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/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 <net/net_namespace.h>
128#include <net/request_sock.h>
129#include <net/sock.h>
130#include <linux/net_tstamp.h>
131#include <net/xfrm.h>
132#include <linux/ipsec.h>
133#include <net/cls_cgroup.h>
134#include <net/netprio_cgroup.h>
135#include <linux/sock_diag.h>
136
137#include <linux/filter.h>
138#include <net/sock_reuseport.h>
139#include <net/bpf_sk_storage.h>
140
141#include <trace/events/sock.h>
142
143#include <net/tcp.h>
144#include <net/busy_poll.h>
145#include <net/phonet/phonet.h>
146
147#include <linux/ethtool.h>
148
149#include "dev.h"
150
151static DEFINE_MUTEX(proto_list_mutex);
152static LIST_HEAD(proto_list);
153
154static void sock_def_write_space_wfree(struct sock *sk);
155static void sock_def_write_space(struct sock *sk);
156
157/**
158 * sk_ns_capable - General socket capability test
159 * @sk: Socket to use a capability on or through
160 * @user_ns: The user namespace of the capability to use
161 * @cap: The capability to use
162 *
163 * Test to see if the opener of the socket had when the socket was
164 * created and the current process has the capability @cap in the user
165 * namespace @user_ns.
166 */
167bool sk_ns_capable(const struct sock *sk,
168 struct user_namespace *user_ns, int cap)
169{
170 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
171 ns_capable(user_ns, cap);
172}
173EXPORT_SYMBOL(sk_ns_capable);
174
175/**
176 * sk_capable - Socket global capability test
177 * @sk: Socket to use a capability on or through
178 * @cap: The global capability to use
179 *
180 * Test to see if the opener of the socket had when the socket was
181 * created and the current process has the capability @cap in all user
182 * namespaces.
183 */
184bool sk_capable(const struct sock *sk, int cap)
185{
186 return sk_ns_capable(sk, &init_user_ns, cap);
187}
188EXPORT_SYMBOL(sk_capable);
189
190/**
191 * sk_net_capable - Network namespace socket capability test
192 * @sk: Socket to use a capability on or through
193 * @cap: The capability to use
194 *
195 * Test to see if the opener of the socket had when the socket was created
196 * and the current process has the capability @cap over the network namespace
197 * the socket is a member of.
198 */
199bool sk_net_capable(const struct sock *sk, int cap)
200{
201 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202}
203EXPORT_SYMBOL(sk_net_capable);
204
205/*
206 * Each address family might have different locking rules, so we have
207 * one slock key per address family and separate keys for internal and
208 * userspace sockets.
209 */
210static struct lock_class_key af_family_keys[AF_MAX];
211static struct lock_class_key af_family_kern_keys[AF_MAX];
212static struct lock_class_key af_family_slock_keys[AF_MAX];
213static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
214
215/*
216 * Make lock validator output more readable. (we pre-construct these
217 * strings build-time, so that runtime initialization of socket
218 * locks is fast):
219 */
220
221#define _sock_locks(x) \
222 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
223 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
224 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
225 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
226 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
227 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
228 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
229 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
230 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
231 x "27" , x "28" , x "AF_CAN" , \
232 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
233 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
234 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
235 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
236 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
237 x "AF_MCTP" , \
238 x "AF_MAX"
239
240static const char *const af_family_key_strings[AF_MAX+1] = {
241 _sock_locks("sk_lock-")
242};
243static const char *const af_family_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("slock-")
245};
246static const char *const af_family_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("clock-")
248};
249
250static const char *const af_family_kern_key_strings[AF_MAX+1] = {
251 _sock_locks("k-sk_lock-")
252};
253static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
254 _sock_locks("k-slock-")
255};
256static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
257 _sock_locks("k-clock-")
258};
259static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
260 _sock_locks("rlock-")
261};
262static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
263 _sock_locks("wlock-")
264};
265static const char *const af_family_elock_key_strings[AF_MAX+1] = {
266 _sock_locks("elock-")
267};
268
269/*
270 * sk_callback_lock and sk queues locking rules are per-address-family,
271 * so split the lock classes by using a per-AF key:
272 */
273static struct lock_class_key af_callback_keys[AF_MAX];
274static struct lock_class_key af_rlock_keys[AF_MAX];
275static struct lock_class_key af_wlock_keys[AF_MAX];
276static struct lock_class_key af_elock_keys[AF_MAX];
277static struct lock_class_key af_kern_callback_keys[AF_MAX];
278
279/* Run time adjustable parameters. */
280__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
281EXPORT_SYMBOL(sysctl_wmem_max);
282__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
283EXPORT_SYMBOL(sysctl_rmem_max);
284__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
285__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286
287int sysctl_tstamp_allow_data __read_mostly = 1;
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) >= 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, 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 if (val) {
824 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
825 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
826 sock_set_flag(sk, SOCK_RCVTSTAMP);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
828 } else {
829 sock_reset_flag(sk, SOCK_RCVTSTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
831 }
832}
833
834void sock_enable_timestamps(struct sock *sk)
835{
836 lock_sock(sk);
837 __sock_set_timestamps(sk, true, false, true);
838 release_sock(sk);
839}
840EXPORT_SYMBOL(sock_enable_timestamps);
841
842void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
843{
844 switch (optname) {
845 case SO_TIMESTAMP_OLD:
846 __sock_set_timestamps(sk, valbool, false, false);
847 break;
848 case SO_TIMESTAMP_NEW:
849 __sock_set_timestamps(sk, valbool, true, false);
850 break;
851 case SO_TIMESTAMPNS_OLD:
852 __sock_set_timestamps(sk, valbool, false, true);
853 break;
854 case SO_TIMESTAMPNS_NEW:
855 __sock_set_timestamps(sk, valbool, true, true);
856 break;
857 }
858}
859
860static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
861{
862 struct net *net = sock_net(sk);
863 struct net_device *dev = NULL;
864 bool match = false;
865 int *vclock_index;
866 int i, num;
867
868 if (sk->sk_bound_dev_if)
869 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
870
871 if (!dev) {
872 pr_err("%s: sock not bind to device\n", __func__);
873 return -EOPNOTSUPP;
874 }
875
876 num = ethtool_get_phc_vclocks(dev, &vclock_index);
877 dev_put(dev);
878
879 for (i = 0; i < num; i++) {
880 if (*(vclock_index + i) == phc_index) {
881 match = true;
882 break;
883 }
884 }
885
886 if (num > 0)
887 kfree(vclock_index);
888
889 if (!match)
890 return -EINVAL;
891
892 WRITE_ONCE(sk->sk_bind_phc, phc_index);
893
894 return 0;
895}
896
897int sock_set_timestamping(struct sock *sk, int optname,
898 struct so_timestamping timestamping)
899{
900 int val = timestamping.flags;
901 int ret;
902
903 if (val & ~SOF_TIMESTAMPING_MASK)
904 return -EINVAL;
905
906 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
907 !(val & SOF_TIMESTAMPING_OPT_ID))
908 return -EINVAL;
909
910 if (val & SOF_TIMESTAMPING_OPT_ID &&
911 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
912 if (sk_is_tcp(sk)) {
913 if ((1 << sk->sk_state) &
914 (TCPF_CLOSE | TCPF_LISTEN))
915 return -EINVAL;
916 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
918 else
919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
920 } else {
921 atomic_set(&sk->sk_tskey, 0);
922 }
923 }
924
925 if (val & SOF_TIMESTAMPING_OPT_STATS &&
926 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
927 return -EINVAL;
928
929 if (val & SOF_TIMESTAMPING_BIND_PHC) {
930 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
931 if (ret)
932 return ret;
933 }
934
935 WRITE_ONCE(sk->sk_tsflags, val);
936 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
937
938 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
939 sock_enable_timestamp(sk,
940 SOCK_TIMESTAMPING_RX_SOFTWARE);
941 else
942 sock_disable_timestamp(sk,
943 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
944 return 0;
945}
946
947void sock_set_keepalive(struct sock *sk)
948{
949 lock_sock(sk);
950 if (sk->sk_prot->keepalive)
951 sk->sk_prot->keepalive(sk, true);
952 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
953 release_sock(sk);
954}
955EXPORT_SYMBOL(sock_set_keepalive);
956
957static void __sock_set_rcvbuf(struct sock *sk, int val)
958{
959 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
960 * as a negative value.
961 */
962 val = min_t(int, val, INT_MAX / 2);
963 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
964
965 /* We double it on the way in to account for "struct sk_buff" etc.
966 * overhead. Applications assume that the SO_RCVBUF setting they make
967 * will allow that much actual data to be received on that socket.
968 *
969 * Applications are unaware that "struct sk_buff" and other overheads
970 * allocate from the receive buffer during socket buffer allocation.
971 *
972 * And after considering the possible alternatives, returning the value
973 * we actually used in getsockopt is the most desirable behavior.
974 */
975 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
976}
977
978void sock_set_rcvbuf(struct sock *sk, int val)
979{
980 lock_sock(sk);
981 __sock_set_rcvbuf(sk, val);
982 release_sock(sk);
983}
984EXPORT_SYMBOL(sock_set_rcvbuf);
985
986static void __sock_set_mark(struct sock *sk, u32 val)
987{
988 if (val != sk->sk_mark) {
989 WRITE_ONCE(sk->sk_mark, val);
990 sk_dst_reset(sk);
991 }
992}
993
994void sock_set_mark(struct sock *sk, u32 val)
995{
996 lock_sock(sk);
997 __sock_set_mark(sk, val);
998 release_sock(sk);
999}
1000EXPORT_SYMBOL(sock_set_mark);
1001
1002static void sock_release_reserved_memory(struct sock *sk, int bytes)
1003{
1004 /* Round down bytes to multiple of pages */
1005 bytes = round_down(bytes, PAGE_SIZE);
1006
1007 WARN_ON(bytes > sk->sk_reserved_mem);
1008 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1009 sk_mem_reclaim(sk);
1010}
1011
1012static int sock_reserve_memory(struct sock *sk, int bytes)
1013{
1014 long allocated;
1015 bool charged;
1016 int pages;
1017
1018 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1019 return -EOPNOTSUPP;
1020
1021 if (!bytes)
1022 return 0;
1023
1024 pages = sk_mem_pages(bytes);
1025
1026 /* pre-charge to memcg */
1027 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1028 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1029 if (!charged)
1030 return -ENOMEM;
1031
1032 /* pre-charge to forward_alloc */
1033 sk_memory_allocated_add(sk, pages);
1034 allocated = sk_memory_allocated(sk);
1035 /* If the system goes into memory pressure with this
1036 * precharge, give up and return error.
1037 */
1038 if (allocated > sk_prot_mem_limits(sk, 1)) {
1039 sk_memory_allocated_sub(sk, pages);
1040 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1041 return -ENOMEM;
1042 }
1043 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1044
1045 WRITE_ONCE(sk->sk_reserved_mem,
1046 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1047
1048 return 0;
1049}
1050
1051void sockopt_lock_sock(struct sock *sk)
1052{
1053 /* When current->bpf_ctx is set, the setsockopt is called from
1054 * a bpf prog. bpf has ensured the sk lock has been
1055 * acquired before calling setsockopt().
1056 */
1057 if (has_current_bpf_ctx())
1058 return;
1059
1060 lock_sock(sk);
1061}
1062EXPORT_SYMBOL(sockopt_lock_sock);
1063
1064void sockopt_release_sock(struct sock *sk)
1065{
1066 if (has_current_bpf_ctx())
1067 return;
1068
1069 release_sock(sk);
1070}
1071EXPORT_SYMBOL(sockopt_release_sock);
1072
1073bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1074{
1075 return has_current_bpf_ctx() || ns_capable(ns, cap);
1076}
1077EXPORT_SYMBOL(sockopt_ns_capable);
1078
1079bool sockopt_capable(int cap)
1080{
1081 return has_current_bpf_ctx() || capable(cap);
1082}
1083EXPORT_SYMBOL(sockopt_capable);
1084
1085/*
1086 * This is meant for all protocols to use and covers goings on
1087 * at the socket level. Everything here is generic.
1088 */
1089
1090int sk_setsockopt(struct sock *sk, int level, int optname,
1091 sockptr_t optval, unsigned int optlen)
1092{
1093 struct so_timestamping timestamping;
1094 struct socket *sock = sk->sk_socket;
1095 struct sock_txtime sk_txtime;
1096 int val;
1097 int valbool;
1098 struct linger ling;
1099 int ret = 0;
1100
1101 /*
1102 * Options without arguments
1103 */
1104
1105 if (optname == SO_BINDTODEVICE)
1106 return sock_setbindtodevice(sk, optval, optlen);
1107
1108 if (optlen < sizeof(int))
1109 return -EINVAL;
1110
1111 if (copy_from_sockptr(&val, optval, sizeof(val)))
1112 return -EFAULT;
1113
1114 valbool = val ? 1 : 0;
1115
1116 /* handle options which do not require locking the socket. */
1117 switch (optname) {
1118 case SO_PRIORITY:
1119 if ((val >= 0 && val <= 6) ||
1120 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1121 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1122 sock_set_priority(sk, val);
1123 return 0;
1124 }
1125 return -EPERM;
1126 case SO_PASSSEC:
1127 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1128 return 0;
1129 case SO_PASSCRED:
1130 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1131 return 0;
1132 case SO_PASSPIDFD:
1133 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1134 return 0;
1135 case SO_TYPE:
1136 case SO_PROTOCOL:
1137 case SO_DOMAIN:
1138 case SO_ERROR:
1139 return -ENOPROTOOPT;
1140#ifdef CONFIG_NET_RX_BUSY_POLL
1141 case SO_BUSY_POLL:
1142 if (val < 0)
1143 return -EINVAL;
1144 WRITE_ONCE(sk->sk_ll_usec, val);
1145 return 0;
1146 case SO_PREFER_BUSY_POLL:
1147 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1148 return -EPERM;
1149 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1150 return 0;
1151 case SO_BUSY_POLL_BUDGET:
1152 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1153 !sockopt_capable(CAP_NET_ADMIN))
1154 return -EPERM;
1155 if (val < 0 || val > U16_MAX)
1156 return -EINVAL;
1157 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1158 return 0;
1159#endif
1160 case SO_MAX_PACING_RATE:
1161 {
1162 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1163 unsigned long pacing_rate;
1164
1165 if (sizeof(ulval) != sizeof(val) &&
1166 optlen >= sizeof(ulval) &&
1167 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1168 return -EFAULT;
1169 }
1170 if (ulval != ~0UL)
1171 cmpxchg(&sk->sk_pacing_status,
1172 SK_PACING_NONE,
1173 SK_PACING_NEEDED);
1174 /* Pairs with READ_ONCE() from sk_getsockopt() */
1175 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1176 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1177 if (ulval < pacing_rate)
1178 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1179 return 0;
1180 }
1181 case SO_TXREHASH:
1182 if (val < -1 || val > 1)
1183 return -EINVAL;
1184 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1185 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1186 /* Paired with READ_ONCE() in tcp_rtx_synack()
1187 * and sk_getsockopt().
1188 */
1189 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1190 return 0;
1191 case SO_PEEK_OFF:
1192 {
1193 int (*set_peek_off)(struct sock *sk, int val);
1194
1195 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1196 if (set_peek_off)
1197 ret = set_peek_off(sk, val);
1198 else
1199 ret = -EOPNOTSUPP;
1200 return ret;
1201 }
1202 }
1203
1204 sockopt_lock_sock(sk);
1205
1206 switch (optname) {
1207 case SO_DEBUG:
1208 if (val && !sockopt_capable(CAP_NET_ADMIN))
1209 ret = -EACCES;
1210 else
1211 sock_valbool_flag(sk, SOCK_DBG, valbool);
1212 break;
1213 case SO_REUSEADDR:
1214 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1215 break;
1216 case SO_REUSEPORT:
1217 sk->sk_reuseport = valbool;
1218 break;
1219 case SO_DONTROUTE:
1220 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1221 sk_dst_reset(sk);
1222 break;
1223 case SO_BROADCAST:
1224 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1225 break;
1226 case SO_SNDBUF:
1227 /* Don't error on this BSD doesn't and if you think
1228 * about it this is right. Otherwise apps have to
1229 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1230 * are treated in BSD as hints
1231 */
1232 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1233set_sndbuf:
1234 /* Ensure val * 2 fits into an int, to prevent max_t()
1235 * from treating it as a negative value.
1236 */
1237 val = min_t(int, val, INT_MAX / 2);
1238 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1239 WRITE_ONCE(sk->sk_sndbuf,
1240 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1241 /* Wake up sending tasks if we upped the value. */
1242 sk->sk_write_space(sk);
1243 break;
1244
1245 case SO_SNDBUFFORCE:
1246 if (!sockopt_capable(CAP_NET_ADMIN)) {
1247 ret = -EPERM;
1248 break;
1249 }
1250
1251 /* No negative values (to prevent underflow, as val will be
1252 * multiplied by 2).
1253 */
1254 if (val < 0)
1255 val = 0;
1256 goto set_sndbuf;
1257
1258 case SO_RCVBUF:
1259 /* Don't error on this BSD doesn't and if you think
1260 * about it this is right. Otherwise apps have to
1261 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1262 * are treated in BSD as hints
1263 */
1264 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1265 break;
1266
1267 case SO_RCVBUFFORCE:
1268 if (!sockopt_capable(CAP_NET_ADMIN)) {
1269 ret = -EPERM;
1270 break;
1271 }
1272
1273 /* No negative values (to prevent underflow, as val will be
1274 * multiplied by 2).
1275 */
1276 __sock_set_rcvbuf(sk, max(val, 0));
1277 break;
1278
1279 case SO_KEEPALIVE:
1280 if (sk->sk_prot->keepalive)
1281 sk->sk_prot->keepalive(sk, valbool);
1282 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1283 break;
1284
1285 case SO_OOBINLINE:
1286 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1287 break;
1288
1289 case SO_NO_CHECK:
1290 sk->sk_no_check_tx = valbool;
1291 break;
1292
1293 case SO_LINGER:
1294 if (optlen < sizeof(ling)) {
1295 ret = -EINVAL; /* 1003.1g */
1296 break;
1297 }
1298 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1299 ret = -EFAULT;
1300 break;
1301 }
1302 if (!ling.l_onoff) {
1303 sock_reset_flag(sk, SOCK_LINGER);
1304 } else {
1305 unsigned long t_sec = ling.l_linger;
1306
1307 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1308 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1309 else
1310 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1311 sock_set_flag(sk, SOCK_LINGER);
1312 }
1313 break;
1314
1315 case SO_BSDCOMPAT:
1316 break;
1317
1318 case SO_TIMESTAMP_OLD:
1319 case SO_TIMESTAMP_NEW:
1320 case SO_TIMESTAMPNS_OLD:
1321 case SO_TIMESTAMPNS_NEW:
1322 sock_set_timestamp(sk, optname, valbool);
1323 break;
1324
1325 case SO_TIMESTAMPING_NEW:
1326 case SO_TIMESTAMPING_OLD:
1327 if (optlen == sizeof(timestamping)) {
1328 if (copy_from_sockptr(×tamping, optval,
1329 sizeof(timestamping))) {
1330 ret = -EFAULT;
1331 break;
1332 }
1333 } else {
1334 memset(×tamping, 0, sizeof(timestamping));
1335 timestamping.flags = val;
1336 }
1337 ret = sock_set_timestamping(sk, optname, timestamping);
1338 break;
1339
1340 case SO_RCVLOWAT:
1341 {
1342 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1343
1344 if (val < 0)
1345 val = INT_MAX;
1346 if (sock)
1347 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1348 if (set_rcvlowat)
1349 ret = set_rcvlowat(sk, val);
1350 else
1351 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1352 break;
1353 }
1354 case SO_RCVTIMEO_OLD:
1355 case SO_RCVTIMEO_NEW:
1356 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1357 optlen, optname == SO_RCVTIMEO_OLD);
1358 break;
1359
1360 case SO_SNDTIMEO_OLD:
1361 case SO_SNDTIMEO_NEW:
1362 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1363 optlen, optname == SO_SNDTIMEO_OLD);
1364 break;
1365
1366 case SO_ATTACH_FILTER: {
1367 struct sock_fprog fprog;
1368
1369 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1370 if (!ret)
1371 ret = sk_attach_filter(&fprog, sk);
1372 break;
1373 }
1374 case SO_ATTACH_BPF:
1375 ret = -EINVAL;
1376 if (optlen == sizeof(u32)) {
1377 u32 ufd;
1378
1379 ret = -EFAULT;
1380 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1381 break;
1382
1383 ret = sk_attach_bpf(ufd, sk);
1384 }
1385 break;
1386
1387 case SO_ATTACH_REUSEPORT_CBPF: {
1388 struct sock_fprog fprog;
1389
1390 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1391 if (!ret)
1392 ret = sk_reuseport_attach_filter(&fprog, sk);
1393 break;
1394 }
1395 case SO_ATTACH_REUSEPORT_EBPF:
1396 ret = -EINVAL;
1397 if (optlen == sizeof(u32)) {
1398 u32 ufd;
1399
1400 ret = -EFAULT;
1401 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1402 break;
1403
1404 ret = sk_reuseport_attach_bpf(ufd, sk);
1405 }
1406 break;
1407
1408 case SO_DETACH_REUSEPORT_BPF:
1409 ret = reuseport_detach_prog(sk);
1410 break;
1411
1412 case SO_DETACH_FILTER:
1413 ret = sk_detach_filter(sk);
1414 break;
1415
1416 case SO_LOCK_FILTER:
1417 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1418 ret = -EPERM;
1419 else
1420 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1421 break;
1422
1423 case SO_MARK:
1424 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1425 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1426 ret = -EPERM;
1427 break;
1428 }
1429
1430 __sock_set_mark(sk, val);
1431 break;
1432 case SO_RCVMARK:
1433 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1434 break;
1435
1436 case SO_RXQ_OVFL:
1437 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1438 break;
1439
1440 case SO_WIFI_STATUS:
1441 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1442 break;
1443
1444 case SO_NOFCS:
1445 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1446 break;
1447
1448 case SO_SELECT_ERR_QUEUE:
1449 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1450 break;
1451
1452
1453 case SO_INCOMING_CPU:
1454 reuseport_update_incoming_cpu(sk, val);
1455 break;
1456
1457 case SO_CNX_ADVICE:
1458 if (val == 1)
1459 dst_negative_advice(sk);
1460 break;
1461
1462 case SO_ZEROCOPY:
1463 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1464 if (!(sk_is_tcp(sk) ||
1465 (sk->sk_type == SOCK_DGRAM &&
1466 sk->sk_protocol == IPPROTO_UDP)))
1467 ret = -EOPNOTSUPP;
1468 } else if (sk->sk_family != PF_RDS) {
1469 ret = -EOPNOTSUPP;
1470 }
1471 if (!ret) {
1472 if (val < 0 || val > 1)
1473 ret = -EINVAL;
1474 else
1475 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1476 }
1477 break;
1478
1479 case SO_TXTIME:
1480 if (optlen != sizeof(struct sock_txtime)) {
1481 ret = -EINVAL;
1482 break;
1483 } else if (copy_from_sockptr(&sk_txtime, optval,
1484 sizeof(struct sock_txtime))) {
1485 ret = -EFAULT;
1486 break;
1487 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1488 ret = -EINVAL;
1489 break;
1490 }
1491 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1492 * scheduler has enough safe guards.
