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 <linux/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/udp.h>
111#include <linux/init.h>
112#include <linux/highmem.h>
113#include <linux/user_namespace.h>
114#include <linux/static_key.h>
115#include <linux/memcontrol.h>
116#include <linux/prefetch.h>
117#include <linux/compat.h>
118#include <linux/mroute.h>
119#include <linux/mroute6.h>
120#include <linux/icmpv6.h>
121
122#include <linux/uaccess.h>
123
124#include <linux/netdevice.h>
125#include <net/protocol.h>
126#include <linux/skbuff.h>
127#include <linux/skbuff_ref.h>
128#include <net/net_namespace.h>
129#include <net/request_sock.h>
130#include <net/sock.h>
131#include <net/proto_memory.h>
132#include <linux/net_tstamp.h>
133#include <net/xfrm.h>
134#include <linux/ipsec.h>
135#include <net/cls_cgroup.h>
136#include <net/netprio_cgroup.h>
137#include <linux/sock_diag.h>
138
139#include <linux/filter.h>
140#include <net/sock_reuseport.h>
141#include <net/bpf_sk_storage.h>
142
143#include <trace/events/sock.h>
144
145#include <net/tcp.h>
146#include <net/busy_poll.h>
147#include <net/phonet/phonet.h>
148
149#include <linux/ethtool.h>
150
151#include "dev.h"
152
153static DEFINE_MUTEX(proto_list_mutex);
154static LIST_HEAD(proto_list);
155
156static void sock_def_write_space_wfree(struct sock *sk);
157static void sock_def_write_space(struct sock *sk);
158
159/**
160 * sk_ns_capable - General socket capability test
161 * @sk: Socket to use a capability on or through
162 * @user_ns: The user namespace of the capability to use
163 * @cap: The capability to use
164 *
165 * Test to see if the opener of the socket had when the socket was
166 * created and the current process has the capability @cap in the user
167 * namespace @user_ns.
168 */
169bool sk_ns_capable(const struct sock *sk,
170 struct user_namespace *user_ns, int cap)
171{
172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
173 ns_capable(user_ns, cap);
174}
175EXPORT_SYMBOL(sk_ns_capable);
176
177/**
178 * sk_capable - Socket global capability test
179 * @sk: Socket to use a capability on or through
180 * @cap: The global capability to use
181 *
182 * Test to see if the opener of the socket had when the socket was
183 * created and the current process has the capability @cap in all user
184 * namespaces.
185 */
186bool sk_capable(const struct sock *sk, int cap)
187{
188 return sk_ns_capable(sk, &init_user_ns, cap);
189}
190EXPORT_SYMBOL(sk_capable);
191
192/**
193 * sk_net_capable - Network namespace socket capability test
194 * @sk: Socket to use a capability on or through
195 * @cap: The capability to use
196 *
197 * Test to see if the opener of the socket had when the socket was created
198 * and the current process has the capability @cap over the network namespace
199 * the socket is a member of.
200 */
201bool sk_net_capable(const struct sock *sk, int cap)
202{
203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
204}
205EXPORT_SYMBOL(sk_net_capable);
206
207/*
208 * Each address family might have different locking rules, so we have
209 * one slock key per address family and separate keys for internal and
210 * userspace sockets.
211 */
212static struct lock_class_key af_family_keys[AF_MAX];
213static struct lock_class_key af_family_kern_keys[AF_MAX];
214static struct lock_class_key af_family_slock_keys[AF_MAX];
215static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
216
217/*
218 * Make lock validator output more readable. (we pre-construct these
219 * strings build-time, so that runtime initialization of socket
220 * locks is fast):
221 */
222
223#define _sock_locks(x) \
224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
233 x "27" , x "28" , x "AF_CAN" , \
234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 x "AF_MCTP" , \
240 x "AF_MAX"
241
242static const char *const af_family_key_strings[AF_MAX+1] = {
243 _sock_locks("sk_lock-")
244};
245static const char *const af_family_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("slock-")
247};
248static const char *const af_family_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("clock-")
250};
251
252static const char *const af_family_kern_key_strings[AF_MAX+1] = {
253 _sock_locks("k-sk_lock-")
254};
255static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-slock-")
257};
258static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
259 _sock_locks("k-clock-")
260};
261static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 _sock_locks("rlock-")
263};
264static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
265 _sock_locks("wlock-")
266};
267static const char *const af_family_elock_key_strings[AF_MAX+1] = {
268 _sock_locks("elock-")
269};
270
271/*
272 * sk_callback_lock and sk queues locking rules are per-address-family,
273 * so split the lock classes by using a per-AF key:
274 */
275static struct lock_class_key af_callback_keys[AF_MAX];
276static struct lock_class_key af_rlock_keys[AF_MAX];
277static struct lock_class_key af_wlock_keys[AF_MAX];
278static struct lock_class_key af_elock_keys[AF_MAX];
279static struct lock_class_key af_kern_callback_keys[AF_MAX];
280
281/* Run time adjustable parameters. */
282__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
283EXPORT_SYMBOL(sysctl_wmem_max);
284__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
285EXPORT_SYMBOL(sysctl_rmem_max);
286__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
287__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
288
289DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
290EXPORT_SYMBOL_GPL(memalloc_socks_key);
291
292/**
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
295 *
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
299 */
300void sk_set_memalloc(struct sock *sk)
301{
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_branch_inc(&memalloc_socks_key);
305}
306EXPORT_SYMBOL_GPL(sk_set_memalloc);
307
308void sk_clear_memalloc(struct sock *sk)
309{
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_branch_dec(&memalloc_socks_key);
313
314 /*
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 */
321 sk_mem_reclaim(sk);
322}
323EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324
325int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
326{
327 int ret;
328 unsigned int noreclaim_flag;
329
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
332
333 noreclaim_flag = memalloc_noreclaim_save();
334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
335 tcp_v6_do_rcv,
336 tcp_v4_do_rcv,
337 sk, skb);
338 memalloc_noreclaim_restore(noreclaim_flag);
339
340 return ret;
341}
342EXPORT_SYMBOL(__sk_backlog_rcv);
343
344void sk_error_report(struct sock *sk)
345{
346 sk->sk_error_report(sk);
347
348 switch (sk->sk_family) {
349 case AF_INET:
350 fallthrough;
351 case AF_INET6:
352 trace_inet_sk_error_report(sk);
353 break;
354 default:
355 break;
356 }
357}
358EXPORT_SYMBOL(sk_error_report);
359
360int sock_get_timeout(long timeo, void *optval, bool old_timeval)
361{
362 struct __kernel_sock_timeval tv;
363
364 if (timeo == MAX_SCHEDULE_TIMEOUT) {
365 tv.tv_sec = 0;
366 tv.tv_usec = 0;
367 } else {
368 tv.tv_sec = timeo / HZ;
369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
370 }
371
372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
374 *(struct old_timeval32 *)optval = tv32;
375 return sizeof(tv32);
376 }
377
378 if (old_timeval) {
379 struct __kernel_old_timeval old_tv;
380 old_tv.tv_sec = tv.tv_sec;
381 old_tv.tv_usec = tv.tv_usec;
382 *(struct __kernel_old_timeval *)optval = old_tv;
383 return sizeof(old_tv);
384 }
385
386 *(struct __kernel_sock_timeval *)optval = tv;
387 return sizeof(tv);
388}
389EXPORT_SYMBOL(sock_get_timeout);
390
391int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
392 sockptr_t optval, int optlen, bool old_timeval)
393{
394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
395 struct old_timeval32 tv32;
396
397 if (optlen < sizeof(tv32))
398 return -EINVAL;
399
400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
401 return -EFAULT;
402 tv->tv_sec = tv32.tv_sec;
403 tv->tv_usec = tv32.tv_usec;
404 } else if (old_timeval) {
405 struct __kernel_old_timeval old_tv;
406
407 if (optlen < sizeof(old_tv))
408 return -EINVAL;
409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
410 return -EFAULT;
411 tv->tv_sec = old_tv.tv_sec;
412 tv->tv_usec = old_tv.tv_usec;
413 } else {
414 if (optlen < sizeof(*tv))
415 return -EINVAL;
416 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
417 return -EFAULT;
418 }
419
420 return 0;
421}
422EXPORT_SYMBOL(sock_copy_user_timeval);
423
424static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
425 bool old_timeval)
426{
427 struct __kernel_sock_timeval tv;
428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
429 long val;
430
431 if (err)
432 return err;
433
434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
435 return -EDOM;
436
437 if (tv.tv_sec < 0) {
438 static int warned __read_mostly;
439
440 WRITE_ONCE(*timeo_p, 0);
441 if (warned < 10 && net_ratelimit()) {
442 warned++;
443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
444 __func__, current->comm, task_pid_nr(current));
445 }
446 return 0;
447 }
448 val = MAX_SCHEDULE_TIMEOUT;
449 if ((tv.tv_sec || tv.tv_usec) &&
450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
452 USEC_PER_SEC / HZ);
453 WRITE_ONCE(*timeo_p, val);
454 return 0;
455}
456
457static bool sock_needs_netstamp(const struct sock *sk)
458{
459 switch (sk->sk_family) {
460 case AF_UNSPEC:
461 case AF_UNIX:
462 return false;
463 default:
464 return true;
465 }
466}
467
468static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
469{
470 if (sk->sk_flags & flags) {
471 sk->sk_flags &= ~flags;
472 if (sock_needs_netstamp(sk) &&
473 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
474 net_disable_timestamp();
475 }
476}
477
478
479int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
480{
481 unsigned long flags;
482 struct sk_buff_head *list = &sk->sk_receive_queue;
483
484 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
485 atomic_inc(&sk->sk_drops);
486 trace_sock_rcvqueue_full(sk, skb);
487 return -ENOMEM;
488 }
489
490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
491 atomic_inc(&sk->sk_drops);
492 return -ENOBUFS;
493 }
494
495 skb->dev = NULL;
496 skb_set_owner_r(skb, sk);
497
498 /* we escape from rcu protected region, make sure we dont leak
499 * a norefcounted dst
500 */
501 skb_dst_force(skb);
502
503 spin_lock_irqsave(&list->lock, flags);
504 sock_skb_set_dropcount(sk, skb);
505 __skb_queue_tail(list, skb);
506 spin_unlock_irqrestore(&list->lock, flags);
507
508 if (!sock_flag(sk, SOCK_DEAD))
509 sk->sk_data_ready(sk);
510 return 0;
511}
512EXPORT_SYMBOL(__sock_queue_rcv_skb);
513
514int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
515 enum skb_drop_reason *reason)
516{
517 enum skb_drop_reason drop_reason;
518 int err;
519
520 err = sk_filter(sk, skb);
521 if (err) {
522 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
523 goto out;
524 }
525 err = __sock_queue_rcv_skb(sk, skb);
526 switch (err) {
527 case -ENOMEM:
528 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
529 break;
530 case -ENOBUFS:
531 drop_reason = SKB_DROP_REASON_PROTO_MEM;
532 break;
533 default:
534 drop_reason = SKB_NOT_DROPPED_YET;
535 break;
536 }
537out:
538 if (reason)
539 *reason = drop_reason;
540 return err;
541}
542EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
543
544int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 const int nested, unsigned int trim_cap, bool refcounted)
546{
547 int rc = NET_RX_SUCCESS;
548
549 if (sk_filter_trim_cap(sk, skb, trim_cap))
550 goto discard_and_relse;
551
552 skb->dev = NULL;
553
554 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
555 atomic_inc(&sk->sk_drops);
556 goto discard_and_relse;
557 }
558 if (nested)
559 bh_lock_sock_nested(sk);
560 else
561 bh_lock_sock(sk);
562 if (!sock_owned_by_user(sk)) {
563 /*
564 * trylock + unlock semantics:
565 */
566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
567
568 rc = sk_backlog_rcv(sk, skb);
569
570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
572 bh_unlock_sock(sk);
573 atomic_inc(&sk->sk_drops);
574 goto discard_and_relse;
575 }
576
577 bh_unlock_sock(sk);
578out:
579 if (refcounted)
580 sock_put(sk);
581 return rc;
582discard_and_relse:
583 kfree_skb(skb);
584 goto out;
585}
586EXPORT_SYMBOL(__sk_receive_skb);
587
588INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
589 u32));
590INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
591 u32));
592struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
593{
594 struct dst_entry *dst = __sk_dst_get(sk);
595
596 if (dst && dst->obsolete &&
597 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
598 dst, cookie) == NULL) {
599 sk_tx_queue_clear(sk);
600 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
601 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
602 dst_release(dst);
603 return NULL;
604 }
605
606 return dst;
607}
608EXPORT_SYMBOL(__sk_dst_check);
609
610struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
611{
612 struct dst_entry *dst = sk_dst_get(sk);
613
614 if (dst && dst->obsolete &&
615 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
616 dst, cookie) == NULL) {
617 sk_dst_reset(sk);
618 dst_release(dst);
619 return NULL;
620 }
621
622 return dst;
623}
624EXPORT_SYMBOL(sk_dst_check);
625
626static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
627{
628 int ret = -ENOPROTOOPT;
629#ifdef CONFIG_NETDEVICES
630 struct net *net = sock_net(sk);
631
632 /* Sorry... */
633 ret = -EPERM;
634 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
635 goto out;
636
637 ret = -EINVAL;
638 if (ifindex < 0)
639 goto out;
640
641 /* Paired with all READ_ONCE() done locklessly. */
642 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
643
644 if (sk->sk_prot->rehash)
645 sk->sk_prot->rehash(sk);
646 sk_dst_reset(sk);
647
648 ret = 0;
649
650out:
651#endif
652
653 return ret;
654}
655
656int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
657{
658 int ret;
659
660 if (lock_sk)
661 lock_sock(sk);
662 ret = sock_bindtoindex_locked(sk, ifindex);
663 if (lock_sk)
664 release_sock(sk);
665
666 return ret;
667}
668EXPORT_SYMBOL(sock_bindtoindex);
669
670static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
671{
672 int ret = -ENOPROTOOPT;
673#ifdef CONFIG_NETDEVICES
674 struct net *net = sock_net(sk);
675 char devname[IFNAMSIZ];
676 int index;
677
678 ret = -EINVAL;
679 if (optlen < 0)
680 goto out;
681
682 /* Bind this socket to a particular device like "eth0",
683 * as specified in the passed interface name. If the
684 * name is "" or the option length is zero the socket
685 * is not bound.
686 */
687 if (optlen > IFNAMSIZ - 1)
688 optlen = IFNAMSIZ - 1;
689 memset(devname, 0, sizeof(devname));
690
691 ret = -EFAULT;
692 if (copy_from_sockptr(devname, optval, optlen))
693 goto out;
694
695 index = 0;
696 if (devname[0] != '\0') {
697 struct net_device *dev;
698
699 rcu_read_lock();
700 dev = dev_get_by_name_rcu(net, devname);
701 if (dev)
702 index = dev->ifindex;
703 rcu_read_unlock();
704 ret = -ENODEV;
705 if (!dev)
706 goto out;
707 }
708
709 sockopt_lock_sock(sk);
710 ret = sock_bindtoindex_locked(sk, index);
711 sockopt_release_sock(sk);
712out:
713#endif
714
715 return ret;
716}
717
718static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
719 sockptr_t optlen, int len)
720{
721 int ret = -ENOPROTOOPT;
722#ifdef CONFIG_NETDEVICES
723 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
724 struct net *net = sock_net(sk);
725 char devname[IFNAMSIZ];
726
727 if (bound_dev_if == 0) {
728 len = 0;
729 goto zero;
730 }
731
732 ret = -EINVAL;
733 if (len < IFNAMSIZ)
734 goto out;
735
736 ret = netdev_get_name(net, devname, bound_dev_if);
737 if (ret)
738 goto out;
739
740 len = strlen(devname) + 1;
741
742 ret = -EFAULT;
743 if (copy_to_sockptr(optval, devname, len))
744 goto out;
745
746zero:
747 ret = -EFAULT;
748 if (copy_to_sockptr(optlen, &len, sizeof(int)))
749 goto out;
750
751 ret = 0;
752
753out:
754#endif
755
756 return ret;
757}
758
759bool sk_mc_loop(const struct sock *sk)
760{
761 if (dev_recursion_level())
762 return false;
763 if (!sk)
764 return true;
765 /* IPV6_ADDRFORM can change sk->sk_family under us. */
766 switch (READ_ONCE(sk->sk_family)) {
767 case AF_INET:
768 return inet_test_bit(MC_LOOP, sk);
769#if IS_ENABLED(CONFIG_IPV6)
770 case AF_INET6:
771 return inet6_test_bit(MC6_LOOP, sk);
772#endif
773 }
774 WARN_ON_ONCE(1);
775 return true;
776}
777EXPORT_SYMBOL(sk_mc_loop);
778
779void sock_set_reuseaddr(struct sock *sk)
780{
781 lock_sock(sk);
782 sk->sk_reuse = SK_CAN_REUSE;
783 release_sock(sk);
784}
785EXPORT_SYMBOL(sock_set_reuseaddr);
786
787void sock_set_reuseport(struct sock *sk)
788{
789 lock_sock(sk);
790 sk->sk_reuseport = true;
791 release_sock(sk);
792}
793EXPORT_SYMBOL(sock_set_reuseport);
794
795void sock_no_linger(struct sock *sk)
796{
797 lock_sock(sk);
798 WRITE_ONCE(sk->sk_lingertime, 0);
799 sock_set_flag(sk, SOCK_LINGER);
800 release_sock(sk);
801}
802EXPORT_SYMBOL(sock_no_linger);
803
804void sock_set_priority(struct sock *sk, u32 priority)
805{
806 WRITE_ONCE(sk->sk_priority, priority);
807}
808EXPORT_SYMBOL(sock_set_priority);
809
810void sock_set_sndtimeo(struct sock *sk, s64 secs)
811{
812 lock_sock(sk);
813 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
814 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
815 else
816 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
817 release_sock(sk);
818}
819EXPORT_SYMBOL(sock_set_sndtimeo);
820
821static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
822{
823 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
824 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
825 if (val) {
826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
828 }
829}
830
831void sock_enable_timestamps(struct sock *sk)
832{
833 lock_sock(sk);
834 __sock_set_timestamps(sk, true, false, true);
835 release_sock(sk);
836}
837EXPORT_SYMBOL(sock_enable_timestamps);
838
839void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
840{
841 switch (optname) {
842 case SO_TIMESTAMP_OLD:
843 __sock_set_timestamps(sk, valbool, false, false);
844 break;
845 case SO_TIMESTAMP_NEW:
846 __sock_set_timestamps(sk, valbool, true, false);
847 break;
848 case SO_TIMESTAMPNS_OLD:
849 __sock_set_timestamps(sk, valbool, false, true);
850 break;
851 case SO_TIMESTAMPNS_NEW:
852 __sock_set_timestamps(sk, valbool, true, true);
853 break;
854 }
855}
856
857static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
858{
859 struct net *net = sock_net(sk);
860 struct net_device *dev = NULL;
861 bool match = false;
862 int *vclock_index;
863 int i, num;
864
865 if (sk->sk_bound_dev_if)
866 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
867
868 if (!dev) {
869 pr_err("%s: sock not bind to device\n", __func__);
870 return -EOPNOTSUPP;
871 }
872
873 num = ethtool_get_phc_vclocks(dev, &vclock_index);
874 dev_put(dev);
875
876 for (i = 0; i < num; i++) {
877 if (*(vclock_index + i) == phc_index) {
878 match = true;
879 break;
880 }
881 }
882
883 if (num > 0)
884 kfree(vclock_index);
885
886 if (!match)
887 return -EINVAL;
888
889 WRITE_ONCE(sk->sk_bind_phc, phc_index);
890
891 return 0;
892}
893
894int sock_set_timestamping(struct sock *sk, int optname,
895 struct so_timestamping timestamping)
896{
897 int val = timestamping.flags;
898 int ret;
899
900 if (val & ~SOF_TIMESTAMPING_MASK)
901 return -EINVAL;
902
903 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
904 !(val & SOF_TIMESTAMPING_OPT_ID))
905 return -EINVAL;
906
907 if (val & SOF_TIMESTAMPING_OPT_ID &&
908 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
909 if (sk_is_tcp(sk)) {
910 if ((1 << sk->sk_state) &
911 (TCPF_CLOSE | TCPF_LISTEN))
912 return -EINVAL;
913 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
915 else
916 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
917 } else {
918 atomic_set(&sk->sk_tskey, 0);
919 }
920 }
921
922 if (val & SOF_TIMESTAMPING_OPT_STATS &&
923 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
924 return -EINVAL;
925
926 if (val & SOF_TIMESTAMPING_BIND_PHC) {
927 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
928 if (ret)
929 return ret;
930 }
931
932 WRITE_ONCE(sk->sk_tsflags, val);
933 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
934
935 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
936 sock_enable_timestamp(sk,
937 SOCK_TIMESTAMPING_RX_SOFTWARE);
938 else
939 sock_disable_timestamp(sk,
940 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
941 return 0;
942}
943
944void sock_set_keepalive(struct sock *sk)
945{
946 lock_sock(sk);
947 if (sk->sk_prot->keepalive)
948 sk->sk_prot->keepalive(sk, true);
949 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
950 release_sock(sk);
951}
952EXPORT_SYMBOL(sock_set_keepalive);
953
954static void __sock_set_rcvbuf(struct sock *sk, int val)
955{
956 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
957 * as a negative value.
958 */
959 val = min_t(int, val, INT_MAX / 2);
960 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
961
962 /* We double it on the way in to account for "struct sk_buff" etc.
963 * overhead. Applications assume that the SO_RCVBUF setting they make
964 * will allow that much actual data to be received on that socket.
965 *
966 * Applications are unaware that "struct sk_buff" and other overheads
967 * allocate from the receive buffer during socket buffer allocation.
968 *
969 * And after considering the possible alternatives, returning the value
970 * we actually used in getsockopt is the most desirable behavior.
971 */
972 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
973}
974
975void sock_set_rcvbuf(struct sock *sk, int val)
976{
977 lock_sock(sk);
978 __sock_set_rcvbuf(sk, val);
979 release_sock(sk);
980}
981EXPORT_SYMBOL(sock_set_rcvbuf);
982
983static void __sock_set_mark(struct sock *sk, u32 val)
984{
985 if (val != sk->sk_mark) {
986 WRITE_ONCE(sk->sk_mark, val);
987 sk_dst_reset(sk);
988 }
989}
990
991void sock_set_mark(struct sock *sk, u32 val)
992{
993 lock_sock(sk);
994 __sock_set_mark(sk, val);
995 release_sock(sk);
996}
997EXPORT_SYMBOL(sock_set_mark);
998
999static void sock_release_reserved_memory(struct sock *sk, int bytes)
1000{
1001 /* Round down bytes to multiple of pages */
1002 bytes = round_down(bytes, PAGE_SIZE);
1003
1004 WARN_ON(bytes > sk->sk_reserved_mem);
1005 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1006 sk_mem_reclaim(sk);
1007}
1008
1009static int sock_reserve_memory(struct sock *sk, int bytes)
1010{
1011 long allocated;
1012 bool charged;
1013 int pages;
1014
1015 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1016 return -EOPNOTSUPP;
1017
1018 if (!bytes)
1019 return 0;
1020
1021 pages = sk_mem_pages(bytes);
1022
1023 /* pre-charge to memcg */
1024 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1025 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1026 if (!charged)
1027 return -ENOMEM;
1028
1029 /* pre-charge to forward_alloc */
1030 sk_memory_allocated_add(sk, pages);
1031 allocated = sk_memory_allocated(sk);
1032 /* If the system goes into memory pressure with this
1033 * precharge, give up and return error.
1034 */
1035 if (allocated > sk_prot_mem_limits(sk, 1)) {
1036 sk_memory_allocated_sub(sk, pages);
1037 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1038 return -ENOMEM;
1039 }
1040 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1041
1042 WRITE_ONCE(sk->sk_reserved_mem,
1043 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1044
1045 return 0;
1046}
1047
1048#ifdef CONFIG_PAGE_POOL
1049
1050/* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1051 * in 1 syscall. The limit exists to limit the amount of memory the kernel
1052 * allocates to copy these tokens, and to prevent looping over the frags for
1053 * too long.
1054 */
1055#define MAX_DONTNEED_TOKENS 128
1056#define MAX_DONTNEED_FRAGS 1024
1057
1058static noinline_for_stack int
1059sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1060{
1061 unsigned int num_tokens, i, j, k, netmem_num = 0;
1062 struct dmabuf_token *tokens;
1063 int ret = 0, num_frags = 0;
1064 netmem_ref netmems[16];
1065
1066 if (!sk_is_tcp(sk))
1067 return -EBADF;
1068
1069 if (optlen % sizeof(*tokens) ||
1070 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1071 return -EINVAL;
1072
1073 num_tokens = optlen / sizeof(*tokens);
1074 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL);
1075 if (!tokens)
1076 return -ENOMEM;
1077
1078 if (copy_from_sockptr(tokens, optval, optlen)) {
1079 kvfree(tokens);
1080 return -EFAULT;
1081 }
1082
1083 xa_lock_bh(&sk->sk_user_frags);
1084 for (i = 0; i < num_tokens; i++) {
1085 for (j = 0; j < tokens[i].token_count; j++) {
1086 if (++num_frags > MAX_DONTNEED_FRAGS)
1087 goto frag_limit_reached;
1088
1089 netmem_ref netmem = (__force netmem_ref)__xa_erase(
1090 &sk->sk_user_frags, tokens[i].token_start + j);
1091
1092 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1093 continue;
1094
1095 netmems[netmem_num++] = netmem;
1096 if (netmem_num == ARRAY_SIZE(netmems)) {
1097 xa_unlock_bh(&sk->sk_user_frags);
1098 for (k = 0; k < netmem_num; k++)
1099 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1100 netmem_num = 0;
1101 xa_lock_bh(&sk->sk_user_frags);
1102 }
1103 ret++;
1104 }
1105 }
1106
1107frag_limit_reached:
1108 xa_unlock_bh(&sk->sk_user_frags);
1109 for (k = 0; k < netmem_num; k++)
1110 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1111
1112 kvfree(tokens);
1113 return ret;
1114}
1115#endif
1116
1117void sockopt_lock_sock(struct sock *sk)
1118{
1119 /* When current->bpf_ctx is set, the setsockopt is called from
1120 * a bpf prog. bpf has ensured the sk lock has been
1121 * acquired before calling setsockopt().
