inuse_add(struct net *net, int val)
3548{
3549}
3550#endif
3551
3552static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3553{
3554	if (!twsk_prot)
3555		return;
3556	kfree(twsk_prot->twsk_slab_name);
3557	twsk_prot->twsk_slab_name = NULL;
3558	kmem_cache_destroy(twsk_prot->twsk_slab);
3559	twsk_prot->twsk_slab = NULL;
3560}
3561
3562static int tw_prot_init(const struct proto *prot)
3563{
3564	struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3565
3566	if (!twsk_prot)
3567		return 0;
3568
3569	twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3570					      prot->name);
3571	if (!twsk_prot->twsk_slab_name)
3572		return -ENOMEM;
3573
3574	twsk_prot->twsk_slab =
3575		kmem_cache_create(twsk_prot->twsk_slab_name,
3576				  twsk_prot->twsk_obj_size, 0,
3577				  SLAB_ACCOUNT | prot->slab_flags,
3578				  NULL);
3579	if (!twsk_prot->twsk_slab) {
3580		pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3581			prot->name);
3582		return -ENOMEM;
3583	}
3584
3585	return 0;
3586}
3587
3588static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3589{
3590	if (!rsk_prot)
3591		return;
3592	kfree(rsk_prot->slab_name);
3593	rsk_prot->slab_name = NULL;
3594	kmem_cache_destroy(rsk_prot->slab);
3595	rsk_prot->slab = NULL;
3596}
3597
3598static int req_prot_init(const struct proto *prot)
3599{
3600	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3601
3602	if (!rsk_prot)
3603		return 0;
3604
3605	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3606					prot->name);
3607	if (!rsk_prot->slab_name)
3608		return -ENOMEM;
3609
3610	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3611					   rsk_prot->obj_size, 0,
3612					   SLAB_ACCOUNT | prot->slab_flags,
3613					   NULL);
3614
3615	if (!rsk_prot->slab) {
3616		pr_crit("%s: Can't create request sock SLAB cache!\n",
3617			prot->name);
3618		return -ENOMEM;
3619	}
3620	return 0;
3621}
3622
3623int proto_register(struct proto *prot, int alloc_slab)
3624{
3625	int ret = -ENOBUFS;
3626
3627	if (alloc_slab) {
3628		prot->slab = kmem_cache_create_usercopy(prot->name,
3629					prot->obj_size, 0,
3630					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3631					prot->slab_flags,
3632					prot->useroffset, prot->usersize,
3633					NULL);
3634
3635		if (prot->slab == NULL) {
3636			pr_crit("%s: Can't create sock SLAB cache!\n",
3637				prot->name);
3638			goto out;
3639		}
3640
3641		if (req_prot_init(prot))
3642			goto out_free_request_sock_slab;
3643
3644		if (tw_prot_init(prot))
3645			goto out_free_timewait_sock_slab;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3646	}
3647
3648	mutex_lock(&proto_list_mutex);
3649	ret = assign_proto_idx(prot);
3650	if (ret) {
3651		mutex_unlock(&proto_list_mutex);
3652		goto out_free_timewait_sock_slab;
3653	}
3654	list_add(&prot->node, &proto_list);
3655	mutex_unlock(&proto_list_mutex);
3656	return ret;
3657
3658out_free_timewait_sock_slab:
3659	if (alloc_slab)
3660		tw_prot_cleanup(prot->twsk_prot);
3661out_free_request_sock_slab:
3662	if (alloc_slab) {
3663		req_prot_cleanup(prot->rsk_prot);
3664
3665		kmem_cache_destroy(prot->slab);
3666		prot->slab = NULL;
3667	}
3668out:
3669	return ret;
3670}
3671EXPORT_SYMBOL(proto_register);
3672
3673void proto_unregister(struct proto *prot)
3674{
3675	mutex_lock(&proto_list_mutex);
3676	release_proto_idx(prot);
3677	list_del(&prot->node);
3678	mutex_unlock(&proto_list_mutex);
3679
3680	kmem_cache_destroy(prot->slab);
3681	prot->slab = NULL;
3682
3683	req_prot_cleanup(prot->rsk_prot);
3684	tw_prot_cleanup(prot->twsk_prot);
3685}
3686EXPORT_SYMBOL(proto_unregister);
3687
3688int sock_load_diag_module(int family, int protocol)
3689{
3690	if (!protocol) {
3691		if (!sock_is_registered(family))
3692			return -ENOENT;
3693
3694		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3695				      NETLINK_SOCK_DIAG, family);
3696	}
3697
3698#ifdef CONFIG_INET
3699	if (family == AF_INET &&
3700	    protocol != IPPROTO_RAW &&
3701	    protocol < MAX_INET_PROTOS &&
3702	    !rcu_access_pointer(inet_protos[protocol]))
3703		return -ENOENT;
3704#endif
3705
3706	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3707			      NETLINK_SOCK_DIAG, family, protocol);
3708}
3709EXPORT_SYMBOL(sock_load_diag_module);
3710
3711#ifdef CONFIG_PROC_FS
3712static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3713	__acquires(proto_list_mutex)
3714{
3715	mutex_lock(&proto_list_mutex);
3716	return seq_list_start_head(&proto_list, *pos);
3717}
3718
3719static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3720{
3721	return seq_list_next(v, &proto_list, pos);
3722}
3723
3724static void proto_seq_stop(struct seq_file *seq, void *v)
3725	__releases(proto_list_mutex)
3726{
3727	mutex_unlock(&proto_list_mutex);
3728}
3729
3730static char proto_method_implemented(const void *method)
3731{
3732	return method == NULL ? 'n' : 'y';
3733}
3734static long sock_prot_memory_allocated(struct proto *proto)
3735{
3736	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3737}
3738
3739static const char *sock_prot_memory_pressure(struct proto *proto)
3740{
3741	return proto->memory_pressure != NULL ?
3742	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3743}
3744
3745static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3746{
3747
3748	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3749			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3750		   proto->name,
3751		   proto->obj_size,
3752		   sock_prot_inuse_get(seq_file_net(seq), proto),
3753		   sock_prot_memory_allocated(proto),
3754		   sock_prot_memory_pressure(proto),
3755		   proto->max_header,
3756		   proto->slab == NULL ? "no" : "yes",
3757		   module_name(proto->owner),
3758		   proto_method_implemented(proto->close),
3759		   proto_method_implemented(proto->connect),
3760		   proto_method_implemented(proto->disconnect),
3761		   proto_method_implemented(proto->accept),
3762		   proto_method_implemented(proto->ioctl),
3763		   proto_method_implemented(proto->init),
3764		   proto_method_implemented(proto->destroy),
3765		   proto_method_implemented(proto->shutdown),
3766		   proto_method_implemented(proto->setsockopt),
3767		   proto_method_implemented(proto->getsockopt),
3768		   proto_method_implemented(proto->sendmsg),
3769		   proto_method_implemented(proto->recvmsg),
3770		   proto_method_implemented(proto->sendpage),
3771		   proto_method_implemented(proto->bind),
3772		   proto_method_implemented(proto->backlog_rcv),
3773		   proto_method_implemented(proto->hash),
3774		   proto_method_implemented(proto->unhash),
3775		   proto_method_implemented(proto->get_port),
3776		   proto_method_implemented(proto->enter_memory_pressure));
3777}
3778
3779static int proto_seq_show(struct seq_file *seq, void *v)
3780{
3781	if (v == &proto_list)
3782		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3783			   "protocol",
3784			   "size",
3785			   "sockets",
3786			   "memory",
3787			   "press",
3788			   "maxhdr",
3789			   "slab",
3790			   "module",
3791			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3792	else
3793		proto_seq_printf(seq, list_entry(v, struct proto, node));
3794	return 0;
3795}
3796
3797static const struct seq_operations proto_seq_ops = {
3798	.start  = proto_seq_start,
3799	.next   = proto_seq_next,
3800	.stop   = proto_seq_stop,
3801	.show   = proto_seq_show,
3802};
3803
3804static __net_init int proto_init_net(struct net *net)
3805{
3806	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3807			sizeof(struct seq_net_private)))
3808		return -ENOMEM;
3809
3810	return 0;
3811}
3812
3813static __net_exit void proto_exit_net(struct net *net)
3814{
3815	remove_proc_entry("protocols", net->proc_net);
3816}
3817
3818
3819static __net_initdata struct pernet_operations proto_net_ops = {
3820	.init = proto_init_net,
3821	.exit = proto_exit_net,
3822};
3823
3824static int __init proto_init(void)
3825{
3826	return register_pernet_subsys(&proto_net_ops);
3827}
3828
3829subsys_initcall(proto_init);
3830
3831#endif /* PROC_FS */
3832
3833#ifdef CONFIG_NET_RX_BUSY_POLL
3834bool sk_busy_loop_end(void *p, unsigned long start_time)
3835{
3836	struct sock *sk = p;
3837
3838	return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3839	       sk_busy_loop_timeout(sk, start_time);
3840}
3841EXPORT_SYMBOL(sk_busy_loop_end);
3842#endif /* CONFIG_NET_RX_BUSY_POLL */
3843
3844int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3845{
3846	if (!sk->sk_prot->bind_add)
3847		return -EOPNOTSUPP;
3848	return sk->sk_prot->bind_add(sk, addr, addr_len);
3849}
3850EXPORT_SYMBOL(sock_bind_add);
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   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.
   6 *
   7 *		Generic socket support routines. Memory allocators, socket lock/release
   8 *		handler for protocols to use and generic option handler.
   9 *
  10 * Authors:	Ross Biro
  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>
  14 *
  15 * Fixes:
  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
  36 *					TCP layer surgery.
  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
  64 *					(compatibility fix)
  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
  82 *
  83 * To Fix:
  84 */
  85
  86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  87
  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>
  94#include <linux/in.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>
 117
 118#include <linux/uaccess.h>
 119
 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>
 132
 133#include <linux/filter.h>
 134#include <net/sock_reuseport.h>
 135#include <net/bpf_sk_storage.h>
 136
 137#include <trace/events/sock.h>
 138
 139#include <net/tcp.h>
 140#include <net/busy_poll.h>
 141
 