1493 */
1494 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1495 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1496 ret = -EPERM;
1497 break;
1498 }
1499 sock_valbool_flag(sk, SOCK_TXTIME, true);
1500 sk->sk_clockid = sk_txtime.clockid;
1501 sk->sk_txtime_deadline_mode =
1502 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1503 sk->sk_txtime_report_errors =
1504 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1505 break;
1506
1507 case SO_BINDTOIFINDEX:
1508 ret = sock_bindtoindex_locked(sk, val);
1509 break;
1510
1511 case SO_BUF_LOCK:
1512 if (val & ~SOCK_BUF_LOCK_MASK) {
1513 ret = -EINVAL;
1514 break;
1515 }
1516 sk->sk_userlocks = val | (sk->sk_userlocks &
1517 ~SOCK_BUF_LOCK_MASK);
1518 break;
1519
1520 case SO_RESERVE_MEM:
1521 {
1522 int delta;
1523
1524 if (val < 0) {
1525 ret = -EINVAL;
1526 break;
1527 }
1528
1529 delta = val - sk->sk_reserved_mem;
1530 if (delta < 0)
1531 sock_release_reserved_memory(sk, -delta);
1532 else
1533 ret = sock_reserve_memory(sk, delta);
1534 break;
1535 }
1536
1537 default:
1538 ret = -ENOPROTOOPT;
1539 break;
1540 }
1541 sockopt_release_sock(sk);
1542 return ret;
1543}
1544
1545int sock_setsockopt(struct socket *sock, int level, int optname,
1546 sockptr_t optval, unsigned int optlen)
1547{
1548 return sk_setsockopt(sock->sk, level, optname,
1549 optval, optlen);
1550}
1551EXPORT_SYMBOL(sock_setsockopt);
1552
1553static const struct cred *sk_get_peer_cred(struct sock *sk)
1554{
1555 const struct cred *cred;
1556
1557 spin_lock(&sk->sk_peer_lock);
1558 cred = get_cred(sk->sk_peer_cred);
1559 spin_unlock(&sk->sk_peer_lock);
1560
1561 return cred;
1562}
1563
1564static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1565 struct ucred *ucred)
1566{
1567 ucred->pid = pid_vnr(pid);
1568 ucred->uid = ucred->gid = -1;
1569 if (cred) {
1570 struct user_namespace *current_ns = current_user_ns();
1571
1572 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1573 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1574 }
1575}
1576
1577static int groups_to_user(sockptr_t dst, const struct group_info *src)
1578{
1579 struct user_namespace *user_ns = current_user_ns();
1580 int i;
1581
1582 for (i = 0; i < src->ngroups; i++) {
1583 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1584
1585 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1586 return -EFAULT;
1587 }
1588
1589 return 0;
1590}
1591
1592int sk_getsockopt(struct sock *sk, int level, int optname,
1593 sockptr_t optval, sockptr_t optlen)
1594{
1595 struct socket *sock = sk->sk_socket;
1596
1597 union {
1598 int val;
1599 u64 val64;
1600 unsigned long ulval;
1601 struct linger ling;
1602 struct old_timeval32 tm32;
1603 struct __kernel_old_timeval tm;
1604 struct __kernel_sock_timeval stm;
1605 struct sock_txtime txtime;
1606 struct so_timestamping timestamping;
1607 } v;
1608
1609 int lv = sizeof(int);
1610 int len;
1611
1612 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1613 return -EFAULT;
1614 if (len < 0)
1615 return -EINVAL;
1616
1617 memset(&v, 0, sizeof(v));
1618
1619 switch (optname) {
1620 case SO_DEBUG:
1621 v.val = sock_flag(sk, SOCK_DBG);
1622 break;
1623
1624 case SO_DONTROUTE:
1625 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1626 break;
1627
1628 case SO_BROADCAST:
1629 v.val = sock_flag(sk, SOCK_BROADCAST);
1630 break;
1631
1632 case SO_SNDBUF:
1633 v.val = READ_ONCE(sk->sk_sndbuf);
1634 break;
1635
1636 case SO_RCVBUF:
1637 v.val = READ_ONCE(sk->sk_rcvbuf);
1638 break;
1639
1640 case SO_REUSEADDR:
1641 v.val = sk->sk_reuse;
1642 break;
1643
1644 case SO_REUSEPORT:
1645 v.val = sk->sk_reuseport;
1646 break;
1647
1648 case SO_KEEPALIVE:
1649 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1650 break;
1651
1652 case SO_TYPE:
1653 v.val = sk->sk_type;
1654 break;
1655
1656 case SO_PROTOCOL:
1657 v.val = sk->sk_protocol;
1658 break;
1659
1660 case SO_DOMAIN:
1661 v.val = sk->sk_family;
1662 break;
1663
1664 case SO_ERROR:
1665 v.val = -sock_error(sk);
1666 if (v.val == 0)
1667 v.val = xchg(&sk->sk_err_soft, 0);
1668 break;
1669
1670 case SO_OOBINLINE:
1671 v.val = sock_flag(sk, SOCK_URGINLINE);
1672 break;
1673
1674 case SO_NO_CHECK:
1675 v.val = sk->sk_no_check_tx;
1676 break;
1677
1678 case SO_PRIORITY:
1679 v.val = READ_ONCE(sk->sk_priority);
1680 break;
1681
1682 case SO_LINGER:
1683 lv = sizeof(v.ling);
1684 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1685 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1686 break;
1687
1688 case SO_BSDCOMPAT:
1689 break;
1690
1691 case SO_TIMESTAMP_OLD:
1692 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1693 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1694 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1695 break;
1696
1697 case SO_TIMESTAMPNS_OLD:
1698 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1699 break;
1700
1701 case SO_TIMESTAMP_NEW:
1702 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1703 break;
1704
1705 case SO_TIMESTAMPNS_NEW:
1706 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1707 break;
1708
1709 case SO_TIMESTAMPING_OLD:
1710 case SO_TIMESTAMPING_NEW:
1711 lv = sizeof(v.timestamping);
1712 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1713 * returning the flags when they were set through the same option.
1714 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1715 */
1716 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1717 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1718 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1719 }
1720 break;
1721
1722 case SO_RCVTIMEO_OLD:
1723 case SO_RCVTIMEO_NEW:
1724 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1725 SO_RCVTIMEO_OLD == optname);
1726 break;
1727
1728 case SO_SNDTIMEO_OLD:
1729 case SO_SNDTIMEO_NEW:
1730 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1731 SO_SNDTIMEO_OLD == optname);
1732 break;
1733
1734 case SO_RCVLOWAT:
1735 v.val = READ_ONCE(sk->sk_rcvlowat);
1736 break;
1737
1738 case SO_SNDLOWAT:
1739 v.val = 1;
1740 break;
1741
1742 case SO_PASSCRED:
1743 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1744 break;
1745
1746 case SO_PASSPIDFD:
1747 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1748 break;
1749
1750 case SO_PEERCRED:
1751 {
1752 struct ucred peercred;
1753 if (len > sizeof(peercred))
1754 len = sizeof(peercred);
1755
1756 spin_lock(&sk->sk_peer_lock);
1757 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1758 spin_unlock(&sk->sk_peer_lock);
1759
1760 if (copy_to_sockptr(optval, &peercred, len))
1761 return -EFAULT;
1762 goto lenout;
1763 }
1764
1765 case SO_PEERPIDFD:
1766 {
1767 struct pid *peer_pid;
1768 struct file *pidfd_file = NULL;
1769 int pidfd;
1770
1771 if (len > sizeof(pidfd))
1772 len = sizeof(pidfd);
1773
1774 spin_lock(&sk->sk_peer_lock);
1775 peer_pid = get_pid(sk->sk_peer_pid);
1776 spin_unlock(&sk->sk_peer_lock);
1777
1778 if (!peer_pid)
1779 return -ENODATA;
1780
1781 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1782 put_pid(peer_pid);
1783 if (pidfd < 0)
1784 return pidfd;
1785
1786 if (copy_to_sockptr(optval, &pidfd, len) ||
1787 copy_to_sockptr(optlen, &len, sizeof(int))) {
1788 put_unused_fd(pidfd);
1789 fput(pidfd_file);
1790
1791 return -EFAULT;
1792 }
1793
1794 fd_install(pidfd, pidfd_file);
1795 return 0;
1796 }
1797
1798 case SO_PEERGROUPS:
1799 {
1800 const struct cred *cred;
1801 int ret, n;
1802
1803 cred = sk_get_peer_cred(sk);
1804 if (!cred)
1805 return -ENODATA;
1806
1807 n = cred->group_info->ngroups;
1808 if (len < n * sizeof(gid_t)) {
1809 len = n * sizeof(gid_t);
1810 put_cred(cred);
1811 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1812 }
1813 len = n * sizeof(gid_t);
1814
1815 ret = groups_to_user(optval, cred->group_info);
1816 put_cred(cred);
1817 if (ret)
1818 return ret;
1819 goto lenout;
1820 }
1821
1822 case SO_PEERNAME:
1823 {
1824 struct sockaddr_storage address;
1825
1826 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1827 if (lv < 0)
1828 return -ENOTCONN;
1829 if (lv < len)
1830 return -EINVAL;
1831 if (copy_to_sockptr(optval, &address, len))
1832 return -EFAULT;
1833 goto lenout;
1834 }
1835
1836 /* Dubious BSD thing... Probably nobody even uses it, but
1837 * the UNIX standard wants it for whatever reason... -DaveM
1838 */
1839 case SO_ACCEPTCONN:
1840 v.val = sk->sk_state == TCP_LISTEN;
1841 break;
1842
1843 case SO_PASSSEC:
1844 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1845 break;
1846
1847 case SO_PEERSEC:
1848 return security_socket_getpeersec_stream(sock,
1849 optval, optlen, len);
1850
1851 case SO_MARK:
1852 v.val = READ_ONCE(sk->sk_mark);
1853 break;
1854
1855 case SO_RCVMARK:
1856 v.val = sock_flag(sk, SOCK_RCVMARK);
1857 break;
1858
1859 case SO_RXQ_OVFL:
1860 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1861 break;
1862
1863 case SO_WIFI_STATUS:
1864 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1865 break;
1866
1867 case SO_PEEK_OFF:
1868 if (!READ_ONCE(sock->ops)->set_peek_off)
1869 return -EOPNOTSUPP;
1870
1871 v.val = READ_ONCE(sk->sk_peek_off);
1872 break;
1873 case SO_NOFCS:
1874 v.val = sock_flag(sk, SOCK_NOFCS);
1875 break;
1876
1877 case SO_BINDTODEVICE:
1878 return sock_getbindtodevice(sk, optval, optlen, len);
1879
1880 case SO_GET_FILTER:
1881 len = sk_get_filter(sk, optval, len);
1882 if (len < 0)
1883 return len;
1884
1885 goto lenout;
1886
1887 case SO_LOCK_FILTER:
1888 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1889 break;
1890
1891 case SO_BPF_EXTENSIONS:
1892 v.val = bpf_tell_extensions();
1893 break;
1894
1895 case SO_SELECT_ERR_QUEUE:
1896 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1897 break;
1898
1899#ifdef CONFIG_NET_RX_BUSY_POLL
1900 case SO_BUSY_POLL:
1901 v.val = READ_ONCE(sk->sk_ll_usec);
1902 break;
1903 case SO_PREFER_BUSY_POLL:
1904 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1905 break;
1906#endif
1907
1908 case SO_MAX_PACING_RATE:
1909 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1910 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1911 lv = sizeof(v.ulval);
1912 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1913 } else {
1914 /* 32bit version */
1915 v.val = min_t(unsigned long, ~0U,
1916 READ_ONCE(sk->sk_max_pacing_rate));
1917 }
1918 break;
1919
1920 case SO_INCOMING_CPU:
1921 v.val = READ_ONCE(sk->sk_incoming_cpu);
1922 break;
1923
1924 case SO_MEMINFO:
1925 {
1926 u32 meminfo[SK_MEMINFO_VARS];
1927
1928 sk_get_meminfo(sk, meminfo);
1929
1930 len = min_t(unsigned int, len, sizeof(meminfo));
1931 if (copy_to_sockptr(optval, &meminfo, len))
1932 return -EFAULT;
1933
1934 goto lenout;
1935 }
1936
1937#ifdef CONFIG_NET_RX_BUSY_POLL
1938 case SO_INCOMING_NAPI_ID:
1939 v.val = READ_ONCE(sk->sk_napi_id);
1940
1941 /* aggregate non-NAPI IDs down to 0 */
1942 if (v.val < MIN_NAPI_ID)
1943 v.val = 0;
1944
1945 break;
1946#endif
1947
1948 case SO_COOKIE:
1949 lv = sizeof(u64);
1950 if (len < lv)
1951 return -EINVAL;
1952 v.val64 = sock_gen_cookie(sk);
1953 break;
1954
1955 case SO_ZEROCOPY:
1956 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1957 break;
1958
1959 case SO_TXTIME:
1960 lv = sizeof(v.txtime);
1961 v.txtime.clockid = sk->sk_clockid;
1962 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1963 SOF_TXTIME_DEADLINE_MODE : 0;
1964 v.txtime.flags |= sk->sk_txtime_report_errors ?
1965 SOF_TXTIME_REPORT_ERRORS : 0;
1966 break;
1967
1968 case SO_BINDTOIFINDEX:
1969 v.val = READ_ONCE(sk->sk_bound_dev_if);
1970 break;
1971
1972 case SO_NETNS_COOKIE:
1973 lv = sizeof(u64);
1974 if (len != lv)
1975 return -EINVAL;
1976 v.val64 = sock_net(sk)->net_cookie;
1977 break;
1978
1979 case SO_BUF_LOCK:
1980 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1981 break;
1982
1983 case SO_RESERVE_MEM:
1984 v.val = READ_ONCE(sk->sk_reserved_mem);
1985 break;
1986
1987 case SO_TXREHASH:
1988 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1989 v.val = READ_ONCE(sk->sk_txrehash);
1990 break;
1991
1992 default:
1993 /* We implement the SO_SNDLOWAT etc to not be settable
1994 * (1003.1g 7).
1995 */
1996 return -ENOPROTOOPT;
1997 }
1998
1999 if (len > lv)
2000 len = lv;
2001 if (copy_to_sockptr(optval, &v, len))
2002 return -EFAULT;
2003lenout:
2004 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2005 return -EFAULT;
2006 return 0;
2007}
2008
2009/*
2010 * Initialize an sk_lock.
2011 *
2012 * (We also register the sk_lock with the lock validator.)
2013 */
2014static inline void sock_lock_init(struct sock *sk)
2015{
2016 if (sk->sk_kern_sock)
2017 sock_lock_init_class_and_name(
2018 sk,
2019 af_family_kern_slock_key_strings[sk->sk_family],
2020 af_family_kern_slock_keys + sk->sk_family,
2021 af_family_kern_key_strings[sk->sk_family],
2022 af_family_kern_keys + sk->sk_family);
2023 else
2024 sock_lock_init_class_and_name(
2025 sk,
2026 af_family_slock_key_strings[sk->sk_family],
2027 af_family_slock_keys + sk->sk_family,
2028 af_family_key_strings[sk->sk_family],
2029 af_family_keys + sk->sk_family);
2030}
2031
2032/*
2033 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2034 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2035 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2036 */
2037static void sock_copy(struct sock *nsk, const struct sock *osk)
2038{
2039 const struct proto *prot = READ_ONCE(osk->sk_prot);
2040#ifdef CONFIG_SECURITY_NETWORK
2041 void *sptr = nsk->sk_security;
2042#endif
2043
2044 /* If we move sk_tx_queue_mapping out of the private section,
2045 * we must check if sk_tx_queue_clear() is called after
2046 * sock_copy() in sk_clone_lock().
2047 */
2048 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2049 offsetof(struct sock, sk_dontcopy_begin) ||
2050 offsetof(struct sock, sk_tx_queue_mapping) >=
2051 offsetof(struct sock, sk_dontcopy_end));
2052
2053 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2054
2055 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2056 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2057
2058#ifdef CONFIG_SECURITY_NETWORK
2059 nsk->sk_security = sptr;
2060 security_sk_clone(osk, nsk);
2061#endif
2062}
2063
2064static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2065 int family)
2066{
2067 struct sock *sk;
2068 struct kmem_cache *slab;
2069
2070 slab = prot->slab;
2071 if (slab != NULL) {
2072 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2073 if (!sk)
2074 return sk;
2075 if (want_init_on_alloc(priority))
2076 sk_prot_clear_nulls(sk, prot->obj_size);
2077 } else
2078 sk = kmalloc(prot->obj_size, priority);
2079
2080 if (sk != NULL) {
2081 if (security_sk_alloc(sk, family, priority))
2082 goto out_free;
2083
2084 if (!try_module_get(prot->owner))
2085 goto out_free_sec;
2086 }
2087
2088 return sk;
2089
2090out_free_sec:
2091 security_sk_free(sk);
2092out_free:
2093 if (slab != NULL)
2094 kmem_cache_free(slab, sk);
2095 else
2096 kfree(sk);
2097 return NULL;
2098}
2099
2100static void sk_prot_free(struct proto *prot, struct sock *sk)
2101{
2102 struct kmem_cache *slab;
2103 struct module *owner;
2104
2105 owner = prot->owner;
2106 slab = prot->slab;
2107
2108 cgroup_sk_free(&sk->sk_cgrp_data);
2109 mem_cgroup_sk_free(sk);
2110 security_sk_free(sk);
2111 if (slab != NULL)
2112 kmem_cache_free(slab, sk);
2113 else
2114 kfree(sk);
2115 module_put(owner);
2116}
2117
2118/**
2119 * sk_alloc - All socket objects are allocated here
2120 * @net: the applicable net namespace
2121 * @family: protocol family
2122 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2123 * @prot: struct proto associated with this new sock instance
2124 * @kern: is this to be a kernel socket?
2125 */
2126struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2127 struct proto *prot, int kern)
2128{
2129 struct sock *sk;
2130
2131 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2132 if (sk) {
2133 sk->sk_family = family;
2134 /*
2135 * See comment in struct sock definition to understand
2136 * why we need sk_prot_creator -acme
2137 */
2138 sk->sk_prot = sk->sk_prot_creator = prot;
2139 sk->sk_kern_sock = kern;
2140 sock_lock_init(sk);
2141 sk->sk_net_refcnt = kern ? 0 : 1;
2142 if (likely(sk->sk_net_refcnt)) {
2143 get_net_track(net, &sk->ns_tracker, priority);
2144 sock_inuse_add(net, 1);
2145 } else {
2146 __netns_tracker_alloc(net, &sk->ns_tracker,
2147 false, priority);
2148 }
2149
2150 sock_net_set(sk, net);
2151 refcount_set(&sk->sk_wmem_alloc, 1);
2152
2153 mem_cgroup_sk_alloc(sk);
2154 cgroup_sk_alloc(&sk->sk_cgrp_data);
2155 sock_update_classid(&sk->sk_cgrp_data);
2156 sock_update_netprioidx(&sk->sk_cgrp_data);
2157 sk_tx_queue_clear(sk);
2158 }
2159
2160 return sk;
2161}
2162EXPORT_SYMBOL(sk_alloc);
2163
2164/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2165 * grace period. This is the case for UDP sockets and TCP listeners.
2166 */
2167static void __sk_destruct(struct rcu_head *head)
2168{
2169 struct sock *sk = container_of(head, struct sock, sk_rcu);
2170 struct sk_filter *filter;
2171
2172 if (sk->sk_destruct)
2173 sk->sk_destruct(sk);
2174
2175 filter = rcu_dereference_check(sk->sk_filter,
2176 refcount_read(&sk->sk_wmem_alloc) == 0);
2177 if (filter) {
2178 sk_filter_uncharge(sk, filter);
2179 RCU_INIT_POINTER(sk->sk_filter, NULL);
2180 }
2181
2182 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2183
2184#ifdef CONFIG_BPF_SYSCALL
2185 bpf_sk_storage_free(sk);
2186#endif
2187
2188 if (atomic_read(&sk->sk_omem_alloc))
2189 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2190 __func__, atomic_read(&sk->sk_omem_alloc));
2191
2192 if (sk->sk_frag.page) {
2193 put_page(sk->sk_frag.page);
2194 sk->sk_frag.page = NULL;
2195 }
2196
2197 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2198 put_cred(sk->sk_peer_cred);
2199 put_pid(sk->sk_peer_pid);
2200
2201 if (likely(sk->sk_net_refcnt))
2202 put_net_track(sock_net(sk), &sk->ns_tracker);
2203 else
2204 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2205
2206 sk_prot_free(sk->sk_prot_creator, sk);
2207}
2208
2209void sk_destruct(struct sock *sk)
2210{
2211 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2212
2213 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2214 reuseport_detach_sock(sk);
2215 use_call_rcu = true;
2216 }
2217
2218 if (use_call_rcu)
2219 call_rcu(&sk->sk_rcu, __sk_destruct);
2220 else
2221 __sk_destruct(&sk->sk_rcu);
2222}
2223
2224static void __sk_free(struct sock *sk)
2225{
2226 if (likely(sk->sk_net_refcnt))
2227 sock_inuse_add(sock_net(sk), -1);
2228
2229 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2230 sock_diag_broadcast_destroy(sk);
2231 else
2232 sk_destruct(sk);
2233}
2234
2235void sk_free(struct sock *sk)
2236{
2237 /*
2238 * We subtract one from sk_wmem_alloc and can know if
2239 * some packets are still in some tx queue.
2240 * If not null, sock_wfree() will call __sk_free(sk) later
2241 */
2242 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2243 __sk_free(sk);
2244}
2245EXPORT_SYMBOL(sk_free);
2246
2247static void sk_init_common(struct sock *sk)
2248{
2249 skb_queue_head_init(&sk->sk_receive_queue);
2250 skb_queue_head_init(&sk->sk_write_queue);
2251 skb_queue_head_init(&sk->sk_error_queue);
2252
2253 rwlock_init(&sk->sk_callback_lock);
2254 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2255 af_rlock_keys + sk->sk_family,
2256 af_family_rlock_key_strings[sk->sk_family]);
2257 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2258 af_wlock_keys + sk->sk_family,
2259 af_family_wlock_key_strings[sk->sk_family]);
2260 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2261 af_elock_keys + sk->sk_family,
2262 af_family_elock_key_strings[sk->sk_family]);
2263 lockdep_set_class_and_name(&sk->sk_callback_lock,
2264 af_callback_keys + sk->sk_family,
2265 af_family_clock_key_strings[sk->sk_family]);
2266}
2267
2268/**
2269 * sk_clone_lock - clone a socket, and lock its clone
2270 * @sk: the socket to clone
2271 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2272 *
2273 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2274 */
2275struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2276{
2277 struct proto *prot = READ_ONCE(sk->sk_prot);
2278 struct sk_filter *filter;
2279 bool is_charged = true;
2280 struct sock *newsk;
2281
2282 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2283 if (!newsk)
2284 goto out;
2285
2286 sock_copy(newsk, sk);
2287
2288 newsk->sk_prot_creator = prot;
2289
2290 /* SANITY */
2291 if (likely(newsk->sk_net_refcnt)) {
2292 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2293 sock_inuse_add(sock_net(newsk), 1);
2294 } else {
2295 /* Kernel sockets are not elevating the struct net refcount.
2296 * Instead, use a tracker to more easily detect if a layer
2297 * is not properly dismantling its kernel sockets at netns
2298 * destroy time.
2299 */
2300 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2301 false, priority);
2302 }
2303 sk_node_init(&newsk->sk_node);
2304 sock_lock_init(newsk);
2305 bh_lock_sock(newsk);
2306 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2307 newsk->sk_backlog.len = 0;
2308
2309 atomic_set(&newsk->sk_rmem_alloc, 0);
2310
2311 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2312 refcount_set(&newsk->sk_wmem_alloc, 1);
2313
2314 atomic_set(&newsk->sk_omem_alloc, 0);
2315 sk_init_common(newsk);
2316
2317 newsk->sk_dst_cache = NULL;
2318 newsk->sk_dst_pending_confirm = 0;
2319 newsk->sk_wmem_queued = 0;
2320 newsk->sk_forward_alloc = 0;
2321 newsk->sk_reserved_mem = 0;
2322 atomic_set(&newsk->sk_drops, 0);
2323 newsk->sk_send_head = NULL;
2324 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2325 atomic_set(&newsk->sk_zckey, 0);
2326
2327 sock_reset_flag(newsk, SOCK_DONE);
2328
2329 /* sk->sk_memcg will be populated at accept() time */
2330 newsk->sk_memcg = NULL;
2331
2332 cgroup_sk_clone(&newsk->sk_cgrp_data);
2333
2334 rcu_read_lock();
2335 filter = rcu_dereference(sk->sk_filter);
2336 if (filter != NULL)
2337 /* though it's an empty new sock, the charging may fail
2338 * if sysctl_optmem_max was changed between creation of
2339 * original socket and cloning
2340 */
2341 is_charged = sk_filter_charge(newsk, filter);
2342 RCU_INIT_POINTER(newsk->sk_filter, filter);
2343 rcu_read_unlock();
2344
2345 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2346 /* We need to make sure that we don't uncharge the new
2347 * socket if we couldn't charge it in the first place
2348 * as otherwise we uncharge the parent's filter.
2349 */
2350 if (!is_charged)
2351 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2352 sk_free_unlock_clone(newsk);
2353 newsk = NULL;
2354 goto out;
2355 }
2356 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2357
2358 if (bpf_sk_storage_clone(sk, newsk)) {
2359 sk_free_unlock_clone(newsk);
2360 newsk = NULL;
2361 goto out;
2362 }
2363
2364 /* Clear sk_user_data if parent had the pointer tagged
2365 * as not suitable for copying when cloning.
2366 */
2367 if (sk_user_data_is_nocopy(newsk))
2368 newsk->sk_user_data = NULL;
2369
2370 newsk->sk_err = 0;
2371 newsk->sk_err_soft = 0;
2372 newsk->sk_priority = 0;
2373 newsk->sk_incoming_cpu = raw_smp_processor_id();
2374
2375 /* Before updating sk_refcnt, we must commit prior changes to memory
2376 * (Documentation/RCU/rculist_nulls.rst for details)
2377 */
2378 smp_wmb();
2379 refcount_set(&newsk->sk_refcnt, 2);
2380
2381 sk_set_socket(newsk, NULL);
2382 sk_tx_queue_clear(newsk);
2383 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2384
2385 if (newsk->sk_prot->sockets_allocated)
2386 sk_sockets_allocated_inc(newsk);
2387
2388 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2389 net_enable_timestamp();
2390out:
2391 return newsk;
2392}
2393EXPORT_SYMBOL_GPL(sk_clone_lock);
2394
2395void sk_free_unlock_clone(struct sock *sk)
2396{
2397 /* It is still raw copy of parent, so invalidate
2398 * destructor and make plain sk_free() */
2399 sk->sk_destruct = NULL;
2400 bh_unlock_sock(sk);
2401 sk_free(sk);
2402}
2403EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2404
2405static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2406{
2407 bool is_ipv6 = false;
2408 u32 max_size;
2409
2410#if IS_ENABLED(CONFIG_IPV6)
2411 is_ipv6 = (sk->sk_family == AF_INET6 &&
2412 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2413#endif
2414 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2415 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2416 READ_ONCE(dst->dev->gso_ipv4_max_size);
2417 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2418 max_size = GSO_LEGACY_MAX_SIZE;
2419
2420 return max_size - (MAX_TCP_HEADER + 1);
2421}
2422
2423void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2424{
2425 u32 max_segs = 1;
2426
2427 sk->sk_route_caps = dst->dev->features;
2428 if (sk_is_tcp(sk))
2429 sk->sk_route_caps |= NETIF_F_GSO;
2430 if (sk->sk_route_caps & NETIF_F_GSO)
2431 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2432 if (unlikely(sk->sk_gso_disabled))
2433 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2434 if (sk_can_gso(sk)) {
2435 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2436 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2437 } else {
2438 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2439 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2440 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2441 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2442 }
2443 }
2444 sk->sk_gso_max_segs = max_segs;
2445 sk_dst_set(sk, dst);
2446}
2447EXPORT_SYMBOL_GPL(sk_setup_caps);
2448
2449/*
2450 * Simple resource managers for sockets.