1122 */
1123 if (has_current_bpf_ctx())
1124 return;
1125
1126 lock_sock(sk);
1127}
1128EXPORT_SYMBOL(sockopt_lock_sock);
1129
1130void sockopt_release_sock(struct sock *sk)
1131{
1132 if (has_current_bpf_ctx())
1133 return;
1134
1135 release_sock(sk);
1136}
1137EXPORT_SYMBOL(sockopt_release_sock);
1138
1139bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1140{
1141 return has_current_bpf_ctx() || ns_capable(ns, cap);
1142}
1143EXPORT_SYMBOL(sockopt_ns_capable);
1144
1145bool sockopt_capable(int cap)
1146{
1147 return has_current_bpf_ctx() || capable(cap);
1148}
1149EXPORT_SYMBOL(sockopt_capable);
1150
1151static int sockopt_validate_clockid(__kernel_clockid_t value)
1152{
1153 switch (value) {
1154 case CLOCK_REALTIME:
1155 case CLOCK_MONOTONIC:
1156 case CLOCK_TAI:
1157 return 0;
1158 }
1159 return -EINVAL;
1160}
1161
1162/*
1163 * This is meant for all protocols to use and covers goings on
1164 * at the socket level. Everything here is generic.
1165 */
1166
1167int sk_setsockopt(struct sock *sk, int level, int optname,
1168 sockptr_t optval, unsigned int optlen)
1169{
1170 struct so_timestamping timestamping;
1171 struct socket *sock = sk->sk_socket;
1172 struct sock_txtime sk_txtime;
1173 int val;
1174 int valbool;
1175 struct linger ling;
1176 int ret = 0;
1177
1178 /*
1179 * Options without arguments
1180 */
1181
1182 if (optname == SO_BINDTODEVICE)
1183 return sock_setbindtodevice(sk, optval, optlen);
1184
1185 if (optlen < sizeof(int))
1186 return -EINVAL;
1187
1188 if (copy_from_sockptr(&val, optval, sizeof(val)))
1189 return -EFAULT;
1190
1191 valbool = val ? 1 : 0;
1192
1193 /* handle options which do not require locking the socket. */
1194 switch (optname) {
1195 case SO_PRIORITY:
1196 if ((val >= 0 && val <= 6) ||
1197 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1198 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1199 sock_set_priority(sk, val);
1200 return 0;
1201 }
1202 return -EPERM;
1203 case SO_PASSSEC:
1204 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1205 return 0;
1206 case SO_PASSCRED:
1207 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1208 return 0;
1209 case SO_PASSPIDFD:
1210 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1211 return 0;
1212 case SO_TYPE:
1213 case SO_PROTOCOL:
1214 case SO_DOMAIN:
1215 case SO_ERROR:
1216 return -ENOPROTOOPT;
1217#ifdef CONFIG_NET_RX_BUSY_POLL
1218 case SO_BUSY_POLL:
1219 if (val < 0)
1220 return -EINVAL;
1221 WRITE_ONCE(sk->sk_ll_usec, val);
1222 return 0;
1223 case SO_PREFER_BUSY_POLL:
1224 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1225 return -EPERM;
1226 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1227 return 0;
1228 case SO_BUSY_POLL_BUDGET:
1229 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1230 !sockopt_capable(CAP_NET_ADMIN))
1231 return -EPERM;
1232 if (val < 0 || val > U16_MAX)
1233 return -EINVAL;
1234 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1235 return 0;
1236#endif
1237 case SO_MAX_PACING_RATE:
1238 {
1239 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1240 unsigned long pacing_rate;
1241
1242 if (sizeof(ulval) != sizeof(val) &&
1243 optlen >= sizeof(ulval) &&
1244 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1245 return -EFAULT;
1246 }
1247 if (ulval != ~0UL)
1248 cmpxchg(&sk->sk_pacing_status,
1249 SK_PACING_NONE,
1250 SK_PACING_NEEDED);
1251 /* Pairs with READ_ONCE() from sk_getsockopt() */
1252 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1253 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1254 if (ulval < pacing_rate)
1255 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1256 return 0;
1257 }
1258 case SO_TXREHASH:
1259 if (val < -1 || val > 1)
1260 return -EINVAL;
1261 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1262 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1263 /* Paired with READ_ONCE() in tcp_rtx_synack()
1264 * and sk_getsockopt().
1265 */
1266 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1267 return 0;
1268 case SO_PEEK_OFF:
1269 {
1270 int (*set_peek_off)(struct sock *sk, int val);
1271
1272 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1273 if (set_peek_off)
1274 ret = set_peek_off(sk, val);
1275 else
1276 ret = -EOPNOTSUPP;
1277 return ret;
1278 }
1279#ifdef CONFIG_PAGE_POOL
1280 case SO_DEVMEM_DONTNEED:
1281 return sock_devmem_dontneed(sk, optval, optlen);
1282#endif
1283 }
1284
1285 sockopt_lock_sock(sk);
1286
1287 switch (optname) {
1288 case SO_DEBUG:
1289 if (val && !sockopt_capable(CAP_NET_ADMIN))
1290 ret = -EACCES;
1291 else
1292 sock_valbool_flag(sk, SOCK_DBG, valbool);
1293 break;
1294 case SO_REUSEADDR:
1295 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1296 break;
1297 case SO_REUSEPORT:
1298 if (valbool && !sk_is_inet(sk))
1299 ret = -EOPNOTSUPP;
1300 else
1301 sk->sk_reuseport = valbool;
1302 break;
1303 case SO_DONTROUTE:
1304 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1305 sk_dst_reset(sk);
1306 break;
1307 case SO_BROADCAST:
1308 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1309 break;
1310 case SO_SNDBUF:
1311 /* Don't error on this BSD doesn't and if you think
1312 * about it this is right. Otherwise apps have to
1313 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1314 * are treated in BSD as hints
1315 */
1316 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1317set_sndbuf:
1318 /* Ensure val * 2 fits into an int, to prevent max_t()
1319 * from treating it as a negative value.
1320 */
1321 val = min_t(int, val, INT_MAX / 2);
1322 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1323 WRITE_ONCE(sk->sk_sndbuf,
1324 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1325 /* Wake up sending tasks if we upped the value. */
1326 sk->sk_write_space(sk);
1327 break;
1328
1329 case SO_SNDBUFFORCE:
1330 if (!sockopt_capable(CAP_NET_ADMIN)) {
1331 ret = -EPERM;
1332 break;
1333 }
1334
1335 /* No negative values (to prevent underflow, as val will be
1336 * multiplied by 2).
1337 */
1338 if (val < 0)
1339 val = 0;
1340 goto set_sndbuf;
1341
1342 case SO_RCVBUF:
1343 /* Don't error on this BSD doesn't and if you think
1344 * about it this is right. Otherwise apps have to
1345 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1346 * are treated in BSD as hints
1347 */
1348 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1349 break;
1350
1351 case SO_RCVBUFFORCE:
1352 if (!sockopt_capable(CAP_NET_ADMIN)) {
1353 ret = -EPERM;
1354 break;
1355 }
1356
1357 /* No negative values (to prevent underflow, as val will be
1358 * multiplied by 2).
1359 */
1360 __sock_set_rcvbuf(sk, max(val, 0));
1361 break;
1362
1363 case SO_KEEPALIVE:
1364 if (sk->sk_prot->keepalive)
1365 sk->sk_prot->keepalive(sk, valbool);
1366 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1367 break;
1368
1369 case SO_OOBINLINE:
1370 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1371 break;
1372
1373 case SO_NO_CHECK:
1374 sk->sk_no_check_tx = valbool;
1375 break;
1376
1377 case SO_LINGER:
1378 if (optlen < sizeof(ling)) {
1379 ret = -EINVAL; /* 1003.1g */
1380 break;
1381 }
1382 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1383 ret = -EFAULT;
1384 break;
1385 }
1386 if (!ling.l_onoff) {
1387 sock_reset_flag(sk, SOCK_LINGER);
1388 } else {
1389 unsigned long t_sec = ling.l_linger;
1390
1391 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1392 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1393 else
1394 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1395 sock_set_flag(sk, SOCK_LINGER);
1396 }
1397 break;
1398
1399 case SO_BSDCOMPAT:
1400 break;
1401
1402 case SO_TIMESTAMP_OLD:
1403 case SO_TIMESTAMP_NEW:
1404 case SO_TIMESTAMPNS_OLD:
1405 case SO_TIMESTAMPNS_NEW:
1406 sock_set_timestamp(sk, optname, valbool);
1407 break;
1408
1409 case SO_TIMESTAMPING_NEW:
1410 case SO_TIMESTAMPING_OLD:
1411 if (optlen == sizeof(timestamping)) {
1412 if (copy_from_sockptr(×tamping, optval,
1413 sizeof(timestamping))) {
1414 ret = -EFAULT;
1415 break;
1416 }
1417 } else {
1418 memset(×tamping, 0, sizeof(timestamping));
1419 timestamping.flags = val;
1420 }
1421 ret = sock_set_timestamping(sk, optname, timestamping);
1422 break;
1423
1424 case SO_RCVLOWAT:
1425 {
1426 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1427
1428 if (val < 0)
1429 val = INT_MAX;
1430 if (sock)
1431 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1432 if (set_rcvlowat)
1433 ret = set_rcvlowat(sk, val);
1434 else
1435 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1436 break;
1437 }
1438 case SO_RCVTIMEO_OLD:
1439 case SO_RCVTIMEO_NEW:
1440 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1441 optlen, optname == SO_RCVTIMEO_OLD);
1442 break;
1443
1444 case SO_SNDTIMEO_OLD:
1445 case SO_SNDTIMEO_NEW:
1446 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1447 optlen, optname == SO_SNDTIMEO_OLD);
1448 break;
1449
1450 case SO_ATTACH_FILTER: {
1451 struct sock_fprog fprog;
1452
1453 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1454 if (!ret)
1455 ret = sk_attach_filter(&fprog, sk);
1456 break;
1457 }
1458 case SO_ATTACH_BPF:
1459 ret = -EINVAL;
1460 if (optlen == sizeof(u32)) {
1461 u32 ufd;
1462
1463 ret = -EFAULT;
1464 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1465 break;
1466
1467 ret = sk_attach_bpf(ufd, sk);
1468 }
1469 break;
1470
1471 case SO_ATTACH_REUSEPORT_CBPF: {
1472 struct sock_fprog fprog;
1473
1474 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1475 if (!ret)
1476 ret = sk_reuseport_attach_filter(&fprog, sk);
1477 break;
1478 }
1479 case SO_ATTACH_REUSEPORT_EBPF:
1480 ret = -EINVAL;
1481 if (optlen == sizeof(u32)) {
1482 u32 ufd;
1483
1484 ret = -EFAULT;
1485 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1486 break;
1487
1488 ret = sk_reuseport_attach_bpf(ufd, sk);
1489 }
1490 break;
1491
1492 case SO_DETACH_REUSEPORT_BPF:
1493 ret = reuseport_detach_prog(sk);
1494 break;
1495
1496 case SO_DETACH_FILTER:
1497 ret = sk_detach_filter(sk);
1498 break;
1499
1500 case SO_LOCK_FILTER:
1501 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1502 ret = -EPERM;
1503 else
1504 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1505 break;
1506
1507 case SO_MARK:
1508 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1509 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1510 ret = -EPERM;
1511 break;
1512 }
1513
1514 __sock_set_mark(sk, val);
1515 break;
1516 case SO_RCVMARK:
1517 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1518 break;
1519
1520 case SO_RXQ_OVFL:
1521 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1522 break;
1523
1524 case SO_WIFI_STATUS:
1525 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1526 break;
1527
1528 case SO_NOFCS:
1529 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1530 break;
1531
1532 case SO_SELECT_ERR_QUEUE:
1533 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1534 break;
1535
1536
1537 case SO_INCOMING_CPU:
1538 reuseport_update_incoming_cpu(sk, val);
1539 break;
1540
1541 case SO_CNX_ADVICE:
1542 if (val == 1)
1543 dst_negative_advice(sk);
1544 break;
1545
1546 case SO_ZEROCOPY:
1547 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1548 if (!(sk_is_tcp(sk) ||
1549 (sk->sk_type == SOCK_DGRAM &&
1550 sk->sk_protocol == IPPROTO_UDP)))
1551 ret = -EOPNOTSUPP;
1552 } else if (sk->sk_family != PF_RDS) {
1553 ret = -EOPNOTSUPP;
1554 }
1555 if (!ret) {
1556 if (val < 0 || val > 1)
1557 ret = -EINVAL;
1558 else
1559 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1560 }
1561 break;
1562
1563 case SO_TXTIME:
1564 if (optlen != sizeof(struct sock_txtime)) {
1565 ret = -EINVAL;
1566 break;
1567 } else if (copy_from_sockptr(&sk_txtime, optval,
1568 sizeof(struct sock_txtime))) {
1569 ret = -EFAULT;
1570 break;
1571 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1572 ret = -EINVAL;
1573 break;
1574 }
1575 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1576 * scheduler has enough safe guards.
1577 */
1578 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1579 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1580 ret = -EPERM;
1581 break;
1582 }
1583
1584 ret = sockopt_validate_clockid(sk_txtime.clockid);
1585 if (ret)
1586 break;
1587
1588 sock_valbool_flag(sk, SOCK_TXTIME, true);
1589 sk->sk_clockid = sk_txtime.clockid;
1590 sk->sk_txtime_deadline_mode =
1591 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1592 sk->sk_txtime_report_errors =
1593 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1594 break;
1595
1596 case SO_BINDTOIFINDEX:
1597 ret = sock_bindtoindex_locked(sk, val);
1598 break;
1599
1600 case SO_BUF_LOCK:
1601 if (val & ~SOCK_BUF_LOCK_MASK) {
1602 ret = -EINVAL;
1603 break;
1604 }
1605 sk->sk_userlocks = val | (sk->sk_userlocks &
1606 ~SOCK_BUF_LOCK_MASK);
1607 break;
1608
1609 case SO_RESERVE_MEM:
1610 {
1611 int delta;
1612
1613 if (val < 0) {
1614 ret = -EINVAL;
1615 break;
1616 }
1617
1618 delta = val - sk->sk_reserved_mem;
1619 if (delta < 0)
1620 sock_release_reserved_memory(sk, -delta);
1621 else
1622 ret = sock_reserve_memory(sk, delta);
1623 break;
1624 }
1625
1626 default:
1627 ret = -ENOPROTOOPT;
1628 break;
1629 }
1630 sockopt_release_sock(sk);
1631 return ret;
1632}
1633
1634int sock_setsockopt(struct socket *sock, int level, int optname,
1635 sockptr_t optval, unsigned int optlen)
1636{
1637 return sk_setsockopt(sock->sk, level, optname,
1638 optval, optlen);
1639}
1640EXPORT_SYMBOL(sock_setsockopt);
1641
1642static const struct cred *sk_get_peer_cred(struct sock *sk)
1643{
1644 const struct cred *cred;
1645
1646 spin_lock(&sk->sk_peer_lock);
1647 cred = get_cred(sk->sk_peer_cred);
1648 spin_unlock(&sk->sk_peer_lock);
1649
1650 return cred;
1651}
1652
1653static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1654 struct ucred *ucred)
1655{
1656 ucred->pid = pid_vnr(pid);
1657 ucred->uid = ucred->gid = -1;
1658 if (cred) {
1659 struct user_namespace *current_ns = current_user_ns();
1660
1661 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1662 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1663 }
1664}
1665
1666static int groups_to_user(sockptr_t dst, const struct group_info *src)
1667{
1668 struct user_namespace *user_ns = current_user_ns();
1669 int i;
1670
1671 for (i = 0; i < src->ngroups; i++) {
1672 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1673
1674 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1675 return -EFAULT;
1676 }
1677
1678 return 0;
1679}
1680
1681int sk_getsockopt(struct sock *sk, int level, int optname,
1682 sockptr_t optval, sockptr_t optlen)
1683{
1684 struct socket *sock = sk->sk_socket;
1685
1686 union {
1687 int val;
1688 u64 val64;
1689 unsigned long ulval;
1690 struct linger ling;
1691 struct old_timeval32 tm32;
1692 struct __kernel_old_timeval tm;
1693 struct __kernel_sock_timeval stm;
1694 struct sock_txtime txtime;
1695 struct so_timestamping timestamping;
1696 } v;
1697
1698 int lv = sizeof(int);
1699 int len;
1700
1701 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1702 return -EFAULT;
1703 if (len < 0)
1704 return -EINVAL;
1705
1706 memset(&v, 0, sizeof(v));
1707
1708 switch (optname) {
1709 case SO_DEBUG:
1710 v.val = sock_flag(sk, SOCK_DBG);
1711 break;
1712
1713 case SO_DONTROUTE:
1714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1715 break;
1716
1717 case SO_BROADCAST:
1718 v.val = sock_flag(sk, SOCK_BROADCAST);
1719 break;
1720
1721 case SO_SNDBUF:
1722 v.val = READ_ONCE(sk->sk_sndbuf);
1723 break;
1724
1725 case SO_RCVBUF:
1726 v.val = READ_ONCE(sk->sk_rcvbuf);
1727 break;
1728
1729 case SO_REUSEADDR:
1730 v.val = sk->sk_reuse;
1731 break;
1732
1733 case SO_REUSEPORT:
1734 v.val = sk->sk_reuseport;
1735 break;
1736
1737 case SO_KEEPALIVE:
1738 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1739 break;
1740
1741 case SO_TYPE:
1742 v.val = sk->sk_type;
1743 break;
1744
1745 case SO_PROTOCOL:
1746 v.val = sk->sk_protocol;
1747 break;
1748
1749 case SO_DOMAIN:
1750 v.val = sk->sk_family;
1751 break;
1752
1753 case SO_ERROR:
1754 v.val = -sock_error(sk);
1755 if (v.val == 0)
1756 v.val = xchg(&sk->sk_err_soft, 0);
1757 break;
1758
1759 case SO_OOBINLINE:
1760 v.val = sock_flag(sk, SOCK_URGINLINE);
1761 break;
1762
1763 case SO_NO_CHECK:
1764 v.val = sk->sk_no_check_tx;
1765 break;
1766
1767 case SO_PRIORITY:
1768 v.val = READ_ONCE(sk->sk_priority);
1769 break;
1770
1771 case SO_LINGER:
1772 lv = sizeof(v.ling);
1773 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1774 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1775 break;
1776
1777 case SO_BSDCOMPAT:
1778 break;
1779
1780 case SO_TIMESTAMP_OLD:
1781 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1782 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1783 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1784 break;
1785
1786 case SO_TIMESTAMPNS_OLD:
1787 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1788 break;
1789
1790 case SO_TIMESTAMP_NEW:
1791 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1792 break;
1793
1794 case SO_TIMESTAMPNS_NEW:
1795 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1796 break;
1797
1798 case SO_TIMESTAMPING_OLD:
1799 case SO_TIMESTAMPING_NEW:
1800 lv = sizeof(v.timestamping);
1801 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1802 * returning the flags when they were set through the same option.
1803 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1804 */
1805 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1806 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1807 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1808 }
1809 break;
1810
1811 case SO_RCVTIMEO_OLD:
1812 case SO_RCVTIMEO_NEW:
1813 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1814 SO_RCVTIMEO_OLD == optname);
1815 break;
1816
1817 case SO_SNDTIMEO_OLD:
1818 case SO_SNDTIMEO_NEW:
1819 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1820 SO_SNDTIMEO_OLD == optname);
1821 break;
1822
1823 case SO_RCVLOWAT:
1824 v.val = READ_ONCE(sk->sk_rcvlowat);
1825 break;
1826
1827 case SO_SNDLOWAT:
1828 v.val = 1;
1829 break;
1830
1831 case SO_PASSCRED:
1832 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1833 break;
1834
1835 case SO_PASSPIDFD:
1836 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1837 break;
1838
1839 case SO_PEERCRED:
1840 {
1841 struct ucred peercred;
1842 if (len > sizeof(peercred))
1843 len = sizeof(peercred);
1844
1845 spin_lock(&sk->sk_peer_lock);
1846 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1847 spin_unlock(&sk->sk_peer_lock);
1848
1849 if (copy_to_sockptr(optval, &peercred, len))
1850 return -EFAULT;
1851 goto lenout;
1852 }
1853
1854 case SO_PEERPIDFD:
1855 {
1856 struct pid *peer_pid;
1857 struct file *pidfd_file = NULL;
1858 int pidfd;
1859
1860 if (len > sizeof(pidfd))
1861 len = sizeof(pidfd);
1862
1863 spin_lock(&sk->sk_peer_lock);
1864 peer_pid = get_pid(sk->sk_peer_pid);
1865 spin_unlock(&sk->sk_peer_lock);
1866
1867 if (!peer_pid)
1868 return -ENODATA;
1869
1870 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1871 put_pid(peer_pid);
1872 if (pidfd < 0)
1873 return pidfd;
1874
1875 if (copy_to_sockptr(optval, &pidfd, len) ||
1876 copy_to_sockptr(optlen, &len, sizeof(int))) {
1877 put_unused_fd(pidfd);
1878 fput(pidfd_file);
1879
1880 return -EFAULT;
1881 }
1882
1883 fd_install(pidfd, pidfd_file);
1884 return 0;
1885 }
1886
1887 case SO_PEERGROUPS:
1888 {
1889 const struct cred *cred;
1890 int ret, n;
1891
1892 cred = sk_get_peer_cred(sk);
1893 if (!cred)
1894 return -ENODATA;
1895
1896 n = cred->group_info->ngroups;
1897 if (len < n * sizeof(gid_t)) {
1898 len = n * sizeof(gid_t);
1899 put_cred(cred);
1900 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1901 }
1902 len = n * sizeof(gid_t);
1903
1904 ret = groups_to_user(optval, cred->group_info);
1905 put_cred(cred);
1906 if (ret)
1907 return ret;
1908 goto lenout;
1909 }
1910
1911 case SO_PEERNAME:
1912 {
1913 struct sockaddr_storage address;
1914
1915 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1916 if (lv < 0)
1917 return -ENOTCONN;
1918 if (lv < len)
1919 return -EINVAL;
1920 if (copy_to_sockptr(optval, &address, len))
1921 return -EFAULT;
1922 goto lenout;
1923 }
1924
1925 /* Dubious BSD thing... Probably nobody even uses it, but
1926 * the UNIX standard wants it for whatever reason... -DaveM
1927 */
1928 case SO_ACCEPTCONN:
1929 v.val = sk->sk_state == TCP_LISTEN;
1930 break;
1931
1932 case SO_PASSSEC:
1933 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1934 break;
1935
1936 case SO_PEERSEC:
1937 return security_socket_getpeersec_stream(sock,
1938 optval, optlen, len);
1939
1940 case SO_MARK:
1941 v.val = READ_ONCE(sk->sk_mark);
1942 break;
1943
1944 case SO_RCVMARK:
1945 v.val = sock_flag(sk, SOCK_RCVMARK);
1946 break;
1947
1948 case SO_RXQ_OVFL:
1949 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1950 break;
1951
1952 case SO_WIFI_STATUS:
1953 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1954 break;
1955
1956 case SO_PEEK_OFF:
1957 if (!READ_ONCE(sock->ops)->set_peek_off)
1958 return -EOPNOTSUPP;
1959
1960 v.val = READ_ONCE(sk->sk_peek_off);
1961 break;
1962 case SO_NOFCS:
1963 v.val = sock_flag(sk, SOCK_NOFCS);
1964 break;
1965
1966 case SO_BINDTODEVICE:
1967 return sock_getbindtodevice(sk, optval, optlen, len);
1968
1969 case SO_GET_FILTER:
1970 len = sk_get_filter(sk, optval, len);
1971 if (len < 0)
1972 return len;
1973
1974 goto lenout;
1975
1976 case SO_LOCK_FILTER:
1977 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1978 break;
1979
1980 case SO_BPF_EXTENSIONS:
1981 v.val = bpf_tell_extensions();
1982 break;
1983
1984 case SO_SELECT_ERR_QUEUE:
1985 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1986 break;
1987
1988#ifdef CONFIG_NET_RX_BUSY_POLL
1989 case SO_BUSY_POLL:
1990 v.val = READ_ONCE(sk->sk_ll_usec);
1991 break;
1992 case SO_PREFER_BUSY_POLL:
1993 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1994 break;
1995#endif
1996
1997 case SO_MAX_PACING_RATE:
1998 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1999 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2000 lv = sizeof(v.ulval);
2001 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2002 } else {
2003 /* 32bit version */
2004 v.val = min_t(unsigned long, ~0U,
2005 READ_ONCE(sk->sk_max_pacing_rate));
2006 }
2007 break;
2008
2009 case SO_INCOMING_CPU:
2010 v.val = READ_ONCE(sk->sk_incoming_cpu);
2011 break;
2012
2013 case SO_MEMINFO:
2014 {
2015 u32 meminfo[SK_MEMINFO_VARS];
2016
2017 sk_get_meminfo(sk, meminfo);
2018
2019 len = min_t(unsigned int, len, sizeof(meminfo));
2020 if (copy_to_sockptr(optval, &meminfo, len))
2021 return -EFAULT;
2022
2023 goto lenout;
2024 }
2025
2026#ifdef CONFIG_NET_RX_BUSY_POLL
2027 case SO_INCOMING_NAPI_ID:
2028 v.val = READ_ONCE(sk->sk_napi_id);
2029
2030 /* aggregate non-NAPI IDs down to 0 */
2031 if (v.val < MIN_NAPI_ID)
2032 v.val = 0;
2033
2034 break;
2035#endif
2036
2037 case SO_COOKIE:
2038 lv = sizeof(u64);
2039 if (len < lv)
2040 return -EINVAL;
2041 v.val64 = sock_gen_cookie(sk);
2042 break;
2043
2044 case SO_ZEROCOPY:
2045 v.val = sock_flag(sk, SOCK_ZEROCOPY);
2046 break;
2047
2048 case SO_TXTIME:
2049 lv = sizeof(v.txtime);
2050 v.txtime.clockid = sk->sk_clockid;
2051 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2052 SOF_TXTIME_DEADLINE_MODE : 0;
2053 v.txtime.flags |= sk->sk_txtime_report_errors ?