 
 142static DEFINE_MUTEX(proto_list_mutex);
 143static LIST_HEAD(proto_list);
 144
 145static void sock_inuse_add(struct net *net, int val);
 146
 147/**
 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
 152 *
 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.
 156 */
 157bool sk_ns_capable(const struct sock *sk,
 158		   struct user_namespace *user_ns, int cap)
 159{
 160	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
 161		ns_capable(user_ns, cap);
 162}
 163EXPORT_SYMBOL(sk_ns_capable);
 164
 165/**
 166 * sk_capable - Socket global capability test
 167 * @sk: Socket to use a capability on or through
 168 * @cap: The global capability to use
 169 *
 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
 172 * namespaces.
 173 */
 174bool sk_capable(const struct sock *sk, int cap)
 175{
 176	return sk_ns_capable(sk, &init_user_ns, cap);
 177}
 178EXPORT_SYMBOL(sk_capable);
 179
 180/**
 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
 184 *
 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.
 188 */
 189bool sk_net_capable(const struct sock *sk, int cap)
 190{
 191	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
 192}
 193EXPORT_SYMBOL(sk_net_capable);
 194
 195/*
 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
 198 * userspace sockets.
 199 */
 200static struct lock_class_key af_family_keys[AF_MAX];
 201static struct lock_class_key af_family_kern_keys[AF_MAX];
 202static struct lock_class_key af_family_slock_keys[AF_MAX];
 203static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
 204
 205/*
 206 * Make lock validator output more readable. (we pre-construct these
 207 * strings build-time, so that runtime initialization of socket
 208 * locks is fast):
 209 */
 210
 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"	, \
 227  x "AF_MAX"
 228
 229static const char *const af_family_key_strings[AF_MAX+1] = {
 230	_sock_locks("sk_lock-")
 231};
 232static const char *const af_family_slock_key_strings[AF_MAX+1] = {
 233	_sock_locks("slock-")
 234};
 235static const char *const af_family_clock_key_strings[AF_MAX+1] = {
 236	_sock_locks("clock-")
 237};
 238
 239static const char *const af_family_kern_key_strings[AF_MAX+1] = {
 240	_sock_locks("k-sk_lock-")
 241};
 242static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
 243	_sock_locks("k-slock-")
 244};
 245static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
 246	_sock_locks("k-clock-")
 247};
 248static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
 249	_sock_locks("rlock-")
 250};
 251static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
 252	_sock_locks("wlock-")
 253};
 254static const char *const af_family_elock_key_strings[AF_MAX+1] = {
 255	_sock_locks("elock-")
 256};
 257
 258/*
 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:
 261 */
 262static struct lock_class_key af_callback_keys[AF_MAX];
 263static struct lock_class_key af_rlock_keys[AF_MAX];
 264static struct lock_class_key af_wlock_keys[AF_MAX];
 265static struct lock_class_key af_elock_keys[AF_MAX];
 266static struct lock_class_key af_kern_callback_keys[AF_MAX];
 267
 268/* Run time adjustable parameters. */
 269__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 270EXPORT_SYMBOL(sysctl_wmem_max);
 271__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 272EXPORT_SYMBOL(sysctl_rmem_max);
 273__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 274__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 275
 276/* Maximal space eaten by iovec or ancillary data plus some space */
 277int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 278EXPORT_SYMBOL(sysctl_optmem_max);
 279
 280int sysctl_tstamp_allow_data __read_mostly = 1;
 281
 282DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
 283EXPORT_SYMBOL_GPL(memalloc_socks_key);
 284
 285/**
 286 * sk_set_memalloc - sets %SOCK_MEMALLOC
 287 * @sk: socket to set it on
 288 *
 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
 292 */
 293void sk_set_memalloc(struct sock *sk)
 294{
 295	sock_set_flag(sk, SOCK_MEMALLOC);
 296	sk->sk_allocation |= __GFP_MEMALLOC;
 297	static_branch_inc(&memalloc_socks_key);
 298}
 299EXPORT_SYMBOL_GPL(sk_set_memalloc);
 300
 301void sk_clear_memalloc(struct sock *sk)
 302{
 303	sock_reset_flag(sk, SOCK_MEMALLOC);
 304	sk->sk_allocation &= ~__GFP_MEMALLOC;
 305	static_branch_dec(&memalloc_socks_key);
 306
 307	/*
 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.
 313	 */
 314	sk_mem_reclaim(sk);
 315}
 316EXPORT_SYMBOL_GPL(sk_clear_memalloc);
 317
 318int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 319{
 320	int ret;
 321	unsigned int noreclaim_flag;
 322
 323	/* these should have been dropped before queueing */
 324	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
 325
 326	noreclaim_flag = memalloc_noreclaim_save();
 327	ret = sk->sk_backlog_rcv(sk, skb);
 328	memalloc_noreclaim_restore(noreclaim_flag);
 329
 330	return ret;
 331}
 332EXPORT_SYMBOL(__sk_backlog_rcv);
 333
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 334static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
 335{
 336	struct __kernel_sock_timeval tv;
 337
 338	if (timeo == MAX_SCHEDULE_TIMEOUT) {
 339		tv.tv_sec = 0;
 340		tv.tv_usec = 0;
 341	} else {
 342		tv.tv_sec = timeo / HZ;
 343		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
 344	}
 345
 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;
 349		return sizeof(tv32);
 350	}
 351
 352	if (old_timeval) {
 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);
 358	}
 359
 360	*(struct __kernel_sock_timeval *)optval = tv;
 361	return sizeof(tv);
 362}
 363
 364static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
 365			    bool old_timeval)
 366{
 367	struct __kernel_sock_timeval tv;
 368
 369	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
 370		struct old_timeval32 tv32;
 371
 372		if (optlen < sizeof(tv32))
 373			return -EINVAL;
 374
 375		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
 376			return -EFAULT;
 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;
 381
 382		if (optlen < sizeof(old_tv))
 383			return -EINVAL;
 384		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
 385			return -EFAULT;
 386		tv.tv_sec = old_tv.tv_sec;
 387		tv.tv_usec = old_tv.tv_usec;
 388	} else {
 389		if (optlen < sizeof(tv))
 390			return -EINVAL;
 391		if (copy_from_sockptr(&tv, optval, sizeof(tv)))
 392			return -EFAULT;
 393	}
 394	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
 395		return -EDOM;
 396
 397	if (tv.tv_sec < 0) {
 398		static int warned __read_mostly;
 399
 400		*timeo_p = 0;
 401		if (warned < 10 && net_ratelimit()) {
 402			warned++;
 403			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
 404				__func__, current->comm, task_pid_nr(current));
 405		}
 406		return 0;
 407	}
 408	*timeo_p = MAX_SCHEDULE_TIMEOUT;
 409	if (tv.tv_sec == 0 && tv.tv_usec == 0)
 410		return 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);
 413	return 0;
 414}
 415
 416static void sock_warn_obsolete_bsdism(const char *name)
 417{
 418	static int warned;
 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",
 423			warncomm, name);
 424		warned++;
 425	}
 426}
 427
 428static bool sock_needs_netstamp(const struct sock *sk)
 429{
 430	switch (sk->sk_family) {
 431	case AF_UNSPEC:
 432	case AF_UNIX:
 433		return false;
 434	default:
 435		return true;
 436	}
 437}
 438
 439static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
 440{
 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();
 446	}
 447}
 448
 449
 450int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 451{
 452	unsigned long flags;
 453	struct sk_buff_head *list = &sk->sk_receive_queue;
 454
 455	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
 456		atomic_inc(&sk->sk_drops);
 457		trace_sock_rcvqueue_full(sk, skb);
 458		return -ENOMEM;
 459	}
 460
 461	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
 462		atomic_inc(&sk->sk_drops);
 463		return -ENOBUFS;
 464	}
 465
 466	skb->dev = NULL;
 467	skb_set_owner_r(skb, sk);
 468
 469	/* we escape from rcu protected region, make sure we dont leak
 470	 * a norefcounted dst
 471	 */
 472	skb_dst_force(skb);
 473
 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);
 478
 479	if (!sock_flag(sk, SOCK_DEAD))
 480		sk->sk_data_ready(sk);
 481	return 0;
 482}
 483EXPORT_SYMBOL(__sock_queue_rcv_skb);
 484
 485int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 486{
 487	int err;
 488
 489	err = sk_filter(sk, skb);
 490	if (err)
 491		return err;
 492
 493	return __sock_queue_rcv_skb(sk, skb);
 494}
 495EXPORT_SYMBOL(sock_queue_rcv_skb);
 496
 497int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
 498		     const int nested, unsigned int trim_cap, bool refcounted)
 499{
 500	int rc = NET_RX_SUCCESS;
 501
 502	if (sk_filter_trim_cap(sk, skb, trim_cap))
 503		goto discard_and_relse;
 504
 505	skb->dev = NULL;
 506
 507	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
 508		atomic_inc(&sk->sk_drops);
 509		goto discard_and_relse;
 510	}
 511	if (nested)
 512		bh_lock_sock_nested(sk);
 513	else
 514		bh_lock_sock(sk);
 515	if (!sock_owned_by_user(sk)) {
 516		/*
 517		 * trylock + unlock semantics:
 518		 */
 519		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 520
 521		rc = sk_backlog_rcv(sk, skb);
 522
 523		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
 524	} else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
 525		bh_unlock_sock(sk);
 526		atomic_inc(&sk->sk_drops);
 527		goto discard_and_relse;
 528	}
 529
 530	bh_unlock_sock(sk);
 531out:
 532	if (refcounted)
 533		sock_put(sk);
 534	return rc;
 535discard_and_relse:
 536	kfree_skb(skb);
 537	goto out;
 538}
 539EXPORT_SYMBOL(__sk_receive_skb);
 540
 
 
 
 
 541struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 542{
 543	struct dst_entry *dst = __sk_dst_get(sk);
 544
 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);
 549		dst_release(dst);
 550		return NULL;
 551	}
 552
 553	return dst;
 554}
 555EXPORT_SYMBOL(__sk_dst_check);
 556
 557struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 558{
 559	struct dst_entry *dst = sk_dst_get(sk);
 560
 561	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 
 
 562		sk_dst_reset(sk);
 563		dst_release(dst);
 564		return NULL;
 565	}
 566
 567	return dst;
 568}
 569EXPORT_SYMBOL(sk_dst_check);
 570
 571static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
 572{
 573	int ret = -ENOPROTOOPT;
 574#ifdef CONFIG_NETDEVICES
 575	struct net *net = sock_net(sk);
 576
 577	/* Sorry... */
 578	ret = -EPERM;
 579	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
 580		goto out;
 581
 582	ret = -EINVAL;
 583	if (ifindex < 0)
 584		goto out;
 585
 586	sk->sk_bound_dev_if = ifindex;
 587	if (sk->sk_prot->rehash)
 588		sk->sk_prot->rehash(sk);
 589	sk_dst_reset(sk);
 590
 591	ret = 0;
 592
 593out:
 594#endif
 595
 596	return ret;
 597}
 598
 599int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
 600{
 601	int ret;
 602
 603	if (lock_sk)
 604		lock_sock(sk);
 605	ret = sock_bindtoindex_locked(sk, ifindex);
 606	if (lock_sk)
 607		release_sock(sk);
 608
 609	return ret;
 610}
 611EXPORT_SYMBOL(sock_bindtoindex);
 612
 613static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
 614{
 615	int ret = -ENOPROTOOPT;
 616#ifdef CONFIG_NETDEVICES
 617	struct net *net = sock_net(sk);
 618	char devname[IFNAMSIZ];
 619	int index;
 620
 621	ret = -EINVAL;
 622	if (optlen < 0)
 623		goto out;
 624
 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
 628	 * is not bound.
 629	 */
 630	if (optlen > IFNAMSIZ - 1)
 631		optlen = IFNAMSIZ - 1;
 632	memset(devname, 0, sizeof(devname));
 633
 634	ret = -EFAULT;
 635	if (copy_from_sockptr(devname, optval, optlen))
 636		goto out;
 637
 638	index = 0;
 639	if (devname[0] != '\0') {
 640		struct net_device *dev;
 641
 642		rcu_read_lock();
 643		dev = dev_get_by_name_rcu(net, devname);
 644		if (dev)
 645			index = dev->ifindex;
 646		rcu_read_unlock();
 647		ret = -ENODEV;
 648		if (!dev)
 649			goto out;
 650	}
 651
 652	return sock_bindtoindex(sk, index, true);
 653out:
 654#endif
 655
 656	return ret;
 657}
 658
 659static int sock_getbindtodevice(struct sock *sk, char __user *optval,
 660				int __user *optlen, int len)
 661{
 662	int ret = -ENOPROTOOPT;
 663#ifdef CONFIG_NETDEVICES
 664	struct net *net = sock_net(sk);
 665	char devname[IFNAMSIZ];
 666
 667	if (sk->sk_bound_dev_if == 0) {
 668		len = 0;
 669		goto zero;
 670	}
 671
 672	ret = -EINVAL;
 673	if (len < IFNAMSIZ)
 674		goto out;
 675
 676	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
 677	if (ret)
 678		goto out;
 679
 680	len = strlen(devname) + 1;
 681
 682	ret = -EFAULT;
 683	if (copy_to_user(optval, devname, len))
 684		goto out;
 685
 686zero:
 687	ret = -EFAULT;
 688	if (put_user(len, optlen))
 689		goto out;
 690
 691	ret = 0;
 692
 693out:
 694#endif
 695
 696	return ret;
 697}
 698
 699bool sk_mc_loop(struct sock *sk)
 700{
 701	if (dev_recursion_level())
 702		return false;
 703	if (!sk)
 704		return true;
 705	switch (sk->sk_family) {
 706	case AF_INET:
 707		return inet_sk(sk)->mc_loop;
 708#if IS_ENABLED(CONFIG_IPV6)
 709	case AF_INET6:
 710		return inet6_sk(sk)->mc_loop;
 711#endif
 712	}
 713	WARN_ON_ONCE(1);
 714	return true;
 715}
 716EXPORT_SYMBOL(sk_mc_loop);
 717
 718void sock_set_reuseaddr(struct sock *sk)
 719{
 720	lock_sock(sk);
 721	sk->sk_reuse = SK_CAN_REUSE;
 722	release_sock(sk);
 723}
 724EXPORT_SYMBOL(sock_set_reuseaddr);
 725
 726void sock_set_reuseport(struct sock *sk)
 727{
 728	lock_sock(sk);
 729	sk->sk_reuseport = true;
 730	release_sock(sk);
 731}
 732EXPORT_SYMBOL(sock_set_reuseport);
 733
 734void sock_no_linger(struct sock *sk)
 735{
 736	lock_sock(sk);
 737	sk->sk_lingertime = 0;
 738	sock_set_flag(sk, SOCK_LINGER);
 739	release_sock(sk);
 740}
 741EXPORT_SYMBOL(sock_no_linger);
 742
 743void sock_set_priority(struct sock *sk, u32 priority)
 744{
 745	lock_sock(sk);
 746	sk->sk_priority = priority;
 747	release_sock(sk);
 748}
 749EXPORT_SYMBOL(sock_set_priority);
 750
 751void sock_set_sndtimeo(struct sock *sk, s64 secs)
 752{
 753	lock_sock(sk);
 754	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
 755		sk->sk_sndtimeo = secs * HZ;
 756	else
 757		sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
 758	release_sock(sk);
 759}
 760EXPORT_SYMBOL(sock_set_sndtimeo);
 761
 762static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
 763{
 764	if (val)  {
 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);
 769	} else {
 770		sock_reset_flag(sk, SOCK_RCVTSTAMP);
 771		sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 772		sock_reset_flag(sk, SOCK_TSTAMP_NEW);
 773	}
 774}
 775
 776void sock_enable_timestamps(struct sock *sk)
 777{
 778	lock_sock(sk);
 779	__sock_set_timestamps(sk, true, false, true);
 780	release_sock(sk);
 781}
 782EXPORT_SYMBOL(sock_enable_timestamps);
 783
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 784void sock_set_keepalive(struct sock *sk)
 785{
 786	lock_sock(sk);
 787	if (sk->sk_prot->keepalive)
 788		sk->sk_prot->keepalive(sk, true);
 789	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
 790	release_sock(sk);
 791}
 792EXPORT_SYMBOL(sock_set_keepalive);
 793
 794static void __sock_set_rcvbuf(struct sock *sk, int val)
 795{
 796	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
 797	 * as a negative value.
 798	 */
 799	val = min_t(int, val, INT_MAX / 2);
 800	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 801
 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.
 805	 *
 806	 * Applications are unaware that "struct sk_buff" and other overheads
 807	 * allocate from the receive buffer during socket buffer allocation.
 808	 *
 809	 * And after considering the possible alternatives, returning the value
 810	 * we actually used in getsockopt is the most desirable behavior.
 811	 */
 812	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
 813}
 814
 815void sock_set_rcvbuf(struct sock *sk, int val)
 816{
 817	lock_sock(sk);
 818	__sock_set_rcvbuf(sk, val);
 819	release_sock(sk);
 820}
 821EXPORT_SYMBOL(sock_set_rcvbuf);
 822
 
 
 
 
 
 
 
 
 823void sock_set_mark(struct sock *sk, u32 val)
 824{
 825	lock_sock(sk);
 826	sk->sk_mark = val;
 827	release_sock(sk);
 828}
 829EXPORT_SYMBOL(sock_set_mark);
 830
 831/*
 832 *	This is meant for all protocols to use and covers goings on
 833 *	at the socket level. Everything here is generic.
 834 */
 835
 836int sock_setsockopt(struct socket *sock, int level, int optname,
 837		    sockptr_t optval, unsigned int optlen)
 838{
 
