Loading...
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(×tamping, optval,
1126 sizeof(timestamping))) {
1127 ret = -EFAULT;
1128 break;
1129 }
1130 } else {
1131 memset(×tamping, 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);
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 <linux/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/udp.h>
111#include <linux/init.h>
112#include <linux/highmem.h>
113#include <linux/user_namespace.h>
114#include <linux/static_key.h>
115#include <linux/memcontrol.h>
116#include <linux/prefetch.h>
117#include <linux/compat.h>
118#include <linux/mroute.h>
119#include <linux/mroute6.h>
120#include <linux/icmpv6.h>
121
122#include <linux/uaccess.h>
123
124#include <linux/netdevice.h>
125#include <net/protocol.h>
126#include <linux/skbuff.h>
127#include <linux/skbuff_ref.h>
128#include <net/net_namespace.h>
129#include <net/request_sock.h>
130#include <net/sock.h>
131#include <net/proto_memory.h>
132#include <linux/net_tstamp.h>
133#include <net/xfrm.h>
134#include <linux/ipsec.h>
135#include <net/cls_cgroup.h>
136#include <net/netprio_cgroup.h>
137#include <linux/sock_diag.h>
138
139#include <linux/filter.h>
140#include <net/sock_reuseport.h>
141#include <net/bpf_sk_storage.h>
142
143#include <trace/events/sock.h>
144
145#include <net/tcp.h>
146#include <net/busy_poll.h>
147#include <net/phonet/phonet.h>
148
149#include <linux/ethtool.h>
150
151#include "dev.h"
152
153static DEFINE_MUTEX(proto_list_mutex);
154static LIST_HEAD(proto_list);
155
156static void sock_def_write_space_wfree(struct sock *sk);
157static void sock_def_write_space(struct sock *sk);
158
159/**
160 * sk_ns_capable - General socket capability test
161 * @sk: Socket to use a capability on or through
162 * @user_ns: The user namespace of the capability to use
163 * @cap: The capability to use
164 *
165 * Test to see if the opener of the socket had when the socket was
166 * created and the current process has the capability @cap in the user
167 * namespace @user_ns.
168 */
169bool sk_ns_capable(const struct sock *sk,
170 struct user_namespace *user_ns, int cap)
171{
172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
173 ns_capable(user_ns, cap);
174}
175EXPORT_SYMBOL(sk_ns_capable);
176
177/**
178 * sk_capable - Socket global capability test
179 * @sk: Socket to use a capability on or through
180 * @cap: The global capability to use
181 *
182 * Test to see if the opener of the socket had when the socket was
183 * created and the current process has the capability @cap in all user
184 * namespaces.
185 */
186bool sk_capable(const struct sock *sk, int cap)
187{
188 return sk_ns_capable(sk, &init_user_ns, cap);
189}
190EXPORT_SYMBOL(sk_capable);
191
192/**
193 * sk_net_capable - Network namespace socket capability test
194 * @sk: Socket to use a capability on or through
195 * @cap: The capability to use
196 *
197 * Test to see if the opener of the socket had when the socket was created
198 * and the current process has the capability @cap over the network namespace
199 * the socket is a member of.
200 */
201bool sk_net_capable(const struct sock *sk, int cap)
202{
203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
204}
205EXPORT_SYMBOL(sk_net_capable);
206
207/*
208 * Each address family might have different locking rules, so we have
209 * one slock key per address family and separate keys for internal and
210 * userspace sockets.
211 */
212static struct lock_class_key af_family_keys[AF_MAX];
213static struct lock_class_key af_family_kern_keys[AF_MAX];
214static struct lock_class_key af_family_slock_keys[AF_MAX];
215static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
216
217/*
218 * Make lock validator output more readable. (we pre-construct these
219 * strings build-time, so that runtime initialization of socket
220 * locks is fast):
221 */
222
223#define _sock_locks(x) \
224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
233 x "27" , x "28" , x "AF_CAN" , \
234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 x "AF_MCTP" , \
240 x "AF_MAX"
241
242static const char *const af_family_key_strings[AF_MAX+1] = {
243 _sock_locks("sk_lock-")
244};
245static const char *const af_family_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("slock-")
247};
248static const char *const af_family_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("clock-")
250};
251
252static const char *const af_family_kern_key_strings[AF_MAX+1] = {
253 _sock_locks("k-sk_lock-")
254};
255static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-slock-")
257};
258static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
259 _sock_locks("k-clock-")
260};
261static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 _sock_locks("rlock-")
263};
264static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
265 _sock_locks("wlock-")
266};
267static const char *const af_family_elock_key_strings[AF_MAX+1] = {
268 _sock_locks("elock-")
269};
270
271/*
272 * sk_callback_lock and sk queues locking rules are per-address-family,
273 * so split the lock classes by using a per-AF key:
274 */
275static struct lock_class_key af_callback_keys[AF_MAX];
276static struct lock_class_key af_rlock_keys[AF_MAX];
277static struct lock_class_key af_wlock_keys[AF_MAX];
278static struct lock_class_key af_elock_keys[AF_MAX];
279static struct lock_class_key af_kern_callback_keys[AF_MAX];
280
281/* Run time adjustable parameters. */
282__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
283EXPORT_SYMBOL(sysctl_wmem_max);
284__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
285EXPORT_SYMBOL(sysctl_rmem_max);
286__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
287__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
288
289DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
290EXPORT_SYMBOL_GPL(memalloc_socks_key);
291
292/**
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
295 *
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
299 */
300void sk_set_memalloc(struct sock *sk)
301{
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_branch_inc(&memalloc_socks_key);
305}
306EXPORT_SYMBOL_GPL(sk_set_memalloc);
307
308void sk_clear_memalloc(struct sock *sk)
309{
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_branch_dec(&memalloc_socks_key);
313
314 /*
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 */
321 sk_mem_reclaim(sk);
322}
323EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324
325int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
326{
327 int ret;
328 unsigned int noreclaim_flag;
329
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
332
333 noreclaim_flag = memalloc_noreclaim_save();
334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
335 tcp_v6_do_rcv,
336 tcp_v4_do_rcv,
337 sk, skb);
338 memalloc_noreclaim_restore(noreclaim_flag);
339
340 return ret;
341}
342EXPORT_SYMBOL(__sk_backlog_rcv);
343
344void sk_error_report(struct sock *sk)
345{
346 sk->sk_error_report(sk);
347
348 switch (sk->sk_family) {
349 case AF_INET:
350 fallthrough;
351 case AF_INET6:
352 trace_inet_sk_error_report(sk);
353 break;
354 default:
355 break;
356 }
357}
358EXPORT_SYMBOL(sk_error_report);
359
360int sock_get_timeout(long timeo, void *optval, bool old_timeval)
361{
362 struct __kernel_sock_timeval tv;
363
364 if (timeo == MAX_SCHEDULE_TIMEOUT) {
365 tv.tv_sec = 0;
366 tv.tv_usec = 0;
367 } else {
368 tv.tv_sec = timeo / HZ;
369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
370 }
371
372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
374 *(struct old_timeval32 *)optval = tv32;
375 return sizeof(tv32);
376 }
377
378 if (old_timeval) {
379 struct __kernel_old_timeval old_tv;
380 old_tv.tv_sec = tv.tv_sec;
381 old_tv.tv_usec = tv.tv_usec;
382 *(struct __kernel_old_timeval *)optval = old_tv;
383 return sizeof(old_tv);
384 }
385
386 *(struct __kernel_sock_timeval *)optval = tv;
387 return sizeof(tv);
388}
389EXPORT_SYMBOL(sock_get_timeout);
390
391int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
392 sockptr_t optval, int optlen, bool old_timeval)
393{
394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
395 struct old_timeval32 tv32;
396
397 if (optlen < sizeof(tv32))
398 return -EINVAL;
399
400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
401 return -EFAULT;
402 tv->tv_sec = tv32.tv_sec;
403 tv->tv_usec = tv32.tv_usec;
404 } else if (old_timeval) {
405 struct __kernel_old_timeval old_tv;
406
407 if (optlen < sizeof(old_tv))
408 return -EINVAL;
409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
410 return -EFAULT;
411 tv->tv_sec = old_tv.tv_sec;
412 tv->tv_usec = old_tv.tv_usec;
413 } else {
414 if (optlen < sizeof(*tv))
415 return -EINVAL;
416 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
417 return -EFAULT;
418 }
419
420 return 0;
421}
422EXPORT_SYMBOL(sock_copy_user_timeval);
423
424static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
425 bool old_timeval)
426{
427 struct __kernel_sock_timeval tv;
428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
429 long val;
430
431 if (err)
432 return err;
433
434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
435 return -EDOM;
436
437 if (tv.tv_sec < 0) {
438 static int warned __read_mostly;
439
440 WRITE_ONCE(*timeo_p, 0);
441 if (warned < 10 && net_ratelimit()) {
442 warned++;
443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
444 __func__, current->comm, task_pid_nr(current));
445 }
446 return 0;
447 }
448 val = MAX_SCHEDULE_TIMEOUT;
449 if ((tv.tv_sec || tv.tv_usec) &&
450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
452 USEC_PER_SEC / HZ);
453 WRITE_ONCE(*timeo_p, val);
454 return 0;
455}
456
457static bool sock_needs_netstamp(const struct sock *sk)
458{
459 switch (sk->sk_family) {
460 case AF_UNSPEC:
461 case AF_UNIX:
462 return false;
463 default:
464 return true;
465 }
466}
467
468static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
469{
470 if (sk->sk_flags & flags) {
471 sk->sk_flags &= ~flags;
472 if (sock_needs_netstamp(sk) &&
473 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
474 net_disable_timestamp();
475 }
476}
477
478
479int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
480{
481 unsigned long flags;
482 struct sk_buff_head *list = &sk->sk_receive_queue;
483
484 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
485 atomic_inc(&sk->sk_drops);
486 trace_sock_rcvqueue_full(sk, skb);
487 return -ENOMEM;
488 }
489
490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
491 atomic_inc(&sk->sk_drops);
492 return -ENOBUFS;
493 }
494
495 skb->dev = NULL;
496 skb_set_owner_r(skb, sk);
497
498 /* we escape from rcu protected region, make sure we dont leak
499 * a norefcounted dst
500 */
501 skb_dst_force(skb);
502
503 spin_lock_irqsave(&list->lock, flags);
504 sock_skb_set_dropcount(sk, skb);
505 __skb_queue_tail(list, skb);
506 spin_unlock_irqrestore(&list->lock, flags);
507
508 if (!sock_flag(sk, SOCK_DEAD))
509 sk->sk_data_ready(sk);
510 return 0;
511}
512EXPORT_SYMBOL(__sock_queue_rcv_skb);
513
514int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
515 enum skb_drop_reason *reason)
516{
517 enum skb_drop_reason drop_reason;
518 int err;
519
520 err = sk_filter(sk, skb);
521 if (err) {
522 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
523 goto out;
524 }
525 err = __sock_queue_rcv_skb(sk, skb);
526 switch (err) {
527 case -ENOMEM:
528 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
529 break;
530 case -ENOBUFS:
531 drop_reason = SKB_DROP_REASON_PROTO_MEM;
532 break;
533 default:
534 drop_reason = SKB_NOT_DROPPED_YET;
535 break;
536 }
537out:
538 if (reason)
539 *reason = drop_reason;
540 return err;
541}
542EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
543
544int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 const int nested, unsigned int trim_cap, bool refcounted)
546{
547 int rc = NET_RX_SUCCESS;
548
549 if (sk_filter_trim_cap(sk, skb, trim_cap))
550 goto discard_and_relse;
551
552 skb->dev = NULL;
553
554 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
555 atomic_inc(&sk->sk_drops);
556 goto discard_and_relse;
557 }
558 if (nested)
559 bh_lock_sock_nested(sk);
560 else
561 bh_lock_sock(sk);
562 if (!sock_owned_by_user(sk)) {
563 /*
564 * trylock + unlock semantics:
565 */
566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
567
568 rc = sk_backlog_rcv(sk, skb);
569
570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
572 bh_unlock_sock(sk);
573 atomic_inc(&sk->sk_drops);
574 goto discard_and_relse;
575 }
576
577 bh_unlock_sock(sk);
578out:
579 if (refcounted)
580 sock_put(sk);
581 return rc;
582discard_and_relse:
583 kfree_skb(skb);
584 goto out;
585}
586EXPORT_SYMBOL(__sk_receive_skb);
587
588INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
589 u32));
590INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
591 u32));
592struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
593{
594 struct dst_entry *dst = __sk_dst_get(sk);
595
596 if (dst && dst->obsolete &&
597 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
598 dst, cookie) == NULL) {
599 sk_tx_queue_clear(sk);
600 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
601 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
602 dst_release(dst);
603 return NULL;
604 }
605
606 return dst;
607}
608EXPORT_SYMBOL(__sk_dst_check);
609
610struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
611{
612 struct dst_entry *dst = sk_dst_get(sk);
613
614 if (dst && dst->obsolete &&
615 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
616 dst, cookie) == NULL) {
617 sk_dst_reset(sk);
618 dst_release(dst);
619 return NULL;
620 }
621
622 return dst;
623}
624EXPORT_SYMBOL(sk_dst_check);
625
626static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
627{
628 int ret = -ENOPROTOOPT;
629#ifdef CONFIG_NETDEVICES
630 struct net *net = sock_net(sk);
631
632 /* Sorry... */
633 ret = -EPERM;
634 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
635 goto out;
636
637 ret = -EINVAL;
638 if (ifindex < 0)
639 goto out;
640
641 /* Paired with all READ_ONCE() done locklessly. */
642 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
643
644 if (sk->sk_prot->rehash)
645 sk->sk_prot->rehash(sk);
646 sk_dst_reset(sk);
647
648 ret = 0;
649
650out:
651#endif
652
653 return ret;
654}
655
656int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
657{
658 int ret;
659
660 if (lock_sk)
661 lock_sock(sk);
662 ret = sock_bindtoindex_locked(sk, ifindex);
663 if (lock_sk)
664 release_sock(sk);
665
666 return ret;
667}
668EXPORT_SYMBOL(sock_bindtoindex);
669
670static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
671{
672 int ret = -ENOPROTOOPT;
673#ifdef CONFIG_NETDEVICES
674 struct net *net = sock_net(sk);
675 char devname[IFNAMSIZ];
676 int index;
677
678 ret = -EINVAL;
679 if (optlen < 0)
680 goto out;
681
682 /* Bind this socket to a particular device like "eth0",
683 * as specified in the passed interface name. If the
684 * name is "" or the option length is zero the socket
685 * is not bound.