2451 */
2452
2453
2454/*
2455 * Write buffer destructor automatically called from kfree_skb.
2456 */
2457void sock_wfree(struct sk_buff *skb)
2458{
2459 struct sock *sk = skb->sk;
2460 unsigned int len = skb->truesize;
2461 bool free;
2462
2463 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2464 if (sock_flag(sk, SOCK_RCU_FREE) &&
2465 sk->sk_write_space == sock_def_write_space) {
2466 rcu_read_lock();
2467 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2468 sock_def_write_space_wfree(sk);
2469 rcu_read_unlock();
2470 if (unlikely(free))
2471 __sk_free(sk);
2472 return;
2473 }
2474
2475 /*
2476 * Keep a reference on sk_wmem_alloc, this will be released
2477 * after sk_write_space() call
2478 */
2479 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2480 sk->sk_write_space(sk);
2481 len = 1;
2482 }
2483 /*
2484 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2485 * could not do because of in-flight packets
2486 */
2487 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2488 __sk_free(sk);
2489}
2490EXPORT_SYMBOL(sock_wfree);
2491
2492/* This variant of sock_wfree() is used by TCP,
2493 * since it sets SOCK_USE_WRITE_QUEUE.
2494 */
2495void __sock_wfree(struct sk_buff *skb)
2496{
2497 struct sock *sk = skb->sk;
2498
2499 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2500 __sk_free(sk);
2501}
2502
2503void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2504{
2505 skb_orphan(skb);
2506 skb->sk = sk;
2507#ifdef CONFIG_INET
2508 if (unlikely(!sk_fullsock(sk))) {
2509 skb->destructor = sock_edemux;
2510 sock_hold(sk);
2511 return;
2512 }
2513#endif
2514 skb->destructor = sock_wfree;
2515 skb_set_hash_from_sk(skb, sk);
2516 /*
2517 * We used to take a refcount on sk, but following operation
2518 * is enough to guarantee sk_free() wont free this sock until
2519 * all in-flight packets are completed
2520 */
2521 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2522}
2523EXPORT_SYMBOL(skb_set_owner_w);
2524
2525static bool can_skb_orphan_partial(const struct sk_buff *skb)
2526{
2527#ifdef CONFIG_TLS_DEVICE
2528 /* Drivers depend on in-order delivery for crypto offload,
2529 * partial orphan breaks out-of-order-OK logic.
2530 */
2531 if (skb->decrypted)
2532 return false;
2533#endif
2534 return (skb->destructor == sock_wfree ||
2535 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2536}
2537
2538/* This helper is used by netem, as it can hold packets in its
2539 * delay queue. We want to allow the owner socket to send more
2540 * packets, as if they were already TX completed by a typical driver.
2541 * But we also want to keep skb->sk set because some packet schedulers
2542 * rely on it (sch_fq for example).
2543 */
2544void skb_orphan_partial(struct sk_buff *skb)
2545{
2546 if (skb_is_tcp_pure_ack(skb))
2547 return;
2548
2549 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2550 return;
2551
2552 skb_orphan(skb);
2553}
2554EXPORT_SYMBOL(skb_orphan_partial);
2555
2556/*
2557 * Read buffer destructor automatically called from kfree_skb.
2558 */
2559void sock_rfree(struct sk_buff *skb)
2560{
2561 struct sock *sk = skb->sk;
2562 unsigned int len = skb->truesize;
2563
2564 atomic_sub(len, &sk->sk_rmem_alloc);
2565 sk_mem_uncharge(sk, len);
2566}
2567EXPORT_SYMBOL(sock_rfree);
2568
2569/*
2570 * Buffer destructor for skbs that are not used directly in read or write
2571 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2572 */
2573void sock_efree(struct sk_buff *skb)
2574{
2575 sock_put(skb->sk);
2576}
2577EXPORT_SYMBOL(sock_efree);
2578
2579/* Buffer destructor for prefetch/receive path where reference count may
2580 * not be held, e.g. for listen sockets.
2581 */
2582#ifdef CONFIG_INET
2583void sock_pfree(struct sk_buff *skb)
2584{
2585 if (sk_is_refcounted(skb->sk))
2586 sock_gen_put(skb->sk);
2587}
2588EXPORT_SYMBOL(sock_pfree);
2589#endif /* CONFIG_INET */
2590
2591kuid_t sock_i_uid(struct sock *sk)
2592{
2593 kuid_t uid;
2594
2595 read_lock_bh(&sk->sk_callback_lock);
2596 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2597 read_unlock_bh(&sk->sk_callback_lock);
2598 return uid;
2599}
2600EXPORT_SYMBOL(sock_i_uid);
2601
2602unsigned long __sock_i_ino(struct sock *sk)
2603{
2604 unsigned long ino;
2605
2606 read_lock(&sk->sk_callback_lock);
2607 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2608 read_unlock(&sk->sk_callback_lock);
2609 return ino;
2610}
2611EXPORT_SYMBOL(__sock_i_ino);
2612
2613unsigned long sock_i_ino(struct sock *sk)
2614{
2615 unsigned long ino;
2616
2617 local_bh_disable();
2618 ino = __sock_i_ino(sk);
2619 local_bh_enable();
2620 return ino;
2621}
2622EXPORT_SYMBOL(sock_i_ino);
2623
2624/*
2625 * Allocate a skb from the socket's send buffer.
2626 */
2627struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2628 gfp_t priority)
2629{
2630 if (force ||
2631 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2632 struct sk_buff *skb = alloc_skb(size, priority);
2633
2634 if (skb) {
2635 skb_set_owner_w(skb, sk);
2636 return skb;
2637 }
2638 }
2639 return NULL;
2640}
2641EXPORT_SYMBOL(sock_wmalloc);
2642
2643static void sock_ofree(struct sk_buff *skb)
2644{
2645 struct sock *sk = skb->sk;
2646
2647 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2648}
2649
2650struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2651 gfp_t priority)
2652{
2653 struct sk_buff *skb;
2654
2655 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2656 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2657 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2658 return NULL;
2659
2660 skb = alloc_skb(size, priority);
2661 if (!skb)
2662 return NULL;
2663
2664 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2665 skb->sk = sk;
2666 skb->destructor = sock_ofree;
2667 return skb;
2668}
2669
2670/*
2671 * Allocate a memory block from the socket's option memory buffer.
2672 */
2673void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2674{
2675 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2676
2677 if ((unsigned int)size <= optmem_max &&
2678 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2679 void *mem;
2680 /* First do the add, to avoid the race if kmalloc
2681 * might sleep.
2682 */
2683 atomic_add(size, &sk->sk_omem_alloc);
2684 mem = kmalloc(size, priority);
2685 if (mem)
2686 return mem;
2687 atomic_sub(size, &sk->sk_omem_alloc);
2688 }
2689 return NULL;
2690}
2691EXPORT_SYMBOL(sock_kmalloc);
2692
2693/* Free an option memory block. Note, we actually want the inline
2694 * here as this allows gcc to detect the nullify and fold away the
2695 * condition entirely.
2696 */
2697static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2698 const bool nullify)
2699{
2700 if (WARN_ON_ONCE(!mem))
2701 return;
2702 if (nullify)
2703 kfree_sensitive(mem);
2704 else
2705 kfree(mem);
2706 atomic_sub(size, &sk->sk_omem_alloc);
2707}
2708
2709void sock_kfree_s(struct sock *sk, void *mem, int size)
2710{
2711 __sock_kfree_s(sk, mem, size, false);
2712}
2713EXPORT_SYMBOL(sock_kfree_s);
2714
2715void sock_kzfree_s(struct sock *sk, void *mem, int size)
2716{
2717 __sock_kfree_s(sk, mem, size, true);
2718}
2719EXPORT_SYMBOL(sock_kzfree_s);
2720
2721/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2722 I think, these locks should be removed for datagram sockets.
2723 */
2724static long sock_wait_for_wmem(struct sock *sk, long timeo)
2725{
2726 DEFINE_WAIT(wait);
2727
2728 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2729 for (;;) {
2730 if (!timeo)
2731 break;
2732 if (signal_pending(current))
2733 break;
2734 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2735 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2736 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2737 break;
2738 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2739 break;
2740 if (READ_ONCE(sk->sk_err))
2741 break;
2742 timeo = schedule_timeout(timeo);
2743 }
2744 finish_wait(sk_sleep(sk), &wait);
2745 return timeo;
2746}
2747
2748
2749/*
2750 * Generic send/receive buffer handlers
2751 */
2752
2753struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2754 unsigned long data_len, int noblock,
2755 int *errcode, int max_page_order)
2756{
2757 struct sk_buff *skb;
2758 long timeo;
2759 int err;
2760
2761 timeo = sock_sndtimeo(sk, noblock);
2762 for (;;) {
2763 err = sock_error(sk);
2764 if (err != 0)
2765 goto failure;
2766
2767 err = -EPIPE;
2768 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2769 goto failure;
2770
2771 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2772 break;
2773
2774 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2775 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2776 err = -EAGAIN;
2777 if (!timeo)
2778 goto failure;
2779 if (signal_pending(current))
2780 goto interrupted;
2781 timeo = sock_wait_for_wmem(sk, timeo);
2782 }
2783 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2784 errcode, sk->sk_allocation);
2785 if (skb)
2786 skb_set_owner_w(skb, sk);
2787 return skb;
2788
2789interrupted:
2790 err = sock_intr_errno(timeo);
2791failure:
2792 *errcode = err;
2793 return NULL;
2794}
2795EXPORT_SYMBOL(sock_alloc_send_pskb);
2796
2797int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2798 struct sockcm_cookie *sockc)
2799{
2800 u32 tsflags;
2801
2802 switch (cmsg->cmsg_type) {
2803 case SO_MARK:
2804 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2805 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2806 return -EPERM;
2807 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2808 return -EINVAL;
2809 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2810 break;
2811 case SO_TIMESTAMPING_OLD:
2812 case SO_TIMESTAMPING_NEW:
2813 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2814 return -EINVAL;
2815
2816 tsflags = *(u32 *)CMSG_DATA(cmsg);
2817 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2818 return -EINVAL;
2819
2820 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2821 sockc->tsflags |= tsflags;
2822 break;
2823 case SCM_TXTIME:
2824 if (!sock_flag(sk, SOCK_TXTIME))
2825 return -EINVAL;
2826 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2827 return -EINVAL;
2828 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2829 break;
2830 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2831 case SCM_RIGHTS:
2832 case SCM_CREDENTIALS:
2833 break;
2834 default:
2835 return -EINVAL;
2836 }
2837 return 0;
2838}
2839EXPORT_SYMBOL(__sock_cmsg_send);
2840
2841int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2842 struct sockcm_cookie *sockc)
2843{
2844 struct cmsghdr *cmsg;
2845 int ret;
2846
2847 for_each_cmsghdr(cmsg, msg) {
2848 if (!CMSG_OK(msg, cmsg))
2849 return -EINVAL;
2850 if (cmsg->cmsg_level != SOL_SOCKET)
2851 continue;
2852 ret = __sock_cmsg_send(sk, cmsg, sockc);
2853 if (ret)
2854 return ret;
2855 }
2856 return 0;
2857}
2858EXPORT_SYMBOL(sock_cmsg_send);
2859
2860static void sk_enter_memory_pressure(struct sock *sk)
2861{
2862 if (!sk->sk_prot->enter_memory_pressure)
2863 return;
2864
2865 sk->sk_prot->enter_memory_pressure(sk);
2866}
2867
2868static void sk_leave_memory_pressure(struct sock *sk)
2869{
2870 if (sk->sk_prot->leave_memory_pressure) {
2871 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2872 tcp_leave_memory_pressure, sk);
2873 } else {
2874 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2875
2876 if (memory_pressure && READ_ONCE(*memory_pressure))
2877 WRITE_ONCE(*memory_pressure, 0);
2878 }
2879}
2880
2881DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2882
2883/**
2884 * skb_page_frag_refill - check that a page_frag contains enough room
2885 * @sz: minimum size of the fragment we want to get
2886 * @pfrag: pointer to page_frag
2887 * @gfp: priority for memory allocation
2888 *
2889 * Note: While this allocator tries to use high order pages, there is
2890 * no guarantee that allocations succeed. Therefore, @sz MUST be
2891 * less or equal than PAGE_SIZE.
2892 */
2893bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2894{
2895 if (pfrag->page) {
2896 if (page_ref_count(pfrag->page) == 1) {
2897 pfrag->offset = 0;
2898 return true;
2899 }
2900 if (pfrag->offset + sz <= pfrag->size)
2901 return true;
2902 put_page(pfrag->page);
2903 }
2904
2905 pfrag->offset = 0;
2906 if (SKB_FRAG_PAGE_ORDER &&
2907 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2908 /* Avoid direct reclaim but allow kswapd to wake */
2909 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2910 __GFP_COMP | __GFP_NOWARN |
2911 __GFP_NORETRY,
2912 SKB_FRAG_PAGE_ORDER);
2913 if (likely(pfrag->page)) {
2914 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2915 return true;
2916 }
2917 }
2918 pfrag->page = alloc_page(gfp);
2919 if (likely(pfrag->page)) {
2920 pfrag->size = PAGE_SIZE;
2921 return true;
2922 }
2923 return false;
2924}
2925EXPORT_SYMBOL(skb_page_frag_refill);
2926
2927bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2928{
2929 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2930 return true;
2931
2932 sk_enter_memory_pressure(sk);
2933 sk_stream_moderate_sndbuf(sk);
2934 return false;
2935}
2936EXPORT_SYMBOL(sk_page_frag_refill);
2937
2938void __lock_sock(struct sock *sk)
2939 __releases(&sk->sk_lock.slock)
2940 __acquires(&sk->sk_lock.slock)
2941{
2942 DEFINE_WAIT(wait);
2943
2944 for (;;) {
2945 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2946 TASK_UNINTERRUPTIBLE);
2947 spin_unlock_bh(&sk->sk_lock.slock);
2948 schedule();
2949 spin_lock_bh(&sk->sk_lock.slock);
2950 if (!sock_owned_by_user(sk))
2951 break;
2952 }
2953 finish_wait(&sk->sk_lock.wq, &wait);
2954}
2955
2956void __release_sock(struct sock *sk)
2957 __releases(&sk->sk_lock.slock)
2958 __acquires(&sk->sk_lock.slock)
2959{
2960 struct sk_buff *skb, *next;
2961
2962 while ((skb = sk->sk_backlog.head) != NULL) {
2963 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2964
2965 spin_unlock_bh(&sk->sk_lock.slock);
2966
2967 do {
2968 next = skb->next;
2969 prefetch(next);
2970 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2971 skb_mark_not_on_list(skb);
2972 sk_backlog_rcv(sk, skb);
2973
2974 cond_resched();
2975
2976 skb = next;
2977 } while (skb != NULL);
2978
2979 spin_lock_bh(&sk->sk_lock.slock);
2980 }
2981
2982 /*
2983 * Doing the zeroing here guarantee we can not loop forever
2984 * while a wild producer attempts to flood us.
2985 */
2986 sk->sk_backlog.len = 0;
2987}
2988
2989void __sk_flush_backlog(struct sock *sk)
2990{
2991 spin_lock_bh(&sk->sk_lock.slock);
2992 __release_sock(sk);
2993
2994 if (sk->sk_prot->release_cb)
2995 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
2996 tcp_release_cb, sk);
2997
2998 spin_unlock_bh(&sk->sk_lock.slock);
2999}
3000EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3001
3002/**
3003 * sk_wait_data - wait for data to arrive at sk_receive_queue
3004 * @sk: sock to wait on
3005 * @timeo: for how long
3006 * @skb: last skb seen on sk_receive_queue
3007 *
3008 * Now socket state including sk->sk_err is changed only under lock,
3009 * hence we may omit checks after joining wait queue.
3010 * We check receive queue before schedule() only as optimization;
3011 * it is very likely that release_sock() added new data.
3012 */
3013int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3014{
3015 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3016 int rc;
3017
3018 add_wait_queue(sk_sleep(sk), &wait);
3019 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3020 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3021 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3022 remove_wait_queue(sk_sleep(sk), &wait);
3023 return rc;
3024}
3025EXPORT_SYMBOL(sk_wait_data);
3026
3027/**
3028 * __sk_mem_raise_allocated - increase memory_allocated
3029 * @sk: socket
3030 * @size: memory size to allocate
3031 * @amt: pages to allocate
3032 * @kind: allocation type
3033 *
3034 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3035 *
3036 * Unlike the globally shared limits among the sockets under same protocol,
3037 * consuming the budget of a memcg won't have direct effect on other ones.
3038 * So be optimistic about memcg's tolerance, and leave the callers to decide
3039 * whether or not to raise allocated through sk_under_memory_pressure() or
3040 * its variants.
3041 */
3042int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3043{
3044 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3045 struct proto *prot = sk->sk_prot;
3046 bool charged = false;
3047 long allocated;
3048
3049 sk_memory_allocated_add(sk, amt);
3050 allocated = sk_memory_allocated(sk);
3051
3052 if (memcg) {
3053 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3054 goto suppress_allocation;
3055 charged = true;
3056 }
3057
3058 /* Under limit. */
3059 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3060 sk_leave_memory_pressure(sk);
3061 return 1;
3062 }
3063
3064 /* Under pressure. */
3065 if (allocated > sk_prot_mem_limits(sk, 1))
3066 sk_enter_memory_pressure(sk);
3067
3068 /* Over hard limit. */
3069 if (allocated > sk_prot_mem_limits(sk, 2))
3070 goto suppress_allocation;
3071
3072 /* Guarantee minimum buffer size under pressure (either global
3073 * or memcg) to make sure features described in RFC 7323 (TCP
3074 * Extensions for High Performance) work properly.
3075 *
3076 * This rule does NOT stand when exceeds global or memcg's hard
3077 * limit, or else a DoS attack can be taken place by spawning
3078 * lots of sockets whose usage are under minimum buffer size.
3079 */
3080 if (kind == SK_MEM_RECV) {
3081 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3082 return 1;
3083
3084 } else { /* SK_MEM_SEND */
3085 int wmem0 = sk_get_wmem0(sk, prot);
3086
3087 if (sk->sk_type == SOCK_STREAM) {
3088 if (sk->sk_wmem_queued < wmem0)
3089 return 1;
3090 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3091 return 1;
3092 }
3093 }
3094
3095 if (sk_has_memory_pressure(sk)) {
3096 u64 alloc;
3097
3098 /* The following 'average' heuristic is within the
3099 * scope of global accounting, so it only makes
3100 * sense for global memory pressure.
3101 */
3102 if (!sk_under_global_memory_pressure(sk))
3103 return 1;
3104
3105 /* Try to be fair among all the sockets under global
3106 * pressure by allowing the ones that below average
3107 * usage to raise.
3108 */
3109 alloc = sk_sockets_allocated_read_positive(sk);
3110 if (sk_prot_mem_limits(sk, 2) > alloc *
3111 sk_mem_pages(sk->sk_wmem_queued +
3112 atomic_read(&sk->sk_rmem_alloc) +
3113 sk->sk_forward_alloc))
3114 return 1;
3115 }
3116
3117suppress_allocation:
3118
3119 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3120 sk_stream_moderate_sndbuf(sk);
3121
3122 /* Fail only if socket is _under_ its sndbuf.
3123 * In this case we cannot block, so that we have to fail.
3124 */
3125 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3126 /* Force charge with __GFP_NOFAIL */
3127 if (memcg && !charged) {
3128 mem_cgroup_charge_skmem(memcg, amt,
3129 gfp_memcg_charge() | __GFP_NOFAIL);
3130 }
3131 return 1;
3132 }
3133 }
3134
3135 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3136 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3137
3138 sk_memory_allocated_sub(sk, amt);
3139
3140 if (charged)
3141 mem_cgroup_uncharge_skmem(memcg, amt);
3142
3143 return 0;
3144}
3145
3146/**
3147 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3148 * @sk: socket
3149 * @size: memory size to allocate
3150 * @kind: allocation type
3151 *
3152 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3153 * rmem allocation. This function assumes that protocols which have
3154 * memory_pressure use sk_wmem_queued as write buffer accounting.
3155 */
3156int __sk_mem_schedule(struct sock *sk, int size, int kind)
3157{
3158 int ret, amt = sk_mem_pages(size);
3159
3160 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3161 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3162 if (!ret)
3163 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3164 return ret;
3165}
3166EXPORT_SYMBOL(__sk_mem_schedule);
3167
3168/**
3169 * __sk_mem_reduce_allocated - reclaim memory_allocated
3170 * @sk: socket
3171 * @amount: number of quanta
3172 *
3173 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3174 */
3175void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3176{
3177 sk_memory_allocated_sub(sk, amount);
3178
3179 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3180 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3181
3182 if (sk_under_global_memory_pressure(sk) &&
3183 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3184 sk_leave_memory_pressure(sk);
3185}
3186
3187/**
3188 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3189 * @sk: socket
3190 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3191 */
3192void __sk_mem_reclaim(struct sock *sk, int amount)
3193{
3194 amount >>= PAGE_SHIFT;
3195 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3196 __sk_mem_reduce_allocated(sk, amount);
3197}
3198EXPORT_SYMBOL(__sk_mem_reclaim);
3199
3200int sk_set_peek_off(struct sock *sk, int val)
3201{
3202 WRITE_ONCE(sk->sk_peek_off, val);
3203 return 0;
3204}
3205EXPORT_SYMBOL_GPL(sk_set_peek_off);
3206
3207/*
3208 * Set of default routines for initialising struct proto_ops when
3209 * the protocol does not support a particular function. In certain
3210 * cases where it makes no sense for a protocol to have a "do nothing"
3211 * function, some default processing is provided.
3212 */
3213
3214int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3215{
3216 return -EOPNOTSUPP;
3217}
3218EXPORT_SYMBOL(sock_no_bind);
3219
3220int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3221 int len, int flags)
3222{
3223 return -EOPNOTSUPP;
3224}
3225EXPORT_SYMBOL(sock_no_connect);
3226
3227int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3228{
3229 return -EOPNOTSUPP;
3230}
3231EXPORT_SYMBOL(sock_no_socketpair);
3232
3233int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3234 bool kern)
3235{
3236 return -EOPNOTSUPP;
3237}
3238EXPORT_SYMBOL(sock_no_accept);
3239
3240int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3241 int peer)
3242{
3243 return -EOPNOTSUPP;
3244}
3245EXPORT_SYMBOL(sock_no_getname);
3246
3247int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3248{
3249 return -EOPNOTSUPP;
3250}
3251EXPORT_SYMBOL(sock_no_ioctl);
3252
3253int sock_no_listen(struct socket *sock, int backlog)
3254{
3255 return -EOPNOTSUPP;
3256}
3257EXPORT_SYMBOL(sock_no_listen);
3258
3259int sock_no_shutdown(struct socket *sock, int how)
3260{
3261 return -EOPNOTSUPP;
3262}
3263EXPORT_SYMBOL(sock_no_shutdown);
3264
3265int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3266{
3267 return -EOPNOTSUPP;
3268}
3269EXPORT_SYMBOL(sock_no_sendmsg);
3270
3271int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3272{
3273 return -EOPNOTSUPP;
3274}
3275EXPORT_SYMBOL(sock_no_sendmsg_locked);
3276
3277int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3278 int flags)
3279{
3280 return -EOPNOTSUPP;
3281}
3282EXPORT_SYMBOL(sock_no_recvmsg);
3283
3284int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3285{
3286 /* Mirror missing mmap method error code */
3287 return -ENODEV;
3288}
3289EXPORT_SYMBOL(sock_no_mmap);
3290
3291/*
3292 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3293 * various sock-based usage counts.