2054 SOF_TXTIME_REPORT_ERRORS : 0;
2055 break;
2056
2057 case SO_BINDTOIFINDEX:
2058 v.val = READ_ONCE(sk->sk_bound_dev_if);
2059 break;
2060
2061 case SO_NETNS_COOKIE:
2062 lv = sizeof(u64);
2063 if (len != lv)
2064 return -EINVAL;
2065 v.val64 = sock_net(sk)->net_cookie;
2066 break;
2067
2068 case SO_BUF_LOCK:
2069 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2070 break;
2071
2072 case SO_RESERVE_MEM:
2073 v.val = READ_ONCE(sk->sk_reserved_mem);
2074 break;
2075
2076 case SO_TXREHASH:
2077 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2078 v.val = READ_ONCE(sk->sk_txrehash);
2079 break;
2080
2081 default:
2082 /* We implement the SO_SNDLOWAT etc to not be settable
2083 * (1003.1g 7).
2084 */
2085 return -ENOPROTOOPT;
2086 }
2087
2088 if (len > lv)
2089 len = lv;
2090 if (copy_to_sockptr(optval, &v, len))
2091 return -EFAULT;
2092lenout:
2093 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2094 return -EFAULT;
2095 return 0;
2096}
2097
2098/*
2099 * Initialize an sk_lock.
2100 *
2101 * (We also register the sk_lock with the lock validator.)
2102 */
2103static inline void sock_lock_init(struct sock *sk)
2104{
2105 if (sk->sk_kern_sock)
2106 sock_lock_init_class_and_name(
2107 sk,
2108 af_family_kern_slock_key_strings[sk->sk_family],
2109 af_family_kern_slock_keys + sk->sk_family,
2110 af_family_kern_key_strings[sk->sk_family],
2111 af_family_kern_keys + sk->sk_family);
2112 else
2113 sock_lock_init_class_and_name(
2114 sk,
2115 af_family_slock_key_strings[sk->sk_family],
2116 af_family_slock_keys + sk->sk_family,
2117 af_family_key_strings[sk->sk_family],
2118 af_family_keys + sk->sk_family);
2119}
2120
2121/*
2122 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2123 * even temporarily, because of RCU lookups. sk_node should also be left as is.
2124 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2125 */
2126static void sock_copy(struct sock *nsk, const struct sock *osk)
2127{
2128 const struct proto *prot = READ_ONCE(osk->sk_prot);
2129#ifdef CONFIG_SECURITY_NETWORK
2130 void *sptr = nsk->sk_security;
2131#endif
2132
2133 /* If we move sk_tx_queue_mapping out of the private section,
2134 * we must check if sk_tx_queue_clear() is called after
2135 * sock_copy() in sk_clone_lock().
2136 */
2137 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2138 offsetof(struct sock, sk_dontcopy_begin) ||
2139 offsetof(struct sock, sk_tx_queue_mapping) >=
2140 offsetof(struct sock, sk_dontcopy_end));
2141
2142 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2143
2144 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2145 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2146 /* alloc is larger than struct, see sk_prot_alloc() */);
2147
2148#ifdef CONFIG_SECURITY_NETWORK
2149 nsk->sk_security = sptr;
2150 security_sk_clone(osk, nsk);
2151#endif
2152}
2153
2154static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2155 int family)
2156{
2157 struct sock *sk;
2158 struct kmem_cache *slab;
2159
2160 slab = prot->slab;
2161 if (slab != NULL) {
2162 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2163 if (!sk)
2164 return sk;
2165 if (want_init_on_alloc(priority))
2166 sk_prot_clear_nulls(sk, prot->obj_size);
2167 } else
2168 sk = kmalloc(prot->obj_size, priority);
2169
2170 if (sk != NULL) {
2171 if (security_sk_alloc(sk, family, priority))
2172 goto out_free;
2173
2174 if (!try_module_get(prot->owner))
2175 goto out_free_sec;
2176 }
2177
2178 return sk;
2179
2180out_free_sec:
2181 security_sk_free(sk);
2182out_free:
2183 if (slab != NULL)
2184 kmem_cache_free(slab, sk);
2185 else
2186 kfree(sk);
2187 return NULL;
2188}
2189
2190static void sk_prot_free(struct proto *prot, struct sock *sk)
2191{
2192 struct kmem_cache *slab;
2193 struct module *owner;
2194
2195 owner = prot->owner;
2196 slab = prot->slab;
2197
2198 cgroup_sk_free(&sk->sk_cgrp_data);
2199 mem_cgroup_sk_free(sk);
2200 security_sk_free(sk);
2201 if (slab != NULL)
2202 kmem_cache_free(slab, sk);
2203 else
2204 kfree(sk);
2205 module_put(owner);
2206}
2207
2208/**
2209 * sk_alloc - All socket objects are allocated here
2210 * @net: the applicable net namespace
2211 * @family: protocol family
2212 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2213 * @prot: struct proto associated with this new sock instance
2214 * @kern: is this to be a kernel socket?
2215 */
2216struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2217 struct proto *prot, int kern)
2218{
2219 struct sock *sk;
2220
2221 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2222 if (sk) {
2223 sk->sk_family = family;
2224 /*
2225 * See comment in struct sock definition to understand
2226 * why we need sk_prot_creator -acme
2227 */
2228 sk->sk_prot = sk->sk_prot_creator = prot;
2229 sk->sk_kern_sock = kern;
2230 sock_lock_init(sk);
2231 sk->sk_net_refcnt = kern ? 0 : 1;
2232 if (likely(sk->sk_net_refcnt)) {
2233 get_net_track(net, &sk->ns_tracker, priority);
2234 sock_inuse_add(net, 1);
2235 } else {
2236 net_passive_inc(net);
2237 __netns_tracker_alloc(net, &sk->ns_tracker,
2238 false, priority);
2239 }
2240
2241 sock_net_set(sk, net);
2242 refcount_set(&sk->sk_wmem_alloc, 1);
2243
2244 mem_cgroup_sk_alloc(sk);
2245 cgroup_sk_alloc(&sk->sk_cgrp_data);
2246 sock_update_classid(&sk->sk_cgrp_data);
2247 sock_update_netprioidx(&sk->sk_cgrp_data);
2248 sk_tx_queue_clear(sk);
2249 }
2250
2251 return sk;
2252}
2253EXPORT_SYMBOL(sk_alloc);
2254
2255/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2256 * grace period. This is the case for UDP sockets and TCP listeners.
2257 */
2258static void __sk_destruct(struct rcu_head *head)
2259{
2260 struct sock *sk = container_of(head, struct sock, sk_rcu);
2261 struct net *net = sock_net(sk);
2262 struct sk_filter *filter;
2263
2264 if (sk->sk_destruct)
2265 sk->sk_destruct(sk);
2266
2267 filter = rcu_dereference_check(sk->sk_filter,
2268 refcount_read(&sk->sk_wmem_alloc) == 0);
2269 if (filter) {
2270 sk_filter_uncharge(sk, filter);
2271 RCU_INIT_POINTER(sk->sk_filter, NULL);
2272 }
2273
2274 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2275
2276#ifdef CONFIG_BPF_SYSCALL
2277 bpf_sk_storage_free(sk);
2278#endif
2279
2280 if (atomic_read(&sk->sk_omem_alloc))
2281 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2282 __func__, atomic_read(&sk->sk_omem_alloc));
2283
2284 if (sk->sk_frag.page) {
2285 put_page(sk->sk_frag.page);
2286 sk->sk_frag.page = NULL;
2287 }
2288
2289 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2290 put_cred(sk->sk_peer_cred);
2291 put_pid(sk->sk_peer_pid);
2292
2293 if (likely(sk->sk_net_refcnt)) {
2294 put_net_track(net, &sk->ns_tracker);
2295 } else {
2296 __netns_tracker_free(net, &sk->ns_tracker, false);
2297 net_passive_dec(net);
2298 }
2299 sk_prot_free(sk->sk_prot_creator, sk);
2300}
2301
2302void sk_net_refcnt_upgrade(struct sock *sk)
2303{
2304 struct net *net = sock_net(sk);
2305
2306 WARN_ON_ONCE(sk->sk_net_refcnt);
2307 __netns_tracker_free(net, &sk->ns_tracker, false);
2308 net_passive_dec(net);
2309 sk->sk_net_refcnt = 1;
2310 get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
2311 sock_inuse_add(net, 1);
2312}
2313EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade);
2314
2315void sk_destruct(struct sock *sk)
2316{
2317 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2318
2319 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2320 reuseport_detach_sock(sk);
2321 use_call_rcu = true;
2322 }
2323
2324 if (use_call_rcu)
2325 call_rcu(&sk->sk_rcu, __sk_destruct);
2326 else
2327 __sk_destruct(&sk->sk_rcu);
2328}
2329
2330static void __sk_free(struct sock *sk)
2331{
2332 if (likely(sk->sk_net_refcnt))
2333 sock_inuse_add(sock_net(sk), -1);
2334
2335 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2336 sock_diag_broadcast_destroy(sk);
2337 else
2338 sk_destruct(sk);
2339}
2340
2341void sk_free(struct sock *sk)
2342{
2343 /*
2344 * We subtract one from sk_wmem_alloc and can know if
2345 * some packets are still in some tx queue.
2346 * If not null, sock_wfree() will call __sk_free(sk) later
2347 */
2348 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2349 __sk_free(sk);
2350}
2351EXPORT_SYMBOL(sk_free);
2352
2353static void sk_init_common(struct sock *sk)
2354{
2355 skb_queue_head_init(&sk->sk_receive_queue);
2356 skb_queue_head_init(&sk->sk_write_queue);
2357 skb_queue_head_init(&sk->sk_error_queue);
2358
2359 rwlock_init(&sk->sk_callback_lock);
2360 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2361 af_rlock_keys + sk->sk_family,
2362 af_family_rlock_key_strings[sk->sk_family]);
2363 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2364 af_wlock_keys + sk->sk_family,
2365 af_family_wlock_key_strings[sk->sk_family]);
2366 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2367 af_elock_keys + sk->sk_family,
2368 af_family_elock_key_strings[sk->sk_family]);
2369 if (sk->sk_kern_sock)
2370 lockdep_set_class_and_name(&sk->sk_callback_lock,
2371 af_kern_callback_keys + sk->sk_family,
2372 af_family_kern_clock_key_strings[sk->sk_family]);
2373 else
2374 lockdep_set_class_and_name(&sk->sk_callback_lock,
2375 af_callback_keys + sk->sk_family,
2376 af_family_clock_key_strings[sk->sk_family]);
2377}
2378
2379/**
2380 * sk_clone_lock - clone a socket, and lock its clone
2381 * @sk: the socket to clone
2382 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2383 *
2384 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2385 */
2386struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2387{
2388 struct proto *prot = READ_ONCE(sk->sk_prot);
2389 struct sk_filter *filter;
2390 bool is_charged = true;
2391 struct sock *newsk;
2392
2393 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2394 if (!newsk)
2395 goto out;
2396
2397 sock_copy(newsk, sk);
2398
2399 newsk->sk_prot_creator = prot;
2400
2401 /* SANITY */
2402 if (likely(newsk->sk_net_refcnt)) {
2403 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2404 sock_inuse_add(sock_net(newsk), 1);
2405 } else {
2406 /* Kernel sockets are not elevating the struct net refcount.
2407 * Instead, use a tracker to more easily detect if a layer
2408 * is not properly dismantling its kernel sockets at netns
2409 * destroy time.
2410 */
2411 net_passive_inc(sock_net(newsk));
2412 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2413 false, priority);
2414 }
2415 sk_node_init(&newsk->sk_node);
2416 sock_lock_init(newsk);
2417 bh_lock_sock(newsk);
2418 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2419 newsk->sk_backlog.len = 0;
2420
2421 atomic_set(&newsk->sk_rmem_alloc, 0);
2422
2423 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2424 refcount_set(&newsk->sk_wmem_alloc, 1);
2425
2426 atomic_set(&newsk->sk_omem_alloc, 0);
2427 sk_init_common(newsk);
2428
2429 newsk->sk_dst_cache = NULL;
2430 newsk->sk_dst_pending_confirm = 0;
2431 newsk->sk_wmem_queued = 0;
2432 newsk->sk_forward_alloc = 0;
2433 newsk->sk_reserved_mem = 0;
2434 atomic_set(&newsk->sk_drops, 0);
2435 newsk->sk_send_head = NULL;
2436 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2437 atomic_set(&newsk->sk_zckey, 0);
2438
2439 sock_reset_flag(newsk, SOCK_DONE);
2440
2441 /* sk->sk_memcg will be populated at accept() time */
2442 newsk->sk_memcg = NULL;
2443
2444 cgroup_sk_clone(&newsk->sk_cgrp_data);
2445
2446 rcu_read_lock();
2447 filter = rcu_dereference(sk->sk_filter);
2448 if (filter != NULL)
2449 /* though it's an empty new sock, the charging may fail
2450 * if sysctl_optmem_max was changed between creation of
2451 * original socket and cloning
2452 */
2453 is_charged = sk_filter_charge(newsk, filter);
2454 RCU_INIT_POINTER(newsk->sk_filter, filter);
2455 rcu_read_unlock();
2456
2457 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2458 /* We need to make sure that we don't uncharge the new
2459 * socket if we couldn't charge it in the first place
2460 * as otherwise we uncharge the parent's filter.
2461 */
2462 if (!is_charged)
2463 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2464 sk_free_unlock_clone(newsk);
2465 newsk = NULL;
2466 goto out;
2467 }
2468 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2469
2470 if (bpf_sk_storage_clone(sk, newsk)) {
2471 sk_free_unlock_clone(newsk);
2472 newsk = NULL;
2473 goto out;
2474 }
2475
2476 /* Clear sk_user_data if parent had the pointer tagged
2477 * as not suitable for copying when cloning.
2478 */
2479 if (sk_user_data_is_nocopy(newsk))
2480 newsk->sk_user_data = NULL;
2481
2482 newsk->sk_err = 0;
2483 newsk->sk_err_soft = 0;
2484 newsk->sk_priority = 0;
2485 newsk->sk_incoming_cpu = raw_smp_processor_id();
2486
2487 /* Before updating sk_refcnt, we must commit prior changes to memory
2488 * (Documentation/RCU/rculist_nulls.rst for details)
2489 */
2490 smp_wmb();
2491 refcount_set(&newsk->sk_refcnt, 2);
2492
2493 sk_set_socket(newsk, NULL);
2494 sk_tx_queue_clear(newsk);
2495 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2496
2497 if (newsk->sk_prot->sockets_allocated)
2498 sk_sockets_allocated_inc(newsk);
2499
2500 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2501 net_enable_timestamp();
2502out:
2503 return newsk;
2504}
2505EXPORT_SYMBOL_GPL(sk_clone_lock);
2506
2507void sk_free_unlock_clone(struct sock *sk)
2508{
2509 /* It is still raw copy of parent, so invalidate
2510 * destructor and make plain sk_free() */
2511 sk->sk_destruct = NULL;
2512 bh_unlock_sock(sk);
2513 sk_free(sk);
2514}
2515EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2516
2517static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2518{
2519 bool is_ipv6 = false;
2520 u32 max_size;
2521
2522#if IS_ENABLED(CONFIG_IPV6)
2523 is_ipv6 = (sk->sk_family == AF_INET6 &&
2524 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2525#endif
2526 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2527 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2528 READ_ONCE(dst->dev->gso_ipv4_max_size);
2529 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2530 max_size = GSO_LEGACY_MAX_SIZE;
2531
2532 return max_size - (MAX_TCP_HEADER + 1);
2533}
2534
2535void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2536{
2537 u32 max_segs = 1;
2538
2539 sk->sk_route_caps = dst->dev->features;
2540 if (sk_is_tcp(sk))
2541 sk->sk_route_caps |= NETIF_F_GSO;
2542 if (sk->sk_route_caps & NETIF_F_GSO)
2543 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2544 if (unlikely(sk->sk_gso_disabled))
2545 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2546 if (sk_can_gso(sk)) {
2547 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2548 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2549 } else {
2550 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2551 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2552 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2553 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2554 }
2555 }
2556 sk->sk_gso_max_segs = max_segs;
2557 sk_dst_set(sk, dst);
2558}
2559EXPORT_SYMBOL_GPL(sk_setup_caps);
2560
2561/*
2562 * Simple resource managers for sockets.
2563 */
2564
2565
2566/*
2567 * Write buffer destructor automatically called from kfree_skb.
2568 */
2569void sock_wfree(struct sk_buff *skb)
2570{
2571 struct sock *sk = skb->sk;
2572 unsigned int len = skb->truesize;
2573 bool free;
2574
2575 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2576 if (sock_flag(sk, SOCK_RCU_FREE) &&
2577 sk->sk_write_space == sock_def_write_space) {
2578 rcu_read_lock();
2579 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2580 sock_def_write_space_wfree(sk);
2581 rcu_read_unlock();
2582 if (unlikely(free))
2583 __sk_free(sk);
2584 return;
2585 }
2586
2587 /*
2588 * Keep a reference on sk_wmem_alloc, this will be released
2589 * after sk_write_space() call
2590 */
2591 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2592 sk->sk_write_space(sk);
2593 len = 1;
2594 }
2595 /*
2596 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2597 * could not do because of in-flight packets
2598 */
2599 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2600 __sk_free(sk);
2601}
2602EXPORT_SYMBOL(sock_wfree);
2603
2604/* This variant of sock_wfree() is used by TCP,
2605 * since it sets SOCK_USE_WRITE_QUEUE.
2606 */
2607void __sock_wfree(struct sk_buff *skb)
2608{
2609 struct sock *sk = skb->sk;
2610
2611 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2612 __sk_free(sk);
2613}
2614
2615void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2616{
2617 skb_orphan(skb);
2618#ifdef CONFIG_INET
2619 if (unlikely(!sk_fullsock(sk)))
2620 return skb_set_owner_edemux(skb, sk);
2621#endif
2622 skb->sk = sk;
2623 skb->destructor = sock_wfree;
2624 skb_set_hash_from_sk(skb, sk);
2625 /*
2626 * We used to take a refcount on sk, but following operation
2627 * is enough to guarantee sk_free() won't free this sock until
2628 * all in-flight packets are completed
2629 */
2630 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2631}
2632EXPORT_SYMBOL(skb_set_owner_w);
2633
2634static bool can_skb_orphan_partial(const struct sk_buff *skb)
2635{
2636 /* Drivers depend on in-order delivery for crypto offload,
2637 * partial orphan breaks out-of-order-OK logic.
2638 */
2639 if (skb_is_decrypted(skb))
2640 return false;
2641
2642 return (skb->destructor == sock_wfree ||
2643 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2644}
2645
2646/* This helper is used by netem, as it can hold packets in its
2647 * delay queue. We want to allow the owner socket to send more
2648 * packets, as if they were already TX completed by a typical driver.
2649 * But we also want to keep skb->sk set because some packet schedulers
2650 * rely on it (sch_fq for example).
2651 */
2652void skb_orphan_partial(struct sk_buff *skb)
2653{
2654 if (skb_is_tcp_pure_ack(skb))
2655 return;
2656
2657 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2658 return;
2659
2660 skb_orphan(skb);
2661}
2662EXPORT_SYMBOL(skb_orphan_partial);
2663
2664/*
2665 * Read buffer destructor automatically called from kfree_skb.
2666 */
2667void sock_rfree(struct sk_buff *skb)
2668{
2669 struct sock *sk = skb->sk;
2670 unsigned int len = skb->truesize;
2671
2672 atomic_sub(len, &sk->sk_rmem_alloc);
2673 sk_mem_uncharge(sk, len);
2674}
2675EXPORT_SYMBOL(sock_rfree);
2676
2677/*
2678 * Buffer destructor for skbs that are not used directly in read or write
2679 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2680 */
2681void sock_efree(struct sk_buff *skb)
2682{
2683 sock_put(skb->sk);
2684}
2685EXPORT_SYMBOL(sock_efree);
2686
2687/* Buffer destructor for prefetch/receive path where reference count may
2688 * not be held, e.g. for listen sockets.
2689 */
2690#ifdef CONFIG_INET
2691void sock_pfree(struct sk_buff *skb)
2692{
2693 struct sock *sk = skb->sk;
2694
2695 if (!sk_is_refcounted(sk))
2696 return;
2697
2698 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2699 inet_reqsk(sk)->rsk_listener = NULL;
2700 reqsk_free(inet_reqsk(sk));
2701 return;
2702 }
2703
2704 sock_gen_put(sk);
2705}
2706EXPORT_SYMBOL(sock_pfree);
2707#endif /* CONFIG_INET */
2708
2709kuid_t sock_i_uid(struct sock *sk)
2710{
2711 kuid_t uid;
2712
2713 read_lock_bh(&sk->sk_callback_lock);
2714 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2715 read_unlock_bh(&sk->sk_callback_lock);
2716 return uid;
2717}
2718EXPORT_SYMBOL(sock_i_uid);
2719
2720unsigned long __sock_i_ino(struct sock *sk)
2721{
2722 unsigned long ino;
2723
2724 read_lock(&sk->sk_callback_lock);
2725 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2726 read_unlock(&sk->sk_callback_lock);
2727 return ino;
2728}
2729EXPORT_SYMBOL(__sock_i_ino);
2730
2731unsigned long sock_i_ino(struct sock *sk)
2732{
2733 unsigned long ino;
2734
2735 local_bh_disable();
2736 ino = __sock_i_ino(sk);
2737 local_bh_enable();
2738 return ino;
2739}
2740EXPORT_SYMBOL(sock_i_ino);
2741
2742/*
2743 * Allocate a skb from the socket's send buffer.
2744 */
2745struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2746 gfp_t priority)
2747{
2748 if (force ||
2749 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2750 struct sk_buff *skb = alloc_skb(size, priority);
2751
2752 if (skb) {
2753 skb_set_owner_w(skb, sk);
2754 return skb;
2755 }
2756 }
2757 return NULL;
2758}
2759EXPORT_SYMBOL(sock_wmalloc);
2760
2761static void sock_ofree(struct sk_buff *skb)
2762{
2763 struct sock *sk = skb->sk;
2764
2765 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2766}
2767
2768struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2769 gfp_t priority)
2770{
2771 struct sk_buff *skb;
2772
2773 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2774 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2775 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2776 return NULL;
2777
2778 skb = alloc_skb(size, priority);
2779 if (!skb)
2780 return NULL;
2781
2782 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2783 skb->sk = sk;
2784 skb->destructor = sock_ofree;
2785 return skb;
2786}
2787
2788/*
2789 * Allocate a memory block from the socket's option memory buffer.
2790 */
2791void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2792{
2793 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2794
2795 if ((unsigned int)size <= optmem_max &&
2796 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2797 void *mem;
2798 /* First do the add, to avoid the race if kmalloc
2799 * might sleep.
2800 */
2801 atomic_add(size, &sk->sk_omem_alloc);
2802 mem = kmalloc(size, priority);
2803 if (mem)
2804 return mem;
2805 atomic_sub(size, &sk->sk_omem_alloc);
2806 }
2807 return NULL;
2808}
2809EXPORT_SYMBOL(sock_kmalloc);
2810
2811/* Free an option memory block. Note, we actually want the inline
2812 * here as this allows gcc to detect the nullify and fold away the
2813 * condition entirely.
2814 */
2815static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2816 const bool nullify)
2817{
2818 if (WARN_ON_ONCE(!mem))
2819 return;
2820 if (nullify)
2821 kfree_sensitive(mem);
2822 else
2823 kfree(mem);
2824 atomic_sub(size, &sk->sk_omem_alloc);
2825}
2826
2827void sock_kfree_s(struct sock *sk, void *mem, int size)
2828{
2829 __sock_kfree_s(sk, mem, size, false);
2830}
2831EXPORT_SYMBOL(sock_kfree_s);
2832
2833void sock_kzfree_s(struct sock *sk, void *mem, int size)
2834{
2835 __sock_kfree_s(sk, mem, size, true);
2836}
2837EXPORT_SYMBOL(sock_kzfree_s);
2838
2839/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2840 I think, these locks should be removed for datagram sockets.