686 */
687 if (optlen > IFNAMSIZ - 1)
688 optlen = IFNAMSIZ - 1;
689 memset(devname, 0, sizeof(devname));
690
691 ret = -EFAULT;
692 if (copy_from_sockptr(devname, optval, optlen))
693 goto out;
694
695 index = 0;
696 if (devname[0] != '\0') {
697 struct net_device *dev;
698
699 rcu_read_lock();
700 dev = dev_get_by_name_rcu(net, devname);
701 if (dev)
702 index = dev->ifindex;
703 rcu_read_unlock();
704 ret = -ENODEV;
705 if (!dev)
706 goto out;
707 }
708
709 sockopt_lock_sock(sk);
710 ret = sock_bindtoindex_locked(sk, index);
711 sockopt_release_sock(sk);
712out:
713#endif
714
715 return ret;
716}
717
718static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
719 sockptr_t optlen, int len)
720{
721 int ret = -ENOPROTOOPT;
722#ifdef CONFIG_NETDEVICES
723 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
724 struct net *net = sock_net(sk);
725 char devname[IFNAMSIZ];
726
727 if (bound_dev_if == 0) {
728 len = 0;
729 goto zero;
730 }
731
732 ret = -EINVAL;
733 if (len < IFNAMSIZ)
734 goto out;
735
736 ret = netdev_get_name(net, devname, bound_dev_if);
737 if (ret)
738 goto out;
739
740 len = strlen(devname) + 1;
741
742 ret = -EFAULT;
743 if (copy_to_sockptr(optval, devname, len))
744 goto out;
745
746zero:
747 ret = -EFAULT;
748 if (copy_to_sockptr(optlen, &len, sizeof(int)))
749 goto out;
750
751 ret = 0;
752
753out:
754#endif
755
756 return ret;
757}
758
759bool sk_mc_loop(const struct sock *sk)
760{
761 if (dev_recursion_level())
762 return false;
763 if (!sk)
764 return true;
765 /* IPV6_ADDRFORM can change sk->sk_family under us. */
766 switch (READ_ONCE(sk->sk_family)) {
767 case AF_INET:
768 return inet_test_bit(MC_LOOP, sk);
769#if IS_ENABLED(CONFIG_IPV6)
770 case AF_INET6:
771 return inet6_test_bit(MC6_LOOP, sk);
772#endif
773 }
774 WARN_ON_ONCE(1);
775 return true;
776}
777EXPORT_SYMBOL(sk_mc_loop);
778
779void sock_set_reuseaddr(struct sock *sk)
780{
781 lock_sock(sk);
782 sk->sk_reuse = SK_CAN_REUSE;
783 release_sock(sk);
784}
785EXPORT_SYMBOL(sock_set_reuseaddr);
786
787void sock_set_reuseport(struct sock *sk)
788{
789 lock_sock(sk);
790 sk->sk_reuseport = true;
791 release_sock(sk);
792}
793EXPORT_SYMBOL(sock_set_reuseport);
794
795void sock_no_linger(struct sock *sk)
796{
797 lock_sock(sk);
798 WRITE_ONCE(sk->sk_lingertime, 0);
799 sock_set_flag(sk, SOCK_LINGER);
800 release_sock(sk);
801}
802EXPORT_SYMBOL(sock_no_linger);
803
804void sock_set_priority(struct sock *sk, u32 priority)
805{
806 WRITE_ONCE(sk->sk_priority, priority);
807}
808EXPORT_SYMBOL(sock_set_priority);
809
810void sock_set_sndtimeo(struct sock *sk, s64 secs)
811{
812 lock_sock(sk);
813 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
814 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
815 else
816 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
817 release_sock(sk);
818}
819EXPORT_SYMBOL(sock_set_sndtimeo);
820
821static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
822{
823 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
824 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
825 if (val) {
826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
828 }
829}
830
831void sock_enable_timestamps(struct sock *sk)
832{
833 lock_sock(sk);
834 __sock_set_timestamps(sk, true, false, true);
835 release_sock(sk);
836}
837EXPORT_SYMBOL(sock_enable_timestamps);
838
839void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
840{
841 switch (optname) {
842 case SO_TIMESTAMP_OLD:
843 __sock_set_timestamps(sk, valbool, false, false);
844 break;
845 case SO_TIMESTAMP_NEW:
846 __sock_set_timestamps(sk, valbool, true, false);
847 break;
848 case SO_TIMESTAMPNS_OLD:
849 __sock_set_timestamps(sk, valbool, false, true);
850 break;
851 case SO_TIMESTAMPNS_NEW:
852 __sock_set_timestamps(sk, valbool, true, true);
853 break;
854 }
855}
856
857static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
858{
859 struct net *net = sock_net(sk);
860 struct net_device *dev = NULL;
861 bool match = false;
862 int *vclock_index;
863 int i, num;
864
865 if (sk->sk_bound_dev_if)
866 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
867
868 if (!dev) {
869 pr_err("%s: sock not bind to device\n", __func__);
870 return -EOPNOTSUPP;
871 }
872
873 num = ethtool_get_phc_vclocks(dev, &vclock_index);
874 dev_put(dev);
875
876 for (i = 0; i < num; i++) {
877 if (*(vclock_index + i) == phc_index) {
878 match = true;
879 break;
880 }
881 }
882
883 if (num > 0)
884 kfree(vclock_index);
885
886 if (!match)
887 return -EINVAL;
888
889 WRITE_ONCE(sk->sk_bind_phc, phc_index);
890
891 return 0;
892}
893
894int sock_set_timestamping(struct sock *sk, int optname,
895 struct so_timestamping timestamping)
896{
897 int val = timestamping.flags;
898 int ret;
899
900 if (val & ~SOF_TIMESTAMPING_MASK)
901 return -EINVAL;
902
903 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
904 !(val & SOF_TIMESTAMPING_OPT_ID))
905 return -EINVAL;
906
907 if (val & SOF_TIMESTAMPING_OPT_ID &&
908 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
909 if (sk_is_tcp(sk)) {
910 if ((1 << sk->sk_state) &
911 (TCPF_CLOSE | TCPF_LISTEN))
912 return -EINVAL;
913 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
915 else
916 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
917 } else {
918 atomic_set(&sk->sk_tskey, 0);
919 }
920 }
921
922 if (val & SOF_TIMESTAMPING_OPT_STATS &&
923 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
924 return -EINVAL;
925
926 if (val & SOF_TIMESTAMPING_BIND_PHC) {
927 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
928 if (ret)
929 return ret;
930 }
931
932 WRITE_ONCE(sk->sk_tsflags, val);
933 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
934
935 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
936 sock_enable_timestamp(sk,
937 SOCK_TIMESTAMPING_RX_SOFTWARE);
938 else
939 sock_disable_timestamp(sk,
940 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
941 return 0;
942}
943
944void sock_set_keepalive(struct sock *sk)
945{
946 lock_sock(sk);
947 if (sk->sk_prot->keepalive)
948 sk->sk_prot->keepalive(sk, true);
949 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
950 release_sock(sk);
951}
952EXPORT_SYMBOL(sock_set_keepalive);
953
954static void __sock_set_rcvbuf(struct sock *sk, int val)
955{
956 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
957 * as a negative value.
958 */
959 val = min_t(int, val, INT_MAX / 2);
960 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
961
962 /* We double it on the way in to account for "struct sk_buff" etc.
963 * overhead. Applications assume that the SO_RCVBUF setting they make
964 * will allow that much actual data to be received on that socket.
965 *
966 * Applications are unaware that "struct sk_buff" and other overheads
967 * allocate from the receive buffer during socket buffer allocation.
968 *
969 * And after considering the possible alternatives, returning the value
970 * we actually used in getsockopt is the most desirable behavior.
971 */
972 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
973}
974
975void sock_set_rcvbuf(struct sock *sk, int val)
976{
977 lock_sock(sk);
978 __sock_set_rcvbuf(sk, val);
979 release_sock(sk);
980}
981EXPORT_SYMBOL(sock_set_rcvbuf);
982
983static void __sock_set_mark(struct sock *sk, u32 val)
984{
985 if (val != sk->sk_mark) {
986 WRITE_ONCE(sk->sk_mark, val);
987 sk_dst_reset(sk);
988 }
989}
990
991void sock_set_mark(struct sock *sk, u32 val)
992{
993 lock_sock(sk);
994 __sock_set_mark(sk, val);
995 release_sock(sk);
996}
997EXPORT_SYMBOL(sock_set_mark);
998
999static void sock_release_reserved_memory(struct sock *sk, int bytes)
1000{
1001 /* Round down bytes to multiple of pages */
1002 bytes = round_down(bytes, PAGE_SIZE);
1003
1004 WARN_ON(bytes > sk->sk_reserved_mem);
1005 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1006 sk_mem_reclaim(sk);
1007}
1008
1009static int sock_reserve_memory(struct sock *sk, int bytes)
1010{
1011 long allocated;
1012 bool charged;
1013 int pages;
1014
1015 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1016 return -EOPNOTSUPP;
1017
1018 if (!bytes)
1019 return 0;
1020
1021 pages = sk_mem_pages(bytes);
1022
1023 /* pre-charge to memcg */
1024 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1025 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1026 if (!charged)
1027 return -ENOMEM;
1028
1029 /* pre-charge to forward_alloc */
1030 sk_memory_allocated_add(sk, pages);
1031 allocated = sk_memory_allocated(sk);
1032 /* If the system goes into memory pressure with this
1033 * precharge, give up and return error.
1034 */
1035 if (allocated > sk_prot_mem_limits(sk, 1)) {
1036 sk_memory_allocated_sub(sk, pages);
1037 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1038 return -ENOMEM;
1039 }
1040 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1041
1042 WRITE_ONCE(sk->sk_reserved_mem,
1043 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1044
1045 return 0;
1046}
1047
1048#ifdef CONFIG_PAGE_POOL
1049
1050/* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1051 * in 1 syscall. The limit exists to limit the amount of memory the kernel
1052 * allocates to copy these tokens, and to prevent looping over the frags for
1053 * too long.
1054 */
1055#define MAX_DONTNEED_TOKENS 128
1056#define MAX_DONTNEED_FRAGS 1024
1057
1058static noinline_for_stack int
1059sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1060{
1061 unsigned int num_tokens, i, j, k, netmem_num = 0;
1062 struct dmabuf_token *tokens;
1063 int ret = 0, num_frags = 0;
1064 netmem_ref netmems[16];
1065
1066 if (!sk_is_tcp(sk))
1067 return -EBADF;
1068
1069 if (optlen % sizeof(*tokens) ||
1070 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1071 return -EINVAL;
1072
1073 num_tokens = optlen / sizeof(*tokens);
1074 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL);
1075 if (!tokens)
1076 return -ENOMEM;
1077
1078 if (copy_from_sockptr(tokens, optval, optlen)) {
1079 kvfree(tokens);
1080 return -EFAULT;
1081 }
1082
1083 xa_lock_bh(&sk->sk_user_frags);
1084 for (i = 0; i < num_tokens; i++) {
1085 for (j = 0; j < tokens[i].token_count; j++) {
1086 if (++num_frags > MAX_DONTNEED_FRAGS)
1087 goto frag_limit_reached;
1088
1089 netmem_ref netmem = (__force netmem_ref)__xa_erase(
1090 &sk->sk_user_frags, tokens[i].token_start + j);
1091
1092 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1093 continue;
1094
1095 netmems[netmem_num++] = netmem;
1096 if (netmem_num == ARRAY_SIZE(netmems)) {
1097 xa_unlock_bh(&sk->sk_user_frags);
1098 for (k = 0; k < netmem_num; k++)
1099 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1100 netmem_num = 0;
1101 xa_lock_bh(&sk->sk_user_frags);
1102 }
1103 ret++;
1104 }
1105 }
1106
1107frag_limit_reached:
1108 xa_unlock_bh(&sk->sk_user_frags);
1109 for (k = 0; k < netmem_num; k++)
1110 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1111
1112 kvfree(tokens);
1113 return ret;
1114}
1115#endif
1116
1117void sockopt_lock_sock(struct sock *sk)
1118{
1119 /* When current->bpf_ctx is set, the setsockopt is called from
1120 * a bpf prog. bpf has ensured the sk lock has been
1121 * acquired before calling setsockopt().
1122 */
1123 if (has_current_bpf_ctx())
1124 return;
1125
1126 lock_sock(sk);
1127}
1128EXPORT_SYMBOL(sockopt_lock_sock);
1129
1130void sockopt_release_sock(struct sock *sk)
1131{
1132 if (has_current_bpf_ctx())
1133 return;
1134
1135 release_sock(sk);
1136}
1137EXPORT_SYMBOL(sockopt_release_sock);
1138
1139bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1140{
1141 return has_current_bpf_ctx() || ns_capable(ns, cap);
1142}
1143EXPORT_SYMBOL(sockopt_ns_capable);
1144
1145bool sockopt_capable(int cap)
1146{
1147 return has_current_bpf_ctx() || capable(cap);
1148}
1149EXPORT_SYMBOL(sockopt_capable);
1150
1151static int sockopt_validate_clockid(__kernel_clockid_t value)
1152{
1153 switch (value) {
1154 case CLOCK_REALTIME:
1155 case CLOCK_MONOTONIC:
1156 case CLOCK_TAI:
1157 return 0;
1158 }
1159 return -EINVAL;
1160}
1161
1162/*
1163 * This is meant for all protocols to use and covers goings on
1164 * at the socket level. Everything here is generic.
1165 */
1166
1167int sk_setsockopt(struct sock *sk, int level, int optname,
1168 sockptr_t optval, unsigned int optlen)
1169{
1170 struct so_timestamping timestamping;
1171 struct socket *sock = sk->sk_socket;
1172 struct sock_txtime sk_txtime;
1173 int val;
1174 int valbool;
1175 struct linger ling;
1176 int ret = 0;
1177
1178 /*
1179 * Options without arguments
1180 */
1181
1182 if (optname == SO_BINDTODEVICE)
1183 return sock_setbindtodevice(sk, optval, optlen);
1184
1185 if (optlen < sizeof(int))
1186 return -EINVAL;
1187
1188 if (copy_from_sockptr(&val, optval, sizeof(val)))
1189 return -EFAULT;
1190
1191 valbool = val ? 1 : 0;
1192
1193 /* handle options which do not require locking the socket. */
1194 switch (optname) {
1195 case SO_PRIORITY:
1196 if ((val >= 0 && val <= 6) ||
1197 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1198 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1199 sock_set_priority(sk, val);
1200 return 0;
1201 }
1202 return -EPERM;
1203 case SO_PASSSEC:
1204 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1205 return 0;
1206 case SO_PASSCRED:
1207 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1208 return 0;
1209 case SO_PASSPIDFD:
1210 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1211 return 0;
1212 case SO_TYPE:
1213 case SO_PROTOCOL:
1214 case SO_DOMAIN:
1215 case SO_ERROR:
1216 return -ENOPROTOOPT;
1217#ifdef CONFIG_NET_RX_BUSY_POLL
1218 case SO_BUSY_POLL:
1219 if (val < 0)
1220 return -EINVAL;
1221 WRITE_ONCE(sk->sk_ll_usec, val);
1222 return 0;
1223 case SO_PREFER_BUSY_POLL:
1224 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1225 return -EPERM;
1226 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1227 return 0;
1228 case SO_BUSY_POLL_BUDGET:
1229 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1230 !sockopt_capable(CAP_NET_ADMIN))
1231 return -EPERM;
1232 if (val < 0 || val > U16_MAX)
1233 return -EINVAL;
1234 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1235 return 0;
1236#endif
1237 case SO_MAX_PACING_RATE:
1238 {
1239 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1240 unsigned long pacing_rate;
1241
1242 if (sizeof(ulval) != sizeof(val) &&
1243 optlen >= sizeof(ulval) &&
1244 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1245 return -EFAULT;
1246 }
1247 if (ulval != ~0UL)
1248 cmpxchg(&sk->sk_pacing_status,
1249 SK_PACING_NONE,
1250 SK_PACING_NEEDED);
1251 /* Pairs with READ_ONCE() from sk_getsockopt() */
1252 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1253 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1254 if (ulval < pacing_rate)
1255 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1256 return 0;
1257 }
1258 case SO_TXREHASH:
1259 if (val < -1 || val > 1)
1260 return -EINVAL;
1261 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1262 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1263 /* Paired with READ_ONCE() in tcp_rtx_synack()
1264 * and sk_getsockopt().