3294 */
3295void __receive_sock(struct file *file)
3296{
3297 struct socket *sock;
3298
3299 sock = sock_from_file(file);
3300 if (sock) {
3301 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3302 sock_update_classid(&sock->sk->sk_cgrp_data);
3303 }
3304}
3305
3306/*
3307 * Default Socket Callbacks
3308 */
3309
3310static void sock_def_wakeup(struct sock *sk)
3311{
3312 struct socket_wq *wq;
3313
3314 rcu_read_lock();
3315 wq = rcu_dereference(sk->sk_wq);
3316 if (skwq_has_sleeper(wq))
3317 wake_up_interruptible_all(&wq->wait);
3318 rcu_read_unlock();
3319}
3320
3321static void sock_def_error_report(struct sock *sk)
3322{
3323 struct socket_wq *wq;
3324
3325 rcu_read_lock();
3326 wq = rcu_dereference(sk->sk_wq);
3327 if (skwq_has_sleeper(wq))
3328 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3329 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3330 rcu_read_unlock();
3331}
3332
3333void sock_def_readable(struct sock *sk)
3334{
3335 struct socket_wq *wq;
3336
3337 trace_sk_data_ready(sk);
3338
3339 rcu_read_lock();
3340 wq = rcu_dereference(sk->sk_wq);
3341 if (skwq_has_sleeper(wq))
3342 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3343 EPOLLRDNORM | EPOLLRDBAND);
3344 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3345 rcu_read_unlock();
3346}
3347
3348static void sock_def_write_space(struct sock *sk)
3349{
3350 struct socket_wq *wq;
3351
3352 rcu_read_lock();
3353
3354 /* Do not wake up a writer until he can make "significant"
3355 * progress. --DaveM
3356 */
3357 if (sock_writeable(sk)) {
3358 wq = rcu_dereference(sk->sk_wq);
3359 if (skwq_has_sleeper(wq))
3360 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3361 EPOLLWRNORM | EPOLLWRBAND);
3362
3363 /* Should agree with poll, otherwise some programs break */
3364 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3365 }
3366
3367 rcu_read_unlock();
3368}
3369
3370/* An optimised version of sock_def_write_space(), should only be called
3371 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3372 * ->sk_wmem_alloc.
3373 */
3374static void sock_def_write_space_wfree(struct sock *sk)
3375{
3376 /* Do not wake up a writer until he can make "significant"
3377 * progress. --DaveM
3378 */
3379 if (sock_writeable(sk)) {
3380 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3381
3382 /* rely on refcount_sub from sock_wfree() */
3383 smp_mb__after_atomic();
3384 if (wq && waitqueue_active(&wq->wait))
3385 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3386 EPOLLWRNORM | EPOLLWRBAND);
3387
3388 /* Should agree with poll, otherwise some programs break */
3389 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3390 }
3391}
3392
3393static void sock_def_destruct(struct sock *sk)
3394{
3395}
3396
3397void sk_send_sigurg(struct sock *sk)
3398{
3399 if (sk->sk_socket && sk->sk_socket->file)
3400 if (send_sigurg(&sk->sk_socket->file->f_owner))
3401 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3402}
3403EXPORT_SYMBOL(sk_send_sigurg);
3404
3405void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3406 unsigned long expires)
3407{
3408 if (!mod_timer(timer, expires))
3409 sock_hold(sk);
3410}
3411EXPORT_SYMBOL(sk_reset_timer);
3412
3413void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3414{
3415 if (del_timer(timer))
3416 __sock_put(sk);
3417}
3418EXPORT_SYMBOL(sk_stop_timer);
3419
3420void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3421{
3422 if (del_timer_sync(timer))
3423 __sock_put(sk);
3424}
3425EXPORT_SYMBOL(sk_stop_timer_sync);
3426
3427void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3428{
3429 sk_init_common(sk);
3430 sk->sk_send_head = NULL;
3431
3432 timer_setup(&sk->sk_timer, NULL, 0);
3433
3434 sk->sk_allocation = GFP_KERNEL;
3435 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3436 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3437 sk->sk_state = TCP_CLOSE;
3438 sk->sk_use_task_frag = true;
3439 sk_set_socket(sk, sock);
3440
3441 sock_set_flag(sk, SOCK_ZAPPED);
3442
3443 if (sock) {
3444 sk->sk_type = sock->type;
3445 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3446 sock->sk = sk;
3447 } else {
3448 RCU_INIT_POINTER(sk->sk_wq, NULL);
3449 }
3450 sk->sk_uid = uid;
3451
3452 rwlock_init(&sk->sk_callback_lock);
3453 if (sk->sk_kern_sock)
3454 lockdep_set_class_and_name(
3455 &sk->sk_callback_lock,
3456 af_kern_callback_keys + sk->sk_family,
3457 af_family_kern_clock_key_strings[sk->sk_family]);
3458 else
3459 lockdep_set_class_and_name(
3460 &sk->sk_callback_lock,
3461 af_callback_keys + sk->sk_family,
3462 af_family_clock_key_strings[sk->sk_family]);
3463
3464 sk->sk_state_change = sock_def_wakeup;
3465 sk->sk_data_ready = sock_def_readable;
3466 sk->sk_write_space = sock_def_write_space;
3467 sk->sk_error_report = sock_def_error_report;
3468 sk->sk_destruct = sock_def_destruct;
3469
3470 sk->sk_frag.page = NULL;
3471 sk->sk_frag.offset = 0;
3472 sk->sk_peek_off = -1;
3473
3474 sk->sk_peer_pid = NULL;
3475 sk->sk_peer_cred = NULL;
3476 spin_lock_init(&sk->sk_peer_lock);
3477
3478 sk->sk_write_pending = 0;
3479 sk->sk_rcvlowat = 1;
3480 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3481 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3482
3483 sk->sk_stamp = SK_DEFAULT_STAMP;
3484#if BITS_PER_LONG==32
3485 seqlock_init(&sk->sk_stamp_seq);
3486#endif
3487 atomic_set(&sk->sk_zckey, 0);
3488
3489#ifdef CONFIG_NET_RX_BUSY_POLL
3490 sk->sk_napi_id = 0;
3491 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3492#endif
3493
3494 sk->sk_max_pacing_rate = ~0UL;
3495 sk->sk_pacing_rate = ~0UL;
3496 WRITE_ONCE(sk->sk_pacing_shift, 10);
3497 sk->sk_incoming_cpu = -1;
3498
3499 sk_rx_queue_clear(sk);
3500 /*
3501 * Before updating sk_refcnt, we must commit prior changes to memory
3502 * (Documentation/RCU/rculist_nulls.rst for details)
3503 */
3504 smp_wmb();
3505 refcount_set(&sk->sk_refcnt, 1);
3506 atomic_set(&sk->sk_drops, 0);
3507}
3508EXPORT_SYMBOL(sock_init_data_uid);
3509
3510void sock_init_data(struct socket *sock, struct sock *sk)
3511{
3512 kuid_t uid = sock ?
3513 SOCK_INODE(sock)->i_uid :
3514 make_kuid(sock_net(sk)->user_ns, 0);
3515
3516 sock_init_data_uid(sock, sk, uid);
3517}
3518EXPORT_SYMBOL(sock_init_data);
3519
3520void lock_sock_nested(struct sock *sk, int subclass)
3521{
3522 /* The sk_lock has mutex_lock() semantics here. */
3523 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3524
3525 might_sleep();
3526 spin_lock_bh(&sk->sk_lock.slock);
3527 if (sock_owned_by_user_nocheck(sk))
3528 __lock_sock(sk);
3529 sk->sk_lock.owned = 1;
3530 spin_unlock_bh(&sk->sk_lock.slock);
3531}
3532EXPORT_SYMBOL(lock_sock_nested);
3533
3534void release_sock(struct sock *sk)
3535{
3536 spin_lock_bh(&sk->sk_lock.slock);
3537 if (sk->sk_backlog.tail)
3538 __release_sock(sk);
3539
3540 if (sk->sk_prot->release_cb)
3541 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3542 tcp_release_cb, sk);
3543
3544 sock_release_ownership(sk);
3545 if (waitqueue_active(&sk->sk_lock.wq))
3546 wake_up(&sk->sk_lock.wq);
3547 spin_unlock_bh(&sk->sk_lock.slock);
3548}
3549EXPORT_SYMBOL(release_sock);
3550
3551bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3552{
3553 might_sleep();
3554 spin_lock_bh(&sk->sk_lock.slock);
3555
3556 if (!sock_owned_by_user_nocheck(sk)) {
3557 /*
3558 * Fast path return with bottom halves disabled and
3559 * sock::sk_lock.slock held.
3560 *
3561 * The 'mutex' is not contended and holding
3562 * sock::sk_lock.slock prevents all other lockers to
3563 * proceed so the corresponding unlock_sock_fast() can
3564 * avoid the slow path of release_sock() completely and
3565 * just release slock.
3566 *
3567 * From a semantical POV this is equivalent to 'acquiring'
3568 * the 'mutex', hence the corresponding lockdep
3569 * mutex_release() has to happen in the fast path of
3570 * unlock_sock_fast().
3571 */
3572 return false;
3573 }
3574
3575 __lock_sock(sk);
3576 sk->sk_lock.owned = 1;
3577 __acquire(&sk->sk_lock.slock);
3578 spin_unlock_bh(&sk->sk_lock.slock);
3579 return true;
3580}
3581EXPORT_SYMBOL(__lock_sock_fast);
3582
3583int sock_gettstamp(struct socket *sock, void __user *userstamp,
3584 bool timeval, bool time32)
3585{
3586 struct sock *sk = sock->sk;
3587 struct timespec64 ts;
3588
3589 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3590 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3591 if (ts.tv_sec == -1)
3592 return -ENOENT;
3593 if (ts.tv_sec == 0) {
3594 ktime_t kt = ktime_get_real();
3595 sock_write_timestamp(sk, kt);
3596 ts = ktime_to_timespec64(kt);
3597 }
3598
3599 if (timeval)
3600 ts.tv_nsec /= 1000;
3601
3602#ifdef CONFIG_COMPAT_32BIT_TIME
3603 if (time32)
3604 return put_old_timespec32(&ts, userstamp);
3605#endif
3606#ifdef CONFIG_SPARC64
3607 /* beware of padding in sparc64 timeval */
3608 if (timeval && !in_compat_syscall()) {
3609 struct __kernel_old_timeval __user tv = {
3610 .tv_sec = ts.tv_sec,
3611 .tv_usec = ts.tv_nsec,
3612 };
3613 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3614 return -EFAULT;
3615 return 0;
3616 }
3617#endif
3618 return put_timespec64(&ts, userstamp);
3619}
3620EXPORT_SYMBOL(sock_gettstamp);
3621
3622void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3623{
3624 if (!sock_flag(sk, flag)) {
3625 unsigned long previous_flags = sk->sk_flags;
3626
3627 sock_set_flag(sk, flag);
3628 /*
3629 * we just set one of the two flags which require net
3630 * time stamping, but time stamping might have been on
3631 * already because of the other one
3632 */
3633 if (sock_needs_netstamp(sk) &&
3634 !(previous_flags & SK_FLAGS_TIMESTAMP))
3635 net_enable_timestamp();
3636 }
3637}
3638
3639int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3640 int level, int type)
3641{
3642 struct sock_exterr_skb *serr;
3643 struct sk_buff *skb;
3644 int copied, err;
3645
3646 err = -EAGAIN;
3647 skb = sock_dequeue_err_skb(sk);
3648 if (skb == NULL)
3649 goto out;
3650
3651 copied = skb->len;
3652 if (copied > len) {
3653 msg->msg_flags |= MSG_TRUNC;
3654 copied = len;
3655 }
3656 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3657 if (err)
3658 goto out_free_skb;
3659
3660 sock_recv_timestamp(msg, sk, skb);
3661
3662 serr = SKB_EXT_ERR(skb);
3663 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3664
3665 msg->msg_flags |= MSG_ERRQUEUE;
3666 err = copied;
3667
3668out_free_skb:
3669 kfree_skb(skb);
3670out:
3671 return err;
3672}
3673EXPORT_SYMBOL(sock_recv_errqueue);
3674
3675/*
3676 * Get a socket option on an socket.
3677 *
3678 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3679 * asynchronous errors should be reported by getsockopt. We assume
3680 * this means if you specify SO_ERROR (otherwise whats the point of it).
3681 */
3682int sock_common_getsockopt(struct socket *sock, int level, int optname,
3683 char __user *optval, int __user *optlen)
3684{
3685 struct sock *sk = sock->sk;
3686
3687 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3688 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3689}
3690EXPORT_SYMBOL(sock_common_getsockopt);
3691
3692int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3693 int flags)
3694{
3695 struct sock *sk = sock->sk;
3696 int addr_len = 0;
3697 int err;
3698
3699 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3700 if (err >= 0)
3701 msg->msg_namelen = addr_len;
3702 return err;
3703}
3704EXPORT_SYMBOL(sock_common_recvmsg);
3705
3706/*
3707 * Set socket options on an inet socket.
3708 */
3709int sock_common_setsockopt(struct socket *sock, int level, int optname,
3710 sockptr_t optval, unsigned int optlen)
3711{
3712 struct sock *sk = sock->sk;
3713
3714 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3715 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3716}
3717EXPORT_SYMBOL(sock_common_setsockopt);
3718
3719void sk_common_release(struct sock *sk)
3720{
3721 if (sk->sk_prot->destroy)
3722 sk->sk_prot->destroy(sk);
3723
3724 /*
3725 * Observation: when sk_common_release is called, processes have
3726 * no access to socket. But net still has.
3727 * Step one, detach it from networking:
3728 *
3729 * A. Remove from hash tables.
3730 */
3731
3732 sk->sk_prot->unhash(sk);
3733
3734 /*
3735 * In this point socket cannot receive new packets, but it is possible
3736 * that some packets are in flight because some CPU runs receiver and
3737 * did hash table lookup before we unhashed socket. They will achieve
3738 * receive queue and will be purged by socket destructor.
3739 *
3740 * Also we still have packets pending on receive queue and probably,
3741 * our own packets waiting in device queues. sock_destroy will drain
3742 * receive queue, but transmitted packets will delay socket destruction
3743 * until the last reference will be released.
3744 */
3745
3746 sock_orphan(sk);
3747
3748 xfrm_sk_free_policy(sk);
3749
3750 sock_put(sk);
3751}
3752EXPORT_SYMBOL(sk_common_release);
3753
3754void sk_get_meminfo(const struct sock *sk, u32 *mem)
3755{
3756 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3757
3758 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3759 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3760 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3761 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3762 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3763 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3764 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3765 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3766 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3767}
3768
3769#ifdef CONFIG_PROC_FS
3770static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3771
3772int sock_prot_inuse_get(struct net *net, struct proto *prot)
3773{
3774 int cpu, idx = prot->inuse_idx;
3775 int res = 0;
3776
3777 for_each_possible_cpu(cpu)
3778 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3779
3780 return res >= 0 ? res : 0;
3781}
3782EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3783
3784int sock_inuse_get(struct net *net)
3785{
3786 int cpu, res = 0;
3787
3788 for_each_possible_cpu(cpu)
3789 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3790
3791 return res;
3792}
3793
3794EXPORT_SYMBOL_GPL(sock_inuse_get);
3795
3796static int __net_init sock_inuse_init_net(struct net *net)
3797{
3798 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3799 if (net->core.prot_inuse == NULL)
3800 return -ENOMEM;
3801 return 0;
3802}
3803
3804static void __net_exit sock_inuse_exit_net(struct net *net)
3805{
3806 free_percpu(net->core.prot_inuse);
3807}
3808
3809static struct pernet_operations net_inuse_ops = {
3810 .init = sock_inuse_init_net,
3811 .exit = sock_inuse_exit_net,
3812};
3813
3814static __init int net_inuse_init(void)
3815{
3816 if (register_pernet_subsys(&net_inuse_ops))
3817 panic("Cannot initialize net inuse counters");
3818
3819 return 0;
3820}
3821
3822core_initcall(net_inuse_init);
3823
3824static int assign_proto_idx(struct proto *prot)
3825{
3826 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3827
3828 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3829 pr_err("PROTO_INUSE_NR exhausted\n");
3830 return -ENOSPC;
3831 }
3832
3833 set_bit(prot->inuse_idx, proto_inuse_idx);
3834 return 0;
3835}
3836
3837static void release_proto_idx(struct proto *prot)
3838{
3839 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3840 clear_bit(prot->inuse_idx, proto_inuse_idx);
3841}
3842#else
3843static inline int assign_proto_idx(struct proto *prot)
3844{
3845 return 0;
3846}
3847
3848static inline void release_proto_idx(struct proto *prot)
3849{
3850}
3851
3852#endif
3853
3854static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3855{
3856 if (!twsk_prot)
3857 return;
3858 kfree(twsk_prot->twsk_slab_name);
3859 twsk_prot->twsk_slab_name = NULL;
3860 kmem_cache_destroy(twsk_prot->twsk_slab);
3861 twsk_prot->twsk_slab = NULL;
3862}
3863
3864static int tw_prot_init(const struct proto *prot)
3865{
3866 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3867
3868 if (!twsk_prot)
3869 return 0;
3870
3871 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3872 prot->name);
3873 if (!twsk_prot->twsk_slab_name)
3874 return -ENOMEM;
3875
3876 twsk_prot->twsk_slab =
3877 kmem_cache_create(twsk_prot->twsk_slab_name,
3878 twsk_prot->twsk_obj_size, 0,
3879 SLAB_ACCOUNT | prot->slab_flags,
3880 NULL);
3881 if (!twsk_prot->twsk_slab) {
3882 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3883 prot->name);
3884 return -ENOMEM;
3885 }
3886
3887 return 0;
3888}
3889
3890static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3891{
3892 if (!rsk_prot)
3893 return;
3894 kfree(rsk_prot->slab_name);
3895 rsk_prot->slab_name = NULL;
3896 kmem_cache_destroy(rsk_prot->slab);
3897 rsk_prot->slab = NULL;
3898}
3899
3900static int req_prot_init(const struct proto *prot)
3901{
3902 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3903
3904 if (!rsk_prot)
3905 return 0;
3906
3907 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3908 prot->name);
3909 if (!rsk_prot->slab_name)
3910 return -ENOMEM;
3911
3912 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3913 rsk_prot->obj_size, 0,
3914 SLAB_ACCOUNT | prot->slab_flags,
3915 NULL);
3916
3917 if (!rsk_prot->slab) {
3918 pr_crit("%s: Can't create request sock SLAB cache!\n",
3919 prot->name);
3920 return -ENOMEM;
3921 }
3922 return 0;
3923}
3924
3925int proto_register(struct proto *prot, int alloc_slab)
3926{
3927 int ret = -ENOBUFS;
3928
3929 if (prot->memory_allocated && !prot->sysctl_mem) {
3930 pr_err("%s: missing sysctl_mem\n", prot->name);
3931 return -EINVAL;
3932 }
3933 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3934 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3935 return -EINVAL;
3936 }
3937 if (alloc_slab) {
3938 prot->slab = kmem_cache_create_usercopy(prot->name,
3939 prot->obj_size, 0,
3940 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3941 prot->slab_flags,
3942 prot->useroffset, prot->usersize,
3943 NULL);
3944
3945 if (prot->slab == NULL) {
3946 pr_crit("%s: Can't create sock SLAB cache!\n",
3947 prot->name);
3948 goto out;
3949 }
3950
3951 if (req_prot_init(prot))
3952 goto out_free_request_sock_slab;
3953
3954 if (tw_prot_init(prot))
3955 goto out_free_timewait_sock_slab;
3956 }
3957
3958 mutex_lock(&proto_list_mutex);
3959 ret = assign_proto_idx(prot);
3960 if (ret) {
3961 mutex_unlock(&proto_list_mutex);
3962 goto out_free_timewait_sock_slab;
3963 }
3964 list_add(&prot->node, &proto_list);
3965 mutex_unlock(&proto_list_mutex);
3966 return ret;
3967
3968out_free_timewait_sock_slab:
3969 if (alloc_slab)
3970 tw_prot_cleanup(prot->twsk_prot);
3971out_free_request_sock_slab:
3972 if (alloc_slab) {
3973 req_prot_cleanup(prot->rsk_prot);
3974
3975 kmem_cache_destroy(prot->slab);
3976 prot->slab = NULL;
3977 }
3978out:
3979 return ret;
3980}
3981EXPORT_SYMBOL(proto_register);
3982
3983void proto_unregister(struct proto *prot)
3984{
3985 mutex_lock(&proto_list_mutex);
3986 release_proto_idx(prot);
3987 list_del(&prot->node);
3988 mutex_unlock(&proto_list_mutex);
3989
3990 kmem_cache_destroy(prot->slab);
3991 prot->slab = NULL;
3992
3993 req_prot_cleanup(prot->rsk_prot);
3994 tw_prot_cleanup(prot->twsk_prot);
3995}
3996EXPORT_SYMBOL(proto_unregister);
3997
3998int sock_load_diag_module(int family, int protocol)
3999{
4000 if (!protocol) {
4001 if (!sock_is_registered(family))
4002 return -ENOENT;
4003
4004 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4005 NETLINK_SOCK_DIAG, family);
4006 }
4007
4008#ifdef CONFIG_INET
4009 if (family == AF_INET &&
4010 protocol != IPPROTO_RAW &&
4011 protocol < MAX_INET_PROTOS &&
4012 !rcu_access_pointer(inet_protos[protocol]))
4013 return -ENOENT;
4014#endif
4015
4016 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4017 NETLINK_SOCK_DIAG, family, protocol);
4018}
4019EXPORT_SYMBOL(sock_load_diag_module);
4020
4021#ifdef CONFIG_PROC_FS
4022static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4023 __acquires(proto_list_mutex)
4024{
4025 mutex_lock(&proto_list_mutex);
4026 return seq_list_start_head(&proto_list, *pos);
4027}
4028
4029static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4030{
4031 return seq_list_next(v, &proto_list, pos);
4032}
4033
4034static void proto_seq_stop(struct seq_file *seq, void *v)
4035 __releases(proto_list_mutex)
4036{
4037 mutex_unlock(&proto_list_mutex);
4038}
4039
4040static char proto_method_implemented(const void *method)
4041{
4042 return method == NULL ? 'n' : 'y';
4043}
4044static long sock_prot_memory_allocated(struct proto *proto)
4045{
4046 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4047}
4048
4049static const char *sock_prot_memory_pressure(struct proto *proto)
4050{
4051 return proto->memory_pressure != NULL ?
4052 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4053}
4054
4055static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4056{
4057
4058 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4059 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4060 proto->name,
4061 proto->obj_size,
4062 sock_prot_inuse_get(seq_file_net(seq), proto),
4063 sock_prot_memory_allocated(proto),
4064 sock_prot_memory_pressure(proto),
4065 proto->max_header,
4066 proto->slab == NULL ? "no" : "yes",
4067 module_name(proto->owner),
4068 proto_method_implemented(proto->close),
4069 proto_method_implemented(proto->connect),
4070 proto_method_implemented(proto->disconnect),
4071 proto_method_implemented(proto->accept),
4072 proto_method_implemented(proto->ioctl),
4073 proto_method_implemented(proto->init),
4074 proto_method_implemented(proto->destroy),
4075 proto_method_implemented(proto->shutdown),
4076 proto_method_implemented(proto->setsockopt),
4077 proto_method_implemented(proto->getsockopt),
4078 proto_method_implemented(proto->sendmsg),
4079 proto_method_implemented(proto->recvmsg),
4080 proto_method_implemented(proto->bind),
4081 proto_method_implemented(proto->backlog_rcv),
4082 proto_method_implemented(proto->hash),
4083 proto_method_implemented(proto->unhash),
4084 proto_method_implemented(proto->get_port),
4085 proto_method_implemented(proto->enter_memory_pressure));
4086}
4087
4088static int proto_seq_show(struct seq_file *seq, void *v)
4089{
4090 if (v == &proto_list)
4091 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4092 "protocol",
4093 "size",
4094 "sockets",
4095 "memory",
4096 "press",
4097 "maxhdr",
4098 "slab",
4099 "module",
4100 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4101 else
4102 proto_seq_printf(seq, list_entry(v, struct proto, node));
4103 return 0;
4104}
4105
4106static const struct seq_operations proto_seq_ops = {
4107 .start = proto_seq_start,
4108 .next = proto_seq_next,
4109 .stop = proto_seq_stop,
4110 .show = proto_seq_show,
4111};
4112
4113static __net_init int proto_init_net(struct net *net)
4114{
4115 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4116 sizeof(struct seq_net_private)))
4117 return -ENOMEM;
4118
4119 return 0;
4120}
4121
4122static __net_exit void proto_exit_net(struct net *net)
4123{
4124 remove_proc_entry("protocols", net->proc_net);
4125}
4126
4127
4128static __net_initdata struct pernet_operations proto_net_ops = {
4129 .init = proto_init_net,
4130 .exit = proto_exit_net,
4131};
4132
4133static int __init proto_init(void)
4134{
4135 return register_pernet_subsys(&proto_net_ops);
4136}
4137
4138subsys_initcall(proto_init);
4139
4140#endif /* PROC_FS */
4141
4142#ifdef CONFIG_NET_RX_BUSY_POLL
4143bool sk_busy_loop_end(void *p, unsigned long start_time)
4144{
4145 struct sock *sk = p;
4146
4147 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4148 return true;
4149
4150 if (sk_is_udp(sk) &&
4151 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4152 return true;
4153
4154 return sk_busy_loop_timeout(sk, start_time);
4155}
4156EXPORT_SYMBOL(sk_busy_loop_end);
4157#endif /* CONFIG_NET_RX_BUSY_POLL */
4158
4159int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4160{
4161 if (!sk->sk_prot->bind_add)
4162 return -EOPNOTSUPP;
4163 return sk->sk_prot->bind_add(sk, addr, addr_len);
4164}
4165EXPORT_SYMBOL(sock_bind_add);
4166
4167/* Copy 'size' bytes from userspace and return `size` back to userspace */
4168int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4169 void __user *arg, void *karg, size_t size)
4170{
4171 int ret;
4172
4173 if (copy_from_user(karg, arg, size))
4174 return -EFAULT;
4175
4176 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4177 if (ret)
4178 return ret;
4179
4180 if (copy_to_user(arg, karg, size))
4181 return -EFAULT;
4182
4183 return 0;
4184}
4185EXPORT_SYMBOL(sock_ioctl_inout);
4186
4187/* This is the most common ioctl prep function, where the result (4 bytes) is
4188 * copied back to userspace if the ioctl() returns successfully. No input is
4189 * copied from userspace as input argument.