2841 */
2842static long sock_wait_for_wmem(struct sock *sk, long timeo)
2843{
2844 DEFINE_WAIT(wait);
2845
2846 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2847 for (;;) {
2848 if (!timeo)
2849 break;
2850 if (signal_pending(current))
2851 break;
2852 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2853 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2854 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2855 break;
2856 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2857 break;
2858 if (READ_ONCE(sk->sk_err))
2859 break;
2860 timeo = schedule_timeout(timeo);
2861 }
2862 finish_wait(sk_sleep(sk), &wait);
2863 return timeo;
2864}
2865
2866
2867/*
2868 * Generic send/receive buffer handlers
2869 */
2870
2871struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2872 unsigned long data_len, int noblock,
2873 int *errcode, int max_page_order)
2874{
2875 struct sk_buff *skb;
2876 long timeo;
2877 int err;
2878
2879 timeo = sock_sndtimeo(sk, noblock);
2880 for (;;) {
2881 err = sock_error(sk);
2882 if (err != 0)
2883 goto failure;
2884
2885 err = -EPIPE;
2886 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2887 goto failure;
2888
2889 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2890 break;
2891
2892 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2893 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2894 err = -EAGAIN;
2895 if (!timeo)
2896 goto failure;
2897 if (signal_pending(current))
2898 goto interrupted;
2899 timeo = sock_wait_for_wmem(sk, timeo);
2900 }
2901 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2902 errcode, sk->sk_allocation);
2903 if (skb)
2904 skb_set_owner_w(skb, sk);
2905 return skb;
2906
2907interrupted:
2908 err = sock_intr_errno(timeo);
2909failure:
2910 *errcode = err;
2911 return NULL;
2912}
2913EXPORT_SYMBOL(sock_alloc_send_pskb);
2914
2915int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2916 struct sockcm_cookie *sockc)
2917{
2918 u32 tsflags;
2919
2920 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
2921
2922 switch (cmsg->cmsg_type) {
2923 case SO_MARK:
2924 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2925 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2926 return -EPERM;
2927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2928 return -EINVAL;
2929 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2930 break;
2931 case SO_TIMESTAMPING_OLD:
2932 case SO_TIMESTAMPING_NEW:
2933 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2934 return -EINVAL;
2935
2936 tsflags = *(u32 *)CMSG_DATA(cmsg);
2937 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2938 return -EINVAL;
2939
2940 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2941 sockc->tsflags |= tsflags;
2942 break;
2943 case SCM_TXTIME:
2944 if (!sock_flag(sk, SOCK_TXTIME))
2945 return -EINVAL;
2946 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2947 return -EINVAL;
2948 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2949 break;
2950 case SCM_TS_OPT_ID:
2951 if (sk_is_tcp(sk))
2952 return -EINVAL;
2953 tsflags = READ_ONCE(sk->sk_tsflags);
2954 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
2955 return -EINVAL;
2956 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2957 return -EINVAL;
2958 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
2959 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
2960 break;
2961 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2962 case SCM_RIGHTS:
2963 case SCM_CREDENTIALS:
2964 break;
2965 default:
2966 return -EINVAL;
2967 }
2968 return 0;
2969}
2970EXPORT_SYMBOL(__sock_cmsg_send);
2971
2972int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2973 struct sockcm_cookie *sockc)
2974{
2975 struct cmsghdr *cmsg;
2976 int ret;
2977
2978 for_each_cmsghdr(cmsg, msg) {
2979 if (!CMSG_OK(msg, cmsg))
2980 return -EINVAL;
2981 if (cmsg->cmsg_level != SOL_SOCKET)
2982 continue;
2983 ret = __sock_cmsg_send(sk, cmsg, sockc);
2984 if (ret)
2985 return ret;
2986 }
2987 return 0;
2988}
2989EXPORT_SYMBOL(sock_cmsg_send);
2990
2991static void sk_enter_memory_pressure(struct sock *sk)
2992{
2993 if (!sk->sk_prot->enter_memory_pressure)
2994 return;
2995
2996 sk->sk_prot->enter_memory_pressure(sk);
2997}
2998
2999static void sk_leave_memory_pressure(struct sock *sk)
3000{
3001 if (sk->sk_prot->leave_memory_pressure) {
3002 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3003 tcp_leave_memory_pressure, sk);
3004 } else {
3005 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3006
3007 if (memory_pressure && READ_ONCE(*memory_pressure))
3008 WRITE_ONCE(*memory_pressure, 0);
3009 }
3010}
3011
3012DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3013
3014/**
3015 * skb_page_frag_refill - check that a page_frag contains enough room
3016 * @sz: minimum size of the fragment we want to get
3017 * @pfrag: pointer to page_frag
3018 * @gfp: priority for memory allocation
3019 *
3020 * Note: While this allocator tries to use high order pages, there is
3021 * no guarantee that allocations succeed. Therefore, @sz MUST be
3022 * less or equal than PAGE_SIZE.
3023 */
3024bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3025{
3026 if (pfrag->page) {
3027 if (page_ref_count(pfrag->page) == 1) {
3028 pfrag->offset = 0;
3029 return true;
3030 }
3031 if (pfrag->offset + sz <= pfrag->size)
3032 return true;
3033 put_page(pfrag->page);
3034 }
3035
3036 pfrag->offset = 0;
3037 if (SKB_FRAG_PAGE_ORDER &&
3038 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3039 /* Avoid direct reclaim but allow kswapd to wake */
3040 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3041 __GFP_COMP | __GFP_NOWARN |
3042 __GFP_NORETRY,
3043 SKB_FRAG_PAGE_ORDER);
3044 if (likely(pfrag->page)) {
3045 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3046 return true;
3047 }
3048 }
3049 pfrag->page = alloc_page(gfp);
3050 if (likely(pfrag->page)) {
3051 pfrag->size = PAGE_SIZE;
3052 return true;
3053 }
3054 return false;
3055}
3056EXPORT_SYMBOL(skb_page_frag_refill);
3057
3058bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3059{
3060 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3061 return true;
3062
3063 sk_enter_memory_pressure(sk);
3064 sk_stream_moderate_sndbuf(sk);
3065 return false;
3066}
3067EXPORT_SYMBOL(sk_page_frag_refill);
3068
3069void __lock_sock(struct sock *sk)
3070 __releases(&sk->sk_lock.slock)
3071 __acquires(&sk->sk_lock.slock)
3072{
3073 DEFINE_WAIT(wait);
3074
3075 for (;;) {
3076 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3077 TASK_UNINTERRUPTIBLE);
3078 spin_unlock_bh(&sk->sk_lock.slock);
3079 schedule();
3080 spin_lock_bh(&sk->sk_lock.slock);
3081 if (!sock_owned_by_user(sk))
3082 break;
3083 }
3084 finish_wait(&sk->sk_lock.wq, &wait);
3085}
3086
3087void __release_sock(struct sock *sk)
3088 __releases(&sk->sk_lock.slock)
3089 __acquires(&sk->sk_lock.slock)
3090{
3091 struct sk_buff *skb, *next;
3092
3093 while ((skb = sk->sk_backlog.head) != NULL) {
3094 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3095
3096 spin_unlock_bh(&sk->sk_lock.slock);
3097
3098 do {
3099 next = skb->next;
3100 prefetch(next);
3101 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3102 skb_mark_not_on_list(skb);
3103 sk_backlog_rcv(sk, skb);
3104
3105 cond_resched();
3106
3107 skb = next;
3108 } while (skb != NULL);
3109
3110 spin_lock_bh(&sk->sk_lock.slock);
3111 }
3112
3113 /*
3114 * Doing the zeroing here guarantee we can not loop forever
3115 * while a wild producer attempts to flood us.
3116 */
3117 sk->sk_backlog.len = 0;
3118}
3119
3120void __sk_flush_backlog(struct sock *sk)
3121{
3122 spin_lock_bh(&sk->sk_lock.slock);
3123 __release_sock(sk);
3124
3125 if (sk->sk_prot->release_cb)
3126 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3127 tcp_release_cb, sk);
3128
3129 spin_unlock_bh(&sk->sk_lock.slock);
3130}
3131EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3132
3133/**
3134 * sk_wait_data - wait for data to arrive at sk_receive_queue
3135 * @sk: sock to wait on
3136 * @timeo: for how long
3137 * @skb: last skb seen on sk_receive_queue
3138 *
3139 * Now socket state including sk->sk_err is changed only under lock,
3140 * hence we may omit checks after joining wait queue.
3141 * We check receive queue before schedule() only as optimization;
3142 * it is very likely that release_sock() added new data.
3143 */
3144int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3145{
3146 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3147 int rc;
3148
3149 add_wait_queue(sk_sleep(sk), &wait);
3150 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3151 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3152 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3153 remove_wait_queue(sk_sleep(sk), &wait);
3154 return rc;
3155}
3156EXPORT_SYMBOL(sk_wait_data);
3157
3158/**
3159 * __sk_mem_raise_allocated - increase memory_allocated
3160 * @sk: socket
3161 * @size: memory size to allocate
3162 * @amt: pages to allocate
3163 * @kind: allocation type
3164 *
3165 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3166 *
3167 * Unlike the globally shared limits among the sockets under same protocol,
3168 * consuming the budget of a memcg won't have direct effect on other ones.
3169 * So be optimistic about memcg's tolerance, and leave the callers to decide
3170 * whether or not to raise allocated through sk_under_memory_pressure() or
3171 * its variants.
3172 */
3173int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3174{
3175 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3176 struct proto *prot = sk->sk_prot;
3177 bool charged = false;
3178 long allocated;
3179
3180 sk_memory_allocated_add(sk, amt);
3181 allocated = sk_memory_allocated(sk);
3182
3183 if (memcg) {
3184 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3185 goto suppress_allocation;
3186 charged = true;
3187 }
3188
3189 /* Under limit. */
3190 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3191 sk_leave_memory_pressure(sk);
3192 return 1;
3193 }
3194
3195 /* Under pressure. */
3196 if (allocated > sk_prot_mem_limits(sk, 1))
3197 sk_enter_memory_pressure(sk);
3198
3199 /* Over hard limit. */
3200 if (allocated > sk_prot_mem_limits(sk, 2))
3201 goto suppress_allocation;
3202
3203 /* Guarantee minimum buffer size under pressure (either global
3204 * or memcg) to make sure features described in RFC 7323 (TCP
3205 * Extensions for High Performance) work properly.
3206 *
3207 * This rule does NOT stand when exceeds global or memcg's hard
3208 * limit, or else a DoS attack can be taken place by spawning
3209 * lots of sockets whose usage are under minimum buffer size.
3210 */
3211 if (kind == SK_MEM_RECV) {
3212 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3213 return 1;
3214
3215 } else { /* SK_MEM_SEND */
3216 int wmem0 = sk_get_wmem0(sk, prot);
3217
3218 if (sk->sk_type == SOCK_STREAM) {
3219 if (sk->sk_wmem_queued < wmem0)
3220 return 1;
3221 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3222 return 1;
3223 }
3224 }
3225
3226 if (sk_has_memory_pressure(sk)) {
3227 u64 alloc;
3228
3229 /* The following 'average' heuristic is within the
3230 * scope of global accounting, so it only makes
3231 * sense for global memory pressure.
3232 */
3233 if (!sk_under_global_memory_pressure(sk))
3234 return 1;
3235
3236 /* Try to be fair among all the sockets under global
3237 * pressure by allowing the ones that below average
3238 * usage to raise.
3239 */
3240 alloc = sk_sockets_allocated_read_positive(sk);
3241 if (sk_prot_mem_limits(sk, 2) > alloc *
3242 sk_mem_pages(sk->sk_wmem_queued +
3243 atomic_read(&sk->sk_rmem_alloc) +
3244 sk->sk_forward_alloc))
3245 return 1;
3246 }
3247
3248suppress_allocation:
3249
3250 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3251 sk_stream_moderate_sndbuf(sk);
3252
3253 /* Fail only if socket is _under_ its sndbuf.
3254 * In this case we cannot block, so that we have to fail.
3255 */
3256 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3257 /* Force charge with __GFP_NOFAIL */
3258 if (memcg && !charged) {
3259 mem_cgroup_charge_skmem(memcg, amt,
3260 gfp_memcg_charge() | __GFP_NOFAIL);
3261 }
3262 return 1;
3263 }
3264 }
3265
3266 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3267 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3268
3269 sk_memory_allocated_sub(sk, amt);
3270
3271 if (charged)
3272 mem_cgroup_uncharge_skmem(memcg, amt);
3273
3274 return 0;
3275}
3276
3277/**
3278 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3279 * @sk: socket
3280 * @size: memory size to allocate
3281 * @kind: allocation type
3282 *
3283 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3284 * rmem allocation. This function assumes that protocols which have
3285 * memory_pressure use sk_wmem_queued as write buffer accounting.
3286 */
3287int __sk_mem_schedule(struct sock *sk, int size, int kind)
3288{
3289 int ret, amt = sk_mem_pages(size);
3290
3291 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3292 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3293 if (!ret)
3294 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3295 return ret;
3296}
3297EXPORT_SYMBOL(__sk_mem_schedule);
3298
3299/**
3300 * __sk_mem_reduce_allocated - reclaim memory_allocated
3301 * @sk: socket
3302 * @amount: number of quanta
3303 *
3304 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3305 */
3306void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3307{
3308 sk_memory_allocated_sub(sk, amount);
3309
3310 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3311 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3312
3313 if (sk_under_global_memory_pressure(sk) &&
3314 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3315 sk_leave_memory_pressure(sk);
3316}
3317
3318/**
3319 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3320 * @sk: socket
3321 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3322 */
3323void __sk_mem_reclaim(struct sock *sk, int amount)
3324{
3325 amount >>= PAGE_SHIFT;
3326 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3327 __sk_mem_reduce_allocated(sk, amount);
3328}
3329EXPORT_SYMBOL(__sk_mem_reclaim);
3330
3331int sk_set_peek_off(struct sock *sk, int val)
3332{
3333 WRITE_ONCE(sk->sk_peek_off, val);
3334 return 0;
3335}
3336EXPORT_SYMBOL_GPL(sk_set_peek_off);
3337
3338/*
3339 * Set of default routines for initialising struct proto_ops when
3340 * the protocol does not support a particular function. In certain
3341 * cases where it makes no sense for a protocol to have a "do nothing"
3342 * function, some default processing is provided.
3343 */
3344
3345int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3346{
3347 return -EOPNOTSUPP;
3348}
3349EXPORT_SYMBOL(sock_no_bind);
3350
3351int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3352 int len, int flags)
3353{
3354 return -EOPNOTSUPP;
3355}
3356EXPORT_SYMBOL(sock_no_connect);
3357
3358int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3359{
3360 return -EOPNOTSUPP;
3361}
3362EXPORT_SYMBOL(sock_no_socketpair);
3363
3364int sock_no_accept(struct socket *sock, struct socket *newsock,
3365 struct proto_accept_arg *arg)
3366{
3367 return -EOPNOTSUPP;
3368}
3369EXPORT_SYMBOL(sock_no_accept);
3370
3371int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3372 int peer)
3373{
3374 return -EOPNOTSUPP;
3375}
3376EXPORT_SYMBOL(sock_no_getname);
3377
3378int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3379{
3380 return -EOPNOTSUPP;
3381}
3382EXPORT_SYMBOL(sock_no_ioctl);
3383
3384int sock_no_listen(struct socket *sock, int backlog)
3385{
3386 return -EOPNOTSUPP;
3387}
3388EXPORT_SYMBOL(sock_no_listen);
3389
3390int sock_no_shutdown(struct socket *sock, int how)
3391{
3392 return -EOPNOTSUPP;
3393}
3394EXPORT_SYMBOL(sock_no_shutdown);
3395
3396int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3397{
3398 return -EOPNOTSUPP;
3399}
3400EXPORT_SYMBOL(sock_no_sendmsg);
3401
3402int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3403{
3404 return -EOPNOTSUPP;
3405}
3406EXPORT_SYMBOL(sock_no_sendmsg_locked);
3407
3408int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3409 int flags)
3410{
3411 return -EOPNOTSUPP;
3412}
3413EXPORT_SYMBOL(sock_no_recvmsg);
3414
3415int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3416{
3417 /* Mirror missing mmap method error code */
3418 return -ENODEV;
3419}
3420EXPORT_SYMBOL(sock_no_mmap);
3421
3422/*
3423 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3424 * various sock-based usage counts.
3425 */
3426void __receive_sock(struct file *file)
3427{
3428 struct socket *sock;
3429
3430 sock = sock_from_file(file);
3431 if (sock) {
3432 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3433 sock_update_classid(&sock->sk->sk_cgrp_data);
3434 }
3435}
3436
3437/*
3438 * Default Socket Callbacks
3439 */
3440
3441static void sock_def_wakeup(struct sock *sk)
3442{
3443 struct socket_wq *wq;
3444
3445 rcu_read_lock();
3446 wq = rcu_dereference(sk->sk_wq);
3447 if (skwq_has_sleeper(wq))
3448 wake_up_interruptible_all(&wq->wait);
3449 rcu_read_unlock();
3450}
3451
3452static void sock_def_error_report(struct sock *sk)
3453{
3454 struct socket_wq *wq;
3455
3456 rcu_read_lock();
3457 wq = rcu_dereference(sk->sk_wq);
3458 if (skwq_has_sleeper(wq))
3459 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3460 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3461 rcu_read_unlock();
3462}
3463
3464void sock_def_readable(struct sock *sk)
3465{
3466 struct socket_wq *wq;
3467
3468 trace_sk_data_ready(sk);
3469
3470 rcu_read_lock();
3471 wq = rcu_dereference(sk->sk_wq);
3472 if (skwq_has_sleeper(wq))
3473 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3474 EPOLLRDNORM | EPOLLRDBAND);
3475 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3476 rcu_read_unlock();
3477}
3478
3479static void sock_def_write_space(struct sock *sk)
3480{
3481 struct socket_wq *wq;
3482
3483 rcu_read_lock();
3484
3485 /* Do not wake up a writer until he can make "significant"
3486 * progress. --DaveM
3487 */
3488 if (sock_writeable(sk)) {
3489 wq = rcu_dereference(sk->sk_wq);
3490 if (skwq_has_sleeper(wq))
3491 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3492 EPOLLWRNORM | EPOLLWRBAND);
3493
3494 /* Should agree with poll, otherwise some programs break */
3495 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3496 }
3497
3498 rcu_read_unlock();
3499}
3500
3501/* An optimised version of sock_def_write_space(), should only be called
3502 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3503 * ->sk_wmem_alloc.
3504 */
3505static void sock_def_write_space_wfree(struct sock *sk)
3506{
3507 /* Do not wake up a writer until he can make "significant"
3508 * progress. --DaveM
3509 */
3510 if (sock_writeable(sk)) {
3511 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3512
3513 /* rely on refcount_sub from sock_wfree() */
3514 smp_mb__after_atomic();
3515 if (wq && waitqueue_active(&wq->wait))
3516 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3517 EPOLLWRNORM | EPOLLWRBAND);
3518
3519 /* Should agree with poll, otherwise some programs break */
3520 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3521 }
3522}
3523
3524static void sock_def_destruct(struct sock *sk)
3525{
3526}
3527
3528void sk_send_sigurg(struct sock *sk)
3529{
3530 if (sk->sk_socket && sk->sk_socket->file)
3531 if (send_sigurg(sk->sk_socket->file))
3532 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3533}
3534EXPORT_SYMBOL(sk_send_sigurg);
3535
3536void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3537 unsigned long expires)
3538{
3539 if (!mod_timer(timer, expires))
3540 sock_hold(sk);
3541}
3542EXPORT_SYMBOL(sk_reset_timer);
3543
3544void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3545{
3546 if (del_timer(timer))
3547 __sock_put(sk);
3548}
3549EXPORT_SYMBOL(sk_stop_timer);
3550
3551void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3552{
3553 if (del_timer_sync(timer))
3554 __sock_put(sk);
3555}
3556EXPORT_SYMBOL(sk_stop_timer_sync);
3557
3558void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3559{
3560 sk_init_common(sk);
3561 sk->sk_send_head = NULL;
3562
3563 timer_setup(&sk->sk_timer, NULL, 0);
3564
3565 sk->sk_allocation = GFP_KERNEL;
3566 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3567 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3568 sk->sk_state = TCP_CLOSE;
3569 sk->sk_use_task_frag = true;
3570 sk_set_socket(sk, sock);
3571
3572 sock_set_flag(sk, SOCK_ZAPPED);
3573
3574 if (sock) {
3575 sk->sk_type = sock->type;
3576 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3577 sock->sk = sk;
3578 } else {
3579 RCU_INIT_POINTER(sk->sk_wq, NULL);
3580 }
3581 sk->sk_uid = uid;
3582
3583 sk->sk_state_change = sock_def_wakeup;
3584 sk->sk_data_ready = sock_def_readable;
3585 sk->sk_write_space = sock_def_write_space;
3586 sk->sk_error_report = sock_def_error_report;
3587 sk->sk_destruct = sock_def_destruct;
3588
3589 sk->sk_frag.page = NULL;
3590 sk->sk_frag.offset = 0;
3591 sk->sk_peek_off = -1;
3592
3593 sk->sk_peer_pid = NULL;
3594 sk->sk_peer_cred = NULL;
3595 spin_lock_init(&sk->sk_peer_lock);
3596
3597 sk->sk_write_pending = 0;
3598 sk->sk_rcvlowat = 1;
3599 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3600 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3601
3602 sk->sk_stamp = SK_DEFAULT_STAMP;
3603#if BITS_PER_LONG==32
3604 seqlock_init(&sk->sk_stamp_seq);
3605#endif
3606 atomic_set(&sk->sk_zckey, 0);
3607
3608#ifdef CONFIG_NET_RX_BUSY_POLL
3609 sk->sk_napi_id = 0;
3610 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3611#endif
3612
3613 sk->sk_max_pacing_rate = ~0UL;
3614 sk->sk_pacing_rate = ~0UL;
3615 WRITE_ONCE(sk->sk_pacing_shift, 10);
3616 sk->sk_incoming_cpu = -1;
3617
3618 sk_rx_queue_clear(sk);
3619 /*
3620 * Before updating sk_refcnt, we must commit prior changes to memory
3621 * (Documentation/RCU/rculist_nulls.rst for details)
3622 */
3623 smp_wmb();
3624 refcount_set(&sk->sk_refcnt, 1);
3625 atomic_set(&sk->sk_drops, 0);
3626}
3627EXPORT_SYMBOL(sock_init_data_uid);
3628
3629void sock_init_data(struct socket *sock, struct sock *sk)
3630{
3631 kuid_t uid = sock ?
3632 SOCK_INODE(sock)->i_uid :
3633 make_kuid(sock_net(sk)->user_ns, 0);
3634
3635 sock_init_data_uid(sock, sk, uid);
3636}
3637EXPORT_SYMBOL(sock_init_data);
3638
3639void lock_sock_nested(struct sock *sk, int subclass)
3640{
3641 /* The sk_lock has mutex_lock() semantics here. */
3642 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3643
3644 might_sleep();
3645 spin_lock_bh(&sk->sk_lock.slock);
3646 if (sock_owned_by_user_nocheck(sk))
3647 __lock_sock(sk);
3648 sk->sk_lock.owned = 1;
3649 spin_unlock_bh(&sk->sk_lock.slock);
3650}
3651EXPORT_SYMBOL(lock_sock_nested);
3652
3653void release_sock(struct sock *sk)
3654{
3655 spin_lock_bh(&sk->sk_lock.slock);
3656 if (sk->sk_backlog.tail)
3657 __release_sock(sk);
3658
3659 if (sk->sk_prot->release_cb)
3660 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3661 tcp_release_cb, sk);
3662
3663 sock_release_ownership(sk);
3664 if (waitqueue_active(&sk->sk_lock.wq))
3665 wake_up(&sk->sk_lock.wq);
3666 spin_unlock_bh(&sk->sk_lock.slock);
3667}
3668EXPORT_SYMBOL(release_sock);
3669
3670bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3671{
3672 might_sleep();
3673 spin_lock_bh(&sk->sk_lock.slock);
3674
3675 if (!sock_owned_by_user_nocheck(sk)) {
3676 /*
3677 * Fast path return with bottom halves disabled and
3678 * sock::sk_lock.slock held.
3679 *
3680 * The 'mutex' is not contended and holding
3681 * sock::sk_lock.slock prevents all other lockers to
3682 * proceed so the corresponding unlock_sock_fast() can
3683 * avoid the slow path of release_sock() completely and
3684 * just release slock.
3685 *
3686 * From a semantical POV this is equivalent to 'acquiring'
3687 * the 'mutex', hence the corresponding lockdep
3688 * mutex_release() has to happen in the fast path of
3689 * unlock_sock_fast().