1265 */
1266 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1267 return 0;
1268 case SO_PEEK_OFF:
1269 {
1270 int (*set_peek_off)(struct sock *sk, int val);
1271
1272 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1273 if (set_peek_off)
1274 ret = set_peek_off(sk, val);
1275 else
1276 ret = -EOPNOTSUPP;
1277 return ret;
1278 }
1279#ifdef CONFIG_PAGE_POOL
1280 case SO_DEVMEM_DONTNEED:
1281 return sock_devmem_dontneed(sk, optval, optlen);
1282#endif
1283 }
1284
1285 sockopt_lock_sock(sk);
1286
1287 switch (optname) {
1288 case SO_DEBUG:
1289 if (val && !sockopt_capable(CAP_NET_ADMIN))
1290 ret = -EACCES;
1291 else
1292 sock_valbool_flag(sk, SOCK_DBG, valbool);
1293 break;
1294 case SO_REUSEADDR:
1295 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1296 break;
1297 case SO_REUSEPORT:
1298 if (valbool && !sk_is_inet(sk))
1299 ret = -EOPNOTSUPP;
1300 else
1301 sk->sk_reuseport = valbool;
1302 break;
1303 case SO_DONTROUTE:
1304 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1305 sk_dst_reset(sk);
1306 break;
1307 case SO_BROADCAST:
1308 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1309 break;
1310 case SO_SNDBUF:
1311 /* Don't error on this BSD doesn't and if you think
1312 * about it this is right. Otherwise apps have to
1313 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1314 * are treated in BSD as hints
1315 */
1316 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1317set_sndbuf:
1318 /* Ensure val * 2 fits into an int, to prevent max_t()
1319 * from treating it as a negative value.
1320 */
1321 val = min_t(int, val, INT_MAX / 2);
1322 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1323 WRITE_ONCE(sk->sk_sndbuf,
1324 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1325 /* Wake up sending tasks if we upped the value. */
1326 sk->sk_write_space(sk);
1327 break;
1328
1329 case SO_SNDBUFFORCE:
1330 if (!sockopt_capable(CAP_NET_ADMIN)) {
1331 ret = -EPERM;
1332 break;
1333 }
1334
1335 /* No negative values (to prevent underflow, as val will be
1336 * multiplied by 2).
1337 */
1338 if (val < 0)
1339 val = 0;
1340 goto set_sndbuf;
1341
1342 case SO_RCVBUF:
1343 /* Don't error on this BSD doesn't and if you think
1344 * about it this is right. Otherwise apps have to
1345 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1346 * are treated in BSD as hints
1347 */
1348 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1349 break;
1350
1351 case SO_RCVBUFFORCE:
1352 if (!sockopt_capable(CAP_NET_ADMIN)) {
1353 ret = -EPERM;
1354 break;
1355 }
1356
1357 /* No negative values (to prevent underflow, as val will be
1358 * multiplied by 2).
1359 */
1360 __sock_set_rcvbuf(sk, max(val, 0));
1361 break;
1362
1363 case SO_KEEPALIVE:
1364 if (sk->sk_prot->keepalive)
1365 sk->sk_prot->keepalive(sk, valbool);
1366 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1367 break;
1368
1369 case SO_OOBINLINE:
1370 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1371 break;
1372
1373 case SO_NO_CHECK:
1374 sk->sk_no_check_tx = valbool;
1375 break;
1376
1377 case SO_LINGER:
1378 if (optlen < sizeof(ling)) {
1379 ret = -EINVAL; /* 1003.1g */
1380 break;
1381 }
1382 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1383 ret = -EFAULT;
1384 break;
1385 }
1386 if (!ling.l_onoff) {
1387 sock_reset_flag(sk, SOCK_LINGER);
1388 } else {
1389 unsigned long t_sec = ling.l_linger;
1390
1391 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1392 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1393 else
1394 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1395 sock_set_flag(sk, SOCK_LINGER);
1396 }
1397 break;
1398
1399 case SO_BSDCOMPAT:
1400 break;
1401
1402 case SO_TIMESTAMP_OLD:
1403 case SO_TIMESTAMP_NEW:
1404 case SO_TIMESTAMPNS_OLD:
1405 case SO_TIMESTAMPNS_NEW:
1406 sock_set_timestamp(sk, optname, valbool);
1407 break;
1408
1409 case SO_TIMESTAMPING_NEW:
1410 case SO_TIMESTAMPING_OLD:
1411 if (optlen == sizeof(timestamping)) {
1412 if (copy_from_sockptr(×tamping, optval,
1413 sizeof(timestamping))) {
1414 ret = -EFAULT;
1415 break;
1416 }
1417 } else {
1418 memset(×tamping, 0, sizeof(timestamping));
1419 timestamping.flags = val;
1420 }
1421 ret = sock_set_timestamping(sk, optname, timestamping);
1422 break;
1423
1424 case SO_RCVLOWAT:
1425 {
1426 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1427
1428 if (val < 0)
1429 val = INT_MAX;
1430 if (sock)
1431 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1432 if (set_rcvlowat)
1433 ret = set_rcvlowat(sk, val);
1434 else
1435 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1436 break;
1437 }
1438 case SO_RCVTIMEO_OLD:
1439 case SO_RCVTIMEO_NEW:
1440 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1441 optlen, optname == SO_RCVTIMEO_OLD);
1442 break;
1443
1444 case SO_SNDTIMEO_OLD:
1445 case SO_SNDTIMEO_NEW:
1446 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1447 optlen, optname == SO_SNDTIMEO_OLD);
1448 break;
1449
1450 case SO_ATTACH_FILTER: {
1451 struct sock_fprog fprog;
1452
1453 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1454 if (!ret)
1455 ret = sk_attach_filter(&fprog, sk);
1456 break;
1457 }
1458 case SO_ATTACH_BPF:
1459 ret = -EINVAL;
1460 if (optlen == sizeof(u32)) {
1461 u32 ufd;
1462
1463 ret = -EFAULT;
1464 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1465 break;
1466
1467 ret = sk_attach_bpf(ufd, sk);
1468 }
1469 break;
1470
1471 case SO_ATTACH_REUSEPORT_CBPF: {
1472 struct sock_fprog fprog;
1473
1474 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1475 if (!ret)
1476 ret = sk_reuseport_attach_filter(&fprog, sk);
1477 break;
1478 }
1479 case SO_ATTACH_REUSEPORT_EBPF:
1480 ret = -EINVAL;
1481 if (optlen == sizeof(u32)) {
1482 u32 ufd;
1483
1484 ret = -EFAULT;
1485 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1486 break;
1487
1488 ret = sk_reuseport_attach_bpf(ufd, sk);
1489 }
1490 break;
1491
1492 case SO_DETACH_REUSEPORT_BPF:
1493 ret = reuseport_detach_prog(sk);
1494 break;
1495
1496 case SO_DETACH_FILTER:
1497 ret = sk_detach_filter(sk);
1498 break;
1499
1500 case SO_LOCK_FILTER:
1501 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1502 ret = -EPERM;
1503 else
1504 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1505 break;
1506
1507 case SO_MARK:
1508 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1509 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1510 ret = -EPERM;
1511 break;
1512 }
1513
1514 __sock_set_mark(sk, val);
1515 break;
1516 case SO_RCVMARK:
1517 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1518 break;
1519
1520 case SO_RXQ_OVFL:
1521 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1522 break;
1523
1524 case SO_WIFI_STATUS:
1525 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1526 break;
1527
1528 case SO_NOFCS:
1529 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1530 break;
1531
1532 case SO_SELECT_ERR_QUEUE:
1533 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1534 break;
1535
1536
1537 case SO_INCOMING_CPU:
1538 reuseport_update_incoming_cpu(sk, val);
1539 break;
1540
1541 case SO_CNX_ADVICE:
1542 if (val == 1)
1543 dst_negative_advice(sk);
1544 break;
1545
1546 case SO_ZEROCOPY:
1547 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1548 if (!(sk_is_tcp(sk) ||
1549 (sk->sk_type == SOCK_DGRAM &&
1550 sk->sk_protocol == IPPROTO_UDP)))
1551 ret = -EOPNOTSUPP;
1552 } else if (sk->sk_family != PF_RDS) {
1553 ret = -EOPNOTSUPP;
1554 }
1555 if (!ret) {
1556 if (val < 0 || val > 1)
1557 ret = -EINVAL;
1558 else
1559 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1560 }
1561 break;
1562
1563 case SO_TXTIME:
1564 if (optlen != sizeof(struct sock_txtime)) {
1565 ret = -EINVAL;
1566 break;
1567 } else if (copy_from_sockptr(&sk_txtime, optval,
1568 sizeof(struct sock_txtime))) {
1569 ret = -EFAULT;
1570 break;
1571 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1572 ret = -EINVAL;
1573 break;
1574 }
1575 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1576 * scheduler has enough safe guards.
1577 */
1578 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1579 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1580 ret = -EPERM;
1581 break;
1582 }
1583
1584 ret = sockopt_validate_clockid(sk_txtime.clockid);
1585 if (ret)
1586 break;
1587
1588 sock_valbool_flag(sk, SOCK_TXTIME, true);
1589 sk->sk_clockid = sk_txtime.clockid;
1590 sk->sk_txtime_deadline_mode =
1591 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1592 sk->sk_txtime_report_errors =
1593 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1594 break;
1595
1596 case SO_BINDTOIFINDEX:
1597 ret = sock_bindtoindex_locked(sk, val);
1598 break;
1599
1600 case SO_BUF_LOCK:
1601 if (val & ~SOCK_BUF_LOCK_MASK) {
1602 ret = -EINVAL;
1603 break;
1604 }
1605 sk->sk_userlocks = val | (sk->sk_userlocks &
1606 ~SOCK_BUF_LOCK_MASK);
1607 break;
1608
1609 case SO_RESERVE_MEM:
1610 {
1611 int delta;
1612
1613 if (val < 0) {
1614 ret = -EINVAL;
1615 break;
1616 }
1617
1618 delta = val - sk->sk_reserved_mem;
1619 if (delta < 0)
1620 sock_release_reserved_memory(sk, -delta);
1621 else
1622 ret = sock_reserve_memory(sk, delta);
1623 break;
1624 }
1625
1626 default:
1627 ret = -ENOPROTOOPT;
1628 break;
1629 }
1630 sockopt_release_sock(sk);
1631 return ret;
1632}
1633
1634int sock_setsockopt(struct socket *sock, int level, int optname,
1635 sockptr_t optval, unsigned int optlen)
1636{
1637 return sk_setsockopt(sock->sk, level, optname,
1638 optval, optlen);
1639}
1640EXPORT_SYMBOL(sock_setsockopt);
1641
1642static const struct cred *sk_get_peer_cred(struct sock *sk)
1643{
1644 const struct cred *cred;
1645
1646 spin_lock(&sk->sk_peer_lock);
1647 cred = get_cred(sk->sk_peer_cred);
1648 spin_unlock(&sk->sk_peer_lock);
1649
1650 return cred;
1651}
1652
1653static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1654 struct ucred *ucred)
1655{
1656 ucred->pid = pid_vnr(pid);
1657 ucred->uid = ucred->gid = -1;
1658 if (cred) {
1659 struct user_namespace *current_ns = current_user_ns();
1660
1661 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1662 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1663 }
1664}
1665
1666static int groups_to_user(sockptr_t dst, const struct group_info *src)
1667{
1668 struct user_namespace *user_ns = current_user_ns();
1669 int i;
1670
1671 for (i = 0; i < src->ngroups; i++) {
1672 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1673
1674 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1675 return -EFAULT;
1676 }
1677
1678 return 0;
1679}
1680
1681int sk_getsockopt(struct sock *sk, int level, int optname,
1682 sockptr_t optval, sockptr_t optlen)
1683{
1684 struct socket *sock = sk->sk_socket;
1685
1686 union {
1687 int val;
1688 u64 val64;
1689 unsigned long ulval;
1690 struct linger ling;
1691 struct old_timeval32 tm32;
1692 struct __kernel_old_timeval tm;
1693 struct __kernel_sock_timeval stm;
1694 struct sock_txtime txtime;
1695 struct so_timestamping timestamping;
1696 } v;
1697
1698 int lv = sizeof(int);
1699 int len;
1700
1701 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1702 return -EFAULT;
1703 if (len < 0)
1704 return -EINVAL;
1705
1706 memset(&v, 0, sizeof(v));
1707
1708 switch (optname) {
1709 case SO_DEBUG:
1710 v.val = sock_flag(sk, SOCK_DBG);
1711 break;
1712
1713 case SO_DONTROUTE:
1714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1715 break;
1716
1717 case SO_BROADCAST:
1718 v.val = sock_flag(sk, SOCK_BROADCAST);
1719 break;
1720
1721 case SO_SNDBUF:
1722 v.val = READ_ONCE(sk->sk_sndbuf);
1723 break;
1724
1725 case SO_RCVBUF:
1726 v.val = READ_ONCE(sk->sk_rcvbuf);
1727 break;
1728
1729 case SO_REUSEADDR:
1730 v.val = sk->sk_reuse;
1731 break;
1732
1733 case SO_REUSEPORT:
1734 v.val = sk->sk_reuseport;
1735 break;
1736
1737 case SO_KEEPALIVE:
1738 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1739 break;
1740
1741 case SO_TYPE:
1742 v.val = sk->sk_type;
1743 break;
1744
1745 case SO_PROTOCOL:
1746 v.val = sk->sk_protocol;
1747 break;
1748
1749 case SO_DOMAIN:
1750 v.val = sk->sk_family;
1751 break;
1752
1753 case SO_ERROR:
1754 v.val = -sock_error(sk);
1755 if (v.val == 0)
1756 v.val = xchg(&sk->sk_err_soft, 0);
1757 break;
1758
1759 case SO_OOBINLINE:
1760 v.val = sock_flag(sk, SOCK_URGINLINE);
1761 break;
1762
1763 case SO_NO_CHECK:
1764 v.val = sk->sk_no_check_tx;
1765 break;
1766
1767 case SO_PRIORITY:
1768 v.val = READ_ONCE(sk->sk_priority);
1769 break;
1770
1771 case SO_LINGER:
1772 lv = sizeof(v.ling);
1773 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1774 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1775 break;
1776
1777 case SO_BSDCOMPAT:
1778 break;
1779
1780 case SO_TIMESTAMP_OLD:
1781 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1782 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1783 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1784 break;
1785
1786 case SO_TIMESTAMPNS_OLD:
1787 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1788 break;
1789
1790 case SO_TIMESTAMP_NEW:
1791 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1792 break;
1793
1794 case SO_TIMESTAMPNS_NEW:
1795 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1796 break;
1797
1798 case SO_TIMESTAMPING_OLD:
1799 case SO_TIMESTAMPING_NEW:
1800 lv = sizeof(v.timestamping);
1801 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1802 * returning the flags when they were set through the same option.