4190 */
4191static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4192{
4193 int ret, karg = 0;
4194
4195 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4196 if (ret)
4197 return ret;
4198
4199 return put_user(karg, (int __user *)arg);
4200}
4201
4202/* A wrapper around sock ioctls, which copies the data from userspace
4203 * (depending on the protocol/ioctl), and copies back the result to userspace.
4204 * The main motivation for this function is to pass kernel memory to the
4205 * protocol ioctl callbacks, instead of userspace memory.
4206 */
4207int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4208{
4209 int rc = 1;
4210
4211 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4212 rc = ipmr_sk_ioctl(sk, cmd, arg);
4213 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4214 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4215 else if (sk_is_phonet(sk))
4216 rc = phonet_sk_ioctl(sk, cmd, arg);
4217
4218 /* If ioctl was processed, returns its value */
4219 if (rc <= 0)
4220 return rc;
4221
4222 /* Otherwise call the default handler */
4223 return sock_ioctl_out(sk, cmd, arg);
4224}
4225EXPORT_SYMBOL(sk_ioctl);
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
117#include <linux/uaccess.h>
118
119#include <linux/netdevice.h>
120#include <net/protocol.h>
121#include <linux/skbuff.h>
122#include <net/net_namespace.h>
123#include <net/request_sock.h>
124#include <net/sock.h>
125#include <linux/net_tstamp.h>
126#include <net/xfrm.h>
127#include <linux/ipsec.h>
128#include <net/cls_cgroup.h>
129#include <net/netprio_cgroup.h>
130#include <linux/sock_diag.h>
131
132#include <linux/filter.h>
133#include <net/sock_reuseport.h>
134#include <net/bpf_sk_storage.h>
135
136#include <trace/events/sock.h>
137
138#include <net/tcp.h>
139#include <net/busy_poll.h>
140
141static DEFINE_MUTEX(proto_list_mutex);
142static LIST_HEAD(proto_list);
143
144static void sock_inuse_add(struct net *net, int val);
145
146/**
147 * sk_ns_capable - General socket capability test
148 * @sk: Socket to use a capability on or through
149 * @user_ns: The user namespace of the capability to use
150 * @cap: The capability to use
151 *
152 * Test to see if the opener of the socket had when the socket was
153 * created and the current process has the capability @cap in the user
154 * namespace @user_ns.
155 */
156bool sk_ns_capable(const struct sock *sk,
157 struct user_namespace *user_ns, int cap)
158{
159 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
160 ns_capable(user_ns, cap);
161}
162EXPORT_SYMBOL(sk_ns_capable);
163
164/**
165 * sk_capable - Socket global capability test
166 * @sk: Socket to use a capability on or through
167 * @cap: The global capability to use
168 *
169 * Test to see if the opener of the socket had when the socket was
170 * created and the current process has the capability @cap in all user
171 * namespaces.
172 */
173bool sk_capable(const struct sock *sk, int cap)
174{
175 return sk_ns_capable(sk, &init_user_ns, cap);
176}
177EXPORT_SYMBOL(sk_capable);
178
179/**
180 * sk_net_capable - Network namespace socket capability test
181 * @sk: Socket to use a capability on or through
182 * @cap: The capability to use
183 *
184 * Test to see if the opener of the socket had when the socket was created
185 * and the current process has the capability @cap over the network namespace
186 * the socket is a member of.
187 */
188bool sk_net_capable(const struct sock *sk, int cap)
189{
190 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
191}
192EXPORT_SYMBOL(sk_net_capable);
193
194/*
195 * Each address family might have different locking rules, so we have
196 * one slock key per address family and separate keys for internal and
197 * userspace sockets.
198 */
199static struct lock_class_key af_family_keys[AF_MAX];
200static struct lock_class_key af_family_kern_keys[AF_MAX];
201static struct lock_class_key af_family_slock_keys[AF_MAX];
202static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
203
204/*
205 * Make lock validator output more readable. (we pre-construct these
206 * strings build-time, so that runtime initialization of socket
207 * locks is fast):
208 */
209
210#define _sock_locks(x) \
211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
220 x "27" , x "28" , x "AF_CAN" , \
221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
226 x "AF_MAX"
227
228static const char *const af_family_key_strings[AF_MAX+1] = {
229 _sock_locks("sk_lock-")
230};
231static const char *const af_family_slock_key_strings[AF_MAX+1] = {
232 _sock_locks("slock-")
233};
234static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 _sock_locks("clock-")
236};
237
238static const char *const af_family_kern_key_strings[AF_MAX+1] = {
239 _sock_locks("k-sk_lock-")
240};
241static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
242 _sock_locks("k-slock-")
243};
244static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-clock-")
246};
247static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
248 _sock_locks("rlock-")
249};
250static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
251 _sock_locks("wlock-")
252};
253static const char *const af_family_elock_key_strings[AF_MAX+1] = {
254 _sock_locks("elock-")
255};
256
257/*
258 * sk_callback_lock and sk queues locking rules are per-address-family,
259 * so split the lock classes by using a per-AF key:
260 */
261static struct lock_class_key af_callback_keys[AF_MAX];
262static struct lock_class_key af_rlock_keys[AF_MAX];
263static struct lock_class_key af_wlock_keys[AF_MAX];
264static struct lock_class_key af_elock_keys[AF_MAX];
265static struct lock_class_key af_kern_callback_keys[AF_MAX];
266
267/* Run time adjustable parameters. */
268__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269EXPORT_SYMBOL(sysctl_wmem_max);
270__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271EXPORT_SYMBOL(sysctl_rmem_max);
272__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
274
275/* Maximal space eaten by iovec or ancillary data plus some space */
276int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277EXPORT_SYMBOL(sysctl_optmem_max);
278
279int sysctl_tstamp_allow_data __read_mostly = 1;
280
281DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
282EXPORT_SYMBOL_GPL(memalloc_socks_key);
283
284/**
285 * sk_set_memalloc - sets %SOCK_MEMALLOC
286 * @sk: socket to set it on
287 *
288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
289 * It's the responsibility of the admin to adjust min_free_kbytes
290 * to meet the requirements
291 */
292void sk_set_memalloc(struct sock *sk)
293{
294 sock_set_flag(sk, SOCK_MEMALLOC);
295 sk->sk_allocation |= __GFP_MEMALLOC;
296 static_branch_inc(&memalloc_socks_key);
297}
298EXPORT_SYMBOL_GPL(sk_set_memalloc);
299
300void sk_clear_memalloc(struct sock *sk)
301{
302 sock_reset_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation &= ~__GFP_MEMALLOC;
304 static_branch_dec(&memalloc_socks_key);
305
306 /*
307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
308 * progress of swapping. SOCK_MEMALLOC may be cleared while
309 * it has rmem allocations due to the last swapfile being deactivated
310 * but there is a risk that the socket is unusable due to exceeding
311 * the rmem limits. Reclaim the reserves and obey rmem limits again.
312 */
313 sk_mem_reclaim(sk);
314}
315EXPORT_SYMBOL_GPL(sk_clear_memalloc);
316
317int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
318{
319 int ret;
320 unsigned int noreclaim_flag;
321
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
324
325 noreclaim_flag = memalloc_noreclaim_save();
326 ret = sk->sk_backlog_rcv(sk, skb);
327 memalloc_noreclaim_restore(noreclaim_flag);
328
329 return ret;
330}
331EXPORT_SYMBOL(__sk_backlog_rcv);
332
333static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
334{
335 struct __kernel_sock_timeval tv;
336 int size;
337
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
339 tv.tv_sec = 0;
340 tv.tv_usec = 0;
341 } else {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344 }
345
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
349 return sizeof(tv32);
350 }
351
352 if (old_timeval) {
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 size = sizeof(old_tv);
358 } else {
359 *(struct __kernel_sock_timeval *)optval = tv;
360 size = sizeof(tv);
361 }
362
363 return size;
364}
365
366static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
367{
368 struct __kernel_sock_timeval tv;
369
370 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
371 struct old_timeval32 tv32;
372
373 if (optlen < sizeof(tv32))
374 return -EINVAL;
375
376 if (copy_from_user(&tv32, optval, sizeof(tv32)))
377 return -EFAULT;
378 tv.tv_sec = tv32.tv_sec;
379 tv.tv_usec = tv32.tv_usec;
380 } else if (old_timeval) {
381 struct __kernel_old_timeval old_tv;
382
383 if (optlen < sizeof(old_tv))
384 return -EINVAL;
385 if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
386 return -EFAULT;
387 tv.tv_sec = old_tv.tv_sec;
388 tv.tv_usec = old_tv.tv_usec;
389 } else {
390 if (optlen < sizeof(tv))
391 return -EINVAL;
392 if (copy_from_user(&tv, optval, sizeof(tv)))
393 return -EFAULT;
394 }
395 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
396 return -EDOM;
397
398 if (tv.tv_sec < 0) {
399 static int warned __read_mostly;
400
401 *timeo_p = 0;
402 if (warned < 10 && net_ratelimit()) {
403 warned++;
404 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
405 __func__, current->comm, task_pid_nr(current));
406 }
407 return 0;
408 }
409 *timeo_p = MAX_SCHEDULE_TIMEOUT;
410 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 return 0;
412 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
413 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
414 return 0;
415}
416
417static void sock_warn_obsolete_bsdism(const char *name)
418{
419 static int warned;
420 static char warncomm[TASK_COMM_LEN];
421 if (strcmp(warncomm, current->comm) && warned < 5) {
422 strcpy(warncomm, current->comm);
423 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
424 warncomm, name);
425 warned++;
426 }
427}
428
429static bool sock_needs_netstamp(const struct sock *sk)
430{
431 switch (sk->sk_family) {
432 case AF_UNSPEC:
433 case AF_UNIX:
434 return false;
435 default:
436 return true;
437 }
438}
439
440static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
441{
442 if (sk->sk_flags & flags) {
443 sk->sk_flags &= ~flags;
444 if (sock_needs_netstamp(sk) &&
445 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
446 net_disable_timestamp();
447 }
448}
449
450
451int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
452{
453 unsigned long flags;
454 struct sk_buff_head *list = &sk->sk_receive_queue;
455
456 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
457 atomic_inc(&sk->sk_drops);
458 trace_sock_rcvqueue_full(sk, skb);
459 return -ENOMEM;
460 }
461
462 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
463 atomic_inc(&sk->sk_drops);
464 return -ENOBUFS;
465 }
466
467 skb->dev = NULL;
468 skb_set_owner_r(skb, sk);
469
470 /* we escape from rcu protected region, make sure we dont leak
471 * a norefcounted dst
472 */
473 skb_dst_force(skb);
474
475 spin_lock_irqsave(&list->lock, flags);
476 sock_skb_set_dropcount(sk, skb);
477 __skb_queue_tail(list, skb);
478 spin_unlock_irqrestore(&list->lock, flags);
479
480 if (!sock_flag(sk, SOCK_DEAD))
481 sk->sk_data_ready(sk);
482 return 0;
483}
484EXPORT_SYMBOL(__sock_queue_rcv_skb);
485
486int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
487{
488 int err;
489
490 err = sk_filter(sk, skb);
491 if (err)
492 return err;
493
494 return __sock_queue_rcv_skb(sk, skb);
495}
496EXPORT_SYMBOL(sock_queue_rcv_skb);
497
498int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
499 const int nested, unsigned int trim_cap, bool refcounted)
500{
501 int rc = NET_RX_SUCCESS;
502
503 if (sk_filter_trim_cap(sk, skb, trim_cap))
504 goto discard_and_relse;
505
506 skb->dev = NULL;
507
508 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
509 atomic_inc(&sk->sk_drops);
510 goto discard_and_relse;
511 }
512 if (nested)
513 bh_lock_sock_nested(sk);
514 else
515 bh_lock_sock(sk);
516 if (!sock_owned_by_user(sk)) {
517 /*
518 * trylock + unlock semantics:
519 */
520 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
521
522 rc = sk_backlog_rcv(sk, skb);
523
524 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
525 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
526 bh_unlock_sock(sk);
527 atomic_inc(&sk->sk_drops);
528 goto discard_and_relse;
529 }
530
531 bh_unlock_sock(sk);
532out:
533 if (refcounted)
534 sock_put(sk);
535 return rc;
536discard_and_relse:
537 kfree_skb(skb);
538 goto out;
539}
540EXPORT_SYMBOL(__sk_receive_skb);
541
542struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
543{
544 struct dst_entry *dst = __sk_dst_get(sk);
545
546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
547 sk_tx_queue_clear(sk);
548 sk->sk_dst_pending_confirm = 0;
549 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
550 dst_release(dst);
551 return NULL;
552 }
553
554 return dst;
555}
556EXPORT_SYMBOL(__sk_dst_check);
557
558struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
559{
560 struct dst_entry *dst = sk_dst_get(sk);
561
562 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
563 sk_dst_reset(sk);
564 dst_release(dst);
565 return NULL;
566 }
567
568 return dst;
569}
570EXPORT_SYMBOL(sk_dst_check);
571
572static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
573{
574 int ret = -ENOPROTOOPT;
575#ifdef CONFIG_NETDEVICES
576 struct net *net = sock_net(sk);
577
578 /* Sorry... */
579 ret = -EPERM;
580 if (!ns_capable(net->user_ns, CAP_NET_RAW))
581 goto out;
582
583 ret = -EINVAL;
584 if (ifindex < 0)
585 goto out;
586
587 sk->sk_bound_dev_if = ifindex;
588 if (sk->sk_prot->rehash)
589 sk->sk_prot->rehash(sk);
590 sk_dst_reset(sk);
591
592 ret = 0;
593
594out:
595#endif
596
597 return ret;
598}
599
600static int sock_setbindtodevice(struct sock *sk, char __user *optval,
601 int optlen)
602{
603 int ret = -ENOPROTOOPT;
604#ifdef CONFIG_NETDEVICES
605 struct net *net = sock_net(sk);
606 char devname[IFNAMSIZ];
607 int index;
608
609 ret = -EINVAL;
610 if (optlen < 0)
611 goto out;
612
613 /* Bind this socket to a particular device like "eth0",
614 * as specified in the passed interface name. If the
615 * name is "" or the option length is zero the socket
616 * is not bound.
617 */
618 if (optlen > IFNAMSIZ - 1)
619 optlen = IFNAMSIZ - 1;
620 memset(devname, 0, sizeof(devname));
621
622 ret = -EFAULT;
623 if (copy_from_user(devname, optval, optlen))
624 goto out;
625
626 index = 0;
627 if (devname[0] != '\0') {
628 struct net_device *dev;
629
630 rcu_read_lock();
631 dev = dev_get_by_name_rcu(net, devname);
632 if (dev)
633 index = dev->ifindex;
634 rcu_read_unlock();
635 ret = -ENODEV;
636 if (!dev)
637 goto out;
638 }
639
640 lock_sock(sk);
641 ret = sock_setbindtodevice_locked(sk, index);
642 release_sock(sk);
643
644out:
645#endif
646
647 return ret;
648}
649
650static int sock_getbindtodevice(struct sock *sk, char __user *optval,
651 int __user *optlen, int len)
652{
653 int ret = -ENOPROTOOPT;
654#ifdef CONFIG_NETDEVICES
655 struct net *net = sock_net(sk);
656 char devname[IFNAMSIZ];
657
658 if (sk->sk_bound_dev_if == 0) {
659 len = 0;
660 goto zero;
661 }
662
663 ret = -EINVAL;
664 if (len < IFNAMSIZ)
665 goto out;
666
667 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
668 if (ret)
669 goto out;
670
671 len = strlen(devname) + 1;
672
673 ret = -EFAULT;
674 if (copy_to_user(optval, devname, len))
675 goto out;
676
677zero:
678 ret = -EFAULT;
679 if (put_user(len, optlen))
680 goto out;
681
682 ret = 0;
683
684out:
685#endif
686
687 return ret;
688}
689
690static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
691{
692 if (valbool)
693 sock_set_flag(sk, bit);
694 else
695 sock_reset_flag(sk, bit);
696}
697
698bool sk_mc_loop(struct sock *sk)
699{
700 if (dev_recursion_level())
701 return false;
702 if (!sk)
703 return true;
704 switch (sk->sk_family) {
705 case AF_INET:
706 return inet_sk(sk)->mc_loop;
707#if IS_ENABLED(CONFIG_IPV6)
708 case AF_INET6:
709 return inet6_sk(sk)->mc_loop;
710#endif
711 }
712 WARN_ON(1);
713 return true;
714}
715EXPORT_SYMBOL(sk_mc_loop);
716
717/*
718 * This is meant for all protocols to use and covers goings on
719 * at the socket level. Everything here is generic.
720 */
721
722int sock_setsockopt(struct socket *sock, int level, int optname,
723 char __user *optval, unsigned int optlen)
724{
725 struct sock_txtime sk_txtime;
726 struct sock *sk = sock->sk;
727 int val;
728 int valbool;
729 struct linger ling;
730 int ret = 0;
731
732 /*
733 * Options without arguments
734 */
735
736 if (optname == SO_BINDTODEVICE)
737 return sock_setbindtodevice(sk, optval, optlen);
738
739 if (optlen < sizeof(int))
740 return -EINVAL;
741
742 if (get_user(val, (int __user *)optval))
743 return -EFAULT;
744
745 valbool = val ? 1 : 0;
746
747 lock_sock(sk);
748
749 switch (optname) {
750 case SO_DEBUG:
751 if (val && !capable(CAP_NET_ADMIN))
752 ret = -EACCES;
753 else
754 sock_valbool_flag(sk, SOCK_DBG, valbool);
755 break;
756 case SO_REUSEADDR:
757 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
758 break;
759 case SO_REUSEPORT:
760 sk->sk_reuseport = valbool;
761 break;
762 case SO_TYPE:
763 case SO_PROTOCOL:
764 case SO_DOMAIN:
765 case SO_ERROR:
766 ret = -ENOPROTOOPT;
767 break;
768 case SO_DONTROUTE:
769 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
770 sk_dst_reset(sk);
771 break;
772 case SO_BROADCAST:
773 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
774 break;
775 case SO_SNDBUF:
776 /* Don't error on this BSD doesn't and if you think
777 * about it this is right. Otherwise apps have to
778 * play 'guess the biggest size' games. RCVBUF/SNDBUF
779 * are treated in BSD as hints
780 */
781 val = min_t(u32, val, sysctl_wmem_max);
782set_sndbuf:
783 /* Ensure val * 2 fits into an int, to prevent max_t()
784 * from treating it as a negative value.
785 */
786 val = min_t(int, val, INT_MAX / 2);
787 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
788 WRITE_ONCE(sk->sk_sndbuf,
789 max_t(int, val * 2, SOCK_MIN_SNDBUF));
790 /* Wake up sending tasks if we upped the value. */
791 sk->sk_write_space(sk);
792 break;
793
794 case SO_SNDBUFFORCE:
795 if (!capable(CAP_NET_ADMIN)) {
796 ret = -EPERM;
797 break;
798 }
799
800 /* No negative values (to prevent underflow, as val will be
801 * multiplied by 2).
802 */
803 if (val < 0)
804 val = 0;
805 goto set_sndbuf;
806
807 case SO_RCVBUF:
808 /* Don't error on this BSD doesn't and if you think
809 * about it this is right. Otherwise apps have to
810 * play 'guess the biggest size' games. RCVBUF/SNDBUF
811 * are treated in BSD as hints
812 */
813 val = min_t(u32, val, sysctl_rmem_max);
814set_rcvbuf:
815 /* Ensure val * 2 fits into an int, to prevent max_t()
816 * from treating it as a negative value.
817 */
818 val = min_t(int, val, INT_MAX / 2);
819 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
820 /*
821 * We double it on the way in to account for
822 * "struct sk_buff" etc. overhead. Applications
823 * assume that the SO_RCVBUF setting they make will
824 * allow that much actual data to be received on that
825 * socket.
826 *
827 * Applications are unaware that "struct sk_buff" and
828 * other overheads allocate from the receive buffer
829 * during socket buffer allocation.
830 *
831 * And after considering the possible alternatives,
832 * returning the value we actually used in getsockopt
833 * is the most desirable behavior.
834 */
835 WRITE_ONCE(sk->sk_rcvbuf,
836 max_t(int, val * 2, SOCK_MIN_RCVBUF));
837 break;
838
839 case SO_RCVBUFFORCE:
840 if (!capable(CAP_NET_ADMIN)) {
841 ret = -EPERM;
842 break;
843 }
844
845 /* No negative values (to prevent underflow, as val will be
846 * multiplied by 2).