 839	struct sock_txtime sk_txtime;
 840	struct sock *sk = sock->sk;
 841	int val;
 842	int valbool;
 843	struct linger ling;
 844	int ret = 0;
 845
 846	/*
 847	 *	Options without arguments
 848	 */
 849
 850	if (optname == SO_BINDTODEVICE)
 851		return sock_setbindtodevice(sk, optval, optlen);
 852
 853	if (optlen < sizeof(int))
 854		return -EINVAL;
 855
 856	if (copy_from_sockptr(&val, optval, sizeof(val)))
 857		return -EFAULT;
 858
 859	valbool = val ? 1 : 0;
 860
 861	lock_sock(sk);
 862
 863	switch (optname) {
 864	case SO_DEBUG:
 865		if (val && !capable(CAP_NET_ADMIN))
 866			ret = -EACCES;
 867		else
 868			sock_valbool_flag(sk, SOCK_DBG, valbool);
 869		break;
 870	case SO_REUSEADDR:
 871		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
 872		break;
 873	case SO_REUSEPORT:
 874		sk->sk_reuseport = valbool;
 875		break;
 876	case SO_TYPE:
 877	case SO_PROTOCOL:
 878	case SO_DOMAIN:
 879	case SO_ERROR:
 880		ret = -ENOPROTOOPT;
 881		break;
 882	case SO_DONTROUTE:
 883		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
 884		sk_dst_reset(sk);
 885		break;
 886	case SO_BROADCAST:
 887		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 888		break;
 889	case SO_SNDBUF:
 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
 894		 */
 895		val = min_t(u32, val, sysctl_wmem_max);
 896set_sndbuf:
 897		/* Ensure val * 2 fits into an int, to prevent max_t()
 898		 * from treating it as a negative value.
 899		 */
 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);
 906		break;
 907
 908	case SO_SNDBUFFORCE:
 909		if (!capable(CAP_NET_ADMIN)) {
 910			ret = -EPERM;
 911			break;
 912		}
 913
 914		/* No negative values (to prevent underflow, as val will be
 915		 * multiplied by 2).
 916		 */
 917		if (val < 0)
 918			val = 0;
 919		goto set_sndbuf;
 920
 921	case SO_RCVBUF:
 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
 926		 */
 927		__sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
 928		break;
 929
 930	case SO_RCVBUFFORCE:
 931		if (!capable(CAP_NET_ADMIN)) {
 932			ret = -EPERM;
 933			break;
 934		}
 935
 936		/* No negative values (to prevent underflow, as val will be
 937		 * multiplied by 2).
 938		 */
 939		__sock_set_rcvbuf(sk, max(val, 0));
 940		break;
 941
 942	case SO_KEEPALIVE:
 943		if (sk->sk_prot->keepalive)
 944			sk->sk_prot->keepalive(sk, valbool);
 945		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 946		break;
 947
 948	case SO_OOBINLINE:
 949		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 950		break;
 951
 952	case SO_NO_CHECK:
 953		sk->sk_no_check_tx = valbool;
 954		break;
 955
 956	case SO_PRIORITY:
 957		if ((val >= 0 && val <= 6) ||
 958		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
 959			sk->sk_priority = val;
 960		else
 961			ret = -EPERM;
 962		break;
 963
 964	case SO_LINGER:
 965		if (optlen < sizeof(ling)) {
 966			ret = -EINVAL;	/* 1003.1g */
 967			break;
 968		}
 969		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
 970			ret = -EFAULT;
 971			break;
 972		}
 973		if (!ling.l_onoff)
 974			sock_reset_flag(sk, SOCK_LINGER);
 975		else {
 976#if (BITS_PER_LONG == 32)
 977			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 978				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 979			else
 980#endif
 981				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 982			sock_set_flag(sk, SOCK_LINGER);
 983		}
 984		break;
 985
 986	case SO_BSDCOMPAT:
 987		sock_warn_obsolete_bsdism("setsockopt");
 988		break;
 989
 990	case SO_PASSCRED:
 991		if (valbool)
 992			set_bit(SOCK_PASSCRED, &sock->flags);
 993		else
 994			clear_bit(SOCK_PASSCRED, &sock->flags);
 995		break;
 996
 997	case SO_TIMESTAMP_OLD:
 998		__sock_set_timestamps(sk, valbool, false, false);
 999		break;
1000	case SO_TIMESTAMP_NEW:
1001		__sock_set_timestamps(sk, valbool, true, false);
1002		break;
1003	case SO_TIMESTAMPNS_OLD:
1004		__sock_set_timestamps(sk, valbool, false, true);
1005		break;
1006	case SO_TIMESTAMPNS_NEW:
1007		__sock_set_timestamps(sk, valbool, true, true);
1008		break;
 
1009	case SO_TIMESTAMPING_NEW:
1010		sock_set_flag(sk, SOCK_TSTAMP_NEW);
1011		fallthrough;
1012	case SO_TIMESTAMPING_OLD:
1013		if (val & ~SOF_TIMESTAMPING_MASK) {
1014			ret = -EINVAL;
1015			break;
1016		}
1017
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)) {
1024					ret = -EINVAL;
1025					break;
1026				}
1027				sk->sk_tskey = tcp_sk(sk)->snd_una;
1028			} else {
1029				sk->sk_tskey = 0;
1030			}
 
 
 
1031		}
1032
1033		if (val & SOF_TIMESTAMPING_OPT_STATS &&
1034		    !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1035			ret = -EINVAL;
1036			break;
1037		}
1038
1039		sk->sk_tsflags = val;
1040		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1041			sock_enable_timestamp(sk,
1042					      SOCK_TIMESTAMPING_RX_SOFTWARE);
1043		else {
1044			if (optname == SO_TIMESTAMPING_NEW)
1045				sock_reset_flag(sk, SOCK_TSTAMP_NEW);
1046
1047			sock_disable_timestamp(sk,
1048					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1049		}
1050		break;
1051
1052	case SO_RCVLOWAT:
1053		if (val < 0)
1054			val = INT_MAX;
1055		if (sock->ops->set_rcvlowat)
1056			ret = sock->ops->set_rcvlowat(sk, val);
1057		else
1058			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1059		break;
1060
1061	case SO_RCVTIMEO_OLD:
1062	case SO_RCVTIMEO_NEW:
1063		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1064				       optlen, optname == SO_RCVTIMEO_OLD);
1065		break;
1066
1067	case SO_SNDTIMEO_OLD:
1068	case SO_SNDTIMEO_NEW:
1069		ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1070				       optlen, optname == SO_SNDTIMEO_OLD);
1071		break;
1072
1073	case SO_ATTACH_FILTER: {
1074		struct sock_fprog fprog;
1075
1076		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1077		if (!ret)
1078			ret = sk_attach_filter(&fprog, sk);
1079		break;
1080	}
1081	case SO_ATTACH_BPF:
1082		ret = -EINVAL;
1083		if (optlen == sizeof(u32)) {
1084			u32 ufd;
1085
1086			ret = -EFAULT;
1087			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1088				break;
1089
1090			ret = sk_attach_bpf(ufd, sk);
1091		}
1092		break;
1093
1094	case SO_ATTACH_REUSEPORT_CBPF: {
1095		struct sock_fprog fprog;
1096
1097		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1098		if (!ret)
1099			ret = sk_reuseport_attach_filter(&fprog, sk);
1100		break;
1101	}
1102	case SO_ATTACH_REUSEPORT_EBPF:
1103		ret = -EINVAL;
1104		if (optlen == sizeof(u32)) {
1105			u32 ufd;
1106
1107			ret = -EFAULT;
1108			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1109				break;
1110
1111			ret = sk_reuseport_attach_bpf(ufd, sk);
1112		}
1113		break;
1114
1115	case SO_DETACH_REUSEPORT_BPF:
1116		ret = reuseport_detach_prog(sk);
1117		break;
1118
1119	case SO_DETACH_FILTER:
1120		ret = sk_detach_filter(sk);
1121		break;
1122
1123	case SO_LOCK_FILTER:
1124		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1125			ret = -EPERM;
1126		else
1127			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1128		break;
1129
1130	case SO_PASSSEC:
1131		if (valbool)
1132			set_bit(SOCK_PASSSEC, &sock->flags);
1133		else
1134			clear_bit(SOCK_PASSSEC, &sock->flags);
1135		break;
1136	case SO_MARK:
1137		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1138			ret = -EPERM;
1139		} else if (val != sk->sk_mark) {
1140			sk->sk_mark = val;
1141			sk_dst_reset(sk);
1142		}
 
 
1143		break;
1144
1145	case SO_RXQ_OVFL:
1146		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1147		break;
1148
1149	case SO_WIFI_STATUS:
1150		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1151		break;
1152
1153	case SO_PEEK_OFF:
1154		if (sock->ops->set_peek_off)
1155			ret = sock->ops->set_peek_off(sk, val);
1156		else
1157			ret = -EOPNOTSUPP;
1158		break;
1159
1160	case SO_NOFCS:
1161		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1162		break;
1163
1164	case SO_SELECT_ERR_QUEUE:
1165		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1166		break;
1167
1168#ifdef CONFIG_NET_RX_BUSY_POLL
1169	case SO_BUSY_POLL:
1170		/* allow unprivileged users to decrease the value */
1171		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1172			ret = -EPERM;
1173		else {
1174			if (val < 0)
1175				ret = -EINVAL;
1176			else
1177				sk->sk_ll_usec = val;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1178		}
1179		break;
1180#endif
1181
1182	case SO_MAX_PACING_RATE:
1183		{
1184		unsigned long ulval = (val == ~0U) ? ~0UL : val;
1185
1186		if (sizeof(ulval) != sizeof(val) &&
1187		    optlen >= sizeof(ulval) &&
1188		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1189			ret = -EFAULT;
1190			break;
1191		}
1192		if (ulval != ~0UL)
1193			cmpxchg(&sk->sk_pacing_status,
1194				SK_PACING_NONE,
1195				SK_PACING_NEEDED);
1196		sk->sk_max_pacing_rate = ulval;
1197		sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1198		break;
1199		}
1200	case SO_INCOMING_CPU:
1201		WRITE_ONCE(sk->sk_incoming_cpu, val);
1202		break;
1203
1204	case SO_CNX_ADVICE:
1205		if (val == 1)
1206			dst_negative_advice(sk);
1207		break;
1208
1209	case SO_ZEROCOPY:
1210		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1211			if (!((sk->sk_type == SOCK_STREAM &&
1212			       sk->sk_protocol == IPPROTO_TCP) ||
1213			      (sk->sk_type == SOCK_DGRAM &&
1214			       sk->sk_protocol == IPPROTO_UDP)))
1215				ret = -ENOTSUPP;
1216		} else if (sk->sk_family != PF_RDS) {
1217			ret = -ENOTSUPP;
1218		}
1219		if (!ret) {
1220			if (val < 0 || val > 1)
1221				ret = -EINVAL;
1222			else
1223				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1224		}
1225		break;
1226
1227	case SO_TXTIME:
1228		if (optlen != sizeof(struct sock_txtime)) {
1229			ret = -EINVAL;
1230			break;
1231		} else if (copy_from_sockptr(&sk_txtime, optval,
1232			   sizeof(struct sock_txtime))) {
1233			ret = -EFAULT;
1234			break;
1235		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1236			ret = -EINVAL;
1237			break;
1238		}
1239		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1240		 * scheduler has enough safe guards.
1241		 */
1242		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1243		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1244			ret = -EPERM;
1245			break;
1246		}
1247		sock_valbool_flag(sk, SOCK_TXTIME, true);
1248		sk->sk_clockid = sk_txtime.clockid;
1249		sk->sk_txtime_deadline_mode =
1250			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1251		sk->sk_txtime_report_errors =
1252			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1253		break;
1254
1255	case SO_BINDTOIFINDEX:
1256		ret = sock_bindtoindex_locked(sk, val);
1257		break;
1258
1259	default:
1260		ret = -ENOPROTOOPT;
1261		break;
1262	}
1263	release_sock(sk);
1264	return ret;
1265}
1266EXPORT_SYMBOL(sock_setsockopt);
1267
 
 
 
 
 
 
 
 
 
 
1268
1269static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1270			  struct ucred *ucred)
1271{
1272	ucred->pid = pid_vnr(pid);
1273	ucred->uid = ucred->gid = -1;
1274	if (cred) {
1275		struct user_namespace *current_ns = current_user_ns();
1276
1277		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1278		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1279	}
1280}
1281
1282static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1283{
1284	struct user_namespace *user_ns = current_user_ns();
1285	int i;
1286
1287	for (i = 0; i < src->ngroups; i++)
1288		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1289			return -EFAULT;
1290
1291	return 0;
1292}
1293
1294int sock_getsockopt(struct socket *sock, int level, int optname,
1295		    char __user *optval, int __user *optlen)
1296{
1297	struct sock *sk = sock->sk;
1298
1299	union {
1300		int val;
1301		u64 val64;
1302		unsigned long ulval;
1303		struct linger ling;
1304		struct old_timeval32 tm32;
1305		struct __kernel_old_timeval tm;
1306		struct  __kernel_sock_timeval stm;
1307		struct sock_txtime txtime;
 