1803 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1804 */
1805 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1806 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1807 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1808 }
1809 break;
1810
1811 case SO_RCVTIMEO_OLD:
1812 case SO_RCVTIMEO_NEW:
1813 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1814 SO_RCVTIMEO_OLD == optname);
1815 break;
1816
1817 case SO_SNDTIMEO_OLD:
1818 case SO_SNDTIMEO_NEW:
1819 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1820 SO_SNDTIMEO_OLD == optname);
1821 break;
1822
1823 case SO_RCVLOWAT:
1824 v.val = READ_ONCE(sk->sk_rcvlowat);
1825 break;
1826
1827 case SO_SNDLOWAT:
1828 v.val = 1;
1829 break;
1830
1831 case SO_PASSCRED:
1832 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1833 break;
1834
1835 case SO_PASSPIDFD:
1836 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1837 break;
1838
1839 case SO_PEERCRED:
1840 {
1841 struct ucred peercred;
1842 if (len > sizeof(peercred))
1843 len = sizeof(peercred);
1844
1845 spin_lock(&sk->sk_peer_lock);
1846 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1847 spin_unlock(&sk->sk_peer_lock);
1848
1849 if (copy_to_sockptr(optval, &peercred, len))
1850 return -EFAULT;
1851 goto lenout;
1852 }
1853
1854 case SO_PEERPIDFD:
1855 {
1856 struct pid *peer_pid;
1857 struct file *pidfd_file = NULL;
1858 int pidfd;
1859
1860 if (len > sizeof(pidfd))
1861 len = sizeof(pidfd);
1862
1863 spin_lock(&sk->sk_peer_lock);
1864 peer_pid = get_pid(sk->sk_peer_pid);
1865 spin_unlock(&sk->sk_peer_lock);
1866
1867 if (!peer_pid)
1868 return -ENODATA;
1869
1870 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1871 put_pid(peer_pid);
1872 if (pidfd < 0)
1873 return pidfd;
1874
1875 if (copy_to_sockptr(optval, &pidfd, len) ||
1876 copy_to_sockptr(optlen, &len, sizeof(int))) {
1877 put_unused_fd(pidfd);
1878 fput(pidfd_file);
1879
1880 return -EFAULT;
1881 }
1882
1883 fd_install(pidfd, pidfd_file);
1884 return 0;
1885 }
1886
1887 case SO_PEERGROUPS:
1888 {
1889 const struct cred *cred;
1890 int ret, n;
1891
1892 cred = sk_get_peer_cred(sk);
1893 if (!cred)
1894 return -ENODATA;
1895
1896 n = cred->group_info->ngroups;
1897 if (len < n * sizeof(gid_t)) {
1898 len = n * sizeof(gid_t);
1899 put_cred(cred);
1900 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1901 }
1902 len = n * sizeof(gid_t);
1903
1904 ret = groups_to_user(optval, cred->group_info);
1905 put_cred(cred);
1906 if (ret)
1907 return ret;
1908 goto lenout;
1909 }
1910
1911 case SO_PEERNAME:
1912 {
1913 struct sockaddr_storage address;
1914
1915 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1916 if (lv < 0)
1917 return -ENOTCONN;
1918 if (lv < len)
1919 return -EINVAL;
1920 if (copy_to_sockptr(optval, &address, len))
1921 return -EFAULT;
1922 goto lenout;
1923 }
1924
1925 /* Dubious BSD thing... Probably nobody even uses it, but
1926 * the UNIX standard wants it for whatever reason... -DaveM
1927 */
1928 case SO_ACCEPTCONN:
1929 v.val = sk->sk_state == TCP_LISTEN;
1930 break;
1931
1932 case SO_PASSSEC:
1933 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1934 break;
1935
1936 case SO_PEERSEC:
1937 return security_socket_getpeersec_stream(sock,
1938 optval, optlen, len);
1939
1940 case SO_MARK:
1941 v.val = READ_ONCE(sk->sk_mark);
1942 break;
1943
1944 case SO_RCVMARK:
1945 v.val = sock_flag(sk, SOCK_RCVMARK);
1946 break;
1947
1948 case SO_RXQ_OVFL:
1949 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1950 break;
1951
1952 case SO_WIFI_STATUS:
1953 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1954 break;
1955
1956 case SO_PEEK_OFF:
1957 if (!READ_ONCE(sock->ops)->set_peek_off)
1958 return -EOPNOTSUPP;
1959
1960 v.val = READ_ONCE(sk->sk_peek_off);
1961 break;
1962 case SO_NOFCS:
1963 v.val = sock_flag(sk, SOCK_NOFCS);
1964 break;
1965
1966 case SO_BINDTODEVICE:
1967 return sock_getbindtodevice(sk, optval, optlen, len);
1968
1969 case SO_GET_FILTER:
1970 len = sk_get_filter(sk, optval, len);
1971 if (len < 0)
1972 return len;
1973
1974 goto lenout;
1975
1976 case SO_LOCK_FILTER:
1977 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1978 break;
1979
1980 case SO_BPF_EXTENSIONS:
1981 v.val = bpf_tell_extensions();
1982 break;
1983
1984 case SO_SELECT_ERR_QUEUE:
1985 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1986 break;
1987
1988#ifdef CONFIG_NET_RX_BUSY_POLL
1989 case SO_BUSY_POLL:
1990 v.val = READ_ONCE(sk->sk_ll_usec);
1991 break;
1992 case SO_PREFER_BUSY_POLL:
1993 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1994 break;
1995#endif
1996
1997 case SO_MAX_PACING_RATE:
1998 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1999 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2000 lv = sizeof(v.ulval);
2001 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2002 } else {
2003 /* 32bit version */
2004 v.val = min_t(unsigned long, ~0U,
2005 READ_ONCE(sk->sk_max_pacing_rate));
2006 }
2007 break;
2008
2009 case SO_INCOMING_CPU:
2010 v.val = READ_ONCE(sk->sk_incoming_cpu);
2011 break;
2012
2013 case SO_MEMINFO:
2014 {
2015 u32 meminfo[SK_MEMINFO_VARS];
2016
2017 sk_get_meminfo(sk, meminfo);
2018
2019 len = min_t(unsigned int, len, sizeof(meminfo));
2020 if (copy_to_sockptr(optval, &meminfo, len))
2021 return -EFAULT;
2022
2023 goto lenout;
2024 }
2025
2026#ifdef CONFIG_NET_RX_BUSY_POLL
2027 case SO_INCOMING_NAPI_ID:
2028 v.val = READ_ONCE(sk->sk_napi_id);
2029
2030 /* aggregate non-NAPI IDs down to 0 */
2031 if (v.val < MIN_NAPI_ID)
2032 v.val = 0;
2033
2034 break;
2035#endif
2036
2037 case SO_COOKIE:
2038 lv = sizeof(u64);
2039 if (len < lv)
2040 return -EINVAL;
2041 v.val64 = sock_gen_cookie(sk);
2042 break;
2043
2044 case SO_ZEROCOPY:
2045 v.val = sock_flag(sk, SOCK_ZEROCOPY);
2046 break;
2047
2048 case SO_TXTIME:
2049 lv = sizeof(v.txtime);
2050 v.txtime.clockid = sk->sk_clockid;
2051 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2052 SOF_TXTIME_DEADLINE_MODE : 0;
2053 v.txtime.flags |= sk->sk_txtime_report_errors ?
2054 SOF_TXTIME_REPORT_ERRORS : 0;
2055 break;
2056
2057 case SO_BINDTOIFINDEX:
2058 v.val = READ_ONCE(sk->sk_bound_dev_if);
2059 break;
2060
2061 case SO_NETNS_COOKIE:
2062 lv = sizeof(u64);
2063 if (len != lv)
2064 return -EINVAL;
2065 v.val64 = sock_net(sk)->net_cookie;
2066 break;
2067
2068 case SO_BUF_LOCK:
2069 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2070 break;
2071
2072 case SO_RESERVE_MEM:
2073 v.val = READ_ONCE(sk->sk_reserved_mem);
2074 break;
2075
2076 case SO_TXREHASH:
2077 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2078 v.val = READ_ONCE(sk->sk_txrehash);
2079 break;
2080
2081 default:
2082 /* We implement the SO_SNDLOWAT etc to not be settable
2083 * (1003.1g 7).
2084 */
2085 return -ENOPROTOOPT;
2086 }
2087
2088 if (len > lv)
2089 len = lv;
2090 if (copy_to_sockptr(optval, &v, len))
2091 return -EFAULT;
2092lenout:
2093 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2094 return -EFAULT;
2095 return 0;
2096}
2097
2098/*
2099 * Initialize an sk_lock.
2100 *
2101 * (We also register the sk_lock with the lock validator.)
2102 */
2103static inline void sock_lock_init(struct sock *sk)
2104{
2105 if (sk->sk_kern_sock)
2106 sock_lock_init_class_and_name(
2107 sk,
2108 af_family_kern_slock_key_strings[sk->sk_family],
2109 af_family_kern_slock_keys + sk->sk_family,
2110 af_family_kern_key_strings[sk->sk_family],
2111 af_family_kern_keys + sk->sk_family);
2112 else
2113 sock_lock_init_class_and_name(
2114 sk,
2115 af_family_slock_key_strings[sk->sk_family],
2116 af_family_slock_keys + sk->sk_family,
2117 af_family_key_strings[sk->sk_family],
2118 af_family_keys + sk->sk_family);
2119}
2120
2121/*
2122 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2123 * even temporarily, because of RCU lookups. sk_node should also be left as is.
2124 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2125 */
2126static void sock_copy(struct sock *nsk, const struct sock *osk)
2127{
2128 const struct proto *prot = READ_ONCE(osk->sk_prot);
2129#ifdef CONFIG_SECURITY_NETWORK
2130 void *sptr = nsk->sk_security;
2131#endif
2132
2133 /* If we move sk_tx_queue_mapping out of the private section,
2134 * we must check if sk_tx_queue_clear() is called after
2135 * sock_copy() in sk_clone_lock().
2136 */
2137 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2138 offsetof(struct sock, sk_dontcopy_begin) ||
2139 offsetof(struct sock, sk_tx_queue_mapping) >=
2140 offsetof(struct sock, sk_dontcopy_end));
2141
2142 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2143
2144 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2145 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2146 /* alloc is larger than struct, see sk_prot_alloc() */);
2147
2148#ifdef CONFIG_SECURITY_NETWORK
2149 nsk->sk_security = sptr;
2150 security_sk_clone(osk, nsk);
2151#endif
2152}
2153
2154static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2155 int family)
2156{
2157 struct sock *sk;
2158 struct kmem_cache *slab;
2159
2160 slab = prot->slab;
2161 if (slab != NULL) {
2162 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2163 if (!sk)
2164 return sk;
2165 if (want_init_on_alloc(priority))
2166 sk_prot_clear_nulls(sk, prot->obj_size);
2167 } else
2168 sk = kmalloc(prot->obj_size, priority);
2169
2170 if (sk != NULL) {
2171 if (security_sk_alloc(sk, family, priority))
2172 goto out_free;
2173
2174 if (!try_module_get(prot->owner))
2175 goto out_free_sec;
2176 }
2177
2178 return sk;
2179
2180out_free_sec:
2181 security_sk_free(sk);
2182out_free:
2183 if (slab != NULL)
2184 kmem_cache_free(slab, sk);
2185 else
2186 kfree(sk);
2187 return NULL;
2188}
2189
2190static void sk_prot_free(struct proto *prot, struct sock *sk)
2191{
2192 struct kmem_cache *slab;
2193 struct module *owner;
2194
2195 owner = prot->owner;
2196 slab = prot->slab;
2197
2198 cgroup_sk_free(&sk->sk_cgrp_data);
2199 mem_cgroup_sk_free(sk);
2200 security_sk_free(sk);
2201 if (slab != NULL)
2202 kmem_cache_free(slab, sk);
2203 else
2204 kfree(sk);
2205 module_put(owner);
2206}
2207
2208/**
2209 * sk_alloc - All socket objects are allocated here
2210 * @net: the applicable net namespace
2211 * @family: protocol family
2212 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2213 * @prot: struct proto associated with this new sock instance
2214 * @kern: is this to be a kernel socket?
2215 */
2216struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2217 struct proto *prot, int kern)
2218{
2219 struct sock *sk;
2220
2221 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2222 if (sk) {
2223 sk->sk_family = family;
2224 /*
2225 * See comment in struct sock definition to understand
2226 * why we need sk_prot_creator -acme
2227 */
2228 sk->sk_prot = sk->sk_prot_creator = prot;
2229 sk->sk_kern_sock = kern;
2230 sock_lock_init(sk);
2231 sk->sk_net_refcnt = kern ? 0 : 1;
2232 if (likely(sk->sk_net_refcnt)) {
2233 get_net_track(net, &sk->ns_tracker, priority);
2234 sock_inuse_add(net, 1);
2235 } else {
2236 net_passive_inc(net);
2237 __netns_tracker_alloc(net, &sk->ns_tracker,
2238 false, priority);
2239 }
2240
2241 sock_net_set(sk, net);
2242 refcount_set(&sk->sk_wmem_alloc, 1);
2243
2244 mem_cgroup_sk_alloc(sk);
2245 cgroup_sk_alloc(&sk->sk_cgrp_data);
2246 sock_update_classid(&sk->sk_cgrp_data);
2247 sock_update_netprioidx(&sk->sk_cgrp_data);
2248 sk_tx_queue_clear(sk);
2249 }
2250
2251 return sk;
2252}
2253EXPORT_SYMBOL(sk_alloc);
2254
2255/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2256 * grace period. This is the case for UDP sockets and TCP listeners.