847 */
848 if (val < 0)
849 val = 0;
850 goto set_rcvbuf;
851
852 case SO_KEEPALIVE:
853 if (sk->sk_prot->keepalive)
854 sk->sk_prot->keepalive(sk, valbool);
855 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
856 break;
857
858 case SO_OOBINLINE:
859 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
860 break;
861
862 case SO_NO_CHECK:
863 sk->sk_no_check_tx = valbool;
864 break;
865
866 case SO_PRIORITY:
867 if ((val >= 0 && val <= 6) ||
868 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
869 sk->sk_priority = val;
870 else
871 ret = -EPERM;
872 break;
873
874 case SO_LINGER:
875 if (optlen < sizeof(ling)) {
876 ret = -EINVAL; /* 1003.1g */
877 break;
878 }
879 if (copy_from_user(&ling, optval, sizeof(ling))) {
880 ret = -EFAULT;
881 break;
882 }
883 if (!ling.l_onoff)
884 sock_reset_flag(sk, SOCK_LINGER);
885 else {
886#if (BITS_PER_LONG == 32)
887 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
888 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
889 else
890#endif
891 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
892 sock_set_flag(sk, SOCK_LINGER);
893 }
894 break;
895
896 case SO_BSDCOMPAT:
897 sock_warn_obsolete_bsdism("setsockopt");
898 break;
899
900 case SO_PASSCRED:
901 if (valbool)
902 set_bit(SOCK_PASSCRED, &sock->flags);
903 else
904 clear_bit(SOCK_PASSCRED, &sock->flags);
905 break;
906
907 case SO_TIMESTAMP_OLD:
908 case SO_TIMESTAMP_NEW:
909 case SO_TIMESTAMPNS_OLD:
910 case SO_TIMESTAMPNS_NEW:
911 if (valbool) {
912 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
913 sock_set_flag(sk, SOCK_TSTAMP_NEW);
914 else
915 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
916
917 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
918 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
919 else
920 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
921 sock_set_flag(sk, SOCK_RCVTSTAMP);
922 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
923 } else {
924 sock_reset_flag(sk, SOCK_RCVTSTAMP);
925 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
926 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
927 }
928 break;
929
930 case SO_TIMESTAMPING_NEW:
931 sock_set_flag(sk, SOCK_TSTAMP_NEW);
932 /* fall through */
933 case SO_TIMESTAMPING_OLD:
934 if (val & ~SOF_TIMESTAMPING_MASK) {
935 ret = -EINVAL;
936 break;
937 }
938
939 if (val & SOF_TIMESTAMPING_OPT_ID &&
940 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
941 if (sk->sk_protocol == IPPROTO_TCP &&
942 sk->sk_type == SOCK_STREAM) {
943 if ((1 << sk->sk_state) &
944 (TCPF_CLOSE | TCPF_LISTEN)) {
945 ret = -EINVAL;
946 break;
947 }
948 sk->sk_tskey = tcp_sk(sk)->snd_una;
949 } else {
950 sk->sk_tskey = 0;
951 }
952 }
953
954 if (val & SOF_TIMESTAMPING_OPT_STATS &&
955 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
956 ret = -EINVAL;
957 break;
958 }
959
960 sk->sk_tsflags = val;
961 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
962 sock_enable_timestamp(sk,
963 SOCK_TIMESTAMPING_RX_SOFTWARE);
964 else {
965 if (optname == SO_TIMESTAMPING_NEW)
966 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
967
968 sock_disable_timestamp(sk,
969 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
970 }
971 break;
972
973 case SO_RCVLOWAT:
974 if (val < 0)
975 val = INT_MAX;
976 if (sock->ops->set_rcvlowat)
977 ret = sock->ops->set_rcvlowat(sk, val);
978 else
979 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
980 break;
981
982 case SO_RCVTIMEO_OLD:
983 case SO_RCVTIMEO_NEW:
984 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
985 break;
986
987 case SO_SNDTIMEO_OLD:
988 case SO_SNDTIMEO_NEW:
989 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
990 break;
991
992 case SO_ATTACH_FILTER:
993 ret = -EINVAL;
994 if (optlen == sizeof(struct sock_fprog)) {
995 struct sock_fprog fprog;
996
997 ret = -EFAULT;
998 if (copy_from_user(&fprog, optval, sizeof(fprog)))
999 break;
1000
1001 ret = sk_attach_filter(&fprog, sk);
1002 }
1003 break;
1004
1005 case SO_ATTACH_BPF:
1006 ret = -EINVAL;
1007 if (optlen == sizeof(u32)) {
1008 u32 ufd;
1009
1010 ret = -EFAULT;
1011 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1012 break;
1013
1014 ret = sk_attach_bpf(ufd, sk);
1015 }
1016 break;
1017
1018 case SO_ATTACH_REUSEPORT_CBPF:
1019 ret = -EINVAL;
1020 if (optlen == sizeof(struct sock_fprog)) {
1021 struct sock_fprog fprog;
1022
1023 ret = -EFAULT;
1024 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1025 break;
1026
1027 ret = sk_reuseport_attach_filter(&fprog, sk);
1028 }
1029 break;
1030
1031 case SO_ATTACH_REUSEPORT_EBPF:
1032 ret = -EINVAL;
1033 if (optlen == sizeof(u32)) {
1034 u32 ufd;
1035
1036 ret = -EFAULT;
1037 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1038 break;
1039
1040 ret = sk_reuseport_attach_bpf(ufd, sk);
1041 }
1042 break;
1043
1044 case SO_DETACH_REUSEPORT_BPF:
1045 ret = reuseport_detach_prog(sk);
1046 break;
1047
1048 case SO_DETACH_FILTER:
1049 ret = sk_detach_filter(sk);
1050 break;
1051
1052 case SO_LOCK_FILTER:
1053 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1054 ret = -EPERM;
1055 else
1056 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1057 break;
1058
1059 case SO_PASSSEC:
1060 if (valbool)
1061 set_bit(SOCK_PASSSEC, &sock->flags);
1062 else
1063 clear_bit(SOCK_PASSSEC, &sock->flags);
1064 break;
1065 case SO_MARK:
1066 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1067 ret = -EPERM;
1068 } else if (val != sk->sk_mark) {
1069 sk->sk_mark = val;
1070 sk_dst_reset(sk);
1071 }
1072 break;
1073
1074 case SO_RXQ_OVFL:
1075 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1076 break;
1077
1078 case SO_WIFI_STATUS:
1079 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1080 break;
1081
1082 case SO_PEEK_OFF:
1083 if (sock->ops->set_peek_off)
1084 ret = sock->ops->set_peek_off(sk, val);
1085 else
1086 ret = -EOPNOTSUPP;
1087 break;
1088
1089 case SO_NOFCS:
1090 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1091 break;
1092
1093 case SO_SELECT_ERR_QUEUE:
1094 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1095 break;
1096
1097#ifdef CONFIG_NET_RX_BUSY_POLL
1098 case SO_BUSY_POLL:
1099 /* allow unprivileged users to decrease the value */
1100 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1101 ret = -EPERM;
1102 else {
1103 if (val < 0)
1104 ret = -EINVAL;
1105 else
1106 sk->sk_ll_usec = val;
1107 }
1108 break;
1109#endif
1110
1111 case SO_MAX_PACING_RATE:
1112 {
1113 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1114
1115 if (sizeof(ulval) != sizeof(val) &&
1116 optlen >= sizeof(ulval) &&
1117 get_user(ulval, (unsigned long __user *)optval)) {
1118 ret = -EFAULT;
1119 break;
1120 }
1121 if (ulval != ~0UL)
1122 cmpxchg(&sk->sk_pacing_status,
1123 SK_PACING_NONE,
1124 SK_PACING_NEEDED);
1125 sk->sk_max_pacing_rate = ulval;
1126 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1127 break;
1128 }
1129 case SO_INCOMING_CPU:
1130 WRITE_ONCE(sk->sk_incoming_cpu, val);
1131 break;
1132
1133 case SO_CNX_ADVICE:
1134 if (val == 1)
1135 dst_negative_advice(sk);
1136 break;
1137
1138 case SO_ZEROCOPY:
1139 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1140 if (!((sk->sk_type == SOCK_STREAM &&
1141 sk->sk_protocol == IPPROTO_TCP) ||
1142 (sk->sk_type == SOCK_DGRAM &&
1143 sk->sk_protocol == IPPROTO_UDP)))
1144 ret = -ENOTSUPP;
1145 } else if (sk->sk_family != PF_RDS) {
1146 ret = -ENOTSUPP;
1147 }
1148 if (!ret) {
1149 if (val < 0 || val > 1)
1150 ret = -EINVAL;
1151 else
1152 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1153 }
1154 break;
1155
1156 case SO_TXTIME:
1157 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1158 ret = -EPERM;
1159 } else if (optlen != sizeof(struct sock_txtime)) {
1160 ret = -EINVAL;
1161 } else if (copy_from_user(&sk_txtime, optval,
1162 sizeof(struct sock_txtime))) {
1163 ret = -EFAULT;
1164 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1165 ret = -EINVAL;
1166 } else {
1167 sock_valbool_flag(sk, SOCK_TXTIME, true);
1168 sk->sk_clockid = sk_txtime.clockid;
1169 sk->sk_txtime_deadline_mode =
1170 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1171 sk->sk_txtime_report_errors =
1172 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1173 }
1174 break;
1175
1176 case SO_BINDTOIFINDEX:
1177 ret = sock_setbindtodevice_locked(sk, val);
1178 break;
1179
1180 default:
1181 ret = -ENOPROTOOPT;
1182 break;
1183 }
1184 release_sock(sk);
1185 return ret;
1186}
1187EXPORT_SYMBOL(sock_setsockopt);
1188
1189
1190static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1191 struct ucred *ucred)
1192{
1193 ucred->pid = pid_vnr(pid);
1194 ucred->uid = ucred->gid = -1;
1195 if (cred) {
1196 struct user_namespace *current_ns = current_user_ns();
1197
1198 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1199 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1200 }
1201}
1202
1203static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1204{
1205 struct user_namespace *user_ns = current_user_ns();
1206 int i;
1207
1208 for (i = 0; i < src->ngroups; i++)
1209 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1210 return -EFAULT;
1211
1212 return 0;
1213}
1214
1215int sock_getsockopt(struct socket *sock, int level, int optname,
1216 char __user *optval, int __user *optlen)
1217{
1218 struct sock *sk = sock->sk;
1219
1220 union {
1221 int val;
1222 u64 val64;
1223 unsigned long ulval;
1224 struct linger ling;
1225 struct old_timeval32 tm32;
1226 struct __kernel_old_timeval tm;
1227 struct __kernel_sock_timeval stm;
1228 struct sock_txtime txtime;
1229 } v;
1230
1231 int lv = sizeof(int);
1232 int len;
1233
1234 if (get_user(len, optlen))
1235 return -EFAULT;
1236 if (len < 0)
1237 return -EINVAL;
1238
1239 memset(&v, 0, sizeof(v));
1240
1241 switch (optname) {
1242 case SO_DEBUG:
1243 v.val = sock_flag(sk, SOCK_DBG);
1244 break;
1245
1246 case SO_DONTROUTE:
1247 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1248 break;
1249
1250 case SO_BROADCAST:
1251 v.val = sock_flag(sk, SOCK_BROADCAST);
1252 break;
1253
1254 case SO_SNDBUF:
1255 v.val = sk->sk_sndbuf;
1256 break;
1257
1258 case SO_RCVBUF:
1259 v.val = sk->sk_rcvbuf;
1260 break;
1261
1262 case SO_REUSEADDR:
1263 v.val = sk->sk_reuse;
1264 break;
1265
1266 case SO_REUSEPORT:
1267 v.val = sk->sk_reuseport;
1268 break;
1269
1270 case SO_KEEPALIVE:
1271 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1272 break;
1273
1274 case SO_TYPE:
1275 v.val = sk->sk_type;
1276 break;
1277
1278 case SO_PROTOCOL:
1279 v.val = sk->sk_protocol;
1280 break;
1281
1282 case SO_DOMAIN:
1283 v.val = sk->sk_family;
1284 break;
1285
1286 case SO_ERROR:
1287 v.val = -sock_error(sk);
1288 if (v.val == 0)
1289 v.val = xchg(&sk->sk_err_soft, 0);
1290 break;
1291
1292 case SO_OOBINLINE:
1293 v.val = sock_flag(sk, SOCK_URGINLINE);
1294 break;
1295
1296 case SO_NO_CHECK:
1297 v.val = sk->sk_no_check_tx;
1298 break;
1299
1300 case SO_PRIORITY:
1301 v.val = sk->sk_priority;
1302 break;
1303
1304 case SO_LINGER:
1305 lv = sizeof(v.ling);
1306 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1307 v.ling.l_linger = sk->sk_lingertime / HZ;
1308 break;
1309
1310 case SO_BSDCOMPAT:
1311 sock_warn_obsolete_bsdism("getsockopt");
1312 break;
1313
1314 case SO_TIMESTAMP_OLD:
1315 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1316 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1317 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1318 break;
1319
1320 case SO_TIMESTAMPNS_OLD:
1321 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1322 break;
1323
1324 case SO_TIMESTAMP_NEW:
1325 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1326 break;
1327
1328 case SO_TIMESTAMPNS_NEW:
1329 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1330 break;
1331
1332 case SO_TIMESTAMPING_OLD:
1333 v.val = sk->sk_tsflags;
1334 break;
1335
1336 case SO_RCVTIMEO_OLD:
1337 case SO_RCVTIMEO_NEW:
1338 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1339 break;
1340
1341 case SO_SNDTIMEO_OLD:
1342 case SO_SNDTIMEO_NEW:
1343 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1344 break;
1345
1346 case SO_RCVLOWAT:
1347 v.val = sk->sk_rcvlowat;
1348 break;
1349
1350 case SO_SNDLOWAT:
1351 v.val = 1;
1352 break;
1353
1354 case SO_PASSCRED:
1355 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1356 break;
1357
1358 case SO_PEERCRED:
1359 {
1360 struct ucred peercred;
1361 if (len > sizeof(peercred))
1362 len = sizeof(peercred);
1363 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1364 if (copy_to_user(optval, &peercred, len))
1365 return -EFAULT;
1366 goto lenout;
1367 }
1368
1369 case SO_PEERGROUPS:
1370 {
1371 int ret, n;
1372
1373 if (!sk->sk_peer_cred)
1374 return -ENODATA;
1375
1376 n = sk->sk_peer_cred->group_info->ngroups;
1377 if (len < n * sizeof(gid_t)) {
1378 len = n * sizeof(gid_t);
1379 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1380 }
1381 len = n * sizeof(gid_t);
1382
1383 ret = groups_to_user((gid_t __user *)optval,
1384 sk->sk_peer_cred->group_info);
1385 if (ret)
1386 return ret;
1387 goto lenout;
1388 }
1389
1390 case SO_PEERNAME:
1391 {
1392 char address[128];
1393
1394 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1395 if (lv < 0)
1396 return -ENOTCONN;
1397 if (lv < len)
1398 return -EINVAL;
1399 if (copy_to_user(optval, address, len))
1400 return -EFAULT;
1401 goto lenout;
1402 }
1403
1404 /* Dubious BSD thing... Probably nobody even uses it, but
1405 * the UNIX standard wants it for whatever reason... -DaveM
1406 */
1407 case SO_ACCEPTCONN:
1408 v.val = sk->sk_state == TCP_LISTEN;
1409 break;
1410
1411 case SO_PASSSEC:
1412 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1413 break;
1414
1415 case SO_PEERSEC:
1416 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1417
1418 case SO_MARK:
1419 v.val = sk->sk_mark;
1420 break;
1421
1422 case SO_RXQ_OVFL:
1423 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1424 break;
1425
1426 case SO_WIFI_STATUS:
1427 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1428 break;
1429
1430 case SO_PEEK_OFF:
1431 if (!sock->ops->set_peek_off)
1432 return -EOPNOTSUPP;
1433
1434 v.val = sk->sk_peek_off;
1435 break;
1436 case SO_NOFCS:
1437 v.val = sock_flag(sk, SOCK_NOFCS);
1438 break;
1439
1440 case SO_BINDTODEVICE:
1441 return sock_getbindtodevice(sk, optval, optlen, len);
1442
1443 case SO_GET_FILTER:
1444 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1445 if (len < 0)
1446 return len;
1447
1448 goto lenout;
1449
1450 case SO_LOCK_FILTER:
1451 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1452 break;
1453
1454 case SO_BPF_EXTENSIONS:
1455 v.val = bpf_tell_extensions();
1456 break;
1457
1458 case SO_SELECT_ERR_QUEUE:
1459 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1460 break;
1461
1462#ifdef CONFIG_NET_RX_BUSY_POLL
1463 case SO_BUSY_POLL:
1464 v.val = sk->sk_ll_usec;
1465 break;
1466#endif
1467
1468 case SO_MAX_PACING_RATE:
1469 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1470 lv = sizeof(v.ulval);
1471 v.ulval = sk->sk_max_pacing_rate;
1472 } else {
1473 /* 32bit version */
1474 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1475 }
1476 break;
1477
1478 case SO_INCOMING_CPU:
1479 v.val = READ_ONCE(sk->sk_incoming_cpu);
1480 break;
1481
1482 case SO_MEMINFO:
1483 {
1484 u32 meminfo[SK_MEMINFO_VARS];
1485
1486 sk_get_meminfo(sk, meminfo);
1487
1488 len = min_t(unsigned int, len, sizeof(meminfo));
1489 if (copy_to_user(optval, &meminfo, len))
1490 return -EFAULT;
1491
1492 goto lenout;
1493 }
1494
1495#ifdef CONFIG_NET_RX_BUSY_POLL
1496 case SO_INCOMING_NAPI_ID:
1497 v.val = READ_ONCE(sk->sk_napi_id);
1498
1499 /* aggregate non-NAPI IDs down to 0 */
1500 if (v.val < MIN_NAPI_ID)
1501 v.val = 0;
1502
1503 break;
1504#endif
1505
1506 case SO_COOKIE:
1507 lv = sizeof(u64);
1508 if (len < lv)
1509 return -EINVAL;
1510 v.val64 = sock_gen_cookie(sk);
1511 break;
1512
1513 case SO_ZEROCOPY:
1514 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1515 break;
1516
1517 case SO_TXTIME:
1518 lv = sizeof(v.txtime);
1519 v.txtime.clockid = sk->sk_clockid;
1520 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1521 SOF_TXTIME_DEADLINE_MODE : 0;
1522 v.txtime.flags |= sk->sk_txtime_report_errors ?
1523 SOF_TXTIME_REPORT_ERRORS : 0;
1524 break;
1525
1526 case SO_BINDTOIFINDEX:
1527 v.val = sk->sk_bound_dev_if;
1528 break;
1529
1530 default:
1531 /* We implement the SO_SNDLOWAT etc to not be settable
1532 * (1003.1g 7).
1533 */
1534 return -ENOPROTOOPT;
1535 }
1536
1537 if (len > lv)
1538 len = lv;
1539 if (copy_to_user(optval, &v, len))
1540 return -EFAULT;
1541lenout:
1542 if (put_user(len, optlen))
1543 return -EFAULT;
1544 return 0;
1545}
1546
1547/*
1548 * Initialize an sk_lock.
1549 *
1550 * (We also register the sk_lock with the lock validator.)
1551 */
1552static inline void sock_lock_init(struct sock *sk)
1553{
1554 if (sk->sk_kern_sock)
1555 sock_lock_init_class_and_name(
1556 sk,
1557 af_family_kern_slock_key_strings[sk->sk_family],
1558 af_family_kern_slock_keys + sk->sk_family,
1559 af_family_kern_key_strings[sk->sk_family],
1560 af_family_kern_keys + sk->sk_family);
1561 else
1562 sock_lock_init_class_and_name(
1563 sk,
1564 af_family_slock_key_strings[sk->sk_family],
1565 af_family_slock_keys + sk->sk_family,
1566 af_family_key_strings[sk->sk_family],
1567 af_family_keys + sk->sk_family);
1568}
1569
1570/*
1571 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1572 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1573 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1574 */
1575static void sock_copy(struct sock *nsk, const struct sock *osk)
1576{
1577#ifdef CONFIG_SECURITY_NETWORK
1578 void *sptr = nsk->sk_security;
1579#endif
1580 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1581
1582 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1583 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1584
1585#ifdef CONFIG_SECURITY_NETWORK
1586 nsk->sk_security = sptr;
1587 security_sk_clone(osk, nsk);
1588#endif
1589}
1590
1591static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1592 int family)
1593{
1594 struct sock *sk;
1595 struct kmem_cache *slab;
1596
1597 slab = prot->slab;
1598 if (slab != NULL) {
1599 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1600 if (!sk)
1601 return sk;
1602 if (want_init_on_alloc(priority))
1603 sk_prot_clear_nulls(sk, prot->obj_size);
1604 } else
1605 sk = kmalloc(prot->obj_size, priority);
1606
1607 if (sk != NULL) {
1608 if (security_sk_alloc(sk, family, priority))
1609 goto out_free;
1610
1611 if (!try_module_get(prot->owner))
1612 goto out_free_sec;
1613 sk_tx_queue_clear(sk);
1614 }
1615
1616 return sk;
1617
1618out_free_sec:
1619 security_sk_free(sk);
1620out_free:
1621 if (slab != NULL)
1622 kmem_cache_free(slab, sk);
1623 else
1624 kfree(sk);
1625 return NULL;
1626}
1627
1628static void sk_prot_free(struct proto *prot, struct sock *sk)
1629{
1630 struct kmem_cache *slab;
1631 struct module *owner;
1632
1633 owner = prot->owner;
1634 slab = prot->slab;
1635
1636 cgroup_sk_free(&sk->sk_cgrp_data);
1637 mem_cgroup_sk_free(sk);
1638 security_sk_free(sk);
1639 if (slab != NULL)
1640 kmem_cache_free(slab, sk);
1641 else
1642 kfree(sk);
1643 module_put(owner);
1644}
1645
1646/**
1647 * sk_alloc - All socket objects are allocated here
1648 * @net: the applicable net namespace
1649 * @family: protocol family
1650 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1651 * @prot: struct proto associated with this new sock instance
1652 * @kern: is this to be a kernel socket?
1653 */
1654struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1655 struct proto *prot, int kern)
1656{
1657 struct sock *sk;
1658
1659 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1660 if (sk) {
1661 sk->sk_family = family;
1662 /*
1663 * See comment in struct sock definition to understand
1664 * why we need sk_prot_creator -acme
1665 */
1666 sk->sk_prot = sk->sk_prot_creator = prot;
1667 sk->sk_kern_sock = kern;
1668 sock_lock_init(sk);
1669 sk->sk_net_refcnt = kern ? 0 : 1;
1670 if (likely(sk->sk_net_refcnt)) {
1671 get_net(net);
1672 sock_inuse_add(net, 1);
1673 }
1674
1675 sock_net_set(sk, net);
1676 refcount_set(&sk->sk_wmem_alloc, 1);
1677
1678 mem_cgroup_sk_alloc(sk);
1679 cgroup_sk_alloc(&sk->sk_cgrp_data);
1680 sock_update_classid(&sk->sk_cgrp_data);
1681 sock_update_netprioidx(&sk->sk_cgrp_data);
1682 }
1683
1684 return sk;
1685}
1686EXPORT_SYMBOL(sk_alloc);
1687
1688/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1689 * grace period. This is the case for UDP sockets and TCP listeners.
1690 */
1691static void __sk_destruct(struct rcu_head *head)
1692{
1693 struct sock *sk = container_of(head, struct sock, sk_rcu);
1694 struct sk_filter *filter;
1695
1696 if (sk->sk_destruct)
1697 sk->sk_destruct(sk);
1698
1699 filter = rcu_dereference_check(sk->sk_filter,
1700 refcount_read(&sk->sk_wmem_alloc) == 0);
1701 if (filter) {
1702 sk_filter_uncharge(sk, filter);
1703 RCU_INIT_POINTER(sk->sk_filter, NULL);
1704 }
1705
1706 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1707
1708#ifdef CONFIG_BPF_SYSCALL
1709 bpf_sk_storage_free(sk);
1710#endif
1711
1712 if (atomic_read(&sk->sk_omem_alloc))
1713 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1714 __func__, atomic_read(&sk->sk_omem_alloc));
1715
1716 if (sk->sk_frag.page) {
1717 put_page(sk->sk_frag.page);
1718 sk->sk_frag.page = NULL;
1719 }
1720
1721 if (sk->sk_peer_cred)
1722 put_cred(sk->sk_peer_cred);
1723 put_pid(sk->sk_peer_pid);
1724 if (likely(sk->sk_net_refcnt))
1725 put_net(sock_net(sk));
1726 sk_prot_free(sk->sk_prot_creator, sk);
1727}
1728
1729void sk_destruct(struct sock *sk)
1730{
1731 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1732
1733 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1734 reuseport_detach_sock(sk);
1735 use_call_rcu = true;
1736 }
1737
1738 if (use_call_rcu)
1739 call_rcu(&sk->sk_rcu, __sk_destruct);
1740 else
1741 __sk_destruct(&sk->sk_rcu);
1742}
1743
1744static void __sk_free(struct sock *sk)
1745{
1746 if (likely(sk->sk_net_refcnt))
1747 sock_inuse_add(sock_net(sk), -1);
1748
1749 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1750 sock_diag_broadcast_destroy(sk);
1751 else
1752 sk_destruct(sk);
1753}
1754
1755void sk_free(struct sock *sk)
1756{
1757 /*
1758 * We subtract one from sk_wmem_alloc and can know if
1759 * some packets are still in some tx queue.
1760 * If not null, sock_wfree() will call __sk_free(sk) later
1761 */
1762 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1763 __sk_free(sk);
1764}
1765EXPORT_SYMBOL(sk_free);
1766
1767static void sk_init_common(struct sock *sk)
1768{
1769 skb_queue_head_init(&sk->sk_receive_queue);
1770 skb_queue_head_init(&sk->sk_write_queue);
1771 skb_queue_head_init(&sk->sk_error_queue);
1772
1773 rwlock_init(&sk->sk_callback_lock);
1774 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1775 af_rlock_keys + sk->sk_family,
1776 af_family_rlock_key_strings[sk->sk_family]);
1777 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1778 af_wlock_keys + sk->sk_family,
1779 af_family_wlock_key_strings[sk->sk_family]);
1780 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1781 af_elock_keys + sk->sk_family,
1782 af_family_elock_key_strings[sk->sk_family]);
1783 lockdep_set_class_and_name(&sk->sk_callback_lock,
1784 af_callback_keys + sk->sk_family,
1785 af_family_clock_key_strings[sk->sk_family]);
1786}
1787
1788/**
1789 * sk_clone_lock - clone a socket, and lock its clone
1790 * @sk: the socket to clone
1791 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1792 *
1793 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1794 */
1795struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1796{
1797 struct sock *newsk;
1798 bool is_charged = true;
1799
1800 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1801 if (newsk != NULL) {
1802 struct sk_filter *filter;
1803
1804 sock_copy(newsk, sk);
1805
1806 newsk->sk_prot_creator = sk->sk_prot;
1807
1808 /* SANITY */
1809 if (likely(newsk->sk_net_refcnt))
1810 get_net(sock_net(newsk));
1811 sk_node_init(&newsk->sk_node);
1812 sock_lock_init(newsk);
1813 bh_lock_sock(newsk);
1814 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1815 newsk->sk_backlog.len = 0;
1816
1817 atomic_set(&newsk->sk_rmem_alloc, 0);
1818 /*
1819 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1820 */
1821 refcount_set(&newsk->sk_wmem_alloc, 1);
1822 atomic_set(&newsk->sk_omem_alloc, 0);
1823 sk_init_common(newsk);
1824
1825 newsk->sk_dst_cache = NULL;
1826 newsk->sk_dst_pending_confirm = 0;
1827 newsk->sk_wmem_queued = 0;
1828 newsk->sk_forward_alloc = 0;
1829 atomic_set(&newsk->sk_drops, 0);
1830 newsk->sk_send_head = NULL;
1831 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1832 atomic_set(&newsk->sk_zckey, 0);
1833
1834 sock_reset_flag(newsk, SOCK_DONE);
1835 mem_cgroup_sk_alloc(newsk);
1836 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1837
1838 rcu_read_lock();
1839 filter = rcu_dereference(sk->sk_filter);
1840 if (filter != NULL)
1841 /* though it's an empty new sock, the charging may fail
1842 * if sysctl_optmem_max was changed between creation of
1843 * original socket and cloning
1844 */
1845 is_charged = sk_filter_charge(newsk, filter);
1846 RCU_INIT_POINTER(newsk->sk_filter, filter);
1847 rcu_read_unlock();
1848
1849 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1850 /* We need to make sure that we don't uncharge the new
1851 * socket if we couldn't charge it in the first place
1852 * as otherwise we uncharge the parent's filter.