1308	} v;
1309
1310	int lv = sizeof(int);
1311	int len;
1312
1313	if (get_user(len, optlen))
1314		return -EFAULT;
1315	if (len < 0)
1316		return -EINVAL;
1317
1318	memset(&v, 0, sizeof(v));
1319
1320	switch (optname) {
1321	case SO_DEBUG:
1322		v.val = sock_flag(sk, SOCK_DBG);
1323		break;
1324
1325	case SO_DONTROUTE:
1326		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1327		break;
1328
1329	case SO_BROADCAST:
1330		v.val = sock_flag(sk, SOCK_BROADCAST);
1331		break;
1332
1333	case SO_SNDBUF:
1334		v.val = sk->sk_sndbuf;
1335		break;
1336
1337	case SO_RCVBUF:
1338		v.val = sk->sk_rcvbuf;
1339		break;
1340
1341	case SO_REUSEADDR:
1342		v.val = sk->sk_reuse;
1343		break;
1344
1345	case SO_REUSEPORT:
1346		v.val = sk->sk_reuseport;
1347		break;
1348
1349	case SO_KEEPALIVE:
1350		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1351		break;
1352
1353	case SO_TYPE:
1354		v.val = sk->sk_type;
1355		break;
1356
1357	case SO_PROTOCOL:
1358		v.val = sk->sk_protocol;
1359		break;
1360
1361	case SO_DOMAIN:
1362		v.val = sk->sk_family;
1363		break;
1364
1365	case SO_ERROR:
1366		v.val = -sock_error(sk);
1367		if (v.val == 0)
1368			v.val = xchg(&sk->sk_err_soft, 0);
1369		break;
1370
1371	case SO_OOBINLINE:
1372		v.val = sock_flag(sk, SOCK_URGINLINE);
1373		break;
1374
1375	case SO_NO_CHECK:
1376		v.val = sk->sk_no_check_tx;
1377		break;
1378
1379	case SO_PRIORITY:
1380		v.val = sk->sk_priority;
1381		break;
1382
1383	case SO_LINGER:
1384		lv		= sizeof(v.ling);
1385		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1386		v.ling.l_linger	= sk->sk_lingertime / HZ;
1387		break;
1388
1389	case SO_BSDCOMPAT:
1390		sock_warn_obsolete_bsdism("getsockopt");
1391		break;
1392
1393	case SO_TIMESTAMP_OLD:
1394		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1395				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1396				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1397		break;
1398
1399	case SO_TIMESTAMPNS_OLD:
1400		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1401		break;
1402
1403	case SO_TIMESTAMP_NEW:
1404		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1405		break;
1406
1407	case SO_TIMESTAMPNS_NEW:
1408		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1409		break;
1410
1411	case SO_TIMESTAMPING_OLD:
1412		v.val = sk->sk_tsflags;
 
 
1413		break;
1414
1415	case SO_RCVTIMEO_OLD:
1416	case SO_RCVTIMEO_NEW:
1417		lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1418		break;
1419
1420	case SO_SNDTIMEO_OLD:
1421	case SO_SNDTIMEO_NEW:
1422		lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1423		break;
1424
1425	case SO_RCVLOWAT:
1426		v.val = sk->sk_rcvlowat;
1427		break;
1428
1429	case SO_SNDLOWAT:
1430		v.val = 1;
1431		break;
1432
1433	case SO_PASSCRED:
1434		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1435		break;
1436
1437	case SO_PEERCRED:
1438	{
1439		struct ucred peercred;
1440		if (len > sizeof(peercred))
1441			len = sizeof(peercred);
 
 
1442		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
 
 
1443		if (copy_to_user(optval, &peercred, len))
1444			return -EFAULT;
1445		goto lenout;
1446	}
1447
1448	case SO_PEERGROUPS:
1449	{
 
1450		int ret, n;
1451
1452		if (!sk->sk_peer_cred)
 
1453			return -ENODATA;
1454
1455		n = sk->sk_peer_cred->group_info->ngroups;
1456		if (len < n * sizeof(gid_t)) {
1457			len = n * sizeof(gid_t);
 
1458			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1459		}
1460		len = n * sizeof(gid_t);
1461
1462		ret = groups_to_user((gid_t __user *)optval,
1463				     sk->sk_peer_cred->group_info);
1464		if (ret)
1465			return ret;
1466		goto lenout;
1467	}
1468
1469	case SO_PEERNAME:
1470	{
1471		char address[128];
1472
1473		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1474		if (lv < 0)
1475			return -ENOTCONN;
1476		if (lv < len)
1477			return -EINVAL;
1478		if (copy_to_user(optval, address, len))
1479			return -EFAULT;
1480		goto lenout;
1481	}
1482
1483	/* Dubious BSD thing... Probably nobody even uses it, but
1484	 * the UNIX standard wants it for whatever reason... -DaveM
1485	 */
1486	case SO_ACCEPTCONN:
1487		v.val = sk->sk_state == TCP_LISTEN;
1488		break;
1489
1490	case SO_PASSSEC:
1491		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1492		break;
1493
1494	case SO_PEERSEC:
1495		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1496
1497	case SO_MARK:
1498		v.val = sk->sk_mark;
1499		break;
1500
1501	case SO_RXQ_OVFL:
1502		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1503		break;
1504
1505	case SO_WIFI_STATUS:
1506		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1507		break;
1508
1509	case SO_PEEK_OFF:
1510		if (!sock->ops->set_peek_off)
1511			return -EOPNOTSUPP;
1512
1513		v.val = sk->sk_peek_off;
1514		break;
1515	case SO_NOFCS:
1516		v.val = sock_flag(sk, SOCK_NOFCS);
1517		break;
1518
1519	case SO_BINDTODEVICE:
1520		return sock_getbindtodevice(sk, optval, optlen, len);
1521
1522	case SO_GET_FILTER:
1523		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1524		if (len < 0)
1525			return len;
1526
1527		goto lenout;
1528
1529	case SO_LOCK_FILTER:
1530		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1531		break;
1532
1533	case SO_BPF_EXTENSIONS:
1534		v.val = bpf_tell_extensions();
1535		break;
1536
1537	case SO_SELECT_ERR_QUEUE:
1538		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1539		break;
1540
1541#ifdef CONFIG_NET_RX_BUSY_POLL
1542	case SO_BUSY_POLL:
1543		v.val = sk->sk_ll_usec;
1544		break;
 
 
 
1545#endif
1546
1547	case SO_MAX_PACING_RATE:
1548		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1549			lv = sizeof(v.ulval);
1550			v.ulval = sk->sk_max_pacing_rate;
1551		} else {
1552			/* 32bit version */
1553			v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1554		}
1555		break;
1556
1557	case SO_INCOMING_CPU:
1558		v.val = READ_ONCE(sk->sk_incoming_cpu);
1559		break;
1560
1561	case SO_MEMINFO:
1562	{
1563		u32 meminfo[SK_MEMINFO_VARS];
1564
1565		sk_get_meminfo(sk, meminfo);
1566
1567		len = min_t(unsigned int, len, sizeof(meminfo));
1568		if (copy_to_user(optval, &meminfo, len))
1569			return -EFAULT;
1570
1571		goto lenout;
1572	}
1573
1574#ifdef CONFIG_NET_RX_BUSY_POLL
1575	case SO_INCOMING_NAPI_ID:
1576		v.val = READ_ONCE(sk->sk_napi_id);
1577
1578		/* aggregate non-NAPI IDs down to 0 */
1579		if (v.val < MIN_NAPI_ID)
1580			v.val = 0;
1581
1582		break;
1583#endif
1584
1585	case SO_COOKIE:
1586		lv = sizeof(u64);
1587		if (len < lv)
1588			return -EINVAL;
1589		v.val64 = sock_gen_cookie(sk);
1590		break;
1591
1592	case SO_ZEROCOPY:
1593		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1594		break;
1595
1596	case SO_TXTIME:
1597		lv = sizeof(v.txtime);
1598		v.txtime.clockid = sk->sk_clockid;
1599		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1600				  SOF_TXTIME_DEADLINE_MODE : 0;
1601		v.txtime.flags |= sk->sk_txtime_report_errors ?
1602				  SOF_TXTIME_REPORT_ERRORS : 0;
1603		break;
1604
1605	case SO_BINDTOIFINDEX:
1606		v.val = sk->sk_bound_dev_if;
1607		break;
1608
 
 
 
 
 
 
 
1609	default:
1610		/* We implement the SO_SNDLOWAT etc to not be settable
1611		 * (1003.1g 7).
1612		 */
1613		return -ENOPROTOOPT;
1614	}
1615
1616	if (len > lv)
1617		len = lv;
1618	if (copy_to_user(optval, &v, len))
1619		return -EFAULT;
1620lenout:
1621	if (put_user(len, optlen))
1622		return -EFAULT;
1623	return 0;
1624}
1625
1626/*
1627 * Initialize an sk_lock.
1628 *
1629 * (We also register the sk_lock with the lock validator.)
1630 */
1631static inline void sock_lock_init(struct sock *sk)
1632{
1633	if (sk->sk_kern_sock)
1634		sock_lock_init_class_and_name(
1635			sk,
1636			af_family_kern_slock_key_strings[sk->sk_family],
1637			af_family_kern_slock_keys + sk->sk_family,
1638			af_family_kern_key_strings[sk->sk_family],
1639			af_family_kern_keys + sk->sk_family);
1640	else
1641		sock_lock_init_class_and_name(
1642			sk,
1643			af_family_slock_key_strings[sk->sk_family],
1644			af_family_slock_keys + sk->sk_family,
1645			af_family_key_strings[sk->sk_family],
1646			af_family_keys + sk->sk_family);
1647}
1648
1649/*
1650 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1651 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1652 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1653 */
1654static void sock_copy(struct sock *nsk, const struct sock *osk)
1655{
1656	const struct proto *prot = READ_ONCE(osk->sk_prot);
1657#ifdef CONFIG_SECURITY_NETWORK
1658	void *sptr = nsk->sk_security;
1659#endif
 
 
 
 
 
 
 
 
 
 
1660	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1661
1662	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1663	       prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1664
1665#ifdef CONFIG_SECURITY_NETWORK
1666	nsk->sk_security = sptr;
1667	security_sk_clone(osk, nsk);
1668#endif
1669}
1670
1671static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1672		int family)
1673{
1674	struct sock *sk;
1675	struct kmem_cache *slab;
1676
1677	slab = prot->slab;
1678	if (slab != NULL) {
1679		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1680		if (!sk)
1681			return sk;
1682		if (want_init_on_alloc(priority))
1683			sk_prot_clear_nulls(sk, prot->obj_size);
1684	} else
1685		sk = kmalloc(prot->obj_size, priority);
1686
1687	if (sk != NULL) {
1688		if (security_sk_alloc(sk, family, priority))
1689			goto out_free;
1690
1691		if (!try_module_get(prot->owner))
1692			goto out_free_sec;
1693		sk_tx_queue_clear(sk);
1694	}
1695
1696	return sk;
1697
1698out_free_sec:
1699	security_sk_free(sk);
1700out_free:
1701	if (slab != NULL)
1702		kmem_cache_free(slab, sk);
1703	else
1704		kfree(sk);
1705	return NULL;
1706}
1707
1708static void sk_prot_free(struct proto *prot, struct sock *sk)
1709{
1710	struct kmem_cache *slab;
1711	struct module *owner;
1712
1713	owner = prot->owner;
1714	slab = prot->slab;
1715
1716	cgroup_sk_free(&sk->sk_cgrp_data);
1717	mem_cgroup_sk_free(sk);
1718	security_sk_free(sk);
1719	if (slab != NULL)
1720		kmem_cache_free(slab, sk);
1721	else
1722		kfree(sk);
1723	module_put(owner);
1724}
1725
1726/**
1727 *	sk_alloc - All socket objects are allocated here
1728 *	@net: the applicable net namespace
1729 *	@family: protocol family
1730 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1731 *	@prot: struct proto associated with this new sock instance
1732 *	@kern: is this to be a kernel socket?
1733 */
1734struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1735		      struct proto *prot, int kern)
1736{
1737	struct sock *sk;
1738
1739	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1740	if (sk) {
1741		sk->sk_family = family;
1742		/*
1743		 * See comment in struct sock definition to understand
1744		 * why we need sk_prot_creator -acme
1745		 */
1746		sk->sk_prot = sk->sk_prot_creator = prot;
1747		sk->sk_kern_sock = kern;
1748		sock_lock_init(sk);
1749		sk->sk_net_refcnt = kern ? 0 : 1;
1750		if (likely(sk->sk_net_refcnt)) {
1751			get_net(net);
1752			sock_inuse_add(net, 1);
1753		}
1754
1755		sock_net_set(sk, net);
1756		refcount_set(&sk->sk_wmem_alloc, 1);
1757
1758		mem_cgroup_sk_alloc(sk);
1759		cgroup_sk_alloc(&sk->sk_cgrp_data);
1760		sock_update_classid(&sk->sk_cgrp_data);
1761		sock_update_netprioidx(&sk->sk_cgrp_data);
1762		sk_tx_queue_clear(sk);
1763	}
1764
1765	return sk;
1766}
1767EXPORT_SYMBOL(sk_alloc);
1768
1769/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1770 * grace period. This is the case for UDP sockets and TCP listeners.
1771 */
1772static void __sk_destruct(struct rcu_head *head)
1773{
1774	struct sock *sk = container_of(head, struct sock, sk_rcu);
1775	struct sk_filter *filter;
1776
1777	if (sk->sk_destruct)
1778		sk->sk_destruct(sk);
1779
1780	filter = rcu_dereference_check(sk->sk_filter,
1781				       refcount_read(&sk->sk_wmem_alloc) == 0);
1782	if (filter) {
1783		sk_filter_uncharge(sk, filter);
1784		RCU_INIT_POINTER(sk->sk_filter, NULL);
1785	}
1786
1787	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1788
1789#ifdef CONFIG_BPF_SYSCALL
1790	bpf_sk_storage_free(sk);
1791#endif
1792
1793	if (atomic_read(&sk->sk_omem_alloc))
1794		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1795			 __func__, atomic_read(&sk->sk_omem_alloc));
1796
1797	if (sk->sk_frag.page) {
1798		put_page(sk->sk_frag.page);
1799		sk->sk_frag.page = NULL;
1800	}
1801
1802	if (sk->sk_peer_cred)
1803		put_cred(sk->sk_peer_cred);
1804	put_pid(sk->sk_peer_pid);
 