3690 */
3691 return false;
3692 }
3693
3694 __lock_sock(sk);
3695 sk->sk_lock.owned = 1;
3696 __acquire(&sk->sk_lock.slock);
3697 spin_unlock_bh(&sk->sk_lock.slock);
3698 return true;
3699}
3700EXPORT_SYMBOL(__lock_sock_fast);
3701
3702int sock_gettstamp(struct socket *sock, void __user *userstamp,
3703 bool timeval, bool time32)
3704{
3705 struct sock *sk = sock->sk;
3706 struct timespec64 ts;
3707
3708 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3709 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3710 if (ts.tv_sec == -1)
3711 return -ENOENT;
3712 if (ts.tv_sec == 0) {
3713 ktime_t kt = ktime_get_real();
3714 sock_write_timestamp(sk, kt);
3715 ts = ktime_to_timespec64(kt);
3716 }
3717
3718 if (timeval)
3719 ts.tv_nsec /= 1000;
3720
3721#ifdef CONFIG_COMPAT_32BIT_TIME
3722 if (time32)
3723 return put_old_timespec32(&ts, userstamp);
3724#endif
3725#ifdef CONFIG_SPARC64
3726 /* beware of padding in sparc64 timeval */
3727 if (timeval && !in_compat_syscall()) {
3728 struct __kernel_old_timeval __user tv = {
3729 .tv_sec = ts.tv_sec,
3730 .tv_usec = ts.tv_nsec,
3731 };
3732 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3733 return -EFAULT;
3734 return 0;
3735 }
3736#endif
3737 return put_timespec64(&ts, userstamp);
3738}
3739EXPORT_SYMBOL(sock_gettstamp);
3740
3741void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3742{
3743 if (!sock_flag(sk, flag)) {
3744 unsigned long previous_flags = sk->sk_flags;
3745
3746 sock_set_flag(sk, flag);
3747 /*
3748 * we just set one of the two flags which require net
3749 * time stamping, but time stamping might have been on
3750 * already because of the other one
3751 */
3752 if (sock_needs_netstamp(sk) &&
3753 !(previous_flags & SK_FLAGS_TIMESTAMP))
3754 net_enable_timestamp();
3755 }
3756}
3757
3758int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3759 int level, int type)
3760{
3761 struct sock_exterr_skb *serr;
3762 struct sk_buff *skb;
3763 int copied, err;
3764
3765 err = -EAGAIN;
3766 skb = sock_dequeue_err_skb(sk);
3767 if (skb == NULL)
3768 goto out;
3769
3770 copied = skb->len;
3771 if (copied > len) {
3772 msg->msg_flags |= MSG_TRUNC;
3773 copied = len;
3774 }
3775 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3776 if (err)
3777 goto out_free_skb;
3778
3779 sock_recv_timestamp(msg, sk, skb);
3780
3781 serr = SKB_EXT_ERR(skb);
3782 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3783
3784 msg->msg_flags |= MSG_ERRQUEUE;
3785 err = copied;
3786
3787out_free_skb:
3788 kfree_skb(skb);
3789out:
3790 return err;
3791}
3792EXPORT_SYMBOL(sock_recv_errqueue);
3793
3794/*
3795 * Get a socket option on an socket.
3796 *
3797 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3798 * asynchronous errors should be reported by getsockopt. We assume
3799 * this means if you specify SO_ERROR (otherwise what is the point of it).
3800 */
3801int sock_common_getsockopt(struct socket *sock, int level, int optname,
3802 char __user *optval, int __user *optlen)
3803{
3804 struct sock *sk = sock->sk;
3805
3806 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3807 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3808}
3809EXPORT_SYMBOL(sock_common_getsockopt);
3810
3811int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3812 int flags)
3813{
3814 struct sock *sk = sock->sk;
3815 int addr_len = 0;
3816 int err;
3817
3818 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3819 if (err >= 0)
3820 msg->msg_namelen = addr_len;
3821 return err;
3822}
3823EXPORT_SYMBOL(sock_common_recvmsg);
3824
3825/*
3826 * Set socket options on an inet socket.
3827 */
3828int sock_common_setsockopt(struct socket *sock, int level, int optname,
3829 sockptr_t optval, unsigned int optlen)
3830{
3831 struct sock *sk = sock->sk;
3832
3833 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3834 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3835}
3836EXPORT_SYMBOL(sock_common_setsockopt);
3837
3838void sk_common_release(struct sock *sk)
3839{
3840 if (sk->sk_prot->destroy)
3841 sk->sk_prot->destroy(sk);
3842
3843 /*
3844 * Observation: when sk_common_release is called, processes have
3845 * no access to socket. But net still has.
3846 * Step one, detach it from networking:
3847 *
3848 * A. Remove from hash tables.
3849 */
3850
3851 sk->sk_prot->unhash(sk);
3852
3853 /*
3854 * In this point socket cannot receive new packets, but it is possible
3855 * that some packets are in flight because some CPU runs receiver and
3856 * did hash table lookup before we unhashed socket. They will achieve
3857 * receive queue and will be purged by socket destructor.
3858 *
3859 * Also we still have packets pending on receive queue and probably,
3860 * our own packets waiting in device queues. sock_destroy will drain
3861 * receive queue, but transmitted packets will delay socket destruction
3862 * until the last reference will be released.
3863 */
3864
3865 sock_orphan(sk);
3866
3867 xfrm_sk_free_policy(sk);
3868
3869 sock_put(sk);
3870}
3871EXPORT_SYMBOL(sk_common_release);
3872
3873void sk_get_meminfo(const struct sock *sk, u32 *mem)
3874{
3875 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3876
3877 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3878 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3879 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3880 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3881 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3882 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3883 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3884 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3885 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3886}
3887
3888#ifdef CONFIG_PROC_FS
3889static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3890
3891int sock_prot_inuse_get(struct net *net, struct proto *prot)
3892{
3893 int cpu, idx = prot->inuse_idx;
3894 int res = 0;
3895
3896 for_each_possible_cpu(cpu)
3897 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3898
3899 return res >= 0 ? res : 0;
3900}
3901EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3902
3903int sock_inuse_get(struct net *net)
3904{
3905 int cpu, res = 0;
3906
3907 for_each_possible_cpu(cpu)
3908 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3909
3910 return res;
3911}
3912
3913EXPORT_SYMBOL_GPL(sock_inuse_get);
3914
3915static int __net_init sock_inuse_init_net(struct net *net)
3916{
3917 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3918 if (net->core.prot_inuse == NULL)
3919 return -ENOMEM;
3920 return 0;
3921}
3922
3923static void __net_exit sock_inuse_exit_net(struct net *net)
3924{
3925 free_percpu(net->core.prot_inuse);
3926}
3927
3928static struct pernet_operations net_inuse_ops = {
3929 .init = sock_inuse_init_net,
3930 .exit = sock_inuse_exit_net,
3931};
3932
3933static __init int net_inuse_init(void)
3934{
3935 if (register_pernet_subsys(&net_inuse_ops))
3936 panic("Cannot initialize net inuse counters");
3937
3938 return 0;
3939}
3940
3941core_initcall(net_inuse_init);
3942
3943static int assign_proto_idx(struct proto *prot)
3944{
3945 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3946
3947 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3948 pr_err("PROTO_INUSE_NR exhausted\n");
3949 return -ENOSPC;
3950 }
3951
3952 set_bit(prot->inuse_idx, proto_inuse_idx);
3953 return 0;
3954}
3955
3956static void release_proto_idx(struct proto *prot)
3957{
3958 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3959 clear_bit(prot->inuse_idx, proto_inuse_idx);
3960}
3961#else
3962static inline int assign_proto_idx(struct proto *prot)
3963{
3964 return 0;
3965}
3966
3967static inline void release_proto_idx(struct proto *prot)
3968{
3969}
3970
3971#endif
3972
3973static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3974{
3975 if (!twsk_prot)
3976 return;
3977 kfree(twsk_prot->twsk_slab_name);
3978 twsk_prot->twsk_slab_name = NULL;
3979 kmem_cache_destroy(twsk_prot->twsk_slab);
3980 twsk_prot->twsk_slab = NULL;
3981}
3982
3983static int tw_prot_init(const struct proto *prot)
3984{
3985 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3986
3987 if (!twsk_prot)
3988 return 0;
3989
3990 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3991 prot->name);
3992 if (!twsk_prot->twsk_slab_name)
3993 return -ENOMEM;
3994
3995 twsk_prot->twsk_slab =
3996 kmem_cache_create(twsk_prot->twsk_slab_name,
3997 twsk_prot->twsk_obj_size, 0,
3998 SLAB_ACCOUNT | prot->slab_flags,
3999 NULL);
4000 if (!twsk_prot->twsk_slab) {
4001 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4002 prot->name);
4003 return -ENOMEM;
4004 }
4005
4006 return 0;
4007}
4008
4009static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4010{
4011 if (!rsk_prot)
4012 return;
4013 kfree(rsk_prot->slab_name);
4014 rsk_prot->slab_name = NULL;
4015 kmem_cache_destroy(rsk_prot->slab);
4016 rsk_prot->slab = NULL;
4017}
4018
4019static int req_prot_init(const struct proto *prot)
4020{
4021 struct request_sock_ops *rsk_prot = prot->rsk_prot;
4022
4023 if (!rsk_prot)
4024 return 0;
4025
4026 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4027 prot->name);
4028 if (!rsk_prot->slab_name)
4029 return -ENOMEM;
4030
4031 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4032 rsk_prot->obj_size, 0,
4033 SLAB_ACCOUNT | prot->slab_flags,
4034 NULL);
4035
4036 if (!rsk_prot->slab) {
4037 pr_crit("%s: Can't create request sock SLAB cache!\n",
4038 prot->name);
4039 return -ENOMEM;
4040 }
4041 return 0;
4042}
4043
4044int proto_register(struct proto *prot, int alloc_slab)
4045{
4046 int ret = -ENOBUFS;
4047
4048 if (prot->memory_allocated && !prot->sysctl_mem) {
4049 pr_err("%s: missing sysctl_mem\n", prot->name);
4050 return -EINVAL;
4051 }
4052 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4053 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4054 return -EINVAL;
4055 }
4056 if (alloc_slab) {
4057 prot->slab = kmem_cache_create_usercopy(prot->name,
4058 prot->obj_size, 0,
4059 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4060 prot->slab_flags,
4061 prot->useroffset, prot->usersize,
4062 NULL);
4063
4064 if (prot->slab == NULL) {
4065 pr_crit("%s: Can't create sock SLAB cache!\n",
4066 prot->name);
4067 goto out;
4068 }
4069
4070 if (req_prot_init(prot))
4071 goto out_free_request_sock_slab;
4072
4073 if (tw_prot_init(prot))
4074 goto out_free_timewait_sock_slab;
4075 }
4076
4077 mutex_lock(&proto_list_mutex);
4078 ret = assign_proto_idx(prot);
4079 if (ret) {
4080 mutex_unlock(&proto_list_mutex);
4081 goto out_free_timewait_sock_slab;
4082 }
4083 list_add(&prot->node, &proto_list);
4084 mutex_unlock(&proto_list_mutex);
4085 return ret;
4086
4087out_free_timewait_sock_slab:
4088 if (alloc_slab)
4089 tw_prot_cleanup(prot->twsk_prot);
4090out_free_request_sock_slab:
4091 if (alloc_slab) {
4092 req_prot_cleanup(prot->rsk_prot);
4093
4094 kmem_cache_destroy(prot->slab);
4095 prot->slab = NULL;
4096 }
4097out:
4098 return ret;
4099}
4100EXPORT_SYMBOL(proto_register);
4101
4102void proto_unregister(struct proto *prot)
4103{
4104 mutex_lock(&proto_list_mutex);
4105 release_proto_idx(prot);
4106 list_del(&prot->node);
4107 mutex_unlock(&proto_list_mutex);
4108
4109 kmem_cache_destroy(prot->slab);
4110 prot->slab = NULL;
4111
4112 req_prot_cleanup(prot->rsk_prot);
4113 tw_prot_cleanup(prot->twsk_prot);
4114}
4115EXPORT_SYMBOL(proto_unregister);
4116
4117int sock_load_diag_module(int family, int protocol)
4118{
4119 if (!protocol) {
4120 if (!sock_is_registered(family))
4121 return -ENOENT;
4122
4123 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4124 NETLINK_SOCK_DIAG, family);
4125 }
4126
4127#ifdef CONFIG_INET
4128 if (family == AF_INET &&
4129 protocol != IPPROTO_RAW &&
4130 protocol < MAX_INET_PROTOS &&
4131 !rcu_access_pointer(inet_protos[protocol]))
4132 return -ENOENT;
4133#endif
4134
4135 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4136 NETLINK_SOCK_DIAG, family, protocol);
4137}
4138EXPORT_SYMBOL(sock_load_diag_module);
4139
4140#ifdef CONFIG_PROC_FS
4141static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4142 __acquires(proto_list_mutex)
4143{
4144 mutex_lock(&proto_list_mutex);
4145 return seq_list_start_head(&proto_list, *pos);
4146}
4147
4148static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4149{
4150 return seq_list_next(v, &proto_list, pos);
4151}
4152
4153static void proto_seq_stop(struct seq_file *seq, void *v)
4154 __releases(proto_list_mutex)
4155{
4156 mutex_unlock(&proto_list_mutex);
4157}
4158
4159static char proto_method_implemented(const void *method)
4160{
4161 return method == NULL ? 'n' : 'y';
4162}
4163static long sock_prot_memory_allocated(struct proto *proto)
4164{
4165 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4166}
4167
4168static const char *sock_prot_memory_pressure(struct proto *proto)
4169{
4170 return proto->memory_pressure != NULL ?
4171 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4172}
4173
4174static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4175{
4176
4177 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4178 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4179 proto->name,
4180 proto->obj_size,
4181 sock_prot_inuse_get(seq_file_net(seq), proto),
4182 sock_prot_memory_allocated(proto),
4183 sock_prot_memory_pressure(proto),
4184 proto->max_header,
4185 proto->slab == NULL ? "no" : "yes",
4186 module_name(proto->owner),
4187 proto_method_implemented(proto->close),
4188 proto_method_implemented(proto->connect),
4189 proto_method_implemented(proto->disconnect),
4190 proto_method_implemented(proto->accept),
4191 proto_method_implemented(proto->ioctl),
4192 proto_method_implemented(proto->init),
4193 proto_method_implemented(proto->destroy),
4194 proto_method_implemented(proto->shutdown),
4195 proto_method_implemented(proto->setsockopt),
4196 proto_method_implemented(proto->getsockopt),
4197 proto_method_implemented(proto->sendmsg),
4198 proto_method_implemented(proto->recvmsg),
4199 proto_method_implemented(proto->bind),
4200 proto_method_implemented(proto->backlog_rcv),
4201 proto_method_implemented(proto->hash),
4202 proto_method_implemented(proto->unhash),
4203 proto_method_implemented(proto->get_port),
4204 proto_method_implemented(proto->enter_memory_pressure));
4205}
4206
4207static int proto_seq_show(struct seq_file *seq, void *v)
4208{
4209 if (v == &proto_list)
4210 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4211 "protocol",
4212 "size",
4213 "sockets",
4214 "memory",
4215 "press",
4216 "maxhdr",
4217 "slab",
4218 "module",
4219 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4220 else
4221 proto_seq_printf(seq, list_entry(v, struct proto, node));
4222 return 0;
4223}
4224
4225static const struct seq_operations proto_seq_ops = {
4226 .start = proto_seq_start,
4227 .next = proto_seq_next,
4228 .stop = proto_seq_stop,
4229 .show = proto_seq_show,
4230};
4231
4232static __net_init int proto_init_net(struct net *net)
4233{
4234 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4235 sizeof(struct seq_net_private)))
4236 return -ENOMEM;
4237
4238 return 0;
4239}
4240
4241static __net_exit void proto_exit_net(struct net *net)
4242{
4243 remove_proc_entry("protocols", net->proc_net);
4244}
4245
4246
4247static __net_initdata struct pernet_operations proto_net_ops = {
4248 .init = proto_init_net,
4249 .exit = proto_exit_net,
4250};
4251
4252static int __init proto_init(void)
4253{
4254 return register_pernet_subsys(&proto_net_ops);
4255}
4256
4257subsys_initcall(proto_init);
4258
4259#endif /* PROC_FS */
4260
4261#ifdef CONFIG_NET_RX_BUSY_POLL
4262bool sk_busy_loop_end(void *p, unsigned long start_time)
4263{
4264 struct sock *sk = p;
4265
4266 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4267 return true;
4268
4269 if (sk_is_udp(sk) &&
4270 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4271 return true;
4272
4273 return sk_busy_loop_timeout(sk, start_time);
4274}
4275EXPORT_SYMBOL(sk_busy_loop_end);
4276#endif /* CONFIG_NET_RX_BUSY_POLL */
4277
4278int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4279{
4280 if (!sk->sk_prot->bind_add)
4281 return -EOPNOTSUPP;
4282 return sk->sk_prot->bind_add(sk, addr, addr_len);
4283}
4284EXPORT_SYMBOL(sock_bind_add);
4285
4286/* Copy 'size' bytes from userspace and return `size` back to userspace */
4287int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4288 void __user *arg, void *karg, size_t size)
4289{
4290 int ret;
4291
4292 if (copy_from_user(karg, arg, size))
4293 return -EFAULT;
4294
4295 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4296 if (ret)
4297 return ret;
4298
4299 if (copy_to_user(arg, karg, size))
4300 return -EFAULT;
4301
4302 return 0;
4303}
4304EXPORT_SYMBOL(sock_ioctl_inout);
4305
4306/* This is the most common ioctl prep function, where the result (4 bytes) is
4307 * copied back to userspace if the ioctl() returns successfully. No input is
4308 * copied from userspace as input argument.
4309 */
4310static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4311{
4312 int ret, karg = 0;
4313
4314 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4315 if (ret)
4316 return ret;
4317
4318 return put_user(karg, (int __user *)arg);
4319}
4320
4321/* A wrapper around sock ioctls, which copies the data from userspace
4322 * (depending on the protocol/ioctl), and copies back the result to userspace.
4323 * The main motivation for this function is to pass kernel memory to the
4324 * protocol ioctl callbacks, instead of userspace memory.
4325 */
4326int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4327{
4328 int rc = 1;
4329
4330 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4331 rc = ipmr_sk_ioctl(sk, cmd, arg);
4332 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4333 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4334 else if (sk_is_phonet(sk))
4335 rc = phonet_sk_ioctl(sk, cmd, arg);
4336
4337 /* If ioctl was processed, returns its value */
4338 if (rc <= 0)
4339 return rc;
4340
4341 /* Otherwise call the default handler */
4342 return sock_ioctl_out(sk, cmd, arg);
4343}
4344EXPORT_SYMBOL(sk_ioctl);
4345
4346static int __init sock_struct_check(void)
4347{
4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4353
4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4363
4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4367
4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4372
4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4385 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4386 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4387 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4388 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4389
4390 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4391 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4392 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4393 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4394 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4395 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4396 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4397 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4398 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4399 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4400 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4401 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4402 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4403 return 0;
4404}
4405
4406core_initcall(sock_struct_check);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88#include <asm/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/init.h>
111#include <linux/highmem.h>
112#include <linux/user_namespace.h>
113#include <linux/static_key.h>
114#include <linux/memcontrol.h>
115#include <linux/prefetch.h>
116#include <linux/compat.h>
117
118#include <linux/uaccess.h>
119
120#include <linux/netdevice.h>
121#include <net/protocol.h>
122#include <linux/skbuff.h>
123#include <net/net_namespace.h>
124#include <net/request_sock.h>
125#include <net/sock.h>
126#include <linux/net_tstamp.h>
127#include <net/xfrm.h>
128#include <linux/ipsec.h>
129#include <net/cls_cgroup.h>
130#include <net/netprio_cgroup.h>
131#include <linux/sock_diag.h>
132
133#include <linux/filter.h>
134#include <net/sock_reuseport.h>
135#include <net/bpf_sk_storage.h>
136
137#include <trace/events/sock.h>
138
139#include <net/tcp.h>
140#include <net/busy_poll.h>
141
142#include <linux/ethtool.h>
143
144#include "dev.h"
145
146static DEFINE_MUTEX(proto_list_mutex);
147static LIST_HEAD(proto_list);
148
149static void sock_def_write_space_wfree(struct sock *sk);
150static void sock_def_write_space(struct sock *sk);
151
152/**
153 * sk_ns_capable - General socket capability test
154 * @sk: Socket to use a capability on or through
155 * @user_ns: The user namespace of the capability to use
156 * @cap: The capability to use
157 *
158 * Test to see if the opener of the socket had when the socket was
159 * created and the current process has the capability @cap in the user
160 * namespace @user_ns.
161 */
162bool sk_ns_capable(const struct sock *sk,
163 struct user_namespace *user_ns, int cap)
164{
165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 ns_capable(user_ns, cap);
167}
168EXPORT_SYMBOL(sk_ns_capable);
169
170/**
171 * sk_capable - Socket global capability test
172 * @sk: Socket to use a capability on or through
173 * @cap: The global capability to use
174 *
175 * Test to see if the opener of the socket had when the socket was
176 * created and the current process has the capability @cap in all user
177 * namespaces.
178 */
179bool sk_capable(const struct sock *sk, int cap)
180{
181 return sk_ns_capable(sk, &init_user_ns, cap);
182}
183EXPORT_SYMBOL(sk_capable);
184
185/**
186 * sk_net_capable - Network namespace socket capability test
187 * @sk: Socket to use a capability on or through
188 * @cap: The capability to use
189 *
190 * Test to see if the opener of the socket had when the socket was created
191 * and the current process has the capability @cap over the network namespace
192 * the socket is a member of.
193 */
194bool sk_net_capable(const struct sock *sk, int cap)
195{
196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
197}
198EXPORT_SYMBOL(sk_net_capable);
199
200/*
201 * Each address family might have different locking rules, so we have
202 * one slock key per address family and separate keys for internal and
203 * userspace sockets.
204 */
205static struct lock_class_key af_family_keys[AF_MAX];
206static struct lock_class_key af_family_kern_keys[AF_MAX];
207static struct lock_class_key af_family_slock_keys[AF_MAX];
208static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209
210/*
211 * Make lock validator output more readable. (we pre-construct these
212 * strings build-time, so that runtime initialization of socket
213 * locks is fast):
214 */
215
216#define _sock_locks(x) \
217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
226 x "27" , x "28" , x "AF_CAN" , \
227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
232 x "AF_MCTP" , \
233 x "AF_MAX"
234
235static const char *const af_family_key_strings[AF_MAX+1] = {
236 _sock_locks("sk_lock-")
237};
238static const char *const af_family_slock_key_strings[AF_MAX+1] = {
239 _sock_locks("slock-")
240};
241static const char *const af_family_clock_key_strings[AF_MAX+1] = {
242 _sock_locks("clock-")
243};
244
245static const char *const af_family_kern_key_strings[AF_MAX+1] = {
246 _sock_locks("k-sk_lock-")
247};
248static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-slock-")
250};
251static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
252 _sock_locks("k-clock-")
253};
254static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
255 _sock_locks("rlock-")
256};
257static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
258 _sock_locks("wlock-")
259};
260static const char *const af_family_elock_key_strings[AF_MAX+1] = {
261 _sock_locks("elock-")
262};
263
264/*
265 * sk_callback_lock and sk queues locking rules are per-address-family,
266 * so split the lock classes by using a per-AF key:
267 */
268static struct lock_class_key af_callback_keys[AF_MAX];
269static struct lock_class_key af_rlock_keys[AF_MAX];
270static struct lock_class_key af_wlock_keys[AF_MAX];
271static struct lock_class_key af_elock_keys[AF_MAX];
272static struct lock_class_key af_kern_callback_keys[AF_MAX];
273
274/* Run time adjustable parameters. */
275__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
276EXPORT_SYMBOL(sysctl_wmem_max);
277__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
278EXPORT_SYMBOL(sysctl_rmem_max);
279__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
280__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
281
282/* Maximal space eaten by iovec or ancillary data plus some space */
283int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
284EXPORT_SYMBOL(sysctl_optmem_max);
285
286int sysctl_tstamp_allow_data __read_mostly = 1;
287
288DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
289EXPORT_SYMBOL_GPL(memalloc_socks_key);
290
291/**
292 * sk_set_memalloc - sets %SOCK_MEMALLOC
293 * @sk: socket to set it on
294 *
295 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
296 * It's the responsibility of the admin to adjust min_free_kbytes
297 * to meet the requirements
298 */
299void sk_set_memalloc(struct sock *sk)
300{
301 sock_set_flag(sk, SOCK_MEMALLOC);
302 sk->sk_allocation |= __GFP_MEMALLOC;
303 static_branch_inc(&memalloc_socks_key);
304}
305EXPORT_SYMBOL_GPL(sk_set_memalloc);
306
307void sk_clear_memalloc(struct sock *sk)
308{
309 sock_reset_flag(sk, SOCK_MEMALLOC);
310 sk->sk_allocation &= ~__GFP_MEMALLOC;
311 static_branch_dec(&memalloc_socks_key);
312
313 /*
314 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
315 * progress of swapping. SOCK_MEMALLOC may be cleared while
316 * it has rmem allocations due to the last swapfile being deactivated
317 * but there is a risk that the socket is unusable due to exceeding
318 * the rmem limits. Reclaim the reserves and obey rmem limits again.