1853 */
1854 if (!is_charged)
1855 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1856 sk_free_unlock_clone(newsk);
1857 newsk = NULL;
1858 goto out;
1859 }
1860 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1861
1862 if (bpf_sk_storage_clone(sk, newsk)) {
1863 sk_free_unlock_clone(newsk);
1864 newsk = NULL;
1865 goto out;
1866 }
1867
1868 newsk->sk_err = 0;
1869 newsk->sk_err_soft = 0;
1870 newsk->sk_priority = 0;
1871 newsk->sk_incoming_cpu = raw_smp_processor_id();
1872 if (likely(newsk->sk_net_refcnt))
1873 sock_inuse_add(sock_net(newsk), 1);
1874
1875 /*
1876 * Before updating sk_refcnt, we must commit prior changes to memory
1877 * (Documentation/RCU/rculist_nulls.txt for details)
1878 */
1879 smp_wmb();
1880 refcount_set(&newsk->sk_refcnt, 2);
1881
1882 /*
1883 * Increment the counter in the same struct proto as the master
1884 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1885 * is the same as sk->sk_prot->socks, as this field was copied
1886 * with memcpy).
1887 *
1888 * This _changes_ the previous behaviour, where
1889 * tcp_create_openreq_child always was incrementing the
1890 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1891 * to be taken into account in all callers. -acme
1892 */
1893 sk_refcnt_debug_inc(newsk);
1894 sk_set_socket(newsk, NULL);
1895 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1896
1897 if (newsk->sk_prot->sockets_allocated)
1898 sk_sockets_allocated_inc(newsk);
1899
1900 if (sock_needs_netstamp(sk) &&
1901 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1902 net_enable_timestamp();
1903 }
1904out:
1905 return newsk;
1906}
1907EXPORT_SYMBOL_GPL(sk_clone_lock);
1908
1909void sk_free_unlock_clone(struct sock *sk)
1910{
1911 /* It is still raw copy of parent, so invalidate
1912 * destructor and make plain sk_free() */
1913 sk->sk_destruct = NULL;
1914 bh_unlock_sock(sk);
1915 sk_free(sk);
1916}
1917EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1918
1919void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1920{
1921 u32 max_segs = 1;
1922
1923 sk_dst_set(sk, dst);
1924 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1925 if (sk->sk_route_caps & NETIF_F_GSO)
1926 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1927 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1928 if (sk_can_gso(sk)) {
1929 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1930 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1931 } else {
1932 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1933 sk->sk_gso_max_size = dst->dev->gso_max_size;
1934 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1935 }
1936 }
1937 sk->sk_gso_max_segs = max_segs;
1938}
1939EXPORT_SYMBOL_GPL(sk_setup_caps);
1940
1941/*
1942 * Simple resource managers for sockets.
1943 */
1944
1945
1946/*
1947 * Write buffer destructor automatically called from kfree_skb.
1948 */
1949void sock_wfree(struct sk_buff *skb)
1950{
1951 struct sock *sk = skb->sk;
1952 unsigned int len = skb->truesize;
1953
1954 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1955 /*
1956 * Keep a reference on sk_wmem_alloc, this will be released
1957 * after sk_write_space() call
1958 */
1959 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1960 sk->sk_write_space(sk);
1961 len = 1;
1962 }
1963 /*
1964 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1965 * could not do because of in-flight packets
1966 */
1967 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1968 __sk_free(sk);
1969}
1970EXPORT_SYMBOL(sock_wfree);
1971
1972/* This variant of sock_wfree() is used by TCP,
1973 * since it sets SOCK_USE_WRITE_QUEUE.
1974 */
1975void __sock_wfree(struct sk_buff *skb)
1976{
1977 struct sock *sk = skb->sk;
1978
1979 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1980 __sk_free(sk);
1981}
1982
1983void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1984{
1985 skb_orphan(skb);
1986 skb->sk = sk;
1987#ifdef CONFIG_INET
1988 if (unlikely(!sk_fullsock(sk))) {
1989 skb->destructor = sock_edemux;
1990 sock_hold(sk);
1991 return;
1992 }
1993#endif
1994 skb->destructor = sock_wfree;
1995 skb_set_hash_from_sk(skb, sk);
1996 /*
1997 * We used to take a refcount on sk, but following operation
1998 * is enough to guarantee sk_free() wont free this sock until
1999 * all in-flight packets are completed
2000 */
2001 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2002}
2003EXPORT_SYMBOL(skb_set_owner_w);
2004
2005static bool can_skb_orphan_partial(const struct sk_buff *skb)
2006{
2007#ifdef CONFIG_TLS_DEVICE
2008 /* Drivers depend on in-order delivery for crypto offload,
2009 * partial orphan breaks out-of-order-OK logic.
2010 */
2011 if (skb->decrypted)
2012 return false;
2013#endif
2014 return (skb->destructor == sock_wfree ||
2015 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2016}
2017
2018/* This helper is used by netem, as it can hold packets in its
2019 * delay queue. We want to allow the owner socket to send more
2020 * packets, as if they were already TX completed by a typical driver.
2021 * But we also want to keep skb->sk set because some packet schedulers
2022 * rely on it (sch_fq for example).
2023 */
2024void skb_orphan_partial(struct sk_buff *skb)
2025{
2026 if (skb_is_tcp_pure_ack(skb))
2027 return;
2028
2029 if (can_skb_orphan_partial(skb)) {
2030 struct sock *sk = skb->sk;
2031
2032 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2033 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2034 skb->destructor = sock_efree;
2035 }
2036 } else {
2037 skb_orphan(skb);
2038 }
2039}
2040EXPORT_SYMBOL(skb_orphan_partial);
2041
2042/*
2043 * Read buffer destructor automatically called from kfree_skb.
2044 */
2045void sock_rfree(struct sk_buff *skb)
2046{
2047 struct sock *sk = skb->sk;
2048 unsigned int len = skb->truesize;
2049
2050 atomic_sub(len, &sk->sk_rmem_alloc);
2051 sk_mem_uncharge(sk, len);
2052}
2053EXPORT_SYMBOL(sock_rfree);
2054
2055/*
2056 * Buffer destructor for skbs that are not used directly in read or write
2057 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2058 */
2059void sock_efree(struct sk_buff *skb)
2060{
2061 sock_put(skb->sk);
2062}
2063EXPORT_SYMBOL(sock_efree);
2064
2065kuid_t sock_i_uid(struct sock *sk)
2066{
2067 kuid_t uid;
2068
2069 read_lock_bh(&sk->sk_callback_lock);
2070 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2071 read_unlock_bh(&sk->sk_callback_lock);
2072 return uid;
2073}
2074EXPORT_SYMBOL(sock_i_uid);
2075
2076unsigned long sock_i_ino(struct sock *sk)
2077{
2078 unsigned long ino;
2079
2080 read_lock_bh(&sk->sk_callback_lock);
2081 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2082 read_unlock_bh(&sk->sk_callback_lock);
2083 return ino;
2084}
2085EXPORT_SYMBOL(sock_i_ino);
2086
2087/*
2088 * Allocate a skb from the socket's send buffer.
2089 */
2090struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2091 gfp_t priority)
2092{
2093 if (force ||
2094 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2095 struct sk_buff *skb = alloc_skb(size, priority);
2096
2097 if (skb) {
2098 skb_set_owner_w(skb, sk);
2099 return skb;
2100 }
2101 }
2102 return NULL;
2103}
2104EXPORT_SYMBOL(sock_wmalloc);
2105
2106static void sock_ofree(struct sk_buff *skb)
2107{
2108 struct sock *sk = skb->sk;
2109
2110 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2111}
2112
2113struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2114 gfp_t priority)
2115{
2116 struct sk_buff *skb;
2117
2118 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2119 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2120 sysctl_optmem_max)
2121 return NULL;
2122
2123 skb = alloc_skb(size, priority);
2124 if (!skb)
2125 return NULL;
2126
2127 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2128 skb->sk = sk;
2129 skb->destructor = sock_ofree;
2130 return skb;
2131}
2132
2133/*
2134 * Allocate a memory block from the socket's option memory buffer.
2135 */
2136void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2137{
2138 if ((unsigned int)size <= sysctl_optmem_max &&
2139 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2140 void *mem;
2141 /* First do the add, to avoid the race if kmalloc
2142 * might sleep.
2143 */
2144 atomic_add(size, &sk->sk_omem_alloc);
2145 mem = kmalloc(size, priority);
2146 if (mem)
2147 return mem;
2148 atomic_sub(size, &sk->sk_omem_alloc);
2149 }
2150 return NULL;
2151}
2152EXPORT_SYMBOL(sock_kmalloc);
2153
2154/* Free an option memory block. Note, we actually want the inline
2155 * here as this allows gcc to detect the nullify and fold away the
2156 * condition entirely.
2157 */
2158static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2159 const bool nullify)
2160{
2161 if (WARN_ON_ONCE(!mem))
2162 return;
2163 if (nullify)
2164 kzfree(mem);
2165 else
2166 kfree(mem);
2167 atomic_sub(size, &sk->sk_omem_alloc);
2168}
2169
2170void sock_kfree_s(struct sock *sk, void *mem, int size)
2171{
2172 __sock_kfree_s(sk, mem, size, false);
2173}
2174EXPORT_SYMBOL(sock_kfree_s);
2175
2176void sock_kzfree_s(struct sock *sk, void *mem, int size)
2177{
2178 __sock_kfree_s(sk, mem, size, true);
2179}
2180EXPORT_SYMBOL(sock_kzfree_s);
2181
2182/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2183 I think, these locks should be removed for datagram sockets.
2184 */
2185static long sock_wait_for_wmem(struct sock *sk, long timeo)
2186{
2187 DEFINE_WAIT(wait);
2188
2189 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2190 for (;;) {
2191 if (!timeo)
2192 break;
2193 if (signal_pending(current))
2194 break;
2195 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2196 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2197 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2198 break;
2199 if (sk->sk_shutdown & SEND_SHUTDOWN)
2200 break;
2201 if (sk->sk_err)
2202 break;
2203 timeo = schedule_timeout(timeo);
2204 }
2205 finish_wait(sk_sleep(sk), &wait);
2206 return timeo;
2207}
2208
2209
2210/*
2211 * Generic send/receive buffer handlers
2212 */
2213
2214struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2215 unsigned long data_len, int noblock,
2216 int *errcode, int max_page_order)
2217{
2218 struct sk_buff *skb;
2219 long timeo;
2220 int err;
2221
2222 timeo = sock_sndtimeo(sk, noblock);
2223 for (;;) {
2224 err = sock_error(sk);
2225 if (err != 0)
2226 goto failure;
2227
2228 err = -EPIPE;
2229 if (sk->sk_shutdown & SEND_SHUTDOWN)
2230 goto failure;
2231
2232 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2233 break;
2234
2235 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2236 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2237 err = -EAGAIN;
2238 if (!timeo)
2239 goto failure;
2240 if (signal_pending(current))
2241 goto interrupted;
2242 timeo = sock_wait_for_wmem(sk, timeo);
2243 }
2244 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2245 errcode, sk->sk_allocation);
2246 if (skb)
2247 skb_set_owner_w(skb, sk);
2248 return skb;
2249
2250interrupted:
2251 err = sock_intr_errno(timeo);
2252failure:
2253 *errcode = err;
2254 return NULL;
2255}
2256EXPORT_SYMBOL(sock_alloc_send_pskb);
2257
2258struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2259 int noblock, int *errcode)
2260{
2261 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2262}
2263EXPORT_SYMBOL(sock_alloc_send_skb);
2264
2265int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2266 struct sockcm_cookie *sockc)
2267{
2268 u32 tsflags;
2269
2270 switch (cmsg->cmsg_type) {
2271 case SO_MARK:
2272 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2273 return -EPERM;
2274 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2275 return -EINVAL;
2276 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2277 break;
2278 case SO_TIMESTAMPING_OLD:
2279 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2280 return -EINVAL;
2281
2282 tsflags = *(u32 *)CMSG_DATA(cmsg);
2283 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2284 return -EINVAL;
2285
2286 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2287 sockc->tsflags |= tsflags;
2288 break;
2289 case SCM_TXTIME:
2290 if (!sock_flag(sk, SOCK_TXTIME))
2291 return -EINVAL;
2292 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2293 return -EINVAL;
2294 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2295 break;
2296 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2297 case SCM_RIGHTS:
2298 case SCM_CREDENTIALS:
2299 break;
2300 default:
2301 return -EINVAL;
2302 }
2303 return 0;
2304}
2305EXPORT_SYMBOL(__sock_cmsg_send);
2306
2307int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2308 struct sockcm_cookie *sockc)
2309{
2310 struct cmsghdr *cmsg;
2311 int ret;
2312
2313 for_each_cmsghdr(cmsg, msg) {
2314 if (!CMSG_OK(msg, cmsg))
2315 return -EINVAL;
2316 if (cmsg->cmsg_level != SOL_SOCKET)
2317 continue;
2318 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2319 if (ret)
2320 return ret;
2321 }
2322 return 0;
2323}
2324EXPORT_SYMBOL(sock_cmsg_send);
2325
2326static void sk_enter_memory_pressure(struct sock *sk)
2327{
2328 if (!sk->sk_prot->enter_memory_pressure)
2329 return;
2330
2331 sk->sk_prot->enter_memory_pressure(sk);
2332}
2333
2334static void sk_leave_memory_pressure(struct sock *sk)
2335{
2336 if (sk->sk_prot->leave_memory_pressure) {
2337 sk->sk_prot->leave_memory_pressure(sk);
2338 } else {
2339 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2340
2341 if (memory_pressure && READ_ONCE(*memory_pressure))
2342 WRITE_ONCE(*memory_pressure, 0);
2343 }
2344}
2345
2346/* On 32bit arches, an skb frag is limited to 2^15 */
2347#define SKB_FRAG_PAGE_ORDER get_order(32768)
2348DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2349
2350/**
2351 * skb_page_frag_refill - check that a page_frag contains enough room
2352 * @sz: minimum size of the fragment we want to get
2353 * @pfrag: pointer to page_frag
2354 * @gfp: priority for memory allocation
2355 *
2356 * Note: While this allocator tries to use high order pages, there is
2357 * no guarantee that allocations succeed. Therefore, @sz MUST be
2358 * less or equal than PAGE_SIZE.
2359 */
2360bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2361{
2362 if (pfrag->page) {
2363 if (page_ref_count(pfrag->page) == 1) {
2364 pfrag->offset = 0;
2365 return true;
2366 }
2367 if (pfrag->offset + sz <= pfrag->size)
2368 return true;
2369 put_page(pfrag->page);
2370 }
2371
2372 pfrag->offset = 0;
2373 if (SKB_FRAG_PAGE_ORDER &&
2374 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2375 /* Avoid direct reclaim but allow kswapd to wake */
2376 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2377 __GFP_COMP | __GFP_NOWARN |
2378 __GFP_NORETRY,
2379 SKB_FRAG_PAGE_ORDER);
2380 if (likely(pfrag->page)) {
2381 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2382 return true;
2383 }
2384 }
2385 pfrag->page = alloc_page(gfp);
2386 if (likely(pfrag->page)) {
2387 pfrag->size = PAGE_SIZE;
2388 return true;
2389 }
2390 return false;
2391}
2392EXPORT_SYMBOL(skb_page_frag_refill);
2393
2394bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2395{
2396 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2397 return true;
2398
2399 sk_enter_memory_pressure(sk);
2400 sk_stream_moderate_sndbuf(sk);
2401 return false;
2402}
2403EXPORT_SYMBOL(sk_page_frag_refill);
2404
2405static void __lock_sock(struct sock *sk)
2406 __releases(&sk->sk_lock.slock)
2407 __acquires(&sk->sk_lock.slock)
2408{
2409 DEFINE_WAIT(wait);
2410
2411 for (;;) {
2412 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2413 TASK_UNINTERRUPTIBLE);
2414 spin_unlock_bh(&sk->sk_lock.slock);
2415 schedule();
2416 spin_lock_bh(&sk->sk_lock.slock);
2417 if (!sock_owned_by_user(sk))
2418 break;
2419 }
2420 finish_wait(&sk->sk_lock.wq, &wait);
2421}
2422
2423void __release_sock(struct sock *sk)
2424 __releases(&sk->sk_lock.slock)
2425 __acquires(&sk->sk_lock.slock)
2426{
2427 struct sk_buff *skb, *next;
2428
2429 while ((skb = sk->sk_backlog.head) != NULL) {
2430 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2431
2432 spin_unlock_bh(&sk->sk_lock.slock);
2433
2434 do {
2435 next = skb->next;
2436 prefetch(next);
2437 WARN_ON_ONCE(skb_dst_is_noref(skb));
2438 skb_mark_not_on_list(skb);
2439 sk_backlog_rcv(sk, skb);
2440
2441 cond_resched();
2442
2443 skb = next;
2444 } while (skb != NULL);
2445
2446 spin_lock_bh(&sk->sk_lock.slock);
2447 }
2448
2449 /*
2450 * Doing the zeroing here guarantee we can not loop forever
2451 * while a wild producer attempts to flood us.
2452 */
2453 sk->sk_backlog.len = 0;
2454}
2455
2456void __sk_flush_backlog(struct sock *sk)
2457{
2458 spin_lock_bh(&sk->sk_lock.slock);
2459 __release_sock(sk);
2460 spin_unlock_bh(&sk->sk_lock.slock);
2461}
2462
2463/**
2464 * sk_wait_data - wait for data to arrive at sk_receive_queue
2465 * @sk: sock to wait on
2466 * @timeo: for how long
2467 * @skb: last skb seen on sk_receive_queue
2468 *
2469 * Now socket state including sk->sk_err is changed only under lock,
2470 * hence we may omit checks after joining wait queue.
2471 * We check receive queue before schedule() only as optimization;
2472 * it is very likely that release_sock() added new data.
2473 */
2474int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2475{
2476 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2477 int rc;
2478
2479 add_wait_queue(sk_sleep(sk), &wait);
2480 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2481 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2482 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2483 remove_wait_queue(sk_sleep(sk), &wait);
2484 return rc;
2485}
2486EXPORT_SYMBOL(sk_wait_data);
2487
2488/**
2489 * __sk_mem_raise_allocated - increase memory_allocated
2490 * @sk: socket
2491 * @size: memory size to allocate
2492 * @amt: pages to allocate
2493 * @kind: allocation type
2494 *
2495 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2496 */
2497int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2498{
2499 struct proto *prot = sk->sk_prot;
2500 long allocated = sk_memory_allocated_add(sk, amt);
2501 bool charged = true;
2502
2503 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2504 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2505 goto suppress_allocation;
2506
2507 /* Under limit. */
2508 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2509 sk_leave_memory_pressure(sk);
2510 return 1;
2511 }
2512
2513 /* Under pressure. */
2514 if (allocated > sk_prot_mem_limits(sk, 1))
2515 sk_enter_memory_pressure(sk);
2516
2517 /* Over hard limit. */
2518 if (allocated > sk_prot_mem_limits(sk, 2))
2519 goto suppress_allocation;
2520
2521 /* guarantee minimum buffer size under pressure */
2522 if (kind == SK_MEM_RECV) {
2523 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2524 return 1;
2525
2526 } else { /* SK_MEM_SEND */
2527 int wmem0 = sk_get_wmem0(sk, prot);
2528
2529 if (sk->sk_type == SOCK_STREAM) {
2530 if (sk->sk_wmem_queued < wmem0)
2531 return 1;
2532 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2533 return 1;
2534 }
2535 }
2536
2537 if (sk_has_memory_pressure(sk)) {
2538 u64 alloc;
2539
2540 if (!sk_under_memory_pressure(sk))
2541 return 1;
2542 alloc = sk_sockets_allocated_read_positive(sk);
2543 if (sk_prot_mem_limits(sk, 2) > alloc *
2544 sk_mem_pages(sk->sk_wmem_queued +
2545 atomic_read(&sk->sk_rmem_alloc) +
2546 sk->sk_forward_alloc))
2547 return 1;
2548 }
2549
2550suppress_allocation:
2551
2552 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2553 sk_stream_moderate_sndbuf(sk);
2554
2555 /* Fail only if socket is _under_ its sndbuf.
2556 * In this case we cannot block, so that we have to fail.
2557 */
2558 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2559 return 1;
2560 }
2561
2562 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2563 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2564
2565 sk_memory_allocated_sub(sk, amt);
2566
2567 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2568 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2569
2570 return 0;
2571}
2572EXPORT_SYMBOL(__sk_mem_raise_allocated);
2573
2574/**
2575 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2576 * @sk: socket
2577 * @size: memory size to allocate
2578 * @kind: allocation type
2579 *
2580 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2581 * rmem allocation. This function assumes that protocols which have
2582 * memory_pressure use sk_wmem_queued as write buffer accounting.
2583 */
2584int __sk_mem_schedule(struct sock *sk, int size, int kind)
2585{
2586 int ret, amt = sk_mem_pages(size);
2587
2588 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2589 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2590 if (!ret)
2591 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2592 return ret;
2593}
2594EXPORT_SYMBOL(__sk_mem_schedule);
2595
2596/**
2597 * __sk_mem_reduce_allocated - reclaim memory_allocated
2598 * @sk: socket
2599 * @amount: number of quanta
2600 *
2601 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2602 */
2603void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2604{
2605 sk_memory_allocated_sub(sk, amount);
2606
2607 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2608 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2609
2610 if (sk_under_memory_pressure(sk) &&
2611 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2612 sk_leave_memory_pressure(sk);
2613}
2614EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2615
2616/**
2617 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2618 * @sk: socket
2619 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2620 */
2621void __sk_mem_reclaim(struct sock *sk, int amount)
2622{
2623 amount >>= SK_MEM_QUANTUM_SHIFT;
2624 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2625 __sk_mem_reduce_allocated(sk, amount);
2626}
2627EXPORT_SYMBOL(__sk_mem_reclaim);
2628
2629int sk_set_peek_off(struct sock *sk, int val)
2630{
2631 sk->sk_peek_off = val;
2632 return 0;
2633}
2634EXPORT_SYMBOL_GPL(sk_set_peek_off);
2635
2636/*
2637 * Set of default routines for initialising struct proto_ops when
2638 * the protocol does not support a particular function. In certain
2639 * cases where it makes no sense for a protocol to have a "do nothing"
2640 * function, some default processing is provided.