1805	if (likely(sk->sk_net_refcnt))
1806		put_net(sock_net(sk));
1807	sk_prot_free(sk->sk_prot_creator, sk);
1808}
1809
1810void sk_destruct(struct sock *sk)
1811{
1812	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1813
1814	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1815		reuseport_detach_sock(sk);
1816		use_call_rcu = true;
1817	}
1818
1819	if (use_call_rcu)
1820		call_rcu(&sk->sk_rcu, __sk_destruct);
1821	else
1822		__sk_destruct(&sk->sk_rcu);
1823}
1824
1825static void __sk_free(struct sock *sk)
1826{
1827	if (likely(sk->sk_net_refcnt))
1828		sock_inuse_add(sock_net(sk), -1);
1829
1830	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1831		sock_diag_broadcast_destroy(sk);
1832	else
1833		sk_destruct(sk);
1834}
1835
1836void sk_free(struct sock *sk)
1837{
1838	/*
1839	 * We subtract one from sk_wmem_alloc and can know if
1840	 * some packets are still in some tx queue.
1841	 * If not null, sock_wfree() will call __sk_free(sk) later
1842	 */
1843	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1844		__sk_free(sk);
1845}
1846EXPORT_SYMBOL(sk_free);
1847
1848static void sk_init_common(struct sock *sk)
1849{
1850	skb_queue_head_init(&sk->sk_receive_queue);
1851	skb_queue_head_init(&sk->sk_write_queue);
1852	skb_queue_head_init(&sk->sk_error_queue);
1853
1854	rwlock_init(&sk->sk_callback_lock);
1855	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1856			af_rlock_keys + sk->sk_family,
1857			af_family_rlock_key_strings[sk->sk_family]);
1858	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1859			af_wlock_keys + sk->sk_family,
1860			af_family_wlock_key_strings[sk->sk_family]);
1861	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1862			af_elock_keys + sk->sk_family,
1863			af_family_elock_key_strings[sk->sk_family]);
1864	lockdep_set_class_and_name(&sk->sk_callback_lock,
1865			af_callback_keys + sk->sk_family,
1866			af_family_clock_key_strings[sk->sk_family]);
1867}
1868
1869/**
1870 *	sk_clone_lock - clone a socket, and lock its clone
1871 *	@sk: the socket to clone
1872 *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1873 *
1874 *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1875 */
1876struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1877{
1878	struct proto *prot = READ_ONCE(sk->sk_prot);
 
 
1879	struct sock *newsk;
1880	bool is_charged = true;
1881
1882	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1883	if (newsk != NULL) {
1884		struct sk_filter *filter;
1885
1886		sock_copy(newsk, sk);
1887
1888		newsk->sk_prot_creator = prot;
1889
1890		/* SANITY */
1891		if (likely(newsk->sk_net_refcnt))
1892			get_net(sock_net(newsk));
1893		sk_node_init(&newsk->sk_node);
1894		sock_lock_init(newsk);
1895		bh_lock_sock(newsk);
1896		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1897		newsk->sk_backlog.len = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1898
1899		atomic_set(&newsk->sk_rmem_alloc, 0);
1900		/*
1901		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1902		 */
1903		refcount_set(&newsk->sk_wmem_alloc, 1);
1904		atomic_set(&newsk->sk_omem_alloc, 0);
1905		sk_init_common(newsk);
1906
1907		newsk->sk_dst_cache	= NULL;
1908		newsk->sk_dst_pending_confirm = 0;
1909		newsk->sk_wmem_queued	= 0;
1910		newsk->sk_forward_alloc = 0;
1911		atomic_set(&newsk->sk_drops, 0);
1912		newsk->sk_send_head	= NULL;
1913		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1914		atomic_set(&newsk->sk_zckey, 0);
1915
1916		sock_reset_flag(newsk, SOCK_DONE);
 
1917
1918		/* sk->sk_memcg will be populated at accept() time */
1919		newsk->sk_memcg = NULL;
1920
1921		cgroup_sk_clone(&newsk->sk_cgrp_data);
 
 
 
 
 
 
 
 
 
1922
1923		rcu_read_lock();
1924		filter = rcu_dereference(sk->sk_filter);
1925		if (filter != NULL)
1926			/* though it's an empty new sock, the charging may fail
1927			 * if sysctl_optmem_max was changed between creation of
1928			 * original socket and cloning
1929			 */
1930			is_charged = sk_filter_charge(newsk, filter);
1931		RCU_INIT_POINTER(newsk->sk_filter, filter);
1932		rcu_read_unlock();
 
 
1933
1934		if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1935			/* We need to make sure that we don't uncharge the new
1936			 * socket if we couldn't charge it in the first place
1937			 * as otherwise we uncharge the parent's filter.
1938			 */
1939			if (!is_charged)
1940				RCU_INIT_POINTER(newsk->sk_filter, NULL);
1941			sk_free_unlock_clone(newsk);
1942			newsk = NULL;
1943			goto out;
1944		}
1945		RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1946
1947		if (bpf_sk_storage_clone(sk, newsk)) {
1948			sk_free_unlock_clone(newsk);
1949			newsk = NULL;
1950			goto out;
1951		}
1952
1953		/* Clear sk_user_data if parent had the pointer tagged
1954		 * as not suitable for copying when cloning.
1955		 */
1956		if (sk_user_data_is_nocopy(newsk))
1957			newsk->sk_user_data = NULL;
 
1958
1959		newsk->sk_err	   = 0;
1960		newsk->sk_err_soft = 0;
1961		newsk->sk_priority = 0;
1962		newsk->sk_incoming_cpu = raw_smp_processor_id();
1963		if (likely(newsk->sk_net_refcnt))
1964			sock_inuse_add(sock_net(newsk), 1);
1965
1966		/*
1967		 * Before updating sk_refcnt, we must commit prior changes to memory
1968		 * (Documentation/RCU/rculist_nulls.rst for details)
1969		 */
1970		smp_wmb();
1971		refcount_set(&newsk->sk_refcnt, 2);
 
 
 
 
 
 
 