2257 */
2258static void __sk_destruct(struct rcu_head *head)
2259{
2260 struct sock *sk = container_of(head, struct sock, sk_rcu);
2261 struct net *net = sock_net(sk);
2262 struct sk_filter *filter;
2263
2264 if (sk->sk_destruct)
2265 sk->sk_destruct(sk);
2266
2267 filter = rcu_dereference_check(sk->sk_filter,
2268 refcount_read(&sk->sk_wmem_alloc) == 0);
2269 if (filter) {
2270 sk_filter_uncharge(sk, filter);
2271 RCU_INIT_POINTER(sk->sk_filter, NULL);
2272 }
2273
2274 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2275
2276#ifdef CONFIG_BPF_SYSCALL
2277 bpf_sk_storage_free(sk);
2278#endif
2279
2280 if (atomic_read(&sk->sk_omem_alloc))
2281 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2282 __func__, atomic_read(&sk->sk_omem_alloc));
2283
2284 if (sk->sk_frag.page) {
2285 put_page(sk->sk_frag.page);
2286 sk->sk_frag.page = NULL;
2287 }
2288
2289 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2290 put_cred(sk->sk_peer_cred);
2291 put_pid(sk->sk_peer_pid);
2292
2293 if (likely(sk->sk_net_refcnt)) {
2294 put_net_track(net, &sk->ns_tracker);
2295 } else {
2296 __netns_tracker_free(net, &sk->ns_tracker, false);
2297 net_passive_dec(net);
2298 }
2299 sk_prot_free(sk->sk_prot_creator, sk);
2300}
2301
2302void sk_net_refcnt_upgrade(struct sock *sk)
2303{
2304 struct net *net = sock_net(sk);
2305
2306 WARN_ON_ONCE(sk->sk_net_refcnt);
2307 __netns_tracker_free(net, &sk->ns_tracker, false);
2308 net_passive_dec(net);
2309 sk->sk_net_refcnt = 1;
2310 get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
2311 sock_inuse_add(net, 1);
2312}
2313EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade);
2314
2315void sk_destruct(struct sock *sk)
2316{
2317 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2318
2319 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2320 reuseport_detach_sock(sk);
2321 use_call_rcu = true;
2322 }
2323
2324 if (use_call_rcu)
2325 call_rcu(&sk->sk_rcu, __sk_destruct);
2326 else
2327 __sk_destruct(&sk->sk_rcu);
2328}
2329
2330static void __sk_free(struct sock *sk)
2331{
2332 if (likely(sk->sk_net_refcnt))
2333 sock_inuse_add(sock_net(sk), -1);
2334
2335 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2336 sock_diag_broadcast_destroy(sk);
2337 else
2338 sk_destruct(sk);
2339}
2340
2341void sk_free(struct sock *sk)
2342{
2343 /*
2344 * We subtract one from sk_wmem_alloc and can know if
2345 * some packets are still in some tx queue.
2346 * If not null, sock_wfree() will call __sk_free(sk) later
2347 */
2348 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2349 __sk_free(sk);
2350}
2351EXPORT_SYMBOL(sk_free);
2352
2353static void sk_init_common(struct sock *sk)
2354{
2355 skb_queue_head_init(&sk->sk_receive_queue);
2356 skb_queue_head_init(&sk->sk_write_queue);
2357 skb_queue_head_init(&sk->sk_error_queue);
2358
2359 rwlock_init(&sk->sk_callback_lock);
2360 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2361 af_rlock_keys + sk->sk_family,
2362 af_family_rlock_key_strings[sk->sk_family]);
2363 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2364 af_wlock_keys + sk->sk_family,
2365 af_family_wlock_key_strings[sk->sk_family]);
2366 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2367 af_elock_keys + sk->sk_family,
2368 af_family_elock_key_strings[sk->sk_family]);
2369 if (sk->sk_kern_sock)
2370 lockdep_set_class_and_name(&sk->sk_callback_lock,
2371 af_kern_callback_keys + sk->sk_family,
2372 af_family_kern_clock_key_strings[sk->sk_family]);
2373 else
2374 lockdep_set_class_and_name(&sk->sk_callback_lock,
2375 af_callback_keys + sk->sk_family,
2376 af_family_clock_key_strings[sk->sk_family]);
2377}
2378
2379/**
2380 * sk_clone_lock - clone a socket, and lock its clone
2381 * @sk: the socket to clone
2382 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2383 *
2384 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2385 */
2386struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2387{
2388 struct proto *prot = READ_ONCE(sk->sk_prot);
2389 struct sk_filter *filter;
2390 bool is_charged = true;
2391 struct sock *newsk;
2392
2393 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2394 if (!newsk)
2395 goto out;
2396
2397 sock_copy(newsk, sk);
2398
2399 newsk->sk_prot_creator = prot;
2400
2401 /* SANITY */
2402 if (likely(newsk->sk_net_refcnt)) {
2403 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2404 sock_inuse_add(sock_net(newsk), 1);
2405 } else {
2406 /* Kernel sockets are not elevating the struct net refcount.
2407 * Instead, use a tracker to more easily detect if a layer
2408 * is not properly dismantling its kernel sockets at netns
2409 * destroy time.
2410 */
2411 net_passive_inc(sock_net(newsk));
2412 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2413 false, priority);
2414 }
2415 sk_node_init(&newsk->sk_node);
2416 sock_lock_init(newsk);
2417 bh_lock_sock(newsk);
2418 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2419 newsk->sk_backlog.len = 0;
2420
2421 atomic_set(&newsk->sk_rmem_alloc, 0);
2422
2423 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2424 refcount_set(&newsk->sk_wmem_alloc, 1);
2425
2426 atomic_set(&newsk->sk_omem_alloc, 0);
2427 sk_init_common(newsk);
2428
2429 newsk->sk_dst_cache = NULL;
2430 newsk->sk_dst_pending_confirm = 0;
2431 newsk->sk_wmem_queued = 0;
2432 newsk->sk_forward_alloc = 0;
2433 newsk->sk_reserved_mem = 0;
2434 atomic_set(&newsk->sk_drops, 0);
2435 newsk->sk_send_head = NULL;
2436 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2437 atomic_set(&newsk->sk_zckey, 0);
2438
2439 sock_reset_flag(newsk, SOCK_DONE);
2440
2441 /* sk->sk_memcg will be populated at accept() time */
2442 newsk->sk_memcg = NULL;
2443
2444 cgroup_sk_clone(&newsk->sk_cgrp_data);
2445
2446 rcu_read_lock();
2447 filter = rcu_dereference(sk->sk_filter);
2448 if (filter != NULL)
2449 /* though it's an empty new sock, the charging may fail
2450 * if sysctl_optmem_max was changed between creation of
2451 * original socket and cloning
2452 */
2453 is_charged = sk_filter_charge(newsk, filter);
2454 RCU_INIT_POINTER(newsk->sk_filter, filter);
2455 rcu_read_unlock();
2456
2457 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2458 /* We need to make sure that we don't uncharge the new
2459 * socket if we couldn't charge it in the first place
2460 * as otherwise we uncharge the parent's filter.
2461 */
2462 if (!is_charged)
2463 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2464 sk_free_unlock_clone(newsk);
2465 newsk = NULL;
2466 goto out;
2467 }
2468 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2469
2470 if (bpf_sk_storage_clone(sk, newsk)) {
2471 sk_free_unlock_clone(newsk);
2472 newsk = NULL;
2473 goto out;
2474 }
2475
2476 /* Clear sk_user_data if parent had the pointer tagged
2477 * as not suitable for copying when cloning.
2478 */
2479 if (sk_user_data_is_nocopy(newsk))
2480 newsk->sk_user_data = NULL;
2481
2482 newsk->sk_err = 0;
2483 newsk->sk_err_soft = 0;
2484 newsk->sk_priority = 0;
2485 newsk->sk_incoming_cpu = raw_smp_processor_id();
2486
2487 /* Before updating sk_refcnt, we must commit prior changes to memory
2488 * (Documentation/RCU/rculist_nulls.rst for details)
2489 */
2490 smp_wmb();
2491 refcount_set(&newsk->sk_refcnt, 2);
2492
2493 sk_set_socket(newsk, NULL);
2494 sk_tx_queue_clear(newsk);
2495 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2496
2497 if (newsk->sk_prot->sockets_allocated)
2498 sk_sockets_allocated_inc(newsk);
2499
2500 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2501 net_enable_timestamp();
2502out:
2503 return newsk;
2504}
2505EXPORT_SYMBOL_GPL(sk_clone_lock);
2506
2507void sk_free_unlock_clone(struct sock *sk)
2508{
2509 /* It is still raw copy of parent, so invalidate
2510 * destructor and make plain sk_free() */
2511 sk->sk_destruct = NULL;
2512 bh_unlock_sock(sk);
2513 sk_free(sk);
2514}
2515EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2516
2517static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2518{
2519 bool is_ipv6 = false;
2520 u32 max_size;
2521
2522#if IS_ENABLED(CONFIG_IPV6)
2523 is_ipv6 = (sk->sk_family == AF_INET6 &&
2524 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2525#endif
2526 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2527 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2528 READ_ONCE(dst->dev->gso_ipv4_max_size);
2529 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2530 max_size = GSO_LEGACY_MAX_SIZE;
2531
2532 return max_size - (MAX_TCP_HEADER + 1);
2533}
2534
2535void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2536{
2537 u32 max_segs = 1;
2538
2539 sk->sk_route_caps = dst->dev->features;
2540 if (sk_is_tcp(sk))
2541 sk->sk_route_caps |= NETIF_F_GSO;
2542 if (sk->sk_route_caps & NETIF_F_GSO)
2543 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2544 if (unlikely(sk->sk_gso_disabled))
2545 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2546 if (sk_can_gso(sk)) {
2547 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2548 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2549 } else {
2550 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2551 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2552 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2553 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2554 }
2555 }
2556 sk->sk_gso_max_segs = max_segs;
2557 sk_dst_set(sk, dst);
2558}
2559EXPORT_SYMBOL_GPL(sk_setup_caps);
2560
2561/*
2562 * Simple resource managers for sockets.
2563 */
2564
2565
2566/*
2567 * Write buffer destructor automatically called from kfree_skb.
2568 */
2569void sock_wfree(struct sk_buff *skb)
2570{
2571 struct sock *sk = skb->sk;
2572 unsigned int len = skb->truesize;
2573 bool free;
2574
2575 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2576 if (sock_flag(sk, SOCK_RCU_FREE) &&
2577 sk->sk_write_space == sock_def_write_space) {
2578 rcu_read_lock();
2579 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2580 sock_def_write_space_wfree(sk);
2581 rcu_read_unlock();
2582 if (unlikely(free))
2583 __sk_free(sk);
2584 return;
2585 }
2586
2587 /*
2588 * Keep a reference on sk_wmem_alloc, this will be released
2589 * after sk_write_space() call
2590 */
2591 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2592 sk->sk_write_space(sk);
2593 len = 1;
2594 }
2595 /*
2596 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2597 * could not do because of in-flight packets
2598 */
2599 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2600 __sk_free(sk);
2601}
2602EXPORT_SYMBOL(sock_wfree);
2603
2604/* This variant of sock_wfree() is used by TCP,
2605 * since it sets SOCK_USE_WRITE_QUEUE.
2606 */
2607void __sock_wfree(struct sk_buff *skb)
2608{
2609 struct sock *sk = skb->sk;
2610
2611 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2612 __sk_free(sk);
2613}
2614
2615void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2616{
2617 skb_orphan(skb);
2618#ifdef CONFIG_INET
2619 if (unlikely(!sk_fullsock(sk)))
2620 return skb_set_owner_edemux(skb, sk);
2621#endif
2622 skb->sk = sk;
2623 skb->destructor = sock_wfree;
2624 skb_set_hash_from_sk(skb, sk);
2625 /*
2626 * We used to take a refcount on sk, but following operation
2627 * is enough to guarantee sk_free() won't free this sock until
2628 * all in-flight packets are completed
2629 */
2630 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2631}
2632EXPORT_SYMBOL(skb_set_owner_w);
2633
2634static bool can_skb_orphan_partial(const struct sk_buff *skb)
2635{
2636 /* Drivers depend on in-order delivery for crypto offload,
2637 * partial orphan breaks out-of-order-OK logic.
2638 */
2639 if (skb_is_decrypted(skb))
2640 return false;
2641
2642 return (skb->destructor == sock_wfree ||
2643 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2644}
2645
2646/* This helper is used by netem, as it can hold packets in its
2647 * delay queue. We want to allow the owner socket to send more
2648 * packets, as if they were already TX completed by a typical driver.
2649 * But we also want to keep skb->sk set because some packet schedulers
2650 * rely on it (sch_fq for example).
2651 */
2652void skb_orphan_partial(struct sk_buff *skb)
2653{
2654 if (skb_is_tcp_pure_ack(skb))
2655 return;
2656
2657 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2658 return;
2659
2660 skb_orphan(skb);
2661}
2662EXPORT_SYMBOL(skb_orphan_partial);
2663
2664/*
2665 * Read buffer destructor automatically called from kfree_skb.
2666 */
2667void sock_rfree(struct sk_buff *skb)
2668{
2669 struct sock *sk = skb->sk;
2670 unsigned int len = skb->truesize;
2671
2672 atomic_sub(len, &sk->sk_rmem_alloc);
2673 sk_mem_uncharge(sk, len);
2674}
2675EXPORT_SYMBOL(sock_rfree);
2676
2677/*
2678 * Buffer destructor for skbs that are not used directly in read or write
2679 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2680 */
2681void sock_efree(struct sk_buff *skb)
2682{
2683 sock_put(skb->sk);
2684}
2685EXPORT_SYMBOL(sock_efree);
2686
2687/* Buffer destructor for prefetch/receive path where reference count may
2688 * not be held, e.g. for listen sockets.
2689 */
2690#ifdef CONFIG_INET
2691void sock_pfree(struct sk_buff *skb)
2692{
2693 struct sock *sk = skb->sk;
2694
2695 if (!sk_is_refcounted(sk))
2696 return;
2697
2698 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2699 inet_reqsk(sk)->rsk_listener = NULL;
2700 reqsk_free(inet_reqsk(sk));
2701 return;
2702 }
2703
2704 sock_gen_put(sk);
2705}
2706EXPORT_SYMBOL(sock_pfree);
2707#endif /* CONFIG_INET */
2708
2709kuid_t sock_i_uid(struct sock *sk)
2710{
2711 kuid_t uid;
2712
2713 read_lock_bh(&sk->sk_callback_lock);
2714 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2715 read_unlock_bh(&sk->sk_callback_lock);
2716 return uid;
2717}
2718EXPORT_SYMBOL(sock_i_uid);
2719
2720unsigned long __sock_i_ino(struct sock *sk)
2721{
2722 unsigned long ino;
2723
2724 read_lock(&sk->sk_callback_lock);
2725 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2726 read_unlock(&sk->sk_callback_lock);
2727 return ino;
2728}
2729EXPORT_SYMBOL(__sock_i_ino);
2730
2731unsigned long sock_i_ino(struct sock *sk)
2732{
2733 unsigned long ino;
2734
2735 local_bh_disable();
2736 ino = __sock_i_ino(sk);
2737 local_bh_enable();
2738 return ino;
2739}
2740EXPORT_SYMBOL(sock_i_ino);
2741
2742/*
2743 * Allocate a skb from the socket's send buffer.
2744 */
2745struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2746 gfp_t priority)
2747{
2748 if (force ||
2749 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2750 struct sk_buff *skb = alloc_skb(size, priority);
2751
2752 if (skb) {
2753 skb_set_owner_w(skb, sk);
2754 return skb;
2755 }
2756 }
2757 return NULL;
2758}
2759EXPORT_SYMBOL(sock_wmalloc);
2760
2761static void sock_ofree(struct sk_buff *skb)
2762{
2763 struct sock *sk = skb->sk;
2764
2765 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2766}
2767
2768struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2769 gfp_t priority)
2770{
2771 struct sk_buff *skb;
2772
2773 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2774 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2775 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2776 return NULL;
2777
2778 skb = alloc_skb(size, priority);
2779 if (!skb)
2780 return NULL;
2781
2782 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2783 skb->sk = sk;
2784 skb->destructor = sock_ofree;
2785 return skb;
2786}
2787
2788/*
2789 * Allocate a memory block from the socket's option memory buffer.