2641 */
2642
2643int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2644{
2645 return -EOPNOTSUPP;
2646}
2647EXPORT_SYMBOL(sock_no_bind);
2648
2649int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2650 int len, int flags)
2651{
2652 return -EOPNOTSUPP;
2653}
2654EXPORT_SYMBOL(sock_no_connect);
2655
2656int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2657{
2658 return -EOPNOTSUPP;
2659}
2660EXPORT_SYMBOL(sock_no_socketpair);
2661
2662int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2663 bool kern)
2664{
2665 return -EOPNOTSUPP;
2666}
2667EXPORT_SYMBOL(sock_no_accept);
2668
2669int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2670 int peer)
2671{
2672 return -EOPNOTSUPP;
2673}
2674EXPORT_SYMBOL(sock_no_getname);
2675
2676int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2677{
2678 return -EOPNOTSUPP;
2679}
2680EXPORT_SYMBOL(sock_no_ioctl);
2681
2682int sock_no_listen(struct socket *sock, int backlog)
2683{
2684 return -EOPNOTSUPP;
2685}
2686EXPORT_SYMBOL(sock_no_listen);
2687
2688int sock_no_shutdown(struct socket *sock, int how)
2689{
2690 return -EOPNOTSUPP;
2691}
2692EXPORT_SYMBOL(sock_no_shutdown);
2693
2694int sock_no_setsockopt(struct socket *sock, int level, int optname,
2695 char __user *optval, unsigned int optlen)
2696{
2697 return -EOPNOTSUPP;
2698}
2699EXPORT_SYMBOL(sock_no_setsockopt);
2700
2701int sock_no_getsockopt(struct socket *sock, int level, int optname,
2702 char __user *optval, int __user *optlen)
2703{
2704 return -EOPNOTSUPP;
2705}
2706EXPORT_SYMBOL(sock_no_getsockopt);
2707
2708int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2709{
2710 return -EOPNOTSUPP;
2711}
2712EXPORT_SYMBOL(sock_no_sendmsg);
2713
2714int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2715{
2716 return -EOPNOTSUPP;
2717}
2718EXPORT_SYMBOL(sock_no_sendmsg_locked);
2719
2720int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2721 int flags)
2722{
2723 return -EOPNOTSUPP;
2724}
2725EXPORT_SYMBOL(sock_no_recvmsg);
2726
2727int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2728{
2729 /* Mirror missing mmap method error code */
2730 return -ENODEV;
2731}
2732EXPORT_SYMBOL(sock_no_mmap);
2733
2734ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2735{
2736 ssize_t res;
2737 struct msghdr msg = {.msg_flags = flags};
2738 struct kvec iov;
2739 char *kaddr = kmap(page);
2740 iov.iov_base = kaddr + offset;
2741 iov.iov_len = size;
2742 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2743 kunmap(page);
2744 return res;
2745}
2746EXPORT_SYMBOL(sock_no_sendpage);
2747
2748ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2749 int offset, size_t size, int flags)
2750{
2751 ssize_t res;
2752 struct msghdr msg = {.msg_flags = flags};
2753 struct kvec iov;
2754 char *kaddr = kmap(page);
2755
2756 iov.iov_base = kaddr + offset;
2757 iov.iov_len = size;
2758 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2759 kunmap(page);
2760 return res;
2761}
2762EXPORT_SYMBOL(sock_no_sendpage_locked);
2763
2764/*
2765 * Default Socket Callbacks
2766 */
2767
2768static void sock_def_wakeup(struct sock *sk)
2769{
2770 struct socket_wq *wq;
2771
2772 rcu_read_lock();
2773 wq = rcu_dereference(sk->sk_wq);
2774 if (skwq_has_sleeper(wq))
2775 wake_up_interruptible_all(&wq->wait);
2776 rcu_read_unlock();
2777}
2778
2779static void sock_def_error_report(struct sock *sk)
2780{
2781 struct socket_wq *wq;
2782
2783 rcu_read_lock();
2784 wq = rcu_dereference(sk->sk_wq);
2785 if (skwq_has_sleeper(wq))
2786 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2787 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2788 rcu_read_unlock();
2789}
2790
2791static void sock_def_readable(struct sock *sk)
2792{
2793 struct socket_wq *wq;
2794
2795 rcu_read_lock();
2796 wq = rcu_dereference(sk->sk_wq);
2797 if (skwq_has_sleeper(wq))
2798 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2799 EPOLLRDNORM | EPOLLRDBAND);
2800 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2801 rcu_read_unlock();
2802}
2803
2804static void sock_def_write_space(struct sock *sk)
2805{
2806 struct socket_wq *wq;
2807
2808 rcu_read_lock();
2809
2810 /* Do not wake up a writer until he can make "significant"
2811 * progress. --DaveM
2812 */
2813 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2814 wq = rcu_dereference(sk->sk_wq);
2815 if (skwq_has_sleeper(wq))
2816 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2817 EPOLLWRNORM | EPOLLWRBAND);
2818
2819 /* Should agree with poll, otherwise some programs break */
2820 if (sock_writeable(sk))
2821 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2822 }
2823
2824 rcu_read_unlock();
2825}
2826
2827static void sock_def_destruct(struct sock *sk)
2828{
2829}
2830
2831void sk_send_sigurg(struct sock *sk)
2832{
2833 if (sk->sk_socket && sk->sk_socket->file)
2834 if (send_sigurg(&sk->sk_socket->file->f_owner))
2835 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2836}
2837EXPORT_SYMBOL(sk_send_sigurg);
2838
2839void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2840 unsigned long expires)
2841{
2842 if (!mod_timer(timer, expires))
2843 sock_hold(sk);
2844}
2845EXPORT_SYMBOL(sk_reset_timer);
2846
2847void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2848{
2849 if (del_timer(timer))
2850 __sock_put(sk);
2851}
2852EXPORT_SYMBOL(sk_stop_timer);
2853
2854void sock_init_data(struct socket *sock, struct sock *sk)
2855{
2856 sk_init_common(sk);
2857 sk->sk_send_head = NULL;
2858
2859 timer_setup(&sk->sk_timer, NULL, 0);
2860
2861 sk->sk_allocation = GFP_KERNEL;
2862 sk->sk_rcvbuf = sysctl_rmem_default;
2863 sk->sk_sndbuf = sysctl_wmem_default;
2864 sk->sk_state = TCP_CLOSE;
2865 sk_set_socket(sk, sock);
2866
2867 sock_set_flag(sk, SOCK_ZAPPED);
2868
2869 if (sock) {
2870 sk->sk_type = sock->type;
2871 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2872 sock->sk = sk;
2873 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2874 } else {
2875 RCU_INIT_POINTER(sk->sk_wq, NULL);
2876 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2877 }
2878
2879 rwlock_init(&sk->sk_callback_lock);
2880 if (sk->sk_kern_sock)
2881 lockdep_set_class_and_name(
2882 &sk->sk_callback_lock,
2883 af_kern_callback_keys + sk->sk_family,
2884 af_family_kern_clock_key_strings[sk->sk_family]);
2885 else
2886 lockdep_set_class_and_name(
2887 &sk->sk_callback_lock,
2888 af_callback_keys + sk->sk_family,
2889 af_family_clock_key_strings[sk->sk_family]);
2890
2891 sk->sk_state_change = sock_def_wakeup;
2892 sk->sk_data_ready = sock_def_readable;
2893 sk->sk_write_space = sock_def_write_space;
2894 sk->sk_error_report = sock_def_error_report;
2895 sk->sk_destruct = sock_def_destruct;
2896
2897 sk->sk_frag.page = NULL;
2898 sk->sk_frag.offset = 0;
2899 sk->sk_peek_off = -1;
2900
2901 sk->sk_peer_pid = NULL;
2902 sk->sk_peer_cred = NULL;
2903 sk->sk_write_pending = 0;
2904 sk->sk_rcvlowat = 1;
2905 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2906 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2907
2908 sk->sk_stamp = SK_DEFAULT_STAMP;
2909#if BITS_PER_LONG==32
2910 seqlock_init(&sk->sk_stamp_seq);
2911#endif
2912 atomic_set(&sk->sk_zckey, 0);
2913
2914#ifdef CONFIG_NET_RX_BUSY_POLL
2915 sk->sk_napi_id = 0;
2916 sk->sk_ll_usec = sysctl_net_busy_read;
2917#endif
2918
2919 sk->sk_max_pacing_rate = ~0UL;
2920 sk->sk_pacing_rate = ~0UL;
2921 sk->sk_pacing_shift = 10;
2922 sk->sk_incoming_cpu = -1;
2923
2924 sk_rx_queue_clear(sk);
2925 /*
2926 * Before updating sk_refcnt, we must commit prior changes to memory
2927 * (Documentation/RCU/rculist_nulls.txt for details)
2928 */
2929 smp_wmb();
2930 refcount_set(&sk->sk_refcnt, 1);
2931 atomic_set(&sk->sk_drops, 0);
2932}
2933EXPORT_SYMBOL(sock_init_data);
2934
2935void lock_sock_nested(struct sock *sk, int subclass)
2936{
2937 might_sleep();
2938 spin_lock_bh(&sk->sk_lock.slock);
2939 if (sk->sk_lock.owned)
2940 __lock_sock(sk);
2941 sk->sk_lock.owned = 1;
2942 spin_unlock(&sk->sk_lock.slock);
2943 /*
2944 * The sk_lock has mutex_lock() semantics here:
2945 */
2946 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2947 local_bh_enable();
2948}
2949EXPORT_SYMBOL(lock_sock_nested);
2950
2951void release_sock(struct sock *sk)
2952{
2953 spin_lock_bh(&sk->sk_lock.slock);
2954 if (sk->sk_backlog.tail)
2955 __release_sock(sk);
2956
2957 /* Warning : release_cb() might need to release sk ownership,
2958 * ie call sock_release_ownership(sk) before us.
2959 */
2960 if (sk->sk_prot->release_cb)
2961 sk->sk_prot->release_cb(sk);
2962
2963 sock_release_ownership(sk);
2964 if (waitqueue_active(&sk->sk_lock.wq))
2965 wake_up(&sk->sk_lock.wq);
2966 spin_unlock_bh(&sk->sk_lock.slock);
2967}
2968EXPORT_SYMBOL(release_sock);
2969
2970/**
2971 * lock_sock_fast - fast version of lock_sock
2972 * @sk: socket
2973 *
2974 * This version should be used for very small section, where process wont block
2975 * return false if fast path is taken:
2976 *
2977 * sk_lock.slock locked, owned = 0, BH disabled
2978 *
2979 * return true if slow path is taken:
2980 *
2981 * sk_lock.slock unlocked, owned = 1, BH enabled
2982 */
2983bool lock_sock_fast(struct sock *sk)
2984{
2985 might_sleep();
2986 spin_lock_bh(&sk->sk_lock.slock);
2987
2988 if (!sk->sk_lock.owned)
2989 /*
2990 * Note : We must disable BH
2991 */
2992 return false;
2993
2994 __lock_sock(sk);
2995 sk->sk_lock.owned = 1;
2996 spin_unlock(&sk->sk_lock.slock);
2997 /*
2998 * The sk_lock has mutex_lock() semantics here:
2999 */
3000 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3001 local_bh_enable();
3002 return true;
3003}
3004EXPORT_SYMBOL(lock_sock_fast);
3005
3006int sock_gettstamp(struct socket *sock, void __user *userstamp,
3007 bool timeval, bool time32)
3008{
3009 struct sock *sk = sock->sk;
3010 struct timespec64 ts;
3011
3012 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3013 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3014 if (ts.tv_sec == -1)
3015 return -ENOENT;
3016 if (ts.tv_sec == 0) {
3017 ktime_t kt = ktime_get_real();
3018 sock_write_timestamp(sk, kt);;
3019 ts = ktime_to_timespec64(kt);
3020 }
3021
3022 if (timeval)
3023 ts.tv_nsec /= 1000;
3024
3025#ifdef CONFIG_COMPAT_32BIT_TIME
3026 if (time32)
3027 return put_old_timespec32(&ts, userstamp);
3028#endif
3029#ifdef CONFIG_SPARC64
3030 /* beware of padding in sparc64 timeval */
3031 if (timeval && !in_compat_syscall()) {
3032 struct __kernel_old_timeval __user tv = {
3033 .tv_sec = ts.tv_sec,
3034 .tv_usec = ts.tv_nsec,
3035 };
3036 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3037 return -EFAULT;
3038 return 0;
3039 }
3040#endif
3041 return put_timespec64(&ts, userstamp);
3042}
3043EXPORT_SYMBOL(sock_gettstamp);
3044
3045void sock_enable_timestamp(struct sock *sk, int flag)
3046{
3047 if (!sock_flag(sk, flag)) {
3048 unsigned long previous_flags = sk->sk_flags;
3049
3050 sock_set_flag(sk, flag);
3051 /*
3052 * we just set one of the two flags which require net
3053 * time stamping, but time stamping might have been on
3054 * already because of the other one
3055 */
3056 if (sock_needs_netstamp(sk) &&
3057 !(previous_flags & SK_FLAGS_TIMESTAMP))
3058 net_enable_timestamp();
3059 }
3060}
3061
3062int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3063 int level, int type)
3064{
3065 struct sock_exterr_skb *serr;
3066 struct sk_buff *skb;
3067 int copied, err;
3068
3069 err = -EAGAIN;
3070 skb = sock_dequeue_err_skb(sk);
3071 if (skb == NULL)
3072 goto out;
3073
3074 copied = skb->len;
3075 if (copied > len) {
3076 msg->msg_flags |= MSG_TRUNC;
3077 copied = len;
3078 }
3079 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3080 if (err)
3081 goto out_free_skb;
3082
3083 sock_recv_timestamp(msg, sk, skb);
3084
3085 serr = SKB_EXT_ERR(skb);
3086 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3087
3088 msg->msg_flags |= MSG_ERRQUEUE;
3089 err = copied;
3090
3091out_free_skb:
3092 kfree_skb(skb);
3093out:
3094 return err;
3095}
3096EXPORT_SYMBOL(sock_recv_errqueue);
3097
3098/*
3099 * Get a socket option on an socket.
3100 *
3101 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3102 * asynchronous errors should be reported by getsockopt. We assume
3103 * this means if you specify SO_ERROR (otherwise whats the point of it).
3104 */
3105int sock_common_getsockopt(struct socket *sock, int level, int optname,
3106 char __user *optval, int __user *optlen)
3107{
3108 struct sock *sk = sock->sk;
3109
3110 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3111}
3112EXPORT_SYMBOL(sock_common_getsockopt);
3113
3114#ifdef CONFIG_COMPAT
3115int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3116 char __user *optval, int __user *optlen)
3117{
3118 struct sock *sk = sock->sk;
3119
3120 if (sk->sk_prot->compat_getsockopt != NULL)
3121 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3122 optval, optlen);
3123 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3124}
3125EXPORT_SYMBOL(compat_sock_common_getsockopt);
3126#endif
3127
3128int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3129 int flags)
3130{
3131 struct sock *sk = sock->sk;
3132 int addr_len = 0;
3133 int err;
3134
3135 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3136 flags & ~MSG_DONTWAIT, &addr_len);
3137 if (err >= 0)
3138 msg->msg_namelen = addr_len;
3139 return err;
3140}
3141EXPORT_SYMBOL(sock_common_recvmsg);
3142
3143/*
3144 * Set socket options on an inet socket.
3145 */
3146int sock_common_setsockopt(struct socket *sock, int level, int optname,
3147 char __user *optval, unsigned int optlen)
3148{
3149 struct sock *sk = sock->sk;
3150
3151 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3152}
3153EXPORT_SYMBOL(sock_common_setsockopt);
3154
3155#ifdef CONFIG_COMPAT
3156int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3157 char __user *optval, unsigned int optlen)
3158{
3159 struct sock *sk = sock->sk;
3160
3161 if (sk->sk_prot->compat_setsockopt != NULL)
3162 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3163 optval, optlen);
3164 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3165}
3166EXPORT_SYMBOL(compat_sock_common_setsockopt);
3167#endif
3168
3169void sk_common_release(struct sock *sk)
3170{
3171 if (sk->sk_prot->destroy)
3172 sk->sk_prot->destroy(sk);
3173
3174 /*
3175 * Observation: when sock_common_release is called, processes have
3176 * no access to socket. But net still has.
3177 * Step one, detach it from networking:
3178 *
3179 * A. Remove from hash tables.
3180 */
3181
3182 sk->sk_prot->unhash(sk);
3183
3184 /*
3185 * In this point socket cannot receive new packets, but it is possible
3186 * that some packets are in flight because some CPU runs receiver and
3187 * did hash table lookup before we unhashed socket. They will achieve
3188 * receive queue and will be purged by socket destructor.
3189 *
3190 * Also we still have packets pending on receive queue and probably,
3191 * our own packets waiting in device queues. sock_destroy will drain
3192 * receive queue, but transmitted packets will delay socket destruction
3193 * until the last reference will be released.
3194 */
3195
3196 sock_orphan(sk);
3197
3198 xfrm_sk_free_policy(sk);
3199
3200 sk_refcnt_debug_release(sk);
3201
3202 sock_put(sk);
3203}
3204EXPORT_SYMBOL(sk_common_release);
3205
3206void sk_get_meminfo(const struct sock *sk, u32 *mem)
3207{
3208 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3209
3210 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3211 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3212 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3213 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3214 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3215 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3216 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3217 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3218 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3219}
3220
3221#ifdef CONFIG_PROC_FS
3222#define PROTO_INUSE_NR 64 /* should be enough for the first time */
3223struct prot_inuse {
3224 int val[PROTO_INUSE_NR];
3225};
3226
3227static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3228
3229void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3230{
3231 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3232}
3233EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3234
3235int sock_prot_inuse_get(struct net *net, struct proto *prot)
3236{
3237 int cpu, idx = prot->inuse_idx;
3238 int res = 0;
3239
3240 for_each_possible_cpu(cpu)
3241 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3242
3243 return res >= 0 ? res : 0;
3244}
3245EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3246
3247static void sock_inuse_add(struct net *net, int val)
3248{
3249 this_cpu_add(*net->core.sock_inuse, val);
3250}
3251
3252int sock_inuse_get(struct net *net)
3253{
3254 int cpu, res = 0;
3255
3256 for_each_possible_cpu(cpu)
3257 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3258
3259 return res;
3260}
3261
3262EXPORT_SYMBOL_GPL(sock_inuse_get);
3263
3264static int __net_init sock_inuse_init_net(struct net *net)
3265{
3266 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3267 if (net->core.prot_inuse == NULL)
3268 return -ENOMEM;
3269
3270 net->core.sock_inuse = alloc_percpu(int);
3271 if (net->core.sock_inuse == NULL)
3272 goto out;
3273
3274 return 0;
3275
3276out:
3277 free_percpu(net->core.prot_inuse);
3278 return -ENOMEM;
3279}
3280
3281static void __net_exit sock_inuse_exit_net(struct net *net)
3282{
3283 free_percpu(net->core.prot_inuse);
3284 free_percpu(net->core.sock_inuse);
3285}
3286
3287static struct pernet_operations net_inuse_ops = {
3288 .init = sock_inuse_init_net,
3289 .exit = sock_inuse_exit_net,
3290};
3291
3292static __init int net_inuse_init(void)
3293{
3294 if (register_pernet_subsys(&net_inuse_ops))
3295 panic("Cannot initialize net inuse counters");
3296
3297 return 0;
3298}
3299
3300core_initcall(net_inuse_init);
3301
3302static int assign_proto_idx(struct proto *prot)
3303{
3304 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3305
3306 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3307 pr_err("PROTO_INUSE_NR exhausted\n");
3308 return -ENOSPC;
3309 }
3310
3311 set_bit(prot->inuse_idx, proto_inuse_idx);
3312 return 0;
3313}
3314
3315static void release_proto_idx(struct proto *prot)
3316{
3317 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3318 clear_bit(prot->inuse_idx, proto_inuse_idx);
3319}
3320#else
3321static inline int assign_proto_idx(struct proto *prot)
3322{
3323 return 0;
3324}
3325
3326static inline void release_proto_idx(struct proto *prot)
3327{
3328}
3329
3330static void sock_inuse_add(struct net *net, int val)
3331{
3332}
3333#endif
3334
3335static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3336{
3337 if (!rsk_prot)
3338 return;
3339 kfree(rsk_prot->slab_name);
3340 rsk_prot->slab_name = NULL;
3341 kmem_cache_destroy(rsk_prot->slab);
3342 rsk_prot->slab = NULL;
3343}
3344
3345static int req_prot_init(const struct proto *prot)
3346{
3347 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3348
3349 if (!rsk_prot)
3350 return 0;
3351
3352 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3353 prot->name);
3354 if (!rsk_prot->slab_name)
3355 return -ENOMEM;
3356
3357 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3358 rsk_prot->obj_size, 0,
3359 SLAB_ACCOUNT | prot->slab_flags,
3360 NULL);
3361
3362 if (!rsk_prot->slab) {
3363 pr_crit("%s: Can't create request sock SLAB cache!\n",
3364 prot->name);
3365 return -ENOMEM;
3366 }
3367 return 0;
3368}
3369
3370int proto_register(struct proto *prot, int alloc_slab)
3371{
3372 int ret = -ENOBUFS;
3373
3374 if (alloc_slab) {
3375 prot->slab = kmem_cache_create_usercopy(prot->name,
3376 prot->obj_size, 0,
3377 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3378 prot->slab_flags,
3379 prot->useroffset, prot->usersize,
3380 NULL);
3381
3382 if (prot->slab == NULL) {
3383 pr_crit("%s: Can't create sock SLAB cache!\n",
3384 prot->name);
3385 goto out;
3386 }
3387
3388 if (req_prot_init(prot))
3389 goto out_free_request_sock_slab;
3390
3391 if (prot->twsk_prot != NULL) {
3392 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3393
3394 if (prot->twsk_prot->twsk_slab_name == NULL)
3395 goto out_free_request_sock_slab;
3396
3397 prot->twsk_prot->twsk_slab =
3398 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3399 prot->twsk_prot->twsk_obj_size,
3400 0,
3401 SLAB_ACCOUNT |
3402 prot->slab_flags,
3403 NULL);
3404 if (prot->twsk_prot->twsk_slab == NULL)
3405 goto out_free_timewait_sock_slab_name;
3406 }
3407 }
3408
3409 mutex_lock(&proto_list_mutex);
3410 ret = assign_proto_idx(prot);
3411 if (ret) {
3412 mutex_unlock(&proto_list_mutex);
3413 goto out_free_timewait_sock_slab_name;
3414 }
3415 list_add(&prot->node, &proto_list);
3416 mutex_unlock(&proto_list_mutex);
3417 return ret;
3418
3419out_free_timewait_sock_slab_name:
3420 if (alloc_slab && prot->twsk_prot)
3421 kfree(prot->twsk_prot->twsk_slab_name);
3422out_free_request_sock_slab:
3423 if (alloc_slab) {
3424 req_prot_cleanup(prot->rsk_prot);
3425
3426 kmem_cache_destroy(prot->slab);
3427 prot->slab = NULL;
3428 }
3429out:
3430 return ret;
3431}
3432EXPORT_SYMBOL(proto_register);
3433
3434void proto_unregister(struct proto *prot)
3435{
3436 mutex_lock(&proto_list_mutex);
3437 release_proto_idx(prot);
3438 list_del(&prot->node);
3439 mutex_unlock(&proto_list_mutex);
3440
3441 kmem_cache_destroy(prot->slab);
3442 prot->slab = NULL;
3443
3444 req_prot_cleanup(prot->rsk_prot);
3445
3446 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3447 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3448 kfree(prot->twsk_prot->twsk_slab_name);
3449 prot->twsk_prot->twsk_slab = NULL;
3450 }
3451}
3452EXPORT_SYMBOL(proto_unregister);
3453
3454int sock_load_diag_module(int family, int protocol)
3455{
3456 if (!protocol) {
3457 if (!sock_is_registered(family))
3458 return -ENOENT;
3459
3460 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3461 NETLINK_SOCK_DIAG, family);
3462 }
3463
3464#ifdef CONFIG_INET
3465 if (family == AF_INET &&
3466 protocol != IPPROTO_RAW &&
3467 !rcu_access_pointer(inet_protos[protocol]))
3468 return -ENOENT;
3469#endif
3470
3471 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3472 NETLINK_SOCK_DIAG, family, protocol);
3473}
3474EXPORT_SYMBOL(sock_load_diag_module);
3475
3476#ifdef CONFIG_PROC_FS
3477static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3478 __acquires(proto_list_mutex)
3479{
3480 mutex_lock(&proto_list_mutex);
3481 return seq_list_start_head(&proto_list, *pos);
3482}
3483
3484static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3485{
3486 return seq_list_next(v, &proto_list, pos);
3487}
3488
3489static void proto_seq_stop(struct seq_file *seq, void *v)
3490 __releases(proto_list_mutex)
3491{
3492 mutex_unlock(&proto_list_mutex);
3493}
3494
3495static char proto_method_implemented(const void *method)
3496{
3497 return method == NULL ? 'n' : 'y';
3498}
3499static long sock_prot_memory_allocated(struct proto *proto)
3500{
3501 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3502}
3503
3504static const char *sock_prot_memory_pressure(struct proto *proto)
3505{
3506 return proto->memory_pressure != NULL ?
3507 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3508}
3509
3510static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3511{
3512
3513 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3514 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3515 proto->name,
3516 proto->obj_size,
3517 sock_prot_inuse_get(seq_file_net(seq), proto),
3518 sock_prot_memory_allocated(proto),
3519 sock_prot_memory_pressure(proto),
3520 proto->max_header,
3521 proto->slab == NULL ? "no" : "yes",
3522 module_name(proto->owner),
3523 proto_method_implemented(proto->close),
3524 proto_method_implemented(proto->connect),
3525 proto_method_implemented(proto->disconnect),
3526 proto_method_implemented(proto->accept),
3527 proto_method_implemented(proto->ioctl),
3528 proto_method_implemented(proto->init),
3529 proto_method_implemented(proto->destroy),
3530 proto_method_implemented(proto->shutdown),
3531 proto_method_implemented(proto->setsockopt),
3532 proto_method_implemented(proto->getsockopt),
3533 proto_method_implemented(proto->sendmsg),
3534 proto_method_implemented(proto->recvmsg),
3535 proto_method_implemented(proto->sendpage),
3536 proto_method_implemented(proto->bind),
3537 proto_method_implemented(proto->backlog_rcv),
3538 proto_method_implemented(proto->hash),
3539 proto_method_implemented(proto->unhash),
3540 proto_method_implemented(proto->get_port),
3541 proto_method_implemented(proto->enter_memory_pressure));
3542}
3543
3544static int proto_seq_show(struct seq_file *seq, void *v)
3545{
3546 if (v == &proto_list)
3547 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3548 "protocol",
3549 "size",
3550 "sockets",
3551 "memory",
3552 "press",
3553 "maxhdr",
3554 "slab",
3555 "module",
3556 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3557 else
3558 proto_seq_printf(seq, list_entry(v, struct proto, node));
3559 return 0;
3560}
3561
3562static const struct seq_operations proto_seq_ops = {
3563 .start = proto_seq_start,
3564 .next = proto_seq_next,
3565 .stop = proto_seq_stop,
3566 .show = proto_seq_show,
3567};
3568
3569static __net_init int proto_init_net(struct net *net)
3570{
3571 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3572 sizeof(struct seq_net_private)))
3573 return -ENOMEM;
3574
3575 return 0;
3576}
3577
3578static __net_exit void proto_exit_net(struct net *net)
3579{
3580 remove_proc_entry("protocols", net->proc_net);
3581}
3582
3583
3584static __net_initdata struct pernet_operations proto_net_ops = {
3585 .init = proto_init_net,
3586 .exit = proto_exit_net,
3587};
3588
3589static int __init proto_init(void)
3590{
3591 return register_pernet_subsys(&proto_net_ops);
3592}
3593
3594subsys_initcall(proto_init);
3595
3596#endif /* PROC_FS */
3597
3598#ifdef CONFIG_NET_RX_BUSY_POLL
3599bool sk_busy_loop_end(void *p, unsigned long start_time)
3600{
3601 struct sock *sk = p;
3602
3603 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3604 sk_busy_loop_timeout(sk, start_time);
3605}
3606EXPORT_SYMBOL(sk_busy_loop_end);
3607#endif /* CONFIG_NET_RX_BUSY_POLL */