 
1972
1973		/*
1974		 * Increment the counter in the same struct proto as the master
1975		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1976		 * is the same as sk->sk_prot->socks, as this field was copied
1977		 * with memcpy).
1978		 *
1979		 * This _changes_ the previous behaviour, where
1980		 * tcp_create_openreq_child always was incrementing the
1981		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1982		 * to be taken into account in all callers. -acme
1983		 */
1984		sk_refcnt_debug_inc(newsk);
1985		sk_set_socket(newsk, NULL);
1986		sk_tx_queue_clear(newsk);
1987		RCU_INIT_POINTER(newsk->sk_wq, NULL);
1988
1989		if (newsk->sk_prot->sockets_allocated)
1990			sk_sockets_allocated_inc(newsk);
1991
1992		if (sock_needs_netstamp(sk) &&
1993		    newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1994			net_enable_timestamp();
1995	}
1996out:
1997	return newsk;
1998}
1999EXPORT_SYMBOL_GPL(sk_clone_lock);
2000
2001void sk_free_unlock_clone(struct sock *sk)
2002{
2003	/* It is still raw copy of parent, so invalidate
2004	 * destructor and make plain sk_free() */
2005	sk->sk_destruct = NULL;
2006	bh_unlock_sock(sk);
2007	sk_free(sk);
2008}
2009EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2010
2011void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2012{
2013	u32 max_segs = 1;
2014
2015	sk_dst_set(sk, dst);
2016	sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2017	if (sk->sk_route_caps & NETIF_F_GSO)
2018		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2019	sk->sk_route_caps &= ~sk->sk_route_nocaps;
2020	if (sk_can_gso(sk)) {
2021		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2022			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2023		} else {
2024			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2025			sk->sk_gso_max_size = dst->dev->gso_max_size;
2026			max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2027		}
2028	}
2029	sk->sk_gso_max_segs = max_segs;
2030}
2031EXPORT_SYMBOL_GPL(sk_setup_caps);
2032
2033/*
2034 *	Simple resource managers for sockets.
2035 */
2036
2037
2038/*
2039 * Write buffer destructor automatically called from kfree_skb.
2040 */
2041void sock_wfree(struct sk_buff *skb)
2042{
2043	struct sock *sk = skb->sk;
2044	unsigned int len = skb->truesize;
2045
2046	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2047		/*
2048		 * Keep a reference on sk_wmem_alloc, this will be released
2049		 * after sk_write_space() call
2050		 */
2051		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2052		sk->sk_write_space(sk);
2053		len = 1;
2054	}
2055	/*
2056	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2057	 * could not do because of in-flight packets
2058	 */
2059	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2060		__sk_free(sk);
2061}
2062EXPORT_SYMBOL(sock_wfree);
2063
2064/* This variant of sock_wfree() is used by TCP,
2065 * since it sets SOCK_USE_WRITE_QUEUE.
2066 */
2067void __sock_wfree(struct sk_buff *skb)
2068{
2069	struct sock *sk = skb->sk;
2070
2071	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2072		__sk_free(sk);
2073}
2074
2075void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2076{
2077	skb_orphan(skb);
2078	skb->sk = sk;
2079#ifdef CONFIG_INET
2080	if (unlikely(!sk_fullsock(sk))) {
2081		skb->destructor = sock_edemux;
2082		sock_hold(sk);
2083		return;
2084	}
2085#endif
2086	skb->destructor = sock_wfree;
2087	skb_set_hash_from_sk(skb, sk);
2088	/*
2089	 * We used to take a refcount on sk, but following operation
2090	 * is enough to guarantee sk_free() wont free this sock until
2091	 * all in-flight packets are completed
2092	 */
2093	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2094}
2095EXPORT_SYMBOL(skb_set_owner_w);
2096
2097static bool can_skb_orphan_partial(const struct sk_buff *skb)
2098{
2099#ifdef CONFIG_TLS_DEVICE
2100	/* Drivers depend on in-order delivery for crypto offload,
2101	 * partial orphan breaks out-of-order-OK logic.
2102	 */
2103	if (skb->decrypted)
2104		return false;
2105#endif
2106	return (skb->destructor == sock_wfree ||
2107		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2108}
2109
2110/* This helper is used by netem, as it can hold packets in its
2111 * delay queue. We want to allow the owner socket to send more
2112 * packets, as if they were already TX completed by a typical driver.
2113 * But we also want to keep skb->sk set because some packet schedulers
2114 * rely on it (sch_fq for example).
2115 */
2116void skb_orphan_partial(struct sk_buff *skb)
2117{
2118	if (skb_is_tcp_pure_ack(skb))
2119		return;
2120
2121	if (can_skb_orphan_partial(skb)) {
2122		struct sock *sk = skb->sk;
2123
2124		if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2125			WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2126			skb->destructor = sock_efree;
2127		}
2128	} else {
2129		skb_orphan(skb);
2130	}
2131}
2132EXPORT_SYMBOL(skb_orphan_partial);
2133
2134/*
2135 * Read buffer destructor automatically called from kfree_skb.
2136 */
2137void sock_rfree(struct sk_buff *skb)
2138{
2139	struct sock *sk = skb->sk;
2140	unsigned int len = skb->truesize;
2141
2142	atomic_sub(len, &sk->sk_rmem_alloc);
2143	sk_mem_uncharge(sk, len);
2144}
2145EXPORT_SYMBOL(sock_rfree);
2146
2147/*
2148 * Buffer destructor for skbs that are not used directly in read or write
2149 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2150 */
2151void sock_efree(struct sk_buff *skb)
2152{
2153	sock_put(skb->sk);
2154}
2155EXPORT_SYMBOL(sock_efree);
2156
2157/* Buffer destructor for prefetch/receive path where reference count may
2158 * not be held, e.g. for listen sockets.
2159 */
2160#ifdef CONFIG_INET
2161void sock_pfree(struct sk_buff *skb)
2162{
2163	if (sk_is_refcounted(skb->sk))
2164		sock_gen_put(skb->sk);
2165}
2166EXPORT_SYMBOL(sock_pfree);
2167#endif /* CONFIG_INET */
2168
2169kuid_t sock_i_uid(struct sock *sk)
2170{
2171	kuid_t uid;
2172
2173	read_lock_bh(&sk->sk_callback_lock);
2174	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2175	read_unlock_bh(&sk->sk_callback_lock);
2176	return uid;
2177}
2178EXPORT_SYMBOL(sock_i_uid);
2179
2180unsigned long sock_i_ino(struct sock *sk)
2181{
2182	unsigned long ino;
2183
2184	read_lock_bh(&sk->sk_callback_lock);
2185	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2186	read_unlock_bh(&sk->sk_callback_lock);
2187	return ino;
2188}
2189EXPORT_SYMBOL(sock_i_ino);
2190
2191/*
2192 * Allocate a skb from the socket's send buffer.
2193 */
2194struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2195			     gfp_t priority)
2196{
2197	if (force ||
2198	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2199		struct sk_buff *skb = alloc_skb(size, priority);
2200
2201		if (skb) {
2202			skb_set_owner_w(skb, sk);
2203			return skb;
2204		}
2205	}
2206	return NULL;
2207}
2208EXPORT_SYMBOL(sock_wmalloc);
2209
2210static void sock_ofree(struct sk_buff *skb)
2211{
2212	struct sock *sk = skb->sk;
2213
2214	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2215}
2216
2217struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2218			     gfp_t priority)
2219{
2220	struct sk_buff *skb;
2221
2222	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2223	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2224	    sysctl_optmem_max)
2225		return NULL;
2226
2227	skb = alloc_skb(size, priority);
2228	if (!skb)
2229		return NULL;
2230
2231	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2232	skb->sk = sk;
2233	skb->destructor = sock_ofree;
2234	return skb;
2235}
2236
2237/*
2238 * Allocate a memory block from the socket's option memory buffer.
2239 */
2240void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2241{
2242	if ((unsigned int)size <= sysctl_optmem_max &&
2243	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2244		void *mem;
2245		/* First do the add, to avoid the race if kmalloc
2246		 * might sleep.
2247		 */
2248		atomic_add(size, &sk->sk_omem_alloc);
2249		mem = kmalloc(size, priority);
2250		if (mem)
2251			return mem;
2252		atomic_sub(size, &sk->sk_omem_alloc);
2253	}
2254	return NULL;
2255}
2256EXPORT_SYMBOL(sock_kmalloc);
2257
2258/* Free an option memory block. Note, we actually want the inline
2259 * here as this allows gcc to detect the nullify and fold away the
2260 * condition entirely.
2261 */
2262static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2263				  const bool nullify)
2264{
2265	if (WARN_ON_ONCE(!mem))
2266		return;
2267	if (nullify)
2268		kfree_sensitive(mem);
2269	else
2270		kfree(mem);
2271	atomic_sub(size, &sk->sk_omem_alloc);
2272}
2273
2274void sock_kfree_s(struct sock *sk, void *mem, int size)
2275{
2276	__sock_kfree_s(sk, mem, size, false);
2277}
2278EXPORT_SYMBOL(sock_kfree_s);
2279
2280void sock_kzfree_s(struct sock *sk, void *mem, int size)
2281{
2282	__sock_kfree_s(sk, mem, size, true);
2283}
2284EXPORT_SYMBOL(sock_kzfree_s);
2285
2286/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2287   I think, these locks should be removed for datagram sockets.
2288 */
2289static long sock_wait_for_wmem(struct sock *sk, long timeo)
2290{
2291	DEFINE_WAIT(wait);
2292
2293	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2294	for (;;) {
2295		if (!timeo)
2296			break;
2297		if (signal_pending(current))
2298			break;
2299		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2300		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2301		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2302			break;
2303		if (sk->sk_shutdown & SEND_SHUTDOWN)
2304			break;
2305		if (sk->sk_err)
2306			break;
2307		timeo = schedule_timeout(timeo);
2308	}
2309	finish_wait(sk_sleep(sk), &wait);
2310	return timeo;
2311}
2312
2313
2314/*
2315 *	Generic send/receive buffer handlers
2316 */
2317
2318struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2319				     unsigned long data_len, int noblock,
2320				     int *errcode, int max_page_order)
2321{
2322	struct sk_buff *skb;
2323	long timeo;
2324	int err;
2325
2326	timeo = sock_sndtimeo(sk, noblock);
2327	for (;;) {
2328		err = sock_error(sk);
2329		if (err != 0)
2330			goto failure;
2331
2332		err = -EPIPE;
2333		if (sk->sk_shutdown & SEND_SHUTDOWN)
2334			goto failure;
2335
2336		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2337			break;
2338
2339		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2340		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2341		err = -EAGAIN;
2342		if (!timeo)
2343			goto failure;
2344		if (signal_pending(current))
2345			goto interrupted;
2346		timeo = sock_wait_for_wmem(sk, timeo);
2347	}
2348	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2349				   errcode, sk->sk_allocation);
2350	if (skb)
2351		skb_set_owner_w(skb, sk);
2352	return skb;
2353
2354interrupted:
2355	err = sock_intr_errno(timeo);
2356failure:
2357	*errcode = err;
2358	return NULL;
2359}
2360EXPORT_SYMBOL(sock_alloc_send_pskb);
2361
2362struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2363				    int noblock, int *errcode)
2364{
2365	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2366}
2367EXPORT_SYMBOL(sock_alloc_send_skb);
2368
2369int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2370		     struct sockcm_cookie *sockc)
2371{
2372	u32 tsflags;
2373
2374	switch (cmsg->cmsg_type) {
2375	case SO_MARK:
2376		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2377			return -EPERM;
2378		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2379			return -EINVAL;
2380		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2381		break;
2382	case SO_TIMESTAMPING_OLD:
2383		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2384			return -EINVAL;
2385
2386		tsflags = *(u32 *)CMSG_DATA(cmsg);
2387		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2388			return -EINVAL;
2389
2390		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2391		sockc->tsflags |= tsflags;
2392		break;
2393	case SCM_TXTIME:
2394		if (!sock_flag(sk, SOCK_TXTIME))
2395			return -EINVAL;
2396		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2397			return -EINVAL;
2398		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2399		break;
2400	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2401	case SCM_RIGHTS:
2402	case SCM_CREDENTIALS:
2403		break;
2404	default:
2405		return -EINVAL;
2406	}
2407	return 0;
2408}
2409EXPORT_SYMBOL(__sock_cmsg_send);
2410
2411int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2412		   struct sockcm_cookie *sockc)
2413{
2414	struct cmsghdr *cmsg;
2415	int ret;
2416
2417	for_each_cmsghdr(cmsg, msg) {
2418		if (!CMSG_OK(msg, cmsg))
2419			return -EINVAL;
2420		if (cmsg->cmsg_level != SOL_SOCKET)
2421			continue;
2422		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2423		if (ret)
2424			return ret;
2425	}
2426	return 0;
2427}
2428EXPORT_SYMBOL(sock_cmsg_send);
2429
2430static void sk_enter_memory_pressure(struct sock *sk)
2431{
2432	if (!sk->sk_prot->enter_memory_pressure)
2433		return;
2434
2435	sk->sk_prot->enter_memory_pressure(sk);
2436}
2437
2438static void sk_leave_memory_pressure(struct sock *sk)
2439{
2440	if (sk->sk_prot->leave_memory_pressure) {
2441		sk->sk_prot->leave_memory_pressure(sk);
2442	} else {
2443		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2444
2445		if (memory_pressure && READ_ONCE(*memory_pressure))
2446			WRITE_ONCE(*memory_pressure, 0);
2447	}
2448}
2449
2450#define SKB_FRAG_PAGE_ORDER	get_order(32768)
2451DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2452
2453/**
2454 * skb_page_frag_refill - check that a page_frag contains enough room
2455 * @sz: minimum size of the fragment we want to get
2456 * @pfrag: pointer to page_frag
2457 * @gfp: priority for memory allocation
2458 *
2459 * Note: While this allocator tries to use high order pages, there is
2460 * no guarantee that allocations succeed. Therefore, @sz MUST be
2461 * less or equal than PAGE_SIZE.
2462 */
2463bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2464{
2465	if (pfrag->page) {
2466		if (page_ref_count(pfrag->page) == 1) {
2467			pfrag->offset = 0;
2468			return true;
2469		}
2470		if (pfrag->offset + sz <= pfrag->size)
2471			return true;
2472		put_page(pfrag->page);
2473	}
2474
2475	pfrag->offset = 0;
2476	if (SKB_FRAG_PAGE_ORDER &&
2477	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2478		/* Avoid direct reclaim but allow kswapd to wake */
2479		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2480					  __GFP_COMP | __GFP_NOWARN |
2481					  __GFP_NORETRY,
2482					  SKB_FRAG_PAGE_ORDER);
2483		if (likely(pfrag->page)) {
2484			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2485			return true;
2486		}
2487	}
2488	pfrag->page = alloc_page(gfp);
2489	if (likely(pfrag->page)) {
2490		pfrag->size = PAGE_SIZE;
2491		return true;
2492	}
2493	return false;
2494}
2495EXPORT_SYMBOL(skb_page_frag_refill);
2496
2497bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2498{
2499	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2500		return true;
2501
2502	sk_enter_memory_pressure(sk);
2503	sk_stream_moderate_sndbuf(sk);
2504	return false;
2505}
2506EXPORT_SYMBOL(sk_page_frag_refill);
2507
2508static void __lock_sock(struct sock *sk)
2509	__releases(&sk->sk_lock.slock)
2510	__acquires(&sk->sk_lock.slock)
2511{
2512	DEFINE_WAIT(wait);
2513
2514	for (;;) {
2515		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2516					TASK_UNINTERRUPTIBLE);
2517		spin_unlock_bh(&sk->sk_lock.slock);
2518		schedule();
2519		spin_lock_bh(&sk->sk_lock.slock);
2520		if (!sock_owned_by_user(sk))
2521			break;
2522	}
2523	finish_wait(&sk->sk_lock.wq, &wait);
2524}
2525
2526void __release_sock(struct sock *sk)
2527	__releases(&sk->sk_lock.slock)
2528	__acquires(&sk->sk_lock.slock)
2529{
2530	struct sk_buff *skb, *next;
2531
2532	while ((skb = sk->sk_backlog.head) != NULL) {
2533		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2534
2535		spin_unlock_bh(&sk->sk_lock.slock);
2536
2537		do {
2538			next = skb->next;
2539			prefetch(next);
2540			WARN_ON_ONCE(skb_dst_is_noref(skb));
2541			skb_mark_not_on_list(skb);
2542			sk_backlog_rcv(sk, skb);
2543
2544			cond_resched();
2545
2546			skb = next;
2547		} while (skb != NULL);
2548
2549		spin_lock_bh(&sk->sk_lock.slock);
2550	}
2551
2552	/*
2553	 * Doing the zeroing here guarantee we can not loop forever
2554	 * while a wild producer attempts to flood us.
2555	 */
2556	sk->sk_backlog.len = 0;
2557}
2558
2559void __sk_flush_backlog(struct sock *sk)
2560{
2561	spin_lock_bh(&sk->sk_lock.slock);
2562	__release_sock(sk);
2563	spin_unlock_bh(&sk->sk_lock.slock);
2564}
2565
2566/**
2567 * sk_wait_data - wait for data to arrive at sk_receive_queue
2568 * @sk:    sock to wait on
2569 * @timeo: for how long
2570 * @skb:   last skb seen on sk_receive_queue
2571 *
2572 * Now socket state including sk->sk_err is changed only under lock,
2573 * hence we may omit checks after joining wait queue.
2574 * We check receive queue before schedule() only as optimization;
2575 * it is very likely that release_sock() added new data.
2576 */
2577int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2578{
2579	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2580	int rc;
2581
2582	add_wait_queue(sk_sleep(sk), &wait);
2583	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2584	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2585	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2586	remove_wait_queue(sk_sleep(sk), &wait);
2587	return rc;
2588}
2589EXPORT_SYMBOL(sk_wait_data);
2590
2591/**
2592 *	__sk_mem_raise_allocated - increase memory_allocated
2593 *	@sk: socket
2594 *	@size: memory size to allocate
2595 *	@amt: pages to allocate
2596 *	@kind: allocation type
2597 *
2598 *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2599 */
2600int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2601{
2602	struct proto *prot = sk->sk_prot;
2603	long allocated = sk_memory_allocated_add(sk, amt);
2604	bool charged = true;
2605
2606	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2607	    !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2608		goto suppress_allocation;
2609
2610	/* Under limit. */
2611	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2612		sk_leave_memory_pressure(sk);
2613		return 1;
2614	}
2615
2616	/* Under pressure. */
2617	if (allocated > sk_prot_mem_limits(sk, 1))
2618		sk_enter_memory_pressure(sk);
2619
2620	/* Over hard limit. */
2621	if (allocated > sk_prot_mem_limits(sk, 2))
2622		goto suppress_allocation;
2623
2624	/* guarantee minimum buffer size under pressure */
2625	if (kind == SK_MEM_RECV) {
2626		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2627			return 1;
2628
2629	} else { /* SK_MEM_SEND */
2630		int wmem0 = sk_get_wmem0(sk, prot);
2631
2632		if (sk->sk_type == SOCK_STREAM) {
2633			if (sk->sk_wmem_queued < wmem0)
2634				return 1;
2635		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2636				return 1;
2637		}
2638	}
2639
2640	if (sk_has_memory_pressure(sk)) {
2641		u64 alloc;
2642
2643		if (!sk_under_memory_pressure(sk))
2644			return 1;
2645		alloc = sk_sockets_allocated_read_positive(sk);
2646		if (sk_prot_mem_limits(sk, 2) > alloc *
2647		    sk_mem_pages(sk->sk_wmem_queued +
2648				 atomic_read(&sk->sk_rmem_alloc) +
2649				 sk->sk_forward_alloc))
2650			return 1;
2651	}
2652
2653suppress_allocation:
2654
2655	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2656		sk_stream_moderate_sndbuf(sk);
2657
2658		/* Fail only if socket is _under_ its sndbuf.
2659		 * In this case we cannot block, so that we have to fail.
2660		 */
2661		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2662			return 1;
2663	}
2664
2665	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2666		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2667
2668	sk_memory_allocated_sub(sk, amt);
2669
2670	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2671		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2672
2673	return 0;
2674}
2675EXPORT_SYMBOL(__sk_mem_raise_allocated);
2676
2677/**
2678 *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2679 *	@sk: socket
2680 *	@size: memory size to allocate
2681 *	@kind: allocation type
2682 *
2683 *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2684 *	rmem allocation. This function assumes that protocols which have
2685 *	memory_pressure use sk_wmem_queued as write buffer accounting.
2686 */
2687int __sk_mem_schedule(struct sock *sk, int size, int kind)
2688{
2689	int ret, amt = sk_mem_pages(size);
2690
2691	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2692	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2693	if (!ret)
2694		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2695	return ret;
2696}
2697EXPORT_SYMBOL(__sk_mem_schedule);
2698
2699/**
2700 *	__sk_mem_reduce_allocated - reclaim memory_allocated
2701 *	@sk: socket
2702 *	@amount: number of quanta
2703 *
2704 *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2705 */
2706void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2707{
2708	sk_memory_allocated_sub(sk, amount);
2709
2710	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2711		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2712
2713	if (sk_under_memory_pressure(sk) &&
2714	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2715		sk_leave_memory_pressure(sk);
2716}
2717EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2718
2719/**
2720 *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2721 *	@sk: socket
2722 *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2723 */
2724void __sk_mem_reclaim(struct sock *sk, int amount)
2725{
2726	amount >>= SK_MEM_QUANTUM_SHIFT;
2727	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2728	__sk_mem_reduce_allocated(sk, amount);
2729}
2730EXPORT_SYMBOL(__sk_mem_reclaim);
2731
2732int sk_set_peek_off(struct sock *sk, int val)
2733{
2734	sk->sk_peek_off = val;
2735	return 0;
2736}
2737EXPORT_SYMBOL_GPL(sk_set_peek_off);
2738
2739/*
2740 * Set of default routines for initialising struct proto_ops when
2741 * the protocol does not support a particular function. In certain
2742 * cases where it makes no sense for a protocol to have a "do nothing"
2743 * function, some default processing is provided.
2744 */
2745
2746int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2747{
2748	return -EOPNOTSUPP;
2749}
2750EXPORT_SYMBOL(sock_no_bind);
2751
2752int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2753		    int len, int flags)
2754{
2755	return -EOPNOTSUPP;
2756}
2757EXPORT_SYMBOL(sock_no_connect);
2758
2759int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2760{
2761	return -EOPNOTSUPP;
2762}
2763EXPORT_SYMBOL(sock_no_socketpair);
2764
2765int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2766		   bool kern)
2767{
2768	return -EOPNOTSUPP;
2769}
2770EXPORT_SYMBOL(sock_no_accept);
2771
2772int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2773		    int peer)
2774{
2775	return -EOPNOTSUPP;
2776}
2777EXPORT_SYMBOL(sock_no_getname);
2778
2779int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2780{
2781	return -EOPNOTSUPP;
2782}
2783EXPORT_SYMBOL(sock_no_ioctl);
2784
2785int sock_no_listen(struct socket *sock, int backlog)
2786{
2787	return -EOPNOTSUPP;
2788}
2789EXPORT_SYMBOL(sock_no_listen);
2790
2791int sock_no_shutdown(struct socket *sock, int how)
2792{
2793	return -EOPNOTSUPP;
2794}
2795EXPORT_SYMBOL(sock_no_shutdown);
2796
2797int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2798{
2799	return -EOPNOTSUPP;
2800}
2801EXPORT_SYMBOL(sock_no_sendmsg);
2802
2803int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2804{
2805	return -EOPNOTSUPP;
2806}
2807EXPORT_SYMBOL(sock_no_sendmsg_locked);
2808
2809int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2810		    int flags)
2811{
2812	return -EOPNOTSUPP;
2813}
2814EXPORT_SYMBOL(sock_no_recvmsg);
2815
2816int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2817{
2818	/* Mirror missing mmap method error code */
2819	return -ENODEV;
2820}
2821EXPORT_SYMBOL(sock_no_mmap);
2822
2823/*
2824 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2825 * various sock-based usage counts.
2826 */
2827void __receive_sock(struct file *file)
2828{
2829	struct socket *sock;
2830	int error;
2831
2832	/*
2833	 * The resulting value of "error" is ignored here since we only
2834	 * need to take action when the file is a socket and testing
2835	 * "sock" for NULL is sufficient.
2836	 */
2837	sock = sock_from_file(file, &error);
2838	if (sock) {
2839		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2840		sock_update_classid(&sock->sk->sk_cgrp_data);
2841	}
2842}
2843
2844ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2845{
2846	ssize_t res;
2847	struct msghdr msg = {.msg_flags = flags};
2848	struct kvec iov;
2849	char *kaddr = kmap(page);
2850	iov.iov_base = kaddr + offset;
2851	iov.iov_len = size;
2852	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2853	kunmap(page);
2854	return res;
2855}
2856EXPORT_SYMBOL(sock_no_sendpage);
2857
2858ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2859				int offset, size_t size, int flags)
2860{
2861	ssize_t res;
2862	struct msghdr msg = {.msg_flags = flags};
2863	struct kvec iov;
2864	char *kaddr = kmap(page);
2865
2866	iov.iov_base = kaddr + offset;
2867	iov.iov_len = size;
2868	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2869	kunmap(page);
2870	return res;
2871}
2872EXPORT_SYMBOL(sock_no_sendpage_locked);
2873
2874/*
2875 *	Default Socket Callbacks
2876 */
2877
2878static void sock_def_wakeup(struct sock *sk)
2879{
2880	struct socket_wq *wq;
2881
2882	rcu_read_lock();
2883	wq = rcu_dereference(sk->sk_wq);
2884	if (skwq_has_sleeper(wq))
2885		wake_up_interruptible_all(&wq->wait);
2886	rcu_read_unlock();
2887}
2888
2889static void sock_def_error_report(struct sock *sk)
2890{
2891	struct socket_wq *wq;
2892
2893	rcu_read_lock();
2894	wq = rcu_dereference(sk->sk_wq);
2895	if (skwq_has_sleeper(wq))
2896		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2897	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2898	rcu_read_unlock();
2899}
2900
2901void sock_def_readable(struct sock *sk)
2902{
2903	struct socket_wq *wq;
2904
2905	rcu_read_lock();
2906	wq = rcu_dereference(sk->sk_wq);
2907	if (skwq_has_sleeper(wq))
2908		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2909						EPOLLRDNORM | EPOLLRDBAND);
2910	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2911	rcu_read_unlock();
2912}
2913
2914static void sock_def_write_space(struct sock *sk)
2915{
2916	struct socket_wq *wq;
2917
2918	rcu_read_lock();
2919
2920	/* Do not wake up a writer until he can make "significant"
2921	 * progress.  --DaveM
2922	 */
2923	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2924		wq = rcu_dereference(sk->sk_wq);
2925		if (skwq_has_sleeper(wq))
2926			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2927						EPOLLWRNORM | EPOLLWRBAND);
2928
2929		/* Should agree with poll, otherwise some programs break */
2930		if (sock_writeable(sk))
2931			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2932	}
2933
2934	rcu_read_unlock();
2935}
2936
2937static void sock_def_destruct(struct sock *sk)
2938{
2939}
2940
2941void sk_send_sigurg(struct sock *sk)
2942{
2943	if (sk->sk_socket && sk->sk_socket->file)
2944		if (send_sigurg(&sk->sk_socket->file->f_owner))
2945			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2946}
2947EXPORT_SYMBOL(sk_send_sigurg);
2948
2949void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2950		    unsigned long expires)
2951{
2952	if (!mod_timer(timer, expires))
2953		sock_hold(sk);
2954}
2955EXPORT_SYMBOL(sk_reset_timer);
2956
2957void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2958{
2959	if (del_timer(timer))
2960		__sock_put(sk);
2961}
2962EXPORT_SYMBOL(sk_stop_timer);
2963
 