2790 */
2791void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2792{
2793 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2794
2795 if ((unsigned int)size <= optmem_max &&
2796 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2797 void *mem;
2798 /* First do the add, to avoid the race if kmalloc
2799 * might sleep.
2800 */
2801 atomic_add(size, &sk->sk_omem_alloc);
2802 mem = kmalloc(size, priority);
2803 if (mem)
2804 return mem;
2805 atomic_sub(size, &sk->sk_omem_alloc);
2806 }
2807 return NULL;
2808}
2809EXPORT_SYMBOL(sock_kmalloc);
2810
2811/* Free an option memory block. Note, we actually want the inline
2812 * here as this allows gcc to detect the nullify and fold away the
2813 * condition entirely.
2814 */
2815static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2816 const bool nullify)
2817{
2818 if (WARN_ON_ONCE(!mem))
2819 return;
2820 if (nullify)
2821 kfree_sensitive(mem);
2822 else
2823 kfree(mem);
2824 atomic_sub(size, &sk->sk_omem_alloc);
2825}
2826
2827void sock_kfree_s(struct sock *sk, void *mem, int size)
2828{
2829 __sock_kfree_s(sk, mem, size, false);
2830}
2831EXPORT_SYMBOL(sock_kfree_s);
2832
2833void sock_kzfree_s(struct sock *sk, void *mem, int size)
2834{
2835 __sock_kfree_s(sk, mem, size, true);
2836}
2837EXPORT_SYMBOL(sock_kzfree_s);
2838
2839/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2840 I think, these locks should be removed for datagram sockets.
2841 */
2842static long sock_wait_for_wmem(struct sock *sk, long timeo)
2843{
2844 DEFINE_WAIT(wait);
2845
2846 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2847 for (;;) {
2848 if (!timeo)
2849 break;
2850 if (signal_pending(current))
2851 break;
2852 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2853 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2854 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2855 break;
2856 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2857 break;
2858 if (READ_ONCE(sk->sk_err))
2859 break;
2860 timeo = schedule_timeout(timeo);
2861 }
2862 finish_wait(sk_sleep(sk), &wait);
2863 return timeo;
2864}
2865
2866
2867/*
2868 * Generic send/receive buffer handlers
2869 */
2870
2871struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2872 unsigned long data_len, int noblock,
2873 int *errcode, int max_page_order)
2874{
2875 struct sk_buff *skb;
2876 long timeo;
2877 int err;
2878
2879 timeo = sock_sndtimeo(sk, noblock);
2880 for (;;) {
2881 err = sock_error(sk);
2882 if (err != 0)
2883 goto failure;
2884
2885 err = -EPIPE;
2886 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2887 goto failure;
2888
2889 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2890 break;
2891
2892 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2893 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2894 err = -EAGAIN;
2895 if (!timeo)
2896 goto failure;
2897 if (signal_pending(current))
2898 goto interrupted;
2899 timeo = sock_wait_for_wmem(sk, timeo);
2900 }
2901 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2902 errcode, sk->sk_allocation);
2903 if (skb)
2904 skb_set_owner_w(skb, sk);
2905 return skb;
2906
2907interrupted:
2908 err = sock_intr_errno(timeo);
2909failure:
2910 *errcode = err;
2911 return NULL;
2912}
2913EXPORT_SYMBOL(sock_alloc_send_pskb);
2914
2915int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2916 struct sockcm_cookie *sockc)
2917{
2918 u32 tsflags;
2919
2920 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
2921
2922 switch (cmsg->cmsg_type) {
2923 case SO_MARK:
2924 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2925 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2926 return -EPERM;
2927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2928 return -EINVAL;
2929 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2930 break;
2931 case SO_TIMESTAMPING_OLD:
2932 case SO_TIMESTAMPING_NEW:
2933 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2934 return -EINVAL;
2935
2936 tsflags = *(u32 *)CMSG_DATA(cmsg);
2937 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2938 return -EINVAL;
2939
2940 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2941 sockc->tsflags |= tsflags;
2942 break;
2943 case SCM_TXTIME:
2944 if (!sock_flag(sk, SOCK_TXTIME))
2945 return -EINVAL;
2946 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2947 return -EINVAL;
2948 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2949 break;
2950 case SCM_TS_OPT_ID:
2951 if (sk_is_tcp(sk))
2952 return -EINVAL;
2953 tsflags = READ_ONCE(sk->sk_tsflags);
2954 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
2955 return -EINVAL;
2956 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2957 return -EINVAL;
2958 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
2959 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
2960 break;
2961 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2962 case SCM_RIGHTS:
2963 case SCM_CREDENTIALS:
2964 break;
2965 default:
2966 return -EINVAL;
2967 }
2968 return 0;
2969}
2970EXPORT_SYMBOL(__sock_cmsg_send);
2971
2972int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2973 struct sockcm_cookie *sockc)
2974{
2975 struct cmsghdr *cmsg;
2976 int ret;
2977
2978 for_each_cmsghdr(cmsg, msg) {
2979 if (!CMSG_OK(msg, cmsg))
2980 return -EINVAL;
2981 if (cmsg->cmsg_level != SOL_SOCKET)
2982 continue;
2983 ret = __sock_cmsg_send(sk, cmsg, sockc);
2984 if (ret)
2985 return ret;
2986 }
2987 return 0;
2988}
2989EXPORT_SYMBOL(sock_cmsg_send);
2990
2991static void sk_enter_memory_pressure(struct sock *sk)
2992{
2993 if (!sk->sk_prot->enter_memory_pressure)
2994 return;
2995
2996 sk->sk_prot->enter_memory_pressure(sk);
2997}
2998
2999static void sk_leave_memory_pressure(struct sock *sk)
3000{
3001 if (sk->sk_prot->leave_memory_pressure) {
3002 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3003 tcp_leave_memory_pressure, sk);
3004 } else {
3005 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3006
3007 if (memory_pressure && READ_ONCE(*memory_pressure))
3008 WRITE_ONCE(*memory_pressure, 0);
3009 }
3010}
3011
3012DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3013
3014/**
3015 * skb_page_frag_refill - check that a page_frag contains enough room
3016 * @sz: minimum size of the fragment we want to get
3017 * @pfrag: pointer to page_frag
3018 * @gfp: priority for memory allocation
3019 *
3020 * Note: While this allocator tries to use high order pages, there is
3021 * no guarantee that allocations succeed. Therefore, @sz MUST be
3022 * less or equal than PAGE_SIZE.
3023 */
3024bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3025{
3026 if (pfrag->page) {
3027 if (page_ref_count(pfrag->page) == 1) {
3028 pfrag->offset = 0;
3029 return true;
3030 }
3031 if (pfrag->offset + sz <= pfrag->size)
3032 return true;
3033 put_page(pfrag->page);
3034 }
3035
3036 pfrag->offset = 0;
3037 if (SKB_FRAG_PAGE_ORDER &&
3038 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3039 /* Avoid direct reclaim but allow kswapd to wake */
3040 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3041 __GFP_COMP | __GFP_NOWARN |
3042 __GFP_NORETRY,
3043 SKB_FRAG_PAGE_ORDER);
3044 if (likely(pfrag->page)) {
3045 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3046 return true;
3047 }
3048 }
3049 pfrag->page = alloc_page(gfp);
3050 if (likely(pfrag->page)) {
3051 pfrag->size = PAGE_SIZE;
3052 return true;
3053 }
3054 return false;
3055}
3056EXPORT_SYMBOL(skb_page_frag_refill);
3057
3058bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3059{
3060 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3061 return true;
3062
3063 sk_enter_memory_pressure(sk);
3064 sk_stream_moderate_sndbuf(sk);
3065 return false;
3066}
3067EXPORT_SYMBOL(sk_page_frag_refill);
3068
3069void __lock_sock(struct sock *sk)
3070 __releases(&sk->sk_lock.slock)
3071 __acquires(&sk->sk_lock.slock)
3072{
3073 DEFINE_WAIT(wait);
3074
3075 for (;;) {
3076 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3077 TASK_UNINTERRUPTIBLE);
3078 spin_unlock_bh(&sk->sk_lock.slock);
3079 schedule();
3080 spin_lock_bh(&sk->sk_lock.slock);
3081 if (!sock_owned_by_user(sk))
3082 break;
3083 }
3084 finish_wait(&sk->sk_lock.wq, &wait);
3085}
3086
3087void __release_sock(struct sock *sk)
3088 __releases(&sk->sk_lock.slock)
3089 __acquires(&sk->sk_lock.slock)
3090{
3091 struct sk_buff *skb, *next;
3092
3093 while ((skb = sk->sk_backlog.head) != NULL) {
3094 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3095
3096 spin_unlock_bh(&sk->sk_lock.slock);
3097
3098 do {
3099 next = skb->next;
3100 prefetch(next);
3101 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3102 skb_mark_not_on_list(skb);
3103 sk_backlog_rcv(sk, skb);
3104
3105 cond_resched();
3106
3107 skb = next;
3108 } while (skb != NULL);
3109
3110 spin_lock_bh(&sk->sk_lock.slock);
3111 }
3112
3113 /*
3114 * Doing the zeroing here guarantee we can not loop forever
3115 * while a wild producer attempts to flood us.
3116 */
3117 sk->sk_backlog.len = 0;
3118}
3119
3120void __sk_flush_backlog(struct sock *sk)
3121{
3122 spin_lock_bh(&sk->sk_lock.slock);
3123 __release_sock(sk);
3124
3125 if (sk->sk_prot->release_cb)
3126 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3127 tcp_release_cb, sk);
3128
3129 spin_unlock_bh(&sk->sk_lock.slock);
3130}
3131EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3132
3133/**
3134 * sk_wait_data - wait for data to arrive at sk_receive_queue
3135 * @sk: sock to wait on
3136 * @timeo: for how long
3137 * @skb: last skb seen on sk_receive_queue
3138 *
3139 * Now socket state including sk->sk_err is changed only under lock,
3140 * hence we may omit checks after joining wait queue.
3141 * We check receive queue before schedule() only as optimization;
3142 * it is very likely that release_sock() added new data.
3143 */
3144int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3145{
3146 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3147 int rc;
3148
3149 add_wait_queue(sk_sleep(sk), &wait);
3150 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3151 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3152 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3153 remove_wait_queue(sk_sleep(sk), &wait);
3154 return rc;
3155}
3156EXPORT_SYMBOL(sk_wait_data);
3157
3158/**
3159 * __sk_mem_raise_allocated - increase memory_allocated
3160 * @sk: socket
3161 * @size: memory size to allocate
3162 * @amt: pages to allocate
3163 * @kind: allocation type
3164 *
3165 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3166 *
3167 * Unlike the globally shared limits among the sockets under same protocol,
3168 * consuming the budget of a memcg won't have direct effect on other ones.
3169 * So be optimistic about memcg's tolerance, and leave the callers to decide
3170 * whether or not to raise allocated through sk_under_memory_pressure() or
3171 * its variants.
3172 */
3173int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3174{
3175 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3176 struct proto *prot = sk->sk_prot;
3177 bool charged = false;
3178 long allocated;
3179
3180 sk_memory_allocated_add(sk, amt);
3181 allocated = sk_memory_allocated(sk);
3182
3183 if (memcg) {
3184 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3185 goto suppress_allocation;
3186 charged = true;
3187 }
3188
3189 /* Under limit. */
3190 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3191 sk_leave_memory_pressure(sk);
3192 return 1;
3193 }
3194
3195 /* Under pressure. */
3196 if (allocated > sk_prot_mem_limits(sk, 1))
3197 sk_enter_memory_pressure(sk);
3198
3199 /* Over hard limit. */
3200 if (allocated > sk_prot_mem_limits(sk, 2))
3201 goto suppress_allocation;
3202
3203 /* Guarantee minimum buffer size under pressure (either global
3204 * or memcg) to make sure features described in RFC 7323 (TCP
3205 * Extensions for High Performance) work properly.
3206 *
3207 * This rule does NOT stand when exceeds global or memcg's hard
3208 * limit, or else a DoS attack can be taken place by spawning
3209 * lots of sockets whose usage are under minimum buffer size.
3210 */
3211 if (kind == SK_MEM_RECV) {
3212 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3213 return 1;
3214
3215 } else { /* SK_MEM_SEND */
3216 int wmem0 = sk_get_wmem0(sk, prot);
3217
3218 if (sk->sk_type == SOCK_STREAM) {
3219 if (sk->sk_wmem_queued < wmem0)
3220 return 1;
3221 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3222 return 1;
3223 }
3224 }
3225
3226 if (sk_has_memory_pressure(sk)) {
3227 u64 alloc;
3228
3229 /* The following 'average' heuristic is within the
3230 * scope of global accounting, so it only makes
3231 * sense for global memory pressure.
3232 */
3233 if (!sk_under_global_memory_pressure(sk))
3234 return 1;
3235
3236 /* Try to be fair among all the sockets under global
3237 * pressure by allowing the ones that below average
3238 * usage to raise.
3239 */
3240 alloc = sk_sockets_allocated_read_positive(sk);
3241 if (sk_prot_mem_limits(sk, 2) > alloc *
3242 sk_mem_pages(sk->sk_wmem_queued +
3243 atomic_read(&sk->sk_rmem_alloc) +
3244 sk->sk_forward_alloc))
3245 return 1;
3246 }
3247
3248suppress_allocation:
3249
3250 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3251 sk_stream_moderate_sndbuf(sk);
3252
3253 /* Fail only if socket is _under_ its sndbuf.
3254 * In this case we cannot block, so that we have to fail.
3255 */
3256 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3257 /* Force charge with __GFP_NOFAIL */
3258 if (memcg && !charged) {
3259 mem_cgroup_charge_skmem(memcg, amt,
3260 gfp_memcg_charge() | __GFP_NOFAIL);
3261 }
3262 return 1;
3263 }
3264 }
3265
3266 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3267 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3268
3269 sk_memory_allocated_sub(sk, amt);
3270
3271 if (charged)
3272 mem_cgroup_uncharge_skmem(memcg, amt);
3273
3274 return 0;
3275}
3276
3277/**
3278 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3279 * @sk: socket
3280 * @size: memory size to allocate
3281 * @kind: allocation type
3282 *
3283 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3284 * rmem allocation. This function assumes that protocols which have
3285 * memory_pressure use sk_wmem_queued as write buffer accounting.
3286 */
3287int __sk_mem_schedule(struct sock *sk, int size, int kind)
3288{
3289 int ret, amt = sk_mem_pages(size);
3290
3291 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3292 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3293 if (!ret)
3294 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3295 return ret;
3296}
3297EXPORT_SYMBOL(__sk_mem_schedule);
3298
3299/**
3300 * __sk_mem_reduce_allocated - reclaim memory_allocated
3301 * @sk: socket
3302 * @amount: number of quanta
3303 *
3304 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3305 */
3306void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3307{
3308 sk_memory_allocated_sub(sk, amount);
3309
3310 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3311 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3312
3313 if (sk_under_global_memory_pressure(sk) &&
3314 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3315 sk_leave_memory_pressure(sk);
3316}
3317
3318/**
3319 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3320 * @sk: socket
3321 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3322 */
3323void __sk_mem_reclaim(struct sock *sk, int amount)
3324{
3325 amount >>= PAGE_SHIFT;
3326 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3327 __sk_mem_reduce_allocated(sk, amount);
3328}
3329EXPORT_SYMBOL(__sk_mem_reclaim);
3330
3331int sk_set_peek_off(struct sock *sk, int val)
3332{
3333 WRITE_ONCE(sk->sk_peek_off, val);
3334 return 0;
3335}
3336EXPORT_SYMBOL_GPL(sk_set_peek_off);
3337
3338/*
3339 * Set of default routines for initialising struct proto_ops when
3340 * the protocol does not support a particular function. In certain
3341 * cases where it makes no sense for a protocol to have a "do nothing"
3342 * function, some default processing is provided.