319 */
320 sk_mem_reclaim(sk);
321}
322EXPORT_SYMBOL_GPL(sk_clear_memalloc);
323
324int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
325{
326 int ret;
327 unsigned int noreclaim_flag;
328
329 /* these should have been dropped before queueing */
330 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
331
332 noreclaim_flag = memalloc_noreclaim_save();
333 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
334 tcp_v6_do_rcv,
335 tcp_v4_do_rcv,
336 sk, skb);
337 memalloc_noreclaim_restore(noreclaim_flag);
338
339 return ret;
340}
341EXPORT_SYMBOL(__sk_backlog_rcv);
342
343void sk_error_report(struct sock *sk)
344{
345 sk->sk_error_report(sk);
346
347 switch (sk->sk_family) {
348 case AF_INET:
349 fallthrough;
350 case AF_INET6:
351 trace_inet_sk_error_report(sk);
352 break;
353 default:
354 break;
355 }
356}
357EXPORT_SYMBOL(sk_error_report);
358
359int sock_get_timeout(long timeo, void *optval, bool old_timeval)
360{
361 struct __kernel_sock_timeval tv;
362
363 if (timeo == MAX_SCHEDULE_TIMEOUT) {
364 tv.tv_sec = 0;
365 tv.tv_usec = 0;
366 } else {
367 tv.tv_sec = timeo / HZ;
368 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
369 }
370
371 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
372 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
373 *(struct old_timeval32 *)optval = tv32;
374 return sizeof(tv32);
375 }
376
377 if (old_timeval) {
378 struct __kernel_old_timeval old_tv;
379 old_tv.tv_sec = tv.tv_sec;
380 old_tv.tv_usec = tv.tv_usec;
381 *(struct __kernel_old_timeval *)optval = old_tv;
382 return sizeof(old_tv);
383 }
384
385 *(struct __kernel_sock_timeval *)optval = tv;
386 return sizeof(tv);
387}
388EXPORT_SYMBOL(sock_get_timeout);
389
390int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
391 sockptr_t optval, int optlen, bool old_timeval)
392{
393 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
394 struct old_timeval32 tv32;
395
396 if (optlen < sizeof(tv32))
397 return -EINVAL;
398
399 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
400 return -EFAULT;
401 tv->tv_sec = tv32.tv_sec;
402 tv->tv_usec = tv32.tv_usec;
403 } else if (old_timeval) {
404 struct __kernel_old_timeval old_tv;
405
406 if (optlen < sizeof(old_tv))
407 return -EINVAL;
408 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
409 return -EFAULT;
410 tv->tv_sec = old_tv.tv_sec;
411 tv->tv_usec = old_tv.tv_usec;
412 } else {
413 if (optlen < sizeof(*tv))
414 return -EINVAL;
415 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
416 return -EFAULT;
417 }
418
419 return 0;
420}
421EXPORT_SYMBOL(sock_copy_user_timeval);
422
423static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
424 bool old_timeval)
425{
426 struct __kernel_sock_timeval tv;
427 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
428
429 if (err)
430 return err;
431
432 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
433 return -EDOM;
434
435 if (tv.tv_sec < 0) {
436 static int warned __read_mostly;
437
438 *timeo_p = 0;
439 if (warned < 10 && net_ratelimit()) {
440 warned++;
441 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
442 __func__, current->comm, task_pid_nr(current));
443 }
444 return 0;
445 }
446 *timeo_p = MAX_SCHEDULE_TIMEOUT;
447 if (tv.tv_sec == 0 && tv.tv_usec == 0)
448 return 0;
449 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
450 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
451 return 0;
452}
453
454static bool sock_needs_netstamp(const struct sock *sk)
455{
456 switch (sk->sk_family) {
457 case AF_UNSPEC:
458 case AF_UNIX:
459 return false;
460 default:
461 return true;
462 }
463}
464
465static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
466{
467 if (sk->sk_flags & flags) {
468 sk->sk_flags &= ~flags;
469 if (sock_needs_netstamp(sk) &&
470 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
471 net_disable_timestamp();
472 }
473}
474
475
476int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
477{
478 unsigned long flags;
479 struct sk_buff_head *list = &sk->sk_receive_queue;
480
481 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
482 atomic_inc(&sk->sk_drops);
483 trace_sock_rcvqueue_full(sk, skb);
484 return -ENOMEM;
485 }
486
487 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
488 atomic_inc(&sk->sk_drops);
489 return -ENOBUFS;
490 }
491
492 skb->dev = NULL;
493 skb_set_owner_r(skb, sk);
494
495 /* we escape from rcu protected region, make sure we dont leak
496 * a norefcounted dst
497 */
498 skb_dst_force(skb);
499
500 spin_lock_irqsave(&list->lock, flags);
501 sock_skb_set_dropcount(sk, skb);
502 __skb_queue_tail(list, skb);
503 spin_unlock_irqrestore(&list->lock, flags);
504
505 if (!sock_flag(sk, SOCK_DEAD))
506 sk->sk_data_ready(sk);
507 return 0;
508}
509EXPORT_SYMBOL(__sock_queue_rcv_skb);
510
511int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
512 enum skb_drop_reason *reason)
513{
514 enum skb_drop_reason drop_reason;
515 int err;
516
517 err = sk_filter(sk, skb);
518 if (err) {
519 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
520 goto out;
521 }
522 err = __sock_queue_rcv_skb(sk, skb);
523 switch (err) {
524 case -ENOMEM:
525 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
526 break;
527 case -ENOBUFS:
528 drop_reason = SKB_DROP_REASON_PROTO_MEM;
529 break;
530 default:
531 drop_reason = SKB_NOT_DROPPED_YET;
532 break;
533 }
534out:
535 if (reason)
536 *reason = drop_reason;
537 return err;
538}
539EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
540
541int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
542 const int nested, unsigned int trim_cap, bool refcounted)
543{
544 int rc = NET_RX_SUCCESS;
545
546 if (sk_filter_trim_cap(sk, skb, trim_cap))
547 goto discard_and_relse;
548
549 skb->dev = NULL;
550
551 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
552 atomic_inc(&sk->sk_drops);
553 goto discard_and_relse;
554 }
555 if (nested)
556 bh_lock_sock_nested(sk);
557 else
558 bh_lock_sock(sk);
559 if (!sock_owned_by_user(sk)) {
560 /*
561 * trylock + unlock semantics:
562 */
563 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
564
565 rc = sk_backlog_rcv(sk, skb);
566
567 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
568 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
569 bh_unlock_sock(sk);
570 atomic_inc(&sk->sk_drops);
571 goto discard_and_relse;
572 }
573
574 bh_unlock_sock(sk);
575out:
576 if (refcounted)
577 sock_put(sk);
578 return rc;
579discard_and_relse:
580 kfree_skb(skb);
581 goto out;
582}
583EXPORT_SYMBOL(__sk_receive_skb);
584
585INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
586 u32));
587INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
588 u32));
589struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
590{
591 struct dst_entry *dst = __sk_dst_get(sk);
592
593 if (dst && dst->obsolete &&
594 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
595 dst, cookie) == NULL) {
596 sk_tx_queue_clear(sk);
597 sk->sk_dst_pending_confirm = 0;
598 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
599 dst_release(dst);
600 return NULL;
601 }
602
603 return dst;
604}
605EXPORT_SYMBOL(__sk_dst_check);
606
607struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
608{
609 struct dst_entry *dst = sk_dst_get(sk);
610
611 if (dst && dst->obsolete &&
612 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
613 dst, cookie) == NULL) {
614 sk_dst_reset(sk);
615 dst_release(dst);
616 return NULL;
617 }
618
619 return dst;
620}
621EXPORT_SYMBOL(sk_dst_check);
622
623static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
624{
625 int ret = -ENOPROTOOPT;
626#ifdef CONFIG_NETDEVICES
627 struct net *net = sock_net(sk);
628
629 /* Sorry... */
630 ret = -EPERM;
631 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
632 goto out;
633
634 ret = -EINVAL;
635 if (ifindex < 0)
636 goto out;
637
638 /* Paired with all READ_ONCE() done locklessly. */
639 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
640
641 if (sk->sk_prot->rehash)
642 sk->sk_prot->rehash(sk);
643 sk_dst_reset(sk);
644
645 ret = 0;
646
647out:
648#endif
649
650 return ret;
651}
652
653int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
654{
655 int ret;
656
657 if (lock_sk)
658 lock_sock(sk);
659 ret = sock_bindtoindex_locked(sk, ifindex);
660 if (lock_sk)
661 release_sock(sk);
662
663 return ret;
664}
665EXPORT_SYMBOL(sock_bindtoindex);
666
667static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
668{
669 int ret = -ENOPROTOOPT;
670#ifdef CONFIG_NETDEVICES
671 struct net *net = sock_net(sk);
672 char devname[IFNAMSIZ];
673 int index;
674
675 ret = -EINVAL;
676 if (optlen < 0)
677 goto out;
678
679 /* Bind this socket to a particular device like "eth0",
680 * as specified in the passed interface name. If the
681 * name is "" or the option length is zero the socket
682 * is not bound.
683 */
684 if (optlen > IFNAMSIZ - 1)
685 optlen = IFNAMSIZ - 1;
686 memset(devname, 0, sizeof(devname));
687
688 ret = -EFAULT;
689 if (copy_from_sockptr(devname, optval, optlen))
690 goto out;
691
692 index = 0;
693 if (devname[0] != '\0') {
694 struct net_device *dev;
695
696 rcu_read_lock();
697 dev = dev_get_by_name_rcu(net, devname);
698 if (dev)
699 index = dev->ifindex;
700 rcu_read_unlock();
701 ret = -ENODEV;
702 if (!dev)
703 goto out;
704 }
705
706 sockopt_lock_sock(sk);
707 ret = sock_bindtoindex_locked(sk, index);
708 sockopt_release_sock(sk);
709out:
710#endif
711
712 return ret;
713}
714
715static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
716 sockptr_t optlen, int len)
717{
718 int ret = -ENOPROTOOPT;
719#ifdef CONFIG_NETDEVICES
720 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
721 struct net *net = sock_net(sk);
722 char devname[IFNAMSIZ];
723
724 if (bound_dev_if == 0) {
725 len = 0;
726 goto zero;
727 }
728
729 ret = -EINVAL;
730 if (len < IFNAMSIZ)
731 goto out;
732
733 ret = netdev_get_name(net, devname, bound_dev_if);
734 if (ret)
735 goto out;
736
737 len = strlen(devname) + 1;
738
739 ret = -EFAULT;
740 if (copy_to_sockptr(optval, devname, len))
741 goto out;
742
743zero:
744 ret = -EFAULT;
745 if (copy_to_sockptr(optlen, &len, sizeof(int)))
746 goto out;
747
748 ret = 0;
749
750out:
751#endif
752
753 return ret;
754}
755
756bool sk_mc_loop(struct sock *sk)
757{
758 if (dev_recursion_level())
759 return false;
760 if (!sk)
761 return true;
762 switch (sk->sk_family) {
763 case AF_INET:
764 return inet_sk(sk)->mc_loop;
765#if IS_ENABLED(CONFIG_IPV6)
766 case AF_INET6:
767 return inet6_sk(sk)->mc_loop;
768#endif
769 }
770 WARN_ON_ONCE(1);
771 return true;
772}
773EXPORT_SYMBOL(sk_mc_loop);
774
775void sock_set_reuseaddr(struct sock *sk)
776{
777 lock_sock(sk);
778 sk->sk_reuse = SK_CAN_REUSE;
779 release_sock(sk);
780}
781EXPORT_SYMBOL(sock_set_reuseaddr);
782
783void sock_set_reuseport(struct sock *sk)
784{
785 lock_sock(sk);
786 sk->sk_reuseport = true;
787 release_sock(sk);
788}
789EXPORT_SYMBOL(sock_set_reuseport);
790
791void sock_no_linger(struct sock *sk)
792{
793 lock_sock(sk);
794 sk->sk_lingertime = 0;
795 sock_set_flag(sk, SOCK_LINGER);
796 release_sock(sk);
797}
798EXPORT_SYMBOL(sock_no_linger);
799
800void sock_set_priority(struct sock *sk, u32 priority)
801{
802 lock_sock(sk);
803 sk->sk_priority = priority;
804 release_sock(sk);
805}
806EXPORT_SYMBOL(sock_set_priority);
807
808void sock_set_sndtimeo(struct sock *sk, s64 secs)
809{
810 lock_sock(sk);
811 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
812 sk->sk_sndtimeo = secs * HZ;
813 else
814 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
815 release_sock(sk);
816}
817EXPORT_SYMBOL(sock_set_sndtimeo);
818
819static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
820{
821 if (val) {
822 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
823 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
824 sock_set_flag(sk, SOCK_RCVTSTAMP);
825 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
826 } else {
827 sock_reset_flag(sk, SOCK_RCVTSTAMP);
828 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
829 }
830}
831
832void sock_enable_timestamps(struct sock *sk)
833{
834 lock_sock(sk);
835 __sock_set_timestamps(sk, true, false, true);
836 release_sock(sk);
837}
838EXPORT_SYMBOL(sock_enable_timestamps);
839
840void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
841{
842 switch (optname) {
843 case SO_TIMESTAMP_OLD:
844 __sock_set_timestamps(sk, valbool, false, false);
845 break;
846 case SO_TIMESTAMP_NEW:
847 __sock_set_timestamps(sk, valbool, true, false);
848 break;
849 case SO_TIMESTAMPNS_OLD:
850 __sock_set_timestamps(sk, valbool, false, true);
851 break;
852 case SO_TIMESTAMPNS_NEW:
853 __sock_set_timestamps(sk, valbool, true, true);
854 break;
855 }
856}
857
858static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
859{
860 struct net *net = sock_net(sk);
861 struct net_device *dev = NULL;
862 bool match = false;
863 int *vclock_index;
864 int i, num;
865
866 if (sk->sk_bound_dev_if)
867 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
868
869 if (!dev) {
870 pr_err("%s: sock not bind to device\n", __func__);
871 return -EOPNOTSUPP;
872 }
873
874 num = ethtool_get_phc_vclocks(dev, &vclock_index);
875 dev_put(dev);
876
877 for (i = 0; i < num; i++) {
878 if (*(vclock_index + i) == phc_index) {
879 match = true;
880 break;
881 }
882 }
883
884 if (num > 0)
885 kfree(vclock_index);
886
887 if (!match)
888 return -EINVAL;
889
890 sk->sk_bind_phc = phc_index;
891
892 return 0;
893}
894
895int sock_set_timestamping(struct sock *sk, int optname,
896 struct so_timestamping timestamping)
897{
898 int val = timestamping.flags;
899 int ret;
900
901 if (val & ~SOF_TIMESTAMPING_MASK)
902 return -EINVAL;
903
904 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
905 !(val & SOF_TIMESTAMPING_OPT_ID))
906 return -EINVAL;
907
908 if (val & SOF_TIMESTAMPING_OPT_ID &&
909 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
910 if (sk_is_tcp(sk)) {
911 if ((1 << sk->sk_state) &
912 (TCPF_CLOSE | TCPF_LISTEN))
913 return -EINVAL;
914 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
915 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
916 else
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
918 } else {
919 atomic_set(&sk->sk_tskey, 0);
920 }
921 }
922
923 if (val & SOF_TIMESTAMPING_OPT_STATS &&
924 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
925 return -EINVAL;
926
927 if (val & SOF_TIMESTAMPING_BIND_PHC) {
928 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
929 if (ret)
930 return ret;
931 }
932
933 sk->sk_tsflags = val;
934 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
935
936 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
937 sock_enable_timestamp(sk,
938 SOCK_TIMESTAMPING_RX_SOFTWARE);
939 else
940 sock_disable_timestamp(sk,
941 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
942 return 0;
943}
944
945void sock_set_keepalive(struct sock *sk)
946{
947 lock_sock(sk);
948 if (sk->sk_prot->keepalive)
949 sk->sk_prot->keepalive(sk, true);
950 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
951 release_sock(sk);
952}
953EXPORT_SYMBOL(sock_set_keepalive);
954
955static void __sock_set_rcvbuf(struct sock *sk, int val)
956{
957 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
958 * as a negative value.
959 */
960 val = min_t(int, val, INT_MAX / 2);
961 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
962
963 /* We double it on the way in to account for "struct sk_buff" etc.
964 * overhead. Applications assume that the SO_RCVBUF setting they make
965 * will allow that much actual data to be received on that socket.
966 *
967 * Applications are unaware that "struct sk_buff" and other overheads
968 * allocate from the receive buffer during socket buffer allocation.
969 *
970 * And after considering the possible alternatives, returning the value
971 * we actually used in getsockopt is the most desirable behavior.
972 */
973 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
974}
975
976void sock_set_rcvbuf(struct sock *sk, int val)
977{
978 lock_sock(sk);
979 __sock_set_rcvbuf(sk, val);
980 release_sock(sk);
981}
982EXPORT_SYMBOL(sock_set_rcvbuf);
983
984static void __sock_set_mark(struct sock *sk, u32 val)
985{
986 if (val != sk->sk_mark) {
987 sk->sk_mark = val;
988 sk_dst_reset(sk);
989 }
990}
991
992void sock_set_mark(struct sock *sk, u32 val)
993{
994 lock_sock(sk);
995 __sock_set_mark(sk, val);
996 release_sock(sk);
997}
998EXPORT_SYMBOL(sock_set_mark);
999
1000static void sock_release_reserved_memory(struct sock *sk, int bytes)
1001{
1002 /* Round down bytes to multiple of pages */
1003 bytes = round_down(bytes, PAGE_SIZE);
1004
1005 WARN_ON(bytes > sk->sk_reserved_mem);
1006 sk->sk_reserved_mem -= bytes;
1007 sk_mem_reclaim(sk);
1008}
1009
1010static int sock_reserve_memory(struct sock *sk, int bytes)
1011{
1012 long allocated;
1013 bool charged;
1014 int pages;
1015
1016 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1017 return -EOPNOTSUPP;
1018
1019 if (!bytes)
1020 return 0;
1021
1022 pages = sk_mem_pages(bytes);
1023
1024 /* pre-charge to memcg */
1025 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1026 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1027 if (!charged)
1028 return -ENOMEM;
1029
1030 /* pre-charge to forward_alloc */
1031 sk_memory_allocated_add(sk, pages);
1032 allocated = sk_memory_allocated(sk);
1033 /* If the system goes into memory pressure with this
1034 * precharge, give up and return error.
1035 */
1036 if (allocated > sk_prot_mem_limits(sk, 1)) {
1037 sk_memory_allocated_sub(sk, pages);
1038 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1039 return -ENOMEM;
1040 }
1041 sk->sk_forward_alloc += pages << PAGE_SHIFT;
1042
1043 sk->sk_reserved_mem += pages << PAGE_SHIFT;
1044
1045 return 0;
1046}
1047
1048void sockopt_lock_sock(struct sock *sk)
1049{
1050 /* When current->bpf_ctx is set, the setsockopt is called from
1051 * a bpf prog. bpf has ensured the sk lock has been
1052 * acquired before calling setsockopt().
1053 */
1054 if (has_current_bpf_ctx())
1055 return;
1056
1057 lock_sock(sk);
1058}
1059EXPORT_SYMBOL(sockopt_lock_sock);
1060
1061void sockopt_release_sock(struct sock *sk)
1062{
1063 if (has_current_bpf_ctx())
1064 return;
1065
1066 release_sock(sk);
1067}
1068EXPORT_SYMBOL(sockopt_release_sock);
1069
1070bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1071{
1072 return has_current_bpf_ctx() || ns_capable(ns, cap);
1073}
1074EXPORT_SYMBOL(sockopt_ns_capable);
1075
1076bool sockopt_capable(int cap)
1077{
1078 return has_current_bpf_ctx() || capable(cap);
1079}
1080EXPORT_SYMBOL(sockopt_capable);
1081
1082/*
1083 * This is meant for all protocols to use and covers goings on
1084 * at the socket level. Everything here is generic.
1085 */
1086
1087int sk_setsockopt(struct sock *sk, int level, int optname,
1088 sockptr_t optval, unsigned int optlen)
1089{
1090 struct so_timestamping timestamping;
1091 struct socket *sock = sk->sk_socket;
1092 struct sock_txtime sk_txtime;
1093 int val;
1094 int valbool;
1095 struct linger ling;
1096 int ret = 0;
1097
1098 /*
1099 * Options without arguments
1100 */
1101
1102 if (optname == SO_BINDTODEVICE)
1103 return sock_setbindtodevice(sk, optval, optlen);
1104
1105 if (optlen < sizeof(int))
1106 return -EINVAL;
1107
1108 if (copy_from_sockptr(&val, optval, sizeof(val)))
1109 return -EFAULT;
1110
1111 valbool = val ? 1 : 0;
1112
1113 sockopt_lock_sock(sk);
1114
1115 switch (optname) {
1116 case SO_DEBUG:
1117 if (val && !sockopt_capable(CAP_NET_ADMIN))
1118 ret = -EACCES;
1119 else
1120 sock_valbool_flag(sk, SOCK_DBG, valbool);
1121 break;
1122 case SO_REUSEADDR:
1123 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1124 break;
1125 case SO_REUSEPORT:
1126 sk->sk_reuseport = valbool;
1127 break;
1128 case SO_TYPE:
1129 case SO_PROTOCOL:
1130 case SO_DOMAIN:
1131 case SO_ERROR:
1132 ret = -ENOPROTOOPT;
1133 break;
1134 case SO_DONTROUTE:
1135 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1136 sk_dst_reset(sk);
1137 break;
1138 case SO_BROADCAST:
1139 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1140 break;
1141 case SO_SNDBUF:
1142 /* Don't error on this BSD doesn't and if you think
1143 * about it this is right. Otherwise apps have to
1144 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1145 * are treated in BSD as hints
1146 */
1147 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1148set_sndbuf:
1149 /* Ensure val * 2 fits into an int, to prevent max_t()
1150 * from treating it as a negative value.
1151 */
1152 val = min_t(int, val, INT_MAX / 2);
1153 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1154 WRITE_ONCE(sk->sk_sndbuf,
1155 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1156 /* Wake up sending tasks if we upped the value. */
1157 sk->sk_write_space(sk);
1158 break;
1159
1160 case SO_SNDBUFFORCE:
1161 if (!sockopt_capable(CAP_NET_ADMIN)) {
1162 ret = -EPERM;
1163 break;
1164 }
1165
1166 /* No negative values (to prevent underflow, as val will be
1167 * multiplied by 2).
1168 */
1169 if (val < 0)
1170 val = 0;
1171 goto set_sndbuf;
1172
1173 case SO_RCVBUF:
1174 /* Don't error on this BSD doesn't and if you think
1175 * about it this is right. Otherwise apps have to
1176 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1177 * are treated in BSD as hints
1178 */
1179 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1180 break;
1181
1182 case SO_RCVBUFFORCE:
1183 if (!sockopt_capable(CAP_NET_ADMIN)) {
1184 ret = -EPERM;
1185 break;
1186 }
1187
1188 /* No negative values (to prevent underflow, as val will be
1189 * multiplied by 2).