 
 
 
 
 
 
2964void sock_init_data(struct socket *sock, struct sock *sk)
2965{
2966	sk_init_common(sk);
2967	sk->sk_send_head	=	NULL;
2968
2969	timer_setup(&sk->sk_timer, NULL, 0);
2970
2971	sk->sk_allocation	=	GFP_KERNEL;
2972	sk->sk_rcvbuf		=	sysctl_rmem_default;
2973	sk->sk_sndbuf		=	sysctl_wmem_default;
2974	sk->sk_state		=	TCP_CLOSE;
2975	sk_set_socket(sk, sock);
2976
2977	sock_set_flag(sk, SOCK_ZAPPED);
2978
2979	if (sock) {
2980		sk->sk_type	=	sock->type;
2981		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2982		sock->sk	=	sk;
2983		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
2984	} else {
2985		RCU_INIT_POINTER(sk->sk_wq, NULL);
2986		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
2987	}
2988
2989	rwlock_init(&sk->sk_callback_lock);
2990	if (sk->sk_kern_sock)
2991		lockdep_set_class_and_name(
2992			&sk->sk_callback_lock,
2993			af_kern_callback_keys + sk->sk_family,
2994			af_family_kern_clock_key_strings[sk->sk_family]);
2995	else
2996		lockdep_set_class_and_name(
2997			&sk->sk_callback_lock,
2998			af_callback_keys + sk->sk_family,
2999			af_family_clock_key_strings[sk->sk_family]);
3000
3001	sk->sk_state_change	=	sock_def_wakeup;
3002	sk->sk_data_ready	=	sock_def_readable;
3003	sk->sk_write_space	=	sock_def_write_space;
3004	sk->sk_error_report	=	sock_def_error_report;
3005	sk->sk_destruct		=	sock_def_destruct;
3006
3007	sk->sk_frag.page	=	NULL;
3008	sk->sk_frag.offset	=	0;
3009	sk->sk_peek_off		=	-1;
3010
3011	sk->sk_peer_pid 	=	NULL;
3012	sk->sk_peer_cred	=	NULL;
 
 
3013	sk->sk_write_pending	=	0;
3014	sk->sk_rcvlowat		=	1;
3015	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3016	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3017
3018	sk->sk_stamp = SK_DEFAULT_STAMP;
3019#if BITS_PER_LONG==32
3020	seqlock_init(&sk->sk_stamp_seq);
3021#endif
3022	atomic_set(&sk->sk_zckey, 0);
3023
3024#ifdef CONFIG_NET_RX_BUSY_POLL
3025	sk->sk_napi_id		=	0;
3026	sk->sk_ll_usec		=	sysctl_net_busy_read;
3027#endif
3028
3029	sk->sk_max_pacing_rate = ~0UL;
3030	sk->sk_pacing_rate = ~0UL;
3031	WRITE_ONCE(sk->sk_pacing_shift, 10);
3032	sk->sk_incoming_cpu = -1;
3033
3034	sk_rx_queue_clear(sk);
3035	/*
3036	 * Before updating sk_refcnt, we must commit prior changes to memory
3037	 * (Documentation/RCU/rculist_nulls.rst for details)
3038	 */
3039	smp_wmb();
3040	refcount_set(&sk->sk_refcnt, 1);
3041	atomic_set(&sk->sk_drops, 0);
3042}
3043EXPORT_SYMBOL(sock_init_data);
3044
3045void lock_sock_nested(struct sock *sk, int subclass)
3046{
3047	might_sleep();
3048	spin_lock_bh(&sk->sk_lock.slock);
3049	if (sk->sk_lock.owned)
3050		__lock_sock(sk);
3051	sk->sk_lock.owned = 1;
3052	spin_unlock(&sk->sk_lock.slock);
3053	/*
3054	 * The sk_lock has mutex_lock() semantics here:
3055	 */
3056	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3057	local_bh_enable();
3058}
3059EXPORT_SYMBOL(lock_sock_nested);
3060
3061void release_sock(struct sock *sk)
3062{
3063	spin_lock_bh(&sk->sk_lock.slock);
3064	if (sk->sk_backlog.tail)
3065		__release_sock(sk);
3066
3067	/* Warning : release_cb() might need to release sk ownership,
3068	 * ie call sock_release_ownership(sk) before us.
3069	 */
3070	if (sk->sk_prot->release_cb)
3071		sk->sk_prot->release_cb(sk);
3072
3073	sock_release_ownership(sk);
3074	if (waitqueue_active(&sk->sk_lock.wq))
3075		wake_up(&sk->sk_lock.wq);
3076	spin_unlock_bh(&sk->sk_lock.slock);
3077}
3078EXPORT_SYMBOL(release_sock);
3079
3080/**
3081 * lock_sock_fast - fast version of lock_sock
3082 * @sk: socket
3083 *
3084 * This version should be used for very small section, where process wont block
3085 * return false if fast path is taken:
3086 *
3087 *   sk_lock.slock locked, owned = 0, BH disabled
3088 *
3089 * return true if slow path is taken:
3090 *
3091 *   sk_lock.slock unlocked, owned = 1, BH enabled
3092 */
3093bool lock_sock_fast(struct sock *sk)
3094{
3095	might_sleep();
3096	spin_lock_bh(&sk->sk_lock.slock);
3097
3098	if (!sk->sk_lock.owned)
3099		/*
3100		 * Note : We must disable BH
3101		 */
3102		return false;
3103
3104	__lock_sock(sk);
3105	sk->sk_lock.owned = 1;
3106	spin_unlock(&sk->sk_lock.slock);
3107	/*
3108	 * The sk_lock has mutex_lock() semantics here:
3109	 */
3110	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
 