3343 */
3344
3345int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3346{
3347 return -EOPNOTSUPP;
3348}
3349EXPORT_SYMBOL(sock_no_bind);
3350
3351int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3352 int len, int flags)
3353{
3354 return -EOPNOTSUPP;
3355}
3356EXPORT_SYMBOL(sock_no_connect);
3357
3358int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3359{
3360 return -EOPNOTSUPP;
3361}
3362EXPORT_SYMBOL(sock_no_socketpair);
3363
3364int sock_no_accept(struct socket *sock, struct socket *newsock,
3365 struct proto_accept_arg *arg)
3366{
3367 return -EOPNOTSUPP;
3368}
3369EXPORT_SYMBOL(sock_no_accept);
3370
3371int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3372 int peer)
3373{
3374 return -EOPNOTSUPP;
3375}
3376EXPORT_SYMBOL(sock_no_getname);
3377
3378int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3379{
3380 return -EOPNOTSUPP;
3381}
3382EXPORT_SYMBOL(sock_no_ioctl);
3383
3384int sock_no_listen(struct socket *sock, int backlog)
3385{
3386 return -EOPNOTSUPP;
3387}
3388EXPORT_SYMBOL(sock_no_listen);
3389
3390int sock_no_shutdown(struct socket *sock, int how)
3391{
3392 return -EOPNOTSUPP;
3393}
3394EXPORT_SYMBOL(sock_no_shutdown);
3395
3396int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3397{
3398 return -EOPNOTSUPP;
3399}
3400EXPORT_SYMBOL(sock_no_sendmsg);
3401
3402int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3403{
3404 return -EOPNOTSUPP;
3405}
3406EXPORT_SYMBOL(sock_no_sendmsg_locked);
3407
3408int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3409 int flags)
3410{
3411 return -EOPNOTSUPP;
3412}
3413EXPORT_SYMBOL(sock_no_recvmsg);
3414
3415int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3416{
3417 /* Mirror missing mmap method error code */
3418 return -ENODEV;
3419}
3420EXPORT_SYMBOL(sock_no_mmap);
3421
3422/*
3423 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3424 * various sock-based usage counts.
3425 */
3426void __receive_sock(struct file *file)
3427{
3428 struct socket *sock;
3429
3430 sock = sock_from_file(file);
3431 if (sock) {
3432 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3433 sock_update_classid(&sock->sk->sk_cgrp_data);
3434 }
3435}
3436
3437/*
3438 * Default Socket Callbacks
3439 */
3440
3441static void sock_def_wakeup(struct sock *sk)
3442{
3443 struct socket_wq *wq;
3444
3445 rcu_read_lock();
3446 wq = rcu_dereference(sk->sk_wq);
3447 if (skwq_has_sleeper(wq))
3448 wake_up_interruptible_all(&wq->wait);
3449 rcu_read_unlock();
3450}
3451
3452static void sock_def_error_report(struct sock *sk)
3453{
3454 struct socket_wq *wq;
3455
3456 rcu_read_lock();
3457 wq = rcu_dereference(sk->sk_wq);
3458 if (skwq_has_sleeper(wq))
3459 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3460 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3461 rcu_read_unlock();
3462}
3463
3464void sock_def_readable(struct sock *sk)
3465{
3466 struct socket_wq *wq;
3467
3468 trace_sk_data_ready(sk);
3469
3470 rcu_read_lock();
3471 wq = rcu_dereference(sk->sk_wq);
3472 if (skwq_has_sleeper(wq))
3473 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3474 EPOLLRDNORM | EPOLLRDBAND);
3475 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3476 rcu_read_unlock();
3477}
3478
3479static void sock_def_write_space(struct sock *sk)
3480{
3481 struct socket_wq *wq;
3482
3483 rcu_read_lock();
3484
3485 /* Do not wake up a writer until he can make "significant"
3486 * progress. --DaveM
3487 */
3488 if (sock_writeable(sk)) {
3489 wq = rcu_dereference(sk->sk_wq);
3490 if (skwq_has_sleeper(wq))
3491 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3492 EPOLLWRNORM | EPOLLWRBAND);
3493
3494 /* Should agree with poll, otherwise some programs break */
3495 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3496 }
3497
3498 rcu_read_unlock();
3499}
3500
3501/* An optimised version of sock_def_write_space(), should only be called
3502 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3503 * ->sk_wmem_alloc.
3504 */
3505static void sock_def_write_space_wfree(struct sock *sk)
3506{
3507 /* Do not wake up a writer until he can make "significant"
3508 * progress. --DaveM
3509 */
3510 if (sock_writeable(sk)) {
3511 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3512
3513 /* rely on refcount_sub from sock_wfree() */
3514 smp_mb__after_atomic();
3515 if (wq && waitqueue_active(&wq->wait))
3516 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3517 EPOLLWRNORM | EPOLLWRBAND);
3518
3519 /* Should agree with poll, otherwise some programs break */
3520 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3521 }
3522}
3523
3524static void sock_def_destruct(struct sock *sk)
3525{
3526}
3527
3528void sk_send_sigurg(struct sock *sk)
3529{
3530 if (sk->sk_socket && sk->sk_socket->file)
3531 if (send_sigurg(sk->sk_socket->file))
3532 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3533}
3534EXPORT_SYMBOL(sk_send_sigurg);
3535
3536void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3537 unsigned long expires)
3538{
3539 if (!mod_timer(timer, expires))
3540 sock_hold(sk);
3541}
3542EXPORT_SYMBOL(sk_reset_timer);
3543
3544void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3545{
3546 if (del_timer(timer))
3547 __sock_put(sk);
3548}
3549EXPORT_SYMBOL(sk_stop_timer);
3550
3551void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3552{
3553 if (del_timer_sync(timer))
3554 __sock_put(sk);
3555}
3556EXPORT_SYMBOL(sk_stop_timer_sync);
3557
3558void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3559{
3560 sk_init_common(sk);
3561 sk->sk_send_head = NULL;
3562
3563 timer_setup(&sk->sk_timer, NULL, 0);
3564
3565 sk->sk_allocation = GFP_KERNEL;
3566 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3567 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3568 sk->sk_state = TCP_CLOSE;
3569 sk->sk_use_task_frag = true;
3570 sk_set_socket(sk, sock);
3571
3572 sock_set_flag(sk, SOCK_ZAPPED);
3573
3574 if (sock) {
3575 sk->sk_type = sock->type;
3576 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3577 sock->sk = sk;
3578 } else {
3579 RCU_INIT_POINTER(sk->sk_wq, NULL);
3580 }
3581 sk->sk_uid = uid;
3582
3583 sk->sk_state_change = sock_def_wakeup;
3584 sk->sk_data_ready = sock_def_readable;
3585 sk->sk_write_space = sock_def_write_space;
3586 sk->sk_error_report = sock_def_error_report;
3587 sk->sk_destruct = sock_def_destruct;
3588
3589 sk->sk_frag.page = NULL;
3590 sk->sk_frag.offset = 0;
3591 sk->sk_peek_off = -1;
3592
3593 sk->sk_peer_pid = NULL;
3594 sk->sk_peer_cred = NULL;
3595 spin_lock_init(&sk->sk_peer_lock);
3596
3597 sk->sk_write_pending = 0;
3598 sk->sk_rcvlowat = 1;
3599 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3600 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3601
3602 sk->sk_stamp = SK_DEFAULT_STAMP;
3603#if BITS_PER_LONG==32
3604 seqlock_init(&sk->sk_stamp_seq);
3605#endif
3606 atomic_set(&sk->sk_zckey, 0);
3607
3608#ifdef CONFIG_NET_RX_BUSY_POLL
3609 sk->sk_napi_id = 0;
3610 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3611#endif
3612
3613 sk->sk_max_pacing_rate = ~0UL;
3614 sk->sk_pacing_rate = ~0UL;
3615 WRITE_ONCE(sk->sk_pacing_shift, 10);
3616 sk->sk_incoming_cpu = -1;
3617
3618 sk_rx_queue_clear(sk);
3619 /*
3620 * Before updating sk_refcnt, we must commit prior changes to memory
3621 * (Documentation/RCU/rculist_nulls.rst for details)
3622 */
3623 smp_wmb();
3624 refcount_set(&sk->sk_refcnt, 1);
3625 atomic_set(&sk->sk_drops, 0);
3626}
3627EXPORT_SYMBOL(sock_init_data_uid);
3628
3629void sock_init_data(struct socket *sock, struct sock *sk)
3630{
3631 kuid_t uid = sock ?
3632 SOCK_INODE(sock)->i_uid :
3633 make_kuid(sock_net(sk)->user_ns, 0);
3634
3635 sock_init_data_uid(sock, sk, uid);
3636}
3637EXPORT_SYMBOL(sock_init_data);
3638
3639void lock_sock_nested(struct sock *sk, int subclass)
3640{
3641 /* The sk_lock has mutex_lock() semantics here. */
3642 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3643
3644 might_sleep();
3645 spin_lock_bh(&sk->sk_lock.slock);
3646 if (sock_owned_by_user_nocheck(sk))
3647 __lock_sock(sk);
3648 sk->sk_lock.owned = 1;
3649 spin_unlock_bh(&sk->sk_lock.slock);
3650}
3651EXPORT_SYMBOL(lock_sock_nested);
3652
3653void release_sock(struct sock *sk)
3654{
3655 spin_lock_bh(&sk->sk_lock.slock);
3656 if (sk->sk_backlog.tail)
3657 __release_sock(sk);
3658
3659 if (sk->sk_prot->release_cb)
3660 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3661 tcp_release_cb, sk);
3662
3663 sock_release_ownership(sk);
3664 if (waitqueue_active(&sk->sk_lock.wq))
3665 wake_up(&sk->sk_lock.wq);
3666 spin_unlock_bh(&sk->sk_lock.slock);
3667}
3668EXPORT_SYMBOL(release_sock);
3669
3670bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3671{
3672 might_sleep();
3673 spin_lock_bh(&sk->sk_lock.slock);
3674
3675 if (!sock_owned_by_user_nocheck(sk)) {
3676 /*
3677 * Fast path return with bottom halves disabled and
3678 * sock::sk_lock.slock held.
3679 *
3680 * The 'mutex' is not contended and holding
3681 * sock::sk_lock.slock prevents all other lockers to
3682 * proceed so the corresponding unlock_sock_fast() can
3683 * avoid the slow path of release_sock() completely and
3684 * just release slock.
3685 *
3686 * From a semantical POV this is equivalent to 'acquiring'
3687 * the 'mutex', hence the corresponding lockdep
3688 * mutex_release() has to happen in the fast path of
3689 * unlock_sock_fast().
3690 */
3691 return false;
3692 }
3693
3694 __lock_sock(sk);
3695 sk->sk_lock.owned = 1;
3696 __acquire(&sk->sk_lock.slock);
3697 spin_unlock_bh(&sk->sk_lock.slock);
3698 return true;
3699}
3700EXPORT_SYMBOL(__lock_sock_fast);
3701
3702int sock_gettstamp(struct socket *sock, void __user *userstamp,
3703 bool timeval, bool time32)
3704{
3705 struct sock *sk = sock->sk;
3706 struct timespec64 ts;
3707
3708 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3709 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3710 if (ts.tv_sec == -1)
3711 return -ENOENT;
3712 if (ts.tv_sec == 0) {
3713 ktime_t kt = ktime_get_real();
3714 sock_write_timestamp(sk, kt);
3715 ts = ktime_to_timespec64(kt);
3716 }
3717
3718 if (timeval)
3719 ts.tv_nsec /= 1000;
3720
3721#ifdef CONFIG_COMPAT_32BIT_TIME
3722 if (time32)
3723 return put_old_timespec32(&ts, userstamp);
3724#endif
3725#ifdef CONFIG_SPARC64
3726 /* beware of padding in sparc64 timeval */
3727 if (timeval && !in_compat_syscall()) {
3728 struct __kernel_old_timeval __user tv = {
3729 .tv_sec = ts.tv_sec,
3730 .tv_usec = ts.tv_nsec,
3731 };
3732 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3733 return -EFAULT;
3734 return 0;
3735 }
3736#endif
3737 return put_timespec64(&ts, userstamp);
3738}
3739EXPORT_SYMBOL(sock_gettstamp);
3740
3741void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3742{
3743 if (!sock_flag(sk, flag)) {
3744 unsigned long previous_flags = sk->sk_flags;
3745
3746 sock_set_flag(sk, flag);
3747 /*
3748 * we just set one of the two flags which require net
3749 * time stamping, but time stamping might have been on
3750 * already because of the other one
3751 */
3752 if (sock_needs_netstamp(sk) &&
3753 !(previous_flags & SK_FLAGS_TIMESTAMP))
3754 net_enable_timestamp();
3755 }
3756}
3757
3758int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3759 int level, int type)
3760{
3761 struct sock_exterr_skb *serr;
3762 struct sk_buff *skb;
3763 int copied, err;
3764
3765 err = -EAGAIN;
3766 skb = sock_dequeue_err_skb(sk);
3767 if (skb == NULL)
3768 goto out;
3769
3770 copied = skb->len;
3771 if (copied > len) {
3772 msg->msg_flags |= MSG_TRUNC;
3773 copied = len;
3774 }
3775 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3776 if (err)
3777 goto out_free_skb;
3778
3779 sock_recv_timestamp(msg, sk, skb);
3780
3781 serr = SKB_EXT_ERR(skb);
3782 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3783
3784 msg->msg_flags |= MSG_ERRQUEUE;
3785 err = copied;
3786
3787out_free_skb:
3788 kfree_skb(skb);
3789out:
3790 return err;
3791}
3792EXPORT_SYMBOL(sock_recv_errqueue);
3793
3794/*
3795 * Get a socket option on an socket.
3796 *
3797 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3798 * asynchronous errors should be reported by getsockopt. We assume
3799 * this means if you specify SO_ERROR (otherwise what is the point of it).
3800 */
3801int sock_common_getsockopt(struct socket *sock, int level, int optname,
3802 char __user *optval, int __user *optlen)
3803{
3804 struct sock *sk = sock->sk;
3805
3806 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3807 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3808}
3809EXPORT_SYMBOL(sock_common_getsockopt);
3810
3811int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3812 int flags)
3813{
3814 struct sock *sk = sock->sk;
3815 int addr_len = 0;
3816 int err;
3817
3818 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3819 if (err >= 0)
3820 msg->msg_namelen = addr_len;
3821 return err;
3822}
3823EXPORT_SYMBOL(sock_common_recvmsg);
3824
3825/*
3826 * Set socket options on an inet socket.