1190 */
1191 __sock_set_rcvbuf(sk, max(val, 0));
1192 break;
1193
1194 case SO_KEEPALIVE:
1195 if (sk->sk_prot->keepalive)
1196 sk->sk_prot->keepalive(sk, valbool);
1197 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1198 break;
1199
1200 case SO_OOBINLINE:
1201 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1202 break;
1203
1204 case SO_NO_CHECK:
1205 sk->sk_no_check_tx = valbool;
1206 break;
1207
1208 case SO_PRIORITY:
1209 if ((val >= 0 && val <= 6) ||
1210 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1211 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1212 sk->sk_priority = val;
1213 else
1214 ret = -EPERM;
1215 break;
1216
1217 case SO_LINGER:
1218 if (optlen < sizeof(ling)) {
1219 ret = -EINVAL; /* 1003.1g */
1220 break;
1221 }
1222 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1223 ret = -EFAULT;
1224 break;
1225 }
1226 if (!ling.l_onoff)
1227 sock_reset_flag(sk, SOCK_LINGER);
1228 else {
1229#if (BITS_PER_LONG == 32)
1230 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1231 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1232 else
1233#endif
1234 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1235 sock_set_flag(sk, SOCK_LINGER);
1236 }
1237 break;
1238
1239 case SO_BSDCOMPAT:
1240 break;
1241
1242 case SO_PASSCRED:
1243 if (valbool)
1244 set_bit(SOCK_PASSCRED, &sock->flags);
1245 else
1246 clear_bit(SOCK_PASSCRED, &sock->flags);
1247 break;
1248
1249 case SO_TIMESTAMP_OLD:
1250 case SO_TIMESTAMP_NEW:
1251 case SO_TIMESTAMPNS_OLD:
1252 case SO_TIMESTAMPNS_NEW:
1253 sock_set_timestamp(sk, optname, valbool);
1254 break;
1255
1256 case SO_TIMESTAMPING_NEW:
1257 case SO_TIMESTAMPING_OLD:
1258 if (optlen == sizeof(timestamping)) {
1259 if (copy_from_sockptr(×tamping, optval,
1260 sizeof(timestamping))) {
1261 ret = -EFAULT;
1262 break;
1263 }
1264 } else {
1265 memset(×tamping, 0, sizeof(timestamping));
1266 timestamping.flags = val;
1267 }
1268 ret = sock_set_timestamping(sk, optname, timestamping);
1269 break;
1270
1271 case SO_RCVLOWAT:
1272 if (val < 0)
1273 val = INT_MAX;
1274 if (sock && sock->ops->set_rcvlowat)
1275 ret = sock->ops->set_rcvlowat(sk, val);
1276 else
1277 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1278 break;
1279
1280 case SO_RCVTIMEO_OLD:
1281 case SO_RCVTIMEO_NEW:
1282 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1283 optlen, optname == SO_RCVTIMEO_OLD);
1284 break;
1285
1286 case SO_SNDTIMEO_OLD:
1287 case SO_SNDTIMEO_NEW:
1288 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1289 optlen, optname == SO_SNDTIMEO_OLD);
1290 break;
1291
1292 case SO_ATTACH_FILTER: {
1293 struct sock_fprog fprog;
1294
1295 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1296 if (!ret)
1297 ret = sk_attach_filter(&fprog, sk);
1298 break;
1299 }
1300 case SO_ATTACH_BPF:
1301 ret = -EINVAL;
1302 if (optlen == sizeof(u32)) {
1303 u32 ufd;
1304
1305 ret = -EFAULT;
1306 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1307 break;
1308
1309 ret = sk_attach_bpf(ufd, sk);
1310 }
1311 break;
1312
1313 case SO_ATTACH_REUSEPORT_CBPF: {
1314 struct sock_fprog fprog;
1315
1316 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1317 if (!ret)
1318 ret = sk_reuseport_attach_filter(&fprog, sk);
1319 break;
1320 }
1321 case SO_ATTACH_REUSEPORT_EBPF:
1322 ret = -EINVAL;
1323 if (optlen == sizeof(u32)) {
1324 u32 ufd;
1325
1326 ret = -EFAULT;
1327 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1328 break;
1329
1330 ret = sk_reuseport_attach_bpf(ufd, sk);
1331 }
1332 break;
1333
1334 case SO_DETACH_REUSEPORT_BPF:
1335 ret = reuseport_detach_prog(sk);
1336 break;
1337
1338 case SO_DETACH_FILTER:
1339 ret = sk_detach_filter(sk);
1340 break;
1341
1342 case SO_LOCK_FILTER:
1343 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1344 ret = -EPERM;
1345 else
1346 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1347 break;
1348
1349 case SO_PASSSEC:
1350 if (valbool)
1351 set_bit(SOCK_PASSSEC, &sock->flags);
1352 else
1353 clear_bit(SOCK_PASSSEC, &sock->flags);
1354 break;
1355 case SO_MARK:
1356 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1357 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1358 ret = -EPERM;
1359 break;
1360 }
1361
1362 __sock_set_mark(sk, val);
1363 break;
1364 case SO_RCVMARK:
1365 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1366 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1367 ret = -EPERM;
1368 break;
1369 }
1370
1371 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1372 break;
1373
1374 case SO_RXQ_OVFL:
1375 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1376 break;
1377
1378 case SO_WIFI_STATUS:
1379 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1380 break;
1381
1382 case SO_PEEK_OFF:
1383 if (sock->ops->set_peek_off)
1384 ret = sock->ops->set_peek_off(sk, val);
1385 else
1386 ret = -EOPNOTSUPP;
1387 break;
1388
1389 case SO_NOFCS:
1390 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1391 break;
1392
1393 case SO_SELECT_ERR_QUEUE:
1394 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1395 break;
1396
1397#ifdef CONFIG_NET_RX_BUSY_POLL
1398 case SO_BUSY_POLL:
1399 /* allow unprivileged users to decrease the value */
1400 if ((val > sk->sk_ll_usec) && !sockopt_capable(CAP_NET_ADMIN))
1401 ret = -EPERM;
1402 else {
1403 if (val < 0)
1404 ret = -EINVAL;
1405 else
1406 WRITE_ONCE(sk->sk_ll_usec, val);
1407 }
1408 break;
1409 case SO_PREFER_BUSY_POLL:
1410 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1411 ret = -EPERM;
1412 else
1413 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1414 break;
1415 case SO_BUSY_POLL_BUDGET:
1416 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1417 ret = -EPERM;
1418 } else {
1419 if (val < 0 || val > U16_MAX)
1420 ret = -EINVAL;
1421 else
1422 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1423 }
1424 break;
1425#endif
1426
1427 case SO_MAX_PACING_RATE:
1428 {
1429 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1430
1431 if (sizeof(ulval) != sizeof(val) &&
1432 optlen >= sizeof(ulval) &&
1433 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1434 ret = -EFAULT;
1435 break;
1436 }
1437 if (ulval != ~0UL)
1438 cmpxchg(&sk->sk_pacing_status,
1439 SK_PACING_NONE,
1440 SK_PACING_NEEDED);
1441 sk->sk_max_pacing_rate = ulval;
1442 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1443 break;
1444 }
1445 case SO_INCOMING_CPU:
1446 reuseport_update_incoming_cpu(sk, val);
1447 break;
1448
1449 case SO_CNX_ADVICE:
1450 if (val == 1)
1451 dst_negative_advice(sk);
1452 break;
1453
1454 case SO_ZEROCOPY:
1455 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1456 if (!(sk_is_tcp(sk) ||
1457 (sk->sk_type == SOCK_DGRAM &&
1458 sk->sk_protocol == IPPROTO_UDP)))
1459 ret = -EOPNOTSUPP;
1460 } else if (sk->sk_family != PF_RDS) {
1461 ret = -EOPNOTSUPP;
1462 }
1463 if (!ret) {
1464 if (val < 0 || val > 1)
1465 ret = -EINVAL;
1466 else
1467 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1468 }
1469 break;
1470
1471 case SO_TXTIME:
1472 if (optlen != sizeof(struct sock_txtime)) {
1473 ret = -EINVAL;
1474 break;
1475 } else if (copy_from_sockptr(&sk_txtime, optval,
1476 sizeof(struct sock_txtime))) {
1477 ret = -EFAULT;
1478 break;
1479 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1480 ret = -EINVAL;
1481 break;
1482 }
1483 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1484 * scheduler has enough safe guards.
1485 */
1486 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1487 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1488 ret = -EPERM;
1489 break;
1490 }
1491 sock_valbool_flag(sk, SOCK_TXTIME, true);
1492 sk->sk_clockid = sk_txtime.clockid;
1493 sk->sk_txtime_deadline_mode =
1494 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1495 sk->sk_txtime_report_errors =
1496 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1497 break;
1498
1499 case SO_BINDTOIFINDEX:
1500 ret = sock_bindtoindex_locked(sk, val);
1501 break;
1502
1503 case SO_BUF_LOCK:
1504 if (val & ~SOCK_BUF_LOCK_MASK) {
1505 ret = -EINVAL;
1506 break;
1507 }
1508 sk->sk_userlocks = val | (sk->sk_userlocks &
1509 ~SOCK_BUF_LOCK_MASK);
1510 break;
1511
1512 case SO_RESERVE_MEM:
1513 {
1514 int delta;
1515
1516 if (val < 0) {
1517 ret = -EINVAL;
1518 break;
1519 }
1520
1521 delta = val - sk->sk_reserved_mem;
1522 if (delta < 0)
1523 sock_release_reserved_memory(sk, -delta);
1524 else
1525 ret = sock_reserve_memory(sk, delta);
1526 break;
1527 }
1528
1529 case SO_TXREHASH:
1530 if (val < -1 || val > 1) {
1531 ret = -EINVAL;
1532 break;
1533 }
1534 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1535 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1536 /* Paired with READ_ONCE() in tcp_rtx_synack() */
1537 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1538 break;
1539
1540 default:
1541 ret = -ENOPROTOOPT;
1542 break;
1543 }
1544 sockopt_release_sock(sk);
1545 return ret;
1546}
1547
1548int sock_setsockopt(struct socket *sock, int level, int optname,
1549 sockptr_t optval, unsigned int optlen)
1550{
1551 return sk_setsockopt(sock->sk, level, optname,
1552 optval, optlen);
1553}
1554EXPORT_SYMBOL(sock_setsockopt);
1555
1556static const struct cred *sk_get_peer_cred(struct sock *sk)
1557{
1558 const struct cred *cred;
1559
1560 spin_lock(&sk->sk_peer_lock);
1561 cred = get_cred(sk->sk_peer_cred);
1562 spin_unlock(&sk->sk_peer_lock);
1563
1564 return cred;
1565}
1566
1567static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1568 struct ucred *ucred)
1569{
1570 ucred->pid = pid_vnr(pid);
1571 ucred->uid = ucred->gid = -1;
1572 if (cred) {
1573 struct user_namespace *current_ns = current_user_ns();
1574
1575 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1576 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1577 }
1578}
1579
1580static int groups_to_user(sockptr_t dst, const struct group_info *src)
1581{
1582 struct user_namespace *user_ns = current_user_ns();
1583 int i;
1584
1585 for (i = 0; i < src->ngroups; i++) {
1586 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1587
1588 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1589 return -EFAULT;
1590 }
1591
1592 return 0;
1593}
1594
1595int sk_getsockopt(struct sock *sk, int level, int optname,
1596 sockptr_t optval, sockptr_t optlen)
1597{
1598 struct socket *sock = sk->sk_socket;
1599
1600 union {
1601 int val;
1602 u64 val64;
1603 unsigned long ulval;
1604 struct linger ling;
1605 struct old_timeval32 tm32;
1606 struct __kernel_old_timeval tm;
1607 struct __kernel_sock_timeval stm;
1608 struct sock_txtime txtime;
1609 struct so_timestamping timestamping;
1610 } v;
1611
1612 int lv = sizeof(int);
1613 int len;
1614
1615 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1616 return -EFAULT;
1617 if (len < 0)
1618 return -EINVAL;
1619
1620 memset(&v, 0, sizeof(v));
1621
1622 switch (optname) {
1623 case SO_DEBUG:
1624 v.val = sock_flag(sk, SOCK_DBG);
1625 break;
1626
1627 case SO_DONTROUTE:
1628 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1629 break;
1630
1631 case SO_BROADCAST:
1632 v.val = sock_flag(sk, SOCK_BROADCAST);
1633 break;
1634
1635 case SO_SNDBUF:
1636 v.val = sk->sk_sndbuf;
1637 break;
1638
1639 case SO_RCVBUF:
1640 v.val = sk->sk_rcvbuf;
1641 break;
1642
1643 case SO_REUSEADDR:
1644 v.val = sk->sk_reuse;
1645 break;
1646
1647 case SO_REUSEPORT:
1648 v.val = sk->sk_reuseport;
1649 break;
1650
1651 case SO_KEEPALIVE:
1652 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1653 break;
1654
1655 case SO_TYPE:
1656 v.val = sk->sk_type;
1657 break;
1658
1659 case SO_PROTOCOL:
1660 v.val = sk->sk_protocol;
1661 break;
1662
1663 case SO_DOMAIN:
1664 v.val = sk->sk_family;
1665 break;
1666
1667 case SO_ERROR:
1668 v.val = -sock_error(sk);
1669 if (v.val == 0)
1670 v.val = xchg(&sk->sk_err_soft, 0);
1671 break;
1672
1673 case SO_OOBINLINE:
1674 v.val = sock_flag(sk, SOCK_URGINLINE);
1675 break;
1676
1677 case SO_NO_CHECK:
1678 v.val = sk->sk_no_check_tx;
1679 break;
1680
1681 case SO_PRIORITY:
1682 v.val = sk->sk_priority;
1683 break;
1684
1685 case SO_LINGER:
1686 lv = sizeof(v.ling);
1687 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1688 v.ling.l_linger = sk->sk_lingertime / HZ;
1689 break;
1690
1691 case SO_BSDCOMPAT:
1692 break;
1693
1694 case SO_TIMESTAMP_OLD:
1695 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1696 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1697 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1698 break;
1699
1700 case SO_TIMESTAMPNS_OLD:
1701 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1702 break;
1703
1704 case SO_TIMESTAMP_NEW:
1705 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1706 break;
1707
1708 case SO_TIMESTAMPNS_NEW:
1709 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1710 break;
1711
1712 case SO_TIMESTAMPING_OLD:
1713 lv = sizeof(v.timestamping);
1714 v.timestamping.flags = sk->sk_tsflags;
1715 v.timestamping.bind_phc = sk->sk_bind_phc;
1716 break;
1717
1718 case SO_RCVTIMEO_OLD:
1719 case SO_RCVTIMEO_NEW:
1720 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1721 break;
1722
1723 case SO_SNDTIMEO_OLD:
1724 case SO_SNDTIMEO_NEW:
1725 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1726 break;
1727
1728 case SO_RCVLOWAT:
1729 v.val = sk->sk_rcvlowat;
1730 break;
1731
1732 case SO_SNDLOWAT:
1733 v.val = 1;
1734 break;
1735
1736 case SO_PASSCRED:
1737 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1738 break;
1739
1740 case SO_PEERCRED:
1741 {
1742 struct ucred peercred;
1743 if (len > sizeof(peercred))
1744 len = sizeof(peercred);
1745
1746 spin_lock(&sk->sk_peer_lock);
1747 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1748 spin_unlock(&sk->sk_peer_lock);
1749
1750 if (copy_to_sockptr(optval, &peercred, len))
1751 return -EFAULT;
1752 goto lenout;
1753 }
1754
1755 case SO_PEERGROUPS:
1756 {
1757 const struct cred *cred;
1758 int ret, n;
1759
1760 cred = sk_get_peer_cred(sk);
1761 if (!cred)
1762 return -ENODATA;
1763
1764 n = cred->group_info->ngroups;
1765 if (len < n * sizeof(gid_t)) {
1766 len = n * sizeof(gid_t);
1767 put_cred(cred);
1768 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1769 }
1770 len = n * sizeof(gid_t);
1771
1772 ret = groups_to_user(optval, cred->group_info);
1773 put_cred(cred);
1774 if (ret)
1775 return ret;
1776 goto lenout;
1777 }
1778
1779 case SO_PEERNAME:
1780 {
1781 char address[128];
1782
1783 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1784 if (lv < 0)
1785 return -ENOTCONN;
1786 if (lv < len)
1787 return -EINVAL;
1788 if (copy_to_sockptr(optval, address, len))
1789 return -EFAULT;
1790 goto lenout;
1791 }
1792
1793 /* Dubious BSD thing... Probably nobody even uses it, but
1794 * the UNIX standard wants it for whatever reason... -DaveM
1795 */
1796 case SO_ACCEPTCONN:
1797 v.val = sk->sk_state == TCP_LISTEN;
1798 break;
1799
1800 case SO_PASSSEC:
1801 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1802 break;
1803
1804 case SO_PEERSEC:
1805 return security_socket_getpeersec_stream(sock,
1806 optval, optlen, len);
1807
1808 case SO_MARK:
1809 v.val = sk->sk_mark;
1810 break;
1811
1812 case SO_RCVMARK:
1813 v.val = sock_flag(sk, SOCK_RCVMARK);
1814 break;
1815
1816 case SO_RXQ_OVFL:
1817 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1818 break;
1819
1820 case SO_WIFI_STATUS:
1821 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1822 break;
1823
1824 case SO_PEEK_OFF:
1825 if (!sock->ops->set_peek_off)
1826 return -EOPNOTSUPP;
1827
1828 v.val = sk->sk_peek_off;
1829 break;
1830 case SO_NOFCS:
1831 v.val = sock_flag(sk, SOCK_NOFCS);
1832 break;
1833
1834 case SO_BINDTODEVICE:
1835 return sock_getbindtodevice(sk, optval, optlen, len);
1836
1837 case SO_GET_FILTER:
1838 len = sk_get_filter(sk, optval, len);
1839 if (len < 0)
1840 return len;
1841
1842 goto lenout;
1843
1844 case SO_LOCK_FILTER:
1845 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1846 break;
1847
1848 case SO_BPF_EXTENSIONS:
1849 v.val = bpf_tell_extensions();
1850 break;
1851
1852 case SO_SELECT_ERR_QUEUE:
1853 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1854 break;
1855
1856#ifdef CONFIG_NET_RX_BUSY_POLL
1857 case SO_BUSY_POLL:
1858 v.val = sk->sk_ll_usec;
1859 break;
1860 case SO_PREFER_BUSY_POLL:
1861 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1862 break;
1863#endif
1864
1865 case SO_MAX_PACING_RATE:
1866 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1867 lv = sizeof(v.ulval);
1868 v.ulval = sk->sk_max_pacing_rate;
1869 } else {
1870 /* 32bit version */
1871 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1872 }
1873 break;
1874
1875 case SO_INCOMING_CPU:
1876 v.val = READ_ONCE(sk->sk_incoming_cpu);
1877 break;
1878
1879 case SO_MEMINFO:
1880 {
1881 u32 meminfo[SK_MEMINFO_VARS];
1882
1883 sk_get_meminfo(sk, meminfo);
1884
1885 len = min_t(unsigned int, len, sizeof(meminfo));
1886 if (copy_to_sockptr(optval, &meminfo, len))
1887 return -EFAULT;
1888
1889 goto lenout;
1890 }
1891
1892#ifdef CONFIG_NET_RX_BUSY_POLL
1893 case SO_INCOMING_NAPI_ID:
1894 v.val = READ_ONCE(sk->sk_napi_id);
1895
1896 /* aggregate non-NAPI IDs down to 0 */
1897 if (v.val < MIN_NAPI_ID)
1898 v.val = 0;
1899
1900 break;
1901#endif
1902
1903 case SO_COOKIE:
1904 lv = sizeof(u64);
1905 if (len < lv)
1906 return -EINVAL;
1907 v.val64 = sock_gen_cookie(sk);
1908 break;
1909
1910 case SO_ZEROCOPY:
1911 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1912 break;
1913
1914 case SO_TXTIME:
1915 lv = sizeof(v.txtime);
1916 v.txtime.clockid = sk->sk_clockid;
1917 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1918 SOF_TXTIME_DEADLINE_MODE : 0;
1919 v.txtime.flags |= sk->sk_txtime_report_errors ?
1920 SOF_TXTIME_REPORT_ERRORS : 0;
1921 break;
1922
1923 case SO_BINDTOIFINDEX:
1924 v.val = READ_ONCE(sk->sk_bound_dev_if);
1925 break;
1926
1927 case SO_NETNS_COOKIE:
1928 lv = sizeof(u64);
1929 if (len != lv)
1930 return -EINVAL;
1931 v.val64 = sock_net(sk)->net_cookie;
1932 break;
1933
1934 case SO_BUF_LOCK:
1935 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1936 break;
1937
1938 case SO_RESERVE_MEM:
1939 v.val = sk->sk_reserved_mem;
1940 break;
1941
1942 case SO_TXREHASH:
1943 v.val = sk->sk_txrehash;
1944 break;
1945
1946 default:
1947 /* We implement the SO_SNDLOWAT etc to not be settable
1948 * (1003.1g 7).
1949 */
1950 return -ENOPROTOOPT;
1951 }
1952
1953 if (len > lv)
1954 len = lv;
1955 if (copy_to_sockptr(optval, &v, len))
1956 return -EFAULT;
1957lenout:
1958 if (copy_to_sockptr(optlen, &len, sizeof(int)))
1959 return -EFAULT;
1960 return 0;
1961}
1962
1963int sock_getsockopt(struct socket *sock, int level, int optname,
1964 char __user *optval, int __user *optlen)
1965{
1966 return sk_getsockopt(sock->sk, level, optname,
1967 USER_SOCKPTR(optval),
1968 USER_SOCKPTR(optlen));
1969}
1970
1971/*
1972 * Initialize an sk_lock.
1973 *
1974 * (We also register the sk_lock with the lock validator.)
1975 */
1976static inline void sock_lock_init(struct sock *sk)
1977{
1978 if (sk->sk_kern_sock)
1979 sock_lock_init_class_and_name(
1980 sk,
1981 af_family_kern_slock_key_strings[sk->sk_family],
1982 af_family_kern_slock_keys + sk->sk_family,
1983 af_family_kern_key_strings[sk->sk_family],
1984 af_family_kern_keys + sk->sk_family);
1985 else
1986 sock_lock_init_class_and_name(
1987 sk,
1988 af_family_slock_key_strings[sk->sk_family],
1989 af_family_slock_keys + sk->sk_family,
1990 af_family_key_strings[sk->sk_family],
1991 af_family_keys + sk->sk_family);
1992}
1993
1994/*
1995 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1996 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1997 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1998 */
1999static void sock_copy(struct sock *nsk, const struct sock *osk)
2000{
2001 const struct proto *prot = READ_ONCE(osk->sk_prot);
2002#ifdef CONFIG_SECURITY_NETWORK
2003 void *sptr = nsk->sk_security;
2004#endif
2005
2006 /* If we move sk_tx_queue_mapping out of the private section,
2007 * we must check if sk_tx_queue_clear() is called after
2008 * sock_copy() in sk_clone_lock().
2009 */
2010 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2011 offsetof(struct sock, sk_dontcopy_begin) ||
2012 offsetof(struct sock, sk_tx_queue_mapping) >=
2013 offsetof(struct sock, sk_dontcopy_end));
2014
2015 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2016
2017 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2018 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2019
2020#ifdef CONFIG_SECURITY_NETWORK
2021 nsk->sk_security = sptr;
2022 security_sk_clone(osk, nsk);
2023#endif
2024}
2025
2026static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2027 int family)
2028{
2029 struct sock *sk;
2030 struct kmem_cache *slab;
2031
2032 slab = prot->slab;
2033 if (slab != NULL) {
2034 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2035 if (!sk)
2036 return sk;
2037 if (want_init_on_alloc(priority))
2038 sk_prot_clear_nulls(sk, prot->obj_size);
2039 } else
2040 sk = kmalloc(prot->obj_size, priority);
2041
2042 if (sk != NULL) {
2043 if (security_sk_alloc(sk, family, priority))
2044 goto out_free;
2045
2046 if (!try_module_get(prot->owner))
2047 goto out_free_sec;
2048 }
2049
2050 return sk;
2051
2052out_free_sec:
2053 security_sk_free(sk);
2054out_free:
2055 if (slab != NULL)
2056 kmem_cache_free(slab, sk);
2057 else
2058 kfree(sk);
2059 return NULL;
2060}
2061
2062static void sk_prot_free(struct proto *prot, struct sock *sk)
2063{
2064 struct kmem_cache *slab;
2065 struct module *owner;
2066
2067 owner = prot->owner;
2068 slab = prot->slab;
2069
2070 cgroup_sk_free(&sk->sk_cgrp_data);
2071 mem_cgroup_sk_free(sk);
2072 security_sk_free(sk);
2073 if (slab != NULL)
2074 kmem_cache_free(slab, sk);
2075 else
2076 kfree(sk);
2077 module_put(owner);
2078}
2079
2080/**
2081 * sk_alloc - All socket objects are allocated here
2082 * @net: the applicable net namespace
2083 * @family: protocol family
2084 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2085 * @prot: struct proto associated with this new sock instance
2086 * @kern: is this to be a kernel socket?
2087 */
2088struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2089 struct proto *prot, int kern)
2090{
2091 struct sock *sk;
2092
2093 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2094 if (sk) {
2095 sk->sk_family = family;
2096 /*
2097 * See comment in struct sock definition to understand
2098 * why we need sk_prot_creator -acme
2099 */
2100 sk->sk_prot = sk->sk_prot_creator = prot;
2101 sk->sk_kern_sock = kern;
2102 sock_lock_init(sk);
2103 sk->sk_net_refcnt = kern ? 0 : 1;
2104 if (likely(sk->sk_net_refcnt)) {
2105 get_net_track(net, &sk->ns_tracker, priority);
2106 sock_inuse_add(net, 1);
2107 } else {
2108 __netns_tracker_alloc(net, &sk->ns_tracker,
2109 false, priority);
2110 }
2111
2112 sock_net_set(sk, net);
2113 refcount_set(&sk->sk_wmem_alloc, 1);
2114
2115 mem_cgroup_sk_alloc(sk);
2116 cgroup_sk_alloc(&sk->sk_cgrp_data);
2117 sock_update_classid(&sk->sk_cgrp_data);
2118 sock_update_netprioidx(&sk->sk_cgrp_data);
2119 sk_tx_queue_clear(sk);
2120 }
2121
2122 return sk;
2123}
2124EXPORT_SYMBOL(sk_alloc);
2125
2126/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2127 * grace period. This is the case for UDP sockets and TCP listeners.
2128 */
2129static void __sk_destruct(struct rcu_head *head)
2130{
2131 struct sock *sk = container_of(head, struct sock, sk_rcu);
2132 struct sk_filter *filter;
2133
2134 if (sk->sk_destruct)
2135 sk->sk_destruct(sk);
2136
2137 filter = rcu_dereference_check(sk->sk_filter,
2138 refcount_read(&sk->sk_wmem_alloc) == 0);
2139 if (filter) {
2140 sk_filter_uncharge(sk, filter);
2141 RCU_INIT_POINTER(sk->sk_filter, NULL);
2142 }
2143
2144 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2145
2146#ifdef CONFIG_BPF_SYSCALL
2147 bpf_sk_storage_free(sk);
2148#endif
2149
2150 if (atomic_read(&sk->sk_omem_alloc))
2151 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2152 __func__, atomic_read(&sk->sk_omem_alloc));
2153
2154 if (sk->sk_frag.page) {
2155 put_page(sk->sk_frag.page);
2156 sk->sk_frag.page = NULL;
2157 }
2158
2159 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2160 put_cred(sk->sk_peer_cred);
2161 put_pid(sk->sk_peer_pid);
2162
2163 if (likely(sk->sk_net_refcnt))
2164 put_net_track(sock_net(sk), &sk->ns_tracker);
2165 else
2166 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2167
2168 sk_prot_free(sk->sk_prot_creator, sk);
2169}
2170
2171void sk_destruct(struct sock *sk)
2172{
2173 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2174
2175 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2176 reuseport_detach_sock(sk);
2177 use_call_rcu = true;
2178 }
2179
2180 if (use_call_rcu)
2181 call_rcu(&sk->sk_rcu, __sk_destruct);
2182 else
2183 __sk_destruct(&sk->sk_rcu);
2184}
2185
2186static void __sk_free(struct sock *sk)
2187{
2188 if (likely(sk->sk_net_refcnt))
2189 sock_inuse_add(sock_net(sk), -1);
2190
2191 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2192 sock_diag_broadcast_destroy(sk);
2193 else
2194 sk_destruct(sk);
2195}
2196
2197void sk_free(struct sock *sk)
2198{
2199 /*
2200 * We subtract one from sk_wmem_alloc and can know if
2201 * some packets are still in some tx queue.