3111	local_bh_enable();
3112	return true;
3113}
3114EXPORT_SYMBOL(lock_sock_fast);
3115
3116int sock_gettstamp(struct socket *sock, void __user *userstamp,
3117		   bool timeval, bool time32)
3118{
3119	struct sock *sk = sock->sk;
3120	struct timespec64 ts;
3121
3122	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3123	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3124	if (ts.tv_sec == -1)
3125		return -ENOENT;
3126	if (ts.tv_sec == 0) {
3127		ktime_t kt = ktime_get_real();
3128		sock_write_timestamp(sk, kt);
3129		ts = ktime_to_timespec64(kt);
3130	}
3131
3132	if (timeval)
3133		ts.tv_nsec /= 1000;
3134
3135#ifdef CONFIG_COMPAT_32BIT_TIME
3136	if (time32)
3137		return put_old_timespec32(&ts, userstamp);
3138#endif
3139#ifdef CONFIG_SPARC64
3140	/* beware of padding in sparc64 timeval */
3141	if (timeval && !in_compat_syscall()) {
3142		struct __kernel_old_timeval __user tv = {
3143			.tv_sec = ts.tv_sec,
3144			.tv_usec = ts.tv_nsec,
3145		};
3146		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3147			return -EFAULT;
3148		return 0;
3149	}
3150#endif
3151	return put_timespec64(&ts, userstamp);
3152}
3153EXPORT_SYMBOL(sock_gettstamp);
3154
3155void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3156{
3157	if (!sock_flag(sk, flag)) {
3158		unsigned long previous_flags = sk->sk_flags;
3159
3160		sock_set_flag(sk, flag);
3161		/*
3162		 * we just set one of the two flags which require net
3163		 * time stamping, but time stamping might have been on
3164		 * already because of the other one
3165		 */
3166		if (sock_needs_netstamp(sk) &&
3167		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3168			net_enable_timestamp();
3169	}
3170}
3171
3172int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3173		       int level, int type)
3174{
3175	struct sock_exterr_skb *serr;
3176	struct sk_buff *skb;
3177	int copied, err;
3178
3179	err = -EAGAIN;
3180	skb = sock_dequeue_err_skb(sk);
3181	if (skb == NULL)
3182		goto out;
3183
3184	copied = skb->len;
3185	if (copied > len) {
3186		msg->msg_flags |= MSG_TRUNC;
3187		copied = len;
3188	}
3189	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3190	if (err)
3191		goto out_free_skb;
3192
3193	sock_recv_timestamp(msg, sk, skb);
3194
3195	serr = SKB_EXT_ERR(skb);
3196	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3197
3198	msg->msg_flags |= MSG_ERRQUEUE;
3199	err = copied;
3200
3201out_free_skb:
3202	kfree_skb(skb);
3203out:
3204	return err;
3205}
3206EXPORT_SYMBOL(sock_recv_errqueue);
3207
3208/*
3209 *	Get a socket option on an socket.
3210 *
3211 *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3212 *	asynchronous errors should be reported by getsockopt. We assume
3213 *	this means if you specify SO_ERROR (otherwise whats the point of it).
3214 */
3215int sock_common_getsockopt(struct socket *sock, int level, int optname,
3216			   char __user *optval, int __user *optlen)
3217{
3218	struct sock *sk = sock->sk;
3219
3220	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3221}
3222EXPORT_SYMBOL(sock_common_getsockopt);
3223
3224int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3225			int flags)
3226{
3227	struct sock *sk = sock->sk;
3228	int addr_len = 0;
3229	int err;
3230
3231	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3232				   flags & ~MSG_DONTWAIT, &addr_len);
3233	if (err >= 0)
3234		msg->msg_namelen = addr_len;
3235	return err;
3236}
3237EXPORT_SYMBOL(sock_common_recvmsg);
3238
3239/*
3240 *	Set socket options on an inet socket.
3241 */
3242int sock_common_setsockopt(struct socket *sock, int level, int optname,
3243			   sockptr_t optval, unsigned int optlen)
3244{
3245	struct sock *sk = sock->sk;
3246
3247	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3248}
3249EXPORT_SYMBOL(sock_common_setsockopt);
3250
3251void sk_common_release(struct sock *sk)
3252{
3253	if (sk->sk_prot->destroy)
3254		sk->sk_prot->destroy(sk);
3255
3256	/*
3257	 * Observation: when sk_common_release is called, processes have
3258	 * no access to socket. But net still has.
3259	 * Step one, detach it from networking:
3260	 *
3261	 * A. Remove from hash tables.
3262	 */
3263
3264	sk->sk_prot->unhash(sk);
3265
3266	/*
3267	 * In this point socket cannot receive new packets, but it is possible
3268	 * that some packets are in flight because some CPU runs receiver and
3269	 * did hash table lookup before we unhashed socket. They will achieve
3270	 * receive queue and will be purged by socket destructor.
3271	 *
3272	 * Also we still have packets pending on receive queue and probably,
3273	 * our own packets waiting in device queues. sock_destroy will drain
3274	 * receive queue, but transmitted packets will delay socket destruction
3275	 * until the last reference will be released.
3276	 */
3277
3278	sock_orphan(sk);
3279
3280	xfrm_sk_free_policy(sk);
3281
3282	sk_refcnt_debug_release(sk);
3283
3284	sock_put(sk);
3285}
3286EXPORT_SYMBOL(sk_common_release);
3287
3288void sk_get_meminfo(const struct sock *sk, u32 *mem)
3289{
3290	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3291
3292	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3293	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3294	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3295	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3296	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3297	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3298	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3299	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3300	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3301}
3302
3303#ifdef CONFIG_PROC_FS
3304#define PROTO_INUSE_NR	64	/* should be enough for the first time */
3305struct prot_inuse {
3306	int val[PROTO_INUSE_NR];
3307};
3308
3309static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3310
3311void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3312{
3313	__this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3314}
3315EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3316
3317int sock_prot_inuse_get(struct net *net, struct proto *prot)
3318{
3319	int cpu, idx = prot->inuse_idx;
3320	int res = 0;
3321
3322	for_each_possible_cpu(cpu)
3323		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3324
3325	return res >= 0 ? res : 0;
3326}
3327EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3328
3329static void sock_inuse_add(struct net *net, int val)
3330{
3331	this_cpu_add(*net->core.sock_inuse, val);
3332}
3333
3334int sock_inuse_get(struct net *net)
3335{
3336	int cpu, res = 0;
3337
3338	for_each_possible_cpu(cpu)
3339		res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3340
3341	return res;
3342}
3343
3344EXPORT_SYMBOL_GPL(sock_inuse_get);
3345
3346static int __net_init sock_inuse_init_net(struct net *net)
3347{
3348	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3349	if (net->core.prot_inuse == NULL)
3350		return -ENOMEM;
3351
3352	net->core.sock_inuse = alloc_percpu(int);
3353	if (net->core.sock_inuse == NULL)
3354		goto out;
3355
3356	return 0;
3357
3358out:
3359	free_percpu(net->core.prot_inuse);
3360	return -ENOMEM;
3361}
3362
3363static void __net_exit sock_inuse_exit_net(struct net *net)
3364{
3365	free_percpu(net->core.prot_inuse);
3366	free_percpu(net->core.sock_inuse);
3367}
3368
3369static struct pernet_operations net_inuse_ops = {
3370	.init = sock_inuse_init_net,
3371	.exit = sock_inuse_exit_net,
3372};
3373
3374static __init int net_inuse_init(void)
3375{
3376	if (register_pernet_subsys(&net_inuse_ops))
3377		panic("Cannot initialize net inuse counters");
3378
3379	return 0;
3380}
3381
3382core_initcall(net_inuse_init);
3383
3384static int assign_proto_idx(struct proto *prot)
3385{
3386	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3387
3388	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3389		pr_err("PROTO_INUSE_NR exhausted\n");
3390		return -ENOSPC;
3391	}
3392
3393	set_bit(prot->inuse_idx, proto_inuse_idx);
3394	return 0;
3395}
3396
3397static void release_proto_idx(struct proto *prot)
3398{
3399	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3400		clear_bit(prot->inuse_idx, proto_inuse_idx);
3401}
3402#else
3403static inline int assign_proto_idx(struct proto *prot)
3404{
3405	return 0;
3406}
3407
3408static inline void release_proto_idx(struct proto *prot)
3409{
3410}
3411
3412static void sock_inuse_add(struct net *net, int val)
3413{
3414}
3415#endif
3416
3417static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3418{
3419	if (!twsk_prot)
3420		return;
3421	kfree(twsk_prot->twsk_slab_name);
3422	twsk_prot->twsk_slab_name = NULL;
3423	kmem_cache_destroy(twsk_prot->twsk_slab);
3424	twsk_prot->twsk_slab = NULL;
3425}
3426
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3427static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3428{
3429	if (!rsk_prot)
3430		return;
3431	kfree(rsk_prot->slab_name);
3432	rsk_prot->slab_name = NULL;
3433	kmem_cache_destroy(rsk_prot->slab);
3434	rsk_prot->slab = NULL;
3435}
3436
3437static int req_prot_init(const struct proto *prot)
3438{
3439	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3440
3441	if (!rsk_prot)
3442		return 0;
3443
3444	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3445					prot->name);
3446	if (!rsk_prot->slab_name)
3447		return -ENOMEM;
3448
3449	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3450					   rsk_prot->obj_size, 0,
3451					   SLAB_ACCOUNT | prot->slab_flags,
3452					   NULL);
3453
3454	if (!rsk_prot->slab) {
3455		pr_crit("%s: Can't create request sock SLAB cache!\n",
3456			prot->name);
3457		return -ENOMEM;
3458	}
3459	return 0;
3460}
3461
3462int proto_register(struct proto *prot, int alloc_slab)
3463{
3464	int ret = -ENOBUFS;
3465
3466	if (alloc_slab) {
3467		prot->slab = kmem_cache_create_usercopy(prot->name,
3468					prot->obj_size, 0,
3469					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3470					prot->slab_flags,
3471					prot->useroffset, prot->usersize,
3472					NULL);
3473
3474		if (prot->slab == NULL) {
3475			pr_crit("%s: Can't create sock SLAB cache!\n",
3476				prot->name);
3477			goto out;
3478		}
3479
3480		if (req_prot_init(prot))
3481			goto out_free_request_sock_slab;
3482
3483		if (prot->twsk_prot != NULL) {
3484			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3485
3486			if (prot->twsk_prot->twsk_slab_name == NULL)
3487				goto out_free_request_sock_slab;
3488
3489			prot->twsk_prot->twsk_slab =
3490				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3491						  prot->twsk_prot->twsk_obj_size,
3492						  0,
3493						  SLAB_ACCOUNT |
3494						  prot->slab_flags,
3495						  NULL);
3496			if (prot->twsk_prot->twsk_slab == NULL)
3497				goto out_free_timewait_sock_slab;
3498		}
3499	}
3500
3501	mutex_lock(&proto_list_mutex);
3502	ret = assign_proto_idx(prot);
3503	if (ret) {
3504		mutex_unlock(&proto_list_mutex);
3505		goto out_free_timewait_sock_slab;
3506	}
3507	list_add(&prot->node, &proto_list);
3508	mutex_unlock(&proto_list_mutex);
3509	return ret;
3510
3511out_free_timewait_sock_slab:
3512	if (alloc_slab && prot->twsk_prot)
3513		tw_prot_cleanup(prot->twsk_prot);
3514out_free_request_sock_slab:
3515	if (alloc_slab) {
3516		req_prot_cleanup(prot->rsk_prot);
3517
3518		kmem_cache_destroy(prot->slab);
3519		prot->slab = NULL;
3520	}
3521out:
3522	return ret;
3523}
3524EXPORT_SYMBOL(proto_register);
3525
3526void proto_unregister(struct proto *prot)
3527{
3528	mutex_lock(&proto_list_mutex);
3529	release_proto_idx(prot);
3530	list_del(&prot->node);
3531	mutex_unlock(&proto_list_mutex);
3532
3533	kmem_cache_destroy(prot->slab);
3534	prot->slab = NULL;
3535
3536	req_prot_cleanup(prot->rsk_prot);
3537	tw_prot_cleanup(prot->twsk_prot);
3538}
3539EXPORT_SYMBOL(proto_unregister);
3540
3541int sock_load_diag_module(int family, int protocol)
3542{
3543	if (!protocol) {
3544		if (!sock_is_registered(family))
3545			return -ENOENT;
3546
3547		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3548				      NETLINK_SOCK_DIAG, family);
3549	}
3550
3551#ifdef CONFIG_INET
3552	if (family == AF_INET &&
3553	    protocol != IPPROTO_RAW &&
3554	    protocol < MAX_INET_PROTOS &&
3555	    !rcu_access_pointer(inet_protos[protocol]))
3556		return -ENOENT;
3557#endif
3558
3559	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3560			      NETLINK_SOCK_DIAG, family, protocol);
3561}
3562EXPORT_SYMBOL(sock_load_diag_module);
3563
3564#ifdef CONFIG_PROC_FS
3565static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3566	__acquires(proto_list_mutex)
3567{
3568	mutex_lock(&proto_list_mutex);
3569	return seq_list_start_head(&proto_list, *pos);
3570}
3571
3572static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3573{
3574	return seq_list_next(v, &proto_list, pos);
3575}
3576
3577static void proto_seq_stop(struct seq_file *seq, void *v)
3578	__releases(proto_list_mutex)
3579{
3580	mutex_unlock(&proto_list_mutex);
3581}
3582
3583static char proto_method_implemented(const void *method)
3584{
3585	return method == NULL ? 'n' : 'y';
3586}
3587static long sock_prot_memory_allocated(struct proto *proto)
3588{
3589	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3590}
3591
3592static const char *sock_prot_memory_pressure(struct proto *proto)
3593{
3594	return proto->memory_pressure != NULL ?
3595	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3596}
3597
3598static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3599{
3600
3601	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3602			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3603		   proto->name,
3604		   proto->obj_size,
3605		   sock_prot_inuse_get(seq_file_net(seq), proto),
3606		   sock_prot_memory_allocated(proto),
3607		   sock_prot_memory_pressure(proto),
3608		   proto->max_header,
3609		   proto->slab == NULL ? "no" : "yes",
3610		   module_name(proto->owner),
3611		   proto_method_implemented(proto->close),
3612		   proto_method_implemented(proto->connect),
3613		   proto_method_implemented(proto->disconnect),
3614		   proto_method_implemented(proto->accept),
3615		   proto_method_implemented(proto->ioctl),
3616		   proto_method_implemented(proto->init),
3617		   proto_method_implemented(proto->destroy),
3618		   proto_method_implemented(proto->shutdown),
3619		   proto_method_implemented(proto->setsockopt),
3620		   proto_method_implemented(proto->getsockopt),
3621		   proto_method_implemented(proto->sendmsg),
3622		   proto_method_implemented(proto->recvmsg),
3623		   proto_method_implemented(proto->sendpage),
3624		   proto_method_implemented(proto->bind),
3625		   proto_method_implemented(proto->backlog_rcv),
3626		   proto_method_implemented(proto->hash),
3627		   proto_method_implemented(proto->unhash),
3628		   proto_method_implemented(proto->get_port),
3629		   proto_method_implemented(proto->enter_memory_pressure));
3630}
3631
3632static int proto_seq_show(struct seq_file *seq, void *v)
3633{
3634	if (v == &proto_list)
3635		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3636			   "protocol",
3637			   "size",
3638			   "sockets",
3639			   "memory",
3640			   "press",
3641			   "maxhdr",
3642			   "slab",
3643			   "module",
3644			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3645	else
3646		proto_seq_printf(seq, list_entry(v, struct proto, node));
3647	return 0;
3648}
3649
3650static const struct seq_operations proto_seq_ops = {
3651	.start  = proto_seq_start,
3652	.next   = proto_seq_next,
3653	.stop   = proto_seq_stop,
3654	.show   = proto_seq_show,
3655};
3656
3657static __net_init int proto_init_net(struct net *net)
3658{
3659	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3660			sizeof(struct seq_net_private)))
3661		return -ENOMEM;
3662
3663	return 0;
3664}
3665
3666static __net_exit void proto_exit_net(struct net *net)
3667{
3668	remove_proc_entry("protocols", net->proc_net);
3669}
3670
3671
3672static __net_initdata struct pernet_operations proto_net_ops = {
3673	.init = proto_init_net,
3674	.exit = proto_exit_net,
3675};
3676
3677static int __init proto_init(void)
3678{
3679	return register_pernet_subsys(&proto_net_ops);
3680}
3681
3682subsys_initcall(proto_init);
3683
3684#endif /* PROC_FS */
3685
3686#ifdef CONFIG_NET_RX_BUSY_POLL
3687bool sk_busy_loop_end(void *p, unsigned long start_time)
3688{
3689	struct sock *sk = p;
3690
3691	return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3692	       sk_busy_loop_timeout(sk, start_time);
3693}
3694EXPORT_SYMBOL(sk_busy_loop_end);
3695#endif /* CONFIG_NET_RX_BUSY_POLL */
3696
3697int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3698{
3699	if (!sk->sk_prot->bind_add)
3700		return -EOPNOTSUPP;
3701	return sk->sk_prot->bind_add(sk, addr, addr_len);
3702}
3703EXPORT_SYMBOL(sock_bind_add);