3827 */
3828int sock_common_setsockopt(struct socket *sock, int level, int optname,
3829 sockptr_t optval, unsigned int optlen)
3830{
3831 struct sock *sk = sock->sk;
3832
3833 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3834 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3835}
3836EXPORT_SYMBOL(sock_common_setsockopt);
3837
3838void sk_common_release(struct sock *sk)
3839{
3840 if (sk->sk_prot->destroy)
3841 sk->sk_prot->destroy(sk);
3842
3843 /*
3844 * Observation: when sk_common_release is called, processes have
3845 * no access to socket. But net still has.
3846 * Step one, detach it from networking:
3847 *
3848 * A. Remove from hash tables.
3849 */
3850
3851 sk->sk_prot->unhash(sk);
3852
3853 /*
3854 * In this point socket cannot receive new packets, but it is possible
3855 * that some packets are in flight because some CPU runs receiver and
3856 * did hash table lookup before we unhashed socket. They will achieve
3857 * receive queue and will be purged by socket destructor.
3858 *
3859 * Also we still have packets pending on receive queue and probably,
3860 * our own packets waiting in device queues. sock_destroy will drain
3861 * receive queue, but transmitted packets will delay socket destruction
3862 * until the last reference will be released.
3863 */
3864
3865 sock_orphan(sk);
3866
3867 xfrm_sk_free_policy(sk);
3868
3869 sock_put(sk);
3870}
3871EXPORT_SYMBOL(sk_common_release);
3872
3873void sk_get_meminfo(const struct sock *sk, u32 *mem)
3874{
3875 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3876
3877 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3878 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3879 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3880 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3881 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3882 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3883 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3884 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3885 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3886}
3887
3888#ifdef CONFIG_PROC_FS
3889static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3890
3891int sock_prot_inuse_get(struct net *net, struct proto *prot)
3892{
3893 int cpu, idx = prot->inuse_idx;
3894 int res = 0;
3895
3896 for_each_possible_cpu(cpu)
3897 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3898
3899 return res >= 0 ? res : 0;
3900}
3901EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3902
3903int sock_inuse_get(struct net *net)
3904{
3905 int cpu, res = 0;
3906
3907 for_each_possible_cpu(cpu)
3908 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3909
3910 return res;
3911}
3912
3913EXPORT_SYMBOL_GPL(sock_inuse_get);
3914
3915static int __net_init sock_inuse_init_net(struct net *net)
3916{
3917 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3918 if (net->core.prot_inuse == NULL)
3919 return -ENOMEM;
3920 return 0;
3921}
3922
3923static void __net_exit sock_inuse_exit_net(struct net *net)
3924{
3925 free_percpu(net->core.prot_inuse);
3926}
3927
3928static struct pernet_operations net_inuse_ops = {
3929 .init = sock_inuse_init_net,
3930 .exit = sock_inuse_exit_net,
3931};
3932
3933static __init int net_inuse_init(void)
3934{
3935 if (register_pernet_subsys(&net_inuse_ops))
3936 panic("Cannot initialize net inuse counters");
3937
3938 return 0;
3939}
3940
3941core_initcall(net_inuse_init);
3942
3943static int assign_proto_idx(struct proto *prot)
3944{
3945 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3946
3947 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3948 pr_err("PROTO_INUSE_NR exhausted\n");
3949 return -ENOSPC;
3950 }
3951
3952 set_bit(prot->inuse_idx, proto_inuse_idx);
3953 return 0;
3954}
3955
3956static void release_proto_idx(struct proto *prot)
3957{
3958 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3959 clear_bit(prot->inuse_idx, proto_inuse_idx);
3960}
3961#else
3962static inline int assign_proto_idx(struct proto *prot)
3963{
3964 return 0;
3965}
3966
3967static inline void release_proto_idx(struct proto *prot)
3968{
3969}
3970
3971#endif
3972
3973static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3974{
3975 if (!twsk_prot)
3976 return;
3977 kfree(twsk_prot->twsk_slab_name);
3978 twsk_prot->twsk_slab_name = NULL;
3979 kmem_cache_destroy(twsk_prot->twsk_slab);
3980 twsk_prot->twsk_slab = NULL;
3981}
3982
3983static int tw_prot_init(const struct proto *prot)
3984{
3985 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3986
3987 if (!twsk_prot)
3988 return 0;
3989
3990 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3991 prot->name);
3992 if (!twsk_prot->twsk_slab_name)
3993 return -ENOMEM;
3994
3995 twsk_prot->twsk_slab =
3996 kmem_cache_create(twsk_prot->twsk_slab_name,
3997 twsk_prot->twsk_obj_size, 0,
3998 SLAB_ACCOUNT | prot->slab_flags,
3999 NULL);
4000 if (!twsk_prot->twsk_slab) {
4001 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4002 prot->name);
4003 return -ENOMEM;
4004 }
4005
4006 return 0;
4007}
4008
4009static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4010{
4011 if (!rsk_prot)
4012 return;
4013 kfree(rsk_prot->slab_name);
4014 rsk_prot->slab_name = NULL;
4015 kmem_cache_destroy(rsk_prot->slab);
4016 rsk_prot->slab = NULL;
4017}
4018
4019static int req_prot_init(const struct proto *prot)
4020{
4021 struct request_sock_ops *rsk_prot = prot->rsk_prot;
4022
4023 if (!rsk_prot)
4024 return 0;
4025
4026 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4027 prot->name);
4028 if (!rsk_prot->slab_name)
4029 return -ENOMEM;
4030
4031 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4032 rsk_prot->obj_size, 0,
4033 SLAB_ACCOUNT | prot->slab_flags,
4034 NULL);
4035
4036 if (!rsk_prot->slab) {
4037 pr_crit("%s: Can't create request sock SLAB cache!\n",
4038 prot->name);
4039 return -ENOMEM;
4040 }
4041 return 0;
4042}
4043
4044int proto_register(struct proto *prot, int alloc_slab)
4045{
4046 int ret = -ENOBUFS;
4047
4048 if (prot->memory_allocated && !prot->sysctl_mem) {
4049 pr_err("%s: missing sysctl_mem\n", prot->name);
4050 return -EINVAL;
4051 }
4052 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4053 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4054 return -EINVAL;
4055 }
4056 if (alloc_slab) {
4057 prot->slab = kmem_cache_create_usercopy(prot->name,
4058 prot->obj_size, 0,
4059 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4060 prot->slab_flags,
4061 prot->useroffset, prot->usersize,
4062 NULL);
4063
4064 if (prot->slab == NULL) {
4065 pr_crit("%s: Can't create sock SLAB cache!\n",
4066 prot->name);
4067 goto out;
4068 }
4069
4070 if (req_prot_init(prot))
4071 goto out_free_request_sock_slab;
4072
4073 if (tw_prot_init(prot))
4074 goto out_free_timewait_sock_slab;
4075 }
4076
4077 mutex_lock(&proto_list_mutex);
4078 ret = assign_proto_idx(prot);
4079 if (ret) {
4080 mutex_unlock(&proto_list_mutex);
4081 goto out_free_timewait_sock_slab;
4082 }
4083 list_add(&prot->node, &proto_list);
4084 mutex_unlock(&proto_list_mutex);
4085 return ret;
4086
4087out_free_timewait_sock_slab:
4088 if (alloc_slab)
4089 tw_prot_cleanup(prot->twsk_prot);
4090out_free_request_sock_slab:
4091 if (alloc_slab) {
4092 req_prot_cleanup(prot->rsk_prot);
4093
4094 kmem_cache_destroy(prot->slab);
4095 prot->slab = NULL;
4096 }
4097out:
4098 return ret;
4099}
4100EXPORT_SYMBOL(proto_register);
4101
4102void proto_unregister(struct proto *prot)
4103{
4104 mutex_lock(&proto_list_mutex);
4105 release_proto_idx(prot);
4106 list_del(&prot->node);
4107 mutex_unlock(&proto_list_mutex);
4108
4109 kmem_cache_destroy(prot->slab);
4110 prot->slab = NULL;
4111
4112 req_prot_cleanup(prot->rsk_prot);
4113 tw_prot_cleanup(prot->twsk_prot);
4114}
4115EXPORT_SYMBOL(proto_unregister);
4116
4117int sock_load_diag_module(int family, int protocol)
4118{
4119 if (!protocol) {
4120 if (!sock_is_registered(family))
4121 return -ENOENT;
4122
4123 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4124 NETLINK_SOCK_DIAG, family);
4125 }
4126
4127#ifdef CONFIG_INET
4128 if (family == AF_INET &&
4129 protocol != IPPROTO_RAW &&
4130 protocol < MAX_INET_PROTOS &&
4131 !rcu_access_pointer(inet_protos[protocol]))
4132 return -ENOENT;
4133#endif
4134
4135 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4136 NETLINK_SOCK_DIAG, family, protocol);
4137}
4138EXPORT_SYMBOL(sock_load_diag_module);
4139
4140#ifdef CONFIG_PROC_FS
4141static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4142 __acquires(proto_list_mutex)
4143{
4144 mutex_lock(&proto_list_mutex);
4145 return seq_list_start_head(&proto_list, *pos);
4146}
4147
4148static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4149{
4150 return seq_list_next(v, &proto_list, pos);
4151}
4152
4153static void proto_seq_stop(struct seq_file *seq, void *v)
4154 __releases(proto_list_mutex)
4155{
4156 mutex_unlock(&proto_list_mutex);
4157}
4158
4159static char proto_method_implemented(const void *method)
4160{
4161 return method == NULL ? 'n' : 'y';
4162}
4163static long sock_prot_memory_allocated(struct proto *proto)
4164{
4165 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4166}
4167
4168static const char *sock_prot_memory_pressure(struct proto *proto)
4169{
4170 return proto->memory_pressure != NULL ?
4171 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4172}
4173
4174static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4175{
4176
4177 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4178 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4179 proto->name,
4180 proto->obj_size,
4181 sock_prot_inuse_get(seq_file_net(seq), proto),
4182 sock_prot_memory_allocated(proto),
4183 sock_prot_memory_pressure(proto),
4184 proto->max_header,
4185 proto->slab == NULL ? "no" : "yes",
4186 module_name(proto->owner),
4187 proto_method_implemented(proto->close),
4188 proto_method_implemented(proto->connect),
4189 proto_method_implemented(proto->disconnect),
4190 proto_method_implemented(proto->accept),
4191 proto_method_implemented(proto->ioctl),
4192 proto_method_implemented(proto->init),
4193 proto_method_implemented(proto->destroy),
4194 proto_method_implemented(proto->shutdown),
4195 proto_method_implemented(proto->setsockopt),
4196 proto_method_implemented(proto->getsockopt),
4197 proto_method_implemented(proto->sendmsg),
4198 proto_method_implemented(proto->recvmsg),
4199 proto_method_implemented(proto->bind),
4200 proto_method_implemented(proto->backlog_rcv),
4201 proto_method_implemented(proto->hash),
4202 proto_method_implemented(proto->unhash),
4203 proto_method_implemented(proto->get_port),
4204 proto_method_implemented(proto->enter_memory_pressure));
4205}
4206
4207static int proto_seq_show(struct seq_file *seq, void *v)
4208{
4209 if (v == &proto_list)
4210 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4211 "protocol",
4212 "size",
4213 "sockets",
4214 "memory",
4215 "press",
4216 "maxhdr",
4217 "slab",
4218 "module",
4219 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4220 else
4221 proto_seq_printf(seq, list_entry(v, struct proto, node));
4222 return 0;
4223}
4224
4225static const struct seq_operations proto_seq_ops = {
4226 .start = proto_seq_start,
4227 .next = proto_seq_next,
4228 .stop = proto_seq_stop,
4229 .show = proto_seq_show,
4230};
4231
4232static __net_init int proto_init_net(struct net *net)
4233{
4234 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4235 sizeof(struct seq_net_private)))
4236 return -ENOMEM;
4237
4238 return 0;
4239}
4240
4241static __net_exit void proto_exit_net(struct net *net)
4242{
4243 remove_proc_entry("protocols", net->proc_net);
4244}
4245
4246
4247static __net_initdata struct pernet_operations proto_net_ops = {
4248 .init = proto_init_net,
4249 .exit = proto_exit_net,
4250};
4251
4252static int __init proto_init(void)
4253{
4254 return register_pernet_subsys(&proto_net_ops);
4255}
4256
4257subsys_initcall(proto_init);
4258
4259#endif /* PROC_FS */
4260
4261#ifdef CONFIG_NET_RX_BUSY_POLL
4262bool sk_busy_loop_end(void *p, unsigned long start_time)
4263{
4264 struct sock *sk = p;
4265
4266 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4267 return true;
4268
4269 if (sk_is_udp(sk) &&
4270 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4271 return true;
4272
4273 return sk_busy_loop_timeout(sk, start_time);
4274}
4275EXPORT_SYMBOL(sk_busy_loop_end);
4276#endif /* CONFIG_NET_RX_BUSY_POLL */
4277
4278int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4279{
4280 if (!sk->sk_prot->bind_add)
4281 return -EOPNOTSUPP;
4282 return sk->sk_prot->bind_add(sk, addr, addr_len);
4283}
4284EXPORT_SYMBOL(sock_bind_add);
4285
4286/* Copy 'size' bytes from userspace and return `size` back to userspace */
4287int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4288 void __user *arg, void *karg, size_t size)
4289{
4290 int ret;
4291
4292 if (copy_from_user(karg, arg, size))
4293 return -EFAULT;
4294
4295 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4296 if (ret)
4297 return ret;
4298
4299 if (copy_to_user(arg, karg, size))
4300 return -EFAULT;
4301
4302 return 0;
4303}
4304EXPORT_SYMBOL(sock_ioctl_inout);
4305
4306/* This is the most common ioctl prep function, where the result (4 bytes) is
4307 * copied back to userspace if the ioctl() returns successfully. No input is
4308 * copied from userspace as input argument.
4309 */
4310static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4311{
4312 int ret, karg = 0;
4313
4314 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4315 if (ret)
4316 return ret;
4317
4318 return put_user(karg, (int __user *)arg);
4319}
4320
4321/* A wrapper around sock ioctls, which copies the data from userspace
4322 * (depending on the protocol/ioctl), and copies back the result to userspace.
4323 * The main motivation for this function is to pass kernel memory to the
4324 * protocol ioctl callbacks, instead of userspace memory.
4325 */
4326int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4327{
4328 int rc = 1;
4329
4330 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4331 rc = ipmr_sk_ioctl(sk, cmd, arg);
4332 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4333 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4334 else if (sk_is_phonet(sk))
4335 rc = phonet_sk_ioctl(sk, cmd, arg);
4336
4337 /* If ioctl was processed, returns its value */
4338 if (rc <= 0)
4339 return rc;
4340
4341 /* Otherwise call the default handler */
4342 return sock_ioctl_out(sk, cmd, arg);
4343}
4344EXPORT_SYMBOL(sk_ioctl);
4345
4346static int __init sock_struct_check(void)
4347{
4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4353
4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4363
4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4367
4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4372
4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4385 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4386 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4387 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4388 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4389
4390 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4391 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4392 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4393 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4394 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4395 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4396 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4397 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4398 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4399 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4400 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4401 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4402 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4403 return 0;
4404}
4405
4406core_initcall(sock_struct_check);