2202 * If not null, sock_wfree() will call __sk_free(sk) later
2203 */
2204 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2205 __sk_free(sk);
2206}
2207EXPORT_SYMBOL(sk_free);
2208
2209static void sk_init_common(struct sock *sk)
2210{
2211 skb_queue_head_init(&sk->sk_receive_queue);
2212 skb_queue_head_init(&sk->sk_write_queue);
2213 skb_queue_head_init(&sk->sk_error_queue);
2214
2215 rwlock_init(&sk->sk_callback_lock);
2216 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2217 af_rlock_keys + sk->sk_family,
2218 af_family_rlock_key_strings[sk->sk_family]);
2219 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2220 af_wlock_keys + sk->sk_family,
2221 af_family_wlock_key_strings[sk->sk_family]);
2222 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2223 af_elock_keys + sk->sk_family,
2224 af_family_elock_key_strings[sk->sk_family]);
2225 lockdep_set_class_and_name(&sk->sk_callback_lock,
2226 af_callback_keys + sk->sk_family,
2227 af_family_clock_key_strings[sk->sk_family]);
2228}
2229
2230/**
2231 * sk_clone_lock - clone a socket, and lock its clone
2232 * @sk: the socket to clone
2233 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2234 *
2235 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2236 */
2237struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2238{
2239 struct proto *prot = READ_ONCE(sk->sk_prot);
2240 struct sk_filter *filter;
2241 bool is_charged = true;
2242 struct sock *newsk;
2243
2244 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2245 if (!newsk)
2246 goto out;
2247
2248 sock_copy(newsk, sk);
2249
2250 newsk->sk_prot_creator = prot;
2251
2252 /* SANITY */
2253 if (likely(newsk->sk_net_refcnt)) {
2254 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2255 sock_inuse_add(sock_net(newsk), 1);
2256 } else {
2257 /* Kernel sockets are not elevating the struct net refcount.
2258 * Instead, use a tracker to more easily detect if a layer
2259 * is not properly dismantling its kernel sockets at netns
2260 * destroy time.
2261 */
2262 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2263 false, priority);
2264 }
2265 sk_node_init(&newsk->sk_node);
2266 sock_lock_init(newsk);
2267 bh_lock_sock(newsk);
2268 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2269 newsk->sk_backlog.len = 0;
2270
2271 atomic_set(&newsk->sk_rmem_alloc, 0);
2272
2273 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2274 refcount_set(&newsk->sk_wmem_alloc, 1);
2275
2276 atomic_set(&newsk->sk_omem_alloc, 0);
2277 sk_init_common(newsk);
2278
2279 newsk->sk_dst_cache = NULL;
2280 newsk->sk_dst_pending_confirm = 0;
2281 newsk->sk_wmem_queued = 0;
2282 newsk->sk_forward_alloc = 0;
2283 newsk->sk_reserved_mem = 0;
2284 atomic_set(&newsk->sk_drops, 0);
2285 newsk->sk_send_head = NULL;
2286 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2287 atomic_set(&newsk->sk_zckey, 0);
2288
2289 sock_reset_flag(newsk, SOCK_DONE);
2290
2291 /* sk->sk_memcg will be populated at accept() time */
2292 newsk->sk_memcg = NULL;
2293
2294 cgroup_sk_clone(&newsk->sk_cgrp_data);
2295
2296 rcu_read_lock();
2297 filter = rcu_dereference(sk->sk_filter);
2298 if (filter != NULL)
2299 /* though it's an empty new sock, the charging may fail
2300 * if sysctl_optmem_max was changed between creation of
2301 * original socket and cloning
2302 */
2303 is_charged = sk_filter_charge(newsk, filter);
2304 RCU_INIT_POINTER(newsk->sk_filter, filter);
2305 rcu_read_unlock();
2306
2307 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2308 /* We need to make sure that we don't uncharge the new
2309 * socket if we couldn't charge it in the first place
2310 * as otherwise we uncharge the parent's filter.
2311 */
2312 if (!is_charged)
2313 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2314 sk_free_unlock_clone(newsk);
2315 newsk = NULL;
2316 goto out;
2317 }
2318 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2319
2320 if (bpf_sk_storage_clone(sk, newsk)) {
2321 sk_free_unlock_clone(newsk);
2322 newsk = NULL;
2323 goto out;
2324 }
2325
2326 /* Clear sk_user_data if parent had the pointer tagged
2327 * as not suitable for copying when cloning.
2328 */
2329 if (sk_user_data_is_nocopy(newsk))
2330 newsk->sk_user_data = NULL;
2331
2332 newsk->sk_err = 0;
2333 newsk->sk_err_soft = 0;
2334 newsk->sk_priority = 0;
2335 newsk->sk_incoming_cpu = raw_smp_processor_id();
2336
2337 /* Before updating sk_refcnt, we must commit prior changes to memory
2338 * (Documentation/RCU/rculist_nulls.rst for details)
2339 */
2340 smp_wmb();
2341 refcount_set(&newsk->sk_refcnt, 2);
2342
2343 /* Increment the counter in the same struct proto as the master
2344 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2345 * is the same as sk->sk_prot->socks, as this field was copied
2346 * with memcpy).
2347 *
2348 * This _changes_ the previous behaviour, where
2349 * tcp_create_openreq_child always was incrementing the
2350 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2351 * to be taken into account in all callers. -acme
2352 */
2353 sk_refcnt_debug_inc(newsk);
2354 sk_set_socket(newsk, NULL);
2355 sk_tx_queue_clear(newsk);
2356 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2357
2358 if (newsk->sk_prot->sockets_allocated)
2359 sk_sockets_allocated_inc(newsk);
2360
2361 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2362 net_enable_timestamp();
2363out:
2364 return newsk;
2365}
2366EXPORT_SYMBOL_GPL(sk_clone_lock);
2367
2368void sk_free_unlock_clone(struct sock *sk)
2369{
2370 /* It is still raw copy of parent, so invalidate
2371 * destructor and make plain sk_free() */
2372 sk->sk_destruct = NULL;
2373 bh_unlock_sock(sk);
2374 sk_free(sk);
2375}
2376EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2377
2378static void sk_trim_gso_size(struct sock *sk)
2379{
2380 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE)
2381 return;
2382#if IS_ENABLED(CONFIG_IPV6)
2383 if (sk->sk_family == AF_INET6 &&
2384 sk_is_tcp(sk) &&
2385 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
2386 return;
2387#endif
2388 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE;
2389}
2390
2391void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2392{
2393 u32 max_segs = 1;
2394
2395 sk_dst_set(sk, dst);
2396 sk->sk_route_caps = dst->dev->features;
2397 if (sk_is_tcp(sk))
2398 sk->sk_route_caps |= NETIF_F_GSO;
2399 if (sk->sk_route_caps & NETIF_F_GSO)
2400 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2401 if (unlikely(sk->sk_gso_disabled))
2402 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2403 if (sk_can_gso(sk)) {
2404 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2405 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2406 } else {
2407 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2408 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2409 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2410 sk_trim_gso_size(sk);
2411 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2412 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2413 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2414 }
2415 }
2416 sk->sk_gso_max_segs = max_segs;
2417}
2418EXPORT_SYMBOL_GPL(sk_setup_caps);
2419
2420/*
2421 * Simple resource managers for sockets.
2422 */
2423
2424
2425/*
2426 * Write buffer destructor automatically called from kfree_skb.
2427 */
2428void sock_wfree(struct sk_buff *skb)
2429{
2430 struct sock *sk = skb->sk;
2431 unsigned int len = skb->truesize;
2432 bool free;
2433
2434 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2435 if (sock_flag(sk, SOCK_RCU_FREE) &&
2436 sk->sk_write_space == sock_def_write_space) {
2437 rcu_read_lock();
2438 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2439 sock_def_write_space_wfree(sk);
2440 rcu_read_unlock();
2441 if (unlikely(free))
2442 __sk_free(sk);
2443 return;
2444 }
2445
2446 /*
2447 * Keep a reference on sk_wmem_alloc, this will be released
2448 * after sk_write_space() call
2449 */
2450 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2451 sk->sk_write_space(sk);
2452 len = 1;
2453 }
2454 /*
2455 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2456 * could not do because of in-flight packets
2457 */
2458 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2459 __sk_free(sk);
2460}
2461EXPORT_SYMBOL(sock_wfree);
2462
2463/* This variant of sock_wfree() is used by TCP,
2464 * since it sets SOCK_USE_WRITE_QUEUE.
2465 */
2466void __sock_wfree(struct sk_buff *skb)
2467{
2468 struct sock *sk = skb->sk;
2469
2470 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2471 __sk_free(sk);
2472}
2473
2474void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2475{
2476 skb_orphan(skb);
2477 skb->sk = sk;
2478#ifdef CONFIG_INET
2479 if (unlikely(!sk_fullsock(sk))) {
2480 skb->destructor = sock_edemux;
2481 sock_hold(sk);
2482 return;
2483 }
2484#endif
2485 skb->destructor = sock_wfree;
2486 skb_set_hash_from_sk(skb, sk);
2487 /*
2488 * We used to take a refcount on sk, but following operation
2489 * is enough to guarantee sk_free() wont free this sock until
2490 * all in-flight packets are completed
2491 */
2492 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2493}
2494EXPORT_SYMBOL(skb_set_owner_w);
2495
2496static bool can_skb_orphan_partial(const struct sk_buff *skb)
2497{
2498#ifdef CONFIG_TLS_DEVICE
2499 /* Drivers depend on in-order delivery for crypto offload,
2500 * partial orphan breaks out-of-order-OK logic.
2501 */
2502 if (skb->decrypted)
2503 return false;
2504#endif
2505 return (skb->destructor == sock_wfree ||
2506 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2507}
2508
2509/* This helper is used by netem, as it can hold packets in its
2510 * delay queue. We want to allow the owner socket to send more
2511 * packets, as if they were already TX completed by a typical driver.
2512 * But we also want to keep skb->sk set because some packet schedulers
2513 * rely on it (sch_fq for example).
2514 */
2515void skb_orphan_partial(struct sk_buff *skb)
2516{
2517 if (skb_is_tcp_pure_ack(skb))
2518 return;
2519
2520 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2521 return;
2522
2523 skb_orphan(skb);
2524}
2525EXPORT_SYMBOL(skb_orphan_partial);
2526
2527/*
2528 * Read buffer destructor automatically called from kfree_skb.
2529 */
2530void sock_rfree(struct sk_buff *skb)
2531{
2532 struct sock *sk = skb->sk;
2533 unsigned int len = skb->truesize;
2534
2535 atomic_sub(len, &sk->sk_rmem_alloc);
2536 sk_mem_uncharge(sk, len);
2537}
2538EXPORT_SYMBOL(sock_rfree);
2539
2540/*
2541 * Buffer destructor for skbs that are not used directly in read or write
2542 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2543 */
2544void sock_efree(struct sk_buff *skb)
2545{
2546 sock_put(skb->sk);
2547}
2548EXPORT_SYMBOL(sock_efree);
2549
2550/* Buffer destructor for prefetch/receive path where reference count may
2551 * not be held, e.g. for listen sockets.
2552 */
2553#ifdef CONFIG_INET
2554void sock_pfree(struct sk_buff *skb)
2555{
2556 if (sk_is_refcounted(skb->sk))
2557 sock_gen_put(skb->sk);
2558}
2559EXPORT_SYMBOL(sock_pfree);
2560#endif /* CONFIG_INET */
2561
2562kuid_t sock_i_uid(struct sock *sk)
2563{
2564 kuid_t uid;
2565
2566 read_lock_bh(&sk->sk_callback_lock);
2567 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2568 read_unlock_bh(&sk->sk_callback_lock);
2569 return uid;
2570}
2571EXPORT_SYMBOL(sock_i_uid);
2572
2573unsigned long sock_i_ino(struct sock *sk)
2574{
2575 unsigned long ino;
2576
2577 read_lock_bh(&sk->sk_callback_lock);
2578 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2579 read_unlock_bh(&sk->sk_callback_lock);
2580 return ino;
2581}
2582EXPORT_SYMBOL(sock_i_ino);
2583
2584/*
2585 * Allocate a skb from the socket's send buffer.
2586 */
2587struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2588 gfp_t priority)
2589{
2590 if (force ||
2591 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2592 struct sk_buff *skb = alloc_skb(size, priority);
2593
2594 if (skb) {
2595 skb_set_owner_w(skb, sk);
2596 return skb;
2597 }
2598 }
2599 return NULL;
2600}
2601EXPORT_SYMBOL(sock_wmalloc);
2602
2603static void sock_ofree(struct sk_buff *skb)
2604{
2605 struct sock *sk = skb->sk;
2606
2607 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2608}
2609
2610struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2611 gfp_t priority)
2612{
2613 struct sk_buff *skb;
2614
2615 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2616 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2617 READ_ONCE(sysctl_optmem_max))
2618 return NULL;
2619
2620 skb = alloc_skb(size, priority);
2621 if (!skb)
2622 return NULL;
2623
2624 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2625 skb->sk = sk;
2626 skb->destructor = sock_ofree;
2627 return skb;
2628}
2629
2630/*
2631 * Allocate a memory block from the socket's option memory buffer.
2632 */
2633void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2634{
2635 int optmem_max = READ_ONCE(sysctl_optmem_max);
2636
2637 if ((unsigned int)size <= optmem_max &&
2638 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2639 void *mem;
2640 /* First do the add, to avoid the race if kmalloc
2641 * might sleep.
2642 */
2643 atomic_add(size, &sk->sk_omem_alloc);
2644 mem = kmalloc(size, priority);
2645 if (mem)
2646 return mem;
2647 atomic_sub(size, &sk->sk_omem_alloc);
2648 }
2649 return NULL;
2650}
2651EXPORT_SYMBOL(sock_kmalloc);
2652
2653/* Free an option memory block. Note, we actually want the inline
2654 * here as this allows gcc to detect the nullify and fold away the
2655 * condition entirely.
2656 */
2657static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2658 const bool nullify)
2659{
2660 if (WARN_ON_ONCE(!mem))
2661 return;
2662 if (nullify)
2663 kfree_sensitive(mem);
2664 else
2665 kfree(mem);
2666 atomic_sub(size, &sk->sk_omem_alloc);
2667}
2668
2669void sock_kfree_s(struct sock *sk, void *mem, int size)
2670{
2671 __sock_kfree_s(sk, mem, size, false);
2672}
2673EXPORT_SYMBOL(sock_kfree_s);
2674
2675void sock_kzfree_s(struct sock *sk, void *mem, int size)
2676{
2677 __sock_kfree_s(sk, mem, size, true);
2678}
2679EXPORT_SYMBOL(sock_kzfree_s);
2680
2681/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2682 I think, these locks should be removed for datagram sockets.
2683 */
2684static long sock_wait_for_wmem(struct sock *sk, long timeo)
2685{
2686 DEFINE_WAIT(wait);
2687
2688 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2689 for (;;) {
2690 if (!timeo)
2691 break;
2692 if (signal_pending(current))
2693 break;
2694 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2695 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2696 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2697 break;
2698 if (sk->sk_shutdown & SEND_SHUTDOWN)
2699 break;
2700 if (sk->sk_err)
2701 break;
2702 timeo = schedule_timeout(timeo);
2703 }
2704 finish_wait(sk_sleep(sk), &wait);
2705 return timeo;
2706}
2707
2708
2709/*
2710 * Generic send/receive buffer handlers
2711 */
2712
2713struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2714 unsigned long data_len, int noblock,
2715 int *errcode, int max_page_order)
2716{
2717 struct sk_buff *skb;
2718 long timeo;
2719 int err;
2720
2721 timeo = sock_sndtimeo(sk, noblock);
2722 for (;;) {
2723 err = sock_error(sk);
2724 if (err != 0)
2725 goto failure;
2726
2727 err = -EPIPE;
2728 if (sk->sk_shutdown & SEND_SHUTDOWN)
2729 goto failure;
2730
2731 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2732 break;
2733
2734 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2735 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2736 err = -EAGAIN;
2737 if (!timeo)
2738 goto failure;
2739 if (signal_pending(current))
2740 goto interrupted;
2741 timeo = sock_wait_for_wmem(sk, timeo);
2742 }
2743 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2744 errcode, sk->sk_allocation);
2745 if (skb)
2746 skb_set_owner_w(skb, sk);
2747 return skb;
2748
2749interrupted:
2750 err = sock_intr_errno(timeo);
2751failure:
2752 *errcode = err;
2753 return NULL;
2754}
2755EXPORT_SYMBOL(sock_alloc_send_pskb);
2756
2757int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2758 struct sockcm_cookie *sockc)
2759{
2760 u32 tsflags;
2761
2762 switch (cmsg->cmsg_type) {
2763 case SO_MARK:
2764 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2765 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2766 return -EPERM;
2767 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2768 return -EINVAL;
2769 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2770 break;
2771 case SO_TIMESTAMPING_OLD:
2772 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2773 return -EINVAL;
2774
2775 tsflags = *(u32 *)CMSG_DATA(cmsg);
2776 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2777 return -EINVAL;
2778
2779 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2780 sockc->tsflags |= tsflags;
2781 break;
2782 case SCM_TXTIME:
2783 if (!sock_flag(sk, SOCK_TXTIME))
2784 return -EINVAL;
2785 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2786 return -EINVAL;
2787 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2788 break;
2789 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2790 case SCM_RIGHTS:
2791 case SCM_CREDENTIALS:
2792 break;
2793 default:
2794 return -EINVAL;
2795 }
2796 return 0;
2797}
2798EXPORT_SYMBOL(__sock_cmsg_send);
2799
2800int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2801 struct sockcm_cookie *sockc)
2802{
2803 struct cmsghdr *cmsg;
2804 int ret;
2805
2806 for_each_cmsghdr(cmsg, msg) {
2807 if (!CMSG_OK(msg, cmsg))
2808 return -EINVAL;
2809 if (cmsg->cmsg_level != SOL_SOCKET)
2810 continue;
2811 ret = __sock_cmsg_send(sk, cmsg, sockc);
2812 if (ret)
2813 return ret;
2814 }
2815 return 0;
2816}
2817EXPORT_SYMBOL(sock_cmsg_send);
2818
2819static void sk_enter_memory_pressure(struct sock *sk)
2820{
2821 if (!sk->sk_prot->enter_memory_pressure)
2822 return;
2823
2824 sk->sk_prot->enter_memory_pressure(sk);
2825}
2826
2827static void sk_leave_memory_pressure(struct sock *sk)
2828{
2829 if (sk->sk_prot->leave_memory_pressure) {
2830 sk->sk_prot->leave_memory_pressure(sk);
2831 } else {
2832 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2833
2834 if (memory_pressure && READ_ONCE(*memory_pressure))
2835 WRITE_ONCE(*memory_pressure, 0);
2836 }
2837}
2838
2839DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2840
2841/**
2842 * skb_page_frag_refill - check that a page_frag contains enough room
2843 * @sz: minimum size of the fragment we want to get
2844 * @pfrag: pointer to page_frag
2845 * @gfp: priority for memory allocation
2846 *
2847 * Note: While this allocator tries to use high order pages, there is
2848 * no guarantee that allocations succeed. Therefore, @sz MUST be
2849 * less or equal than PAGE_SIZE.
2850 */
2851bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2852{
2853 if (pfrag->page) {
2854 if (page_ref_count(pfrag->page) == 1) {
2855 pfrag->offset = 0;
2856 return true;
2857 }
2858 if (pfrag->offset + sz <= pfrag->size)
2859 return true;
2860 put_page(pfrag->page);
2861 }
2862
2863 pfrag->offset = 0;
2864 if (SKB_FRAG_PAGE_ORDER &&
2865 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2866 /* Avoid direct reclaim but allow kswapd to wake */
2867 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2868 __GFP_COMP | __GFP_NOWARN |
2869 __GFP_NORETRY,
2870 SKB_FRAG_PAGE_ORDER);
2871 if (likely(pfrag->page)) {
2872 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2873 return true;
2874 }
2875 }
2876 pfrag->page = alloc_page(gfp);
2877 if (likely(pfrag->page)) {
2878 pfrag->size = PAGE_SIZE;
2879 return true;
2880 }
2881 return false;
2882}
2883EXPORT_SYMBOL(skb_page_frag_refill);
2884
2885bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2886{
2887 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2888 return true;
2889
2890 sk_enter_memory_pressure(sk);
2891 sk_stream_moderate_sndbuf(sk);
2892 return false;
2893}
2894EXPORT_SYMBOL(sk_page_frag_refill);
2895
2896void __lock_sock(struct sock *sk)
2897 __releases(&sk->sk_lock.slock)
2898 __acquires(&sk->sk_lock.slock)
2899{
2900 DEFINE_WAIT(wait);
2901
2902 for (;;) {
2903 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2904 TASK_UNINTERRUPTIBLE);
2905 spin_unlock_bh(&sk->sk_lock.slock);
2906 schedule();
2907 spin_lock_bh(&sk->sk_lock.slock);
2908 if (!sock_owned_by_user(sk))
2909 break;
2910 }
2911 finish_wait(&sk->sk_lock.wq, &wait);
2912}
2913
2914void __release_sock(struct sock *sk)
2915 __releases(&sk->sk_lock.slock)
2916 __acquires(&sk->sk_lock.slock)
2917{
2918 struct sk_buff *skb, *next;
2919
2920 while ((skb = sk->sk_backlog.head) != NULL) {
2921 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2922
2923 spin_unlock_bh(&sk->sk_lock.slock);
2924
2925 do {
2926 next = skb->next;
2927 prefetch(next);
2928 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2929 skb_mark_not_on_list(skb);
2930 sk_backlog_rcv(sk, skb);
2931
2932 cond_resched();
2933
2934 skb = next;
2935 } while (skb != NULL);
2936
2937 spin_lock_bh(&sk->sk_lock.slock);
2938 }
2939
2940 /*
2941 * Doing the zeroing here guarantee we can not loop forever
2942 * while a wild producer attempts to flood us.
2943 */
2944 sk->sk_backlog.len = 0;
2945}
2946
2947void __sk_flush_backlog(struct sock *sk)
2948{
2949 spin_lock_bh(&sk->sk_lock.slock);
2950 __release_sock(sk);
2951 spin_unlock_bh(&sk->sk_lock.slock);
2952}
2953EXPORT_SYMBOL_GPL(__sk_flush_backlog);
2954
2955/**
2956 * sk_wait_data - wait for data to arrive at sk_receive_queue
2957 * @sk: sock to wait on
2958 * @timeo: for how long
2959 * @skb: last skb seen on sk_receive_queue
2960 *
2961 * Now socket state including sk->sk_err is changed only under lock,
2962 * hence we may omit checks after joining wait queue.
2963 * We check receive queue before schedule() only as optimization;
2964 * it is very likely that release_sock() added new data.
2965 */
2966int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2967{
2968 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2969 int rc;
2970
2971 add_wait_queue(sk_sleep(sk), &wait);
2972 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2973 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2974 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2975 remove_wait_queue(sk_sleep(sk), &wait);
2976 return rc;
2977}
2978EXPORT_SYMBOL(sk_wait_data);
2979
2980/**
2981 * __sk_mem_raise_allocated - increase memory_allocated
2982 * @sk: socket
2983 * @size: memory size to allocate
2984 * @amt: pages to allocate
2985 * @kind: allocation type
2986 *
2987 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2988 */
2989int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2990{
2991 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2992 struct proto *prot = sk->sk_prot;
2993 bool charged = true;
2994 long allocated;
2995
2996 sk_memory_allocated_add(sk, amt);
2997 allocated = sk_memory_allocated(sk);
2998 if (memcg_charge &&
2999 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3000 gfp_memcg_charge())))
3001 goto suppress_allocation;
3002
3003 /* Under limit. */
3004 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3005 sk_leave_memory_pressure(sk);
3006 return 1;
3007 }
3008
3009 /* Under pressure. */
3010 if (allocated > sk_prot_mem_limits(sk, 1))
3011 sk_enter_memory_pressure(sk);
3012
3013 /* Over hard limit. */
3014 if (allocated > sk_prot_mem_limits(sk, 2))
3015 goto suppress_allocation;
3016
3017 /* guarantee minimum buffer size under pressure */
3018 if (kind == SK_MEM_RECV) {
3019 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3020 return 1;
3021
3022 } else { /* SK_MEM_SEND */
3023 int wmem0 = sk_get_wmem0(sk, prot);
3024
3025 if (sk->sk_type == SOCK_STREAM) {
3026 if (sk->sk_wmem_queued < wmem0)
3027 return 1;
3028 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3029 return 1;
3030 }
3031 }
3032
3033 if (sk_has_memory_pressure(sk)) {
3034 u64 alloc;
3035
3036 if (!sk_under_memory_pressure(sk))
3037 return 1;
3038 alloc = sk_sockets_allocated_read_positive(sk);
3039 if (sk_prot_mem_limits(sk, 2) > alloc *
3040 sk_mem_pages(sk->sk_wmem_queued +
3041 atomic_read(&sk->sk_rmem_alloc) +
3042 sk->sk_forward_alloc))
3043 return 1;
3044 }
3045
3046suppress_allocation:
3047
3048 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3049 sk_stream_moderate_sndbuf(sk);
3050
3051 /* Fail only if socket is _under_ its sndbuf.
3052 * In this case we cannot block, so that we have to fail.
3053 */
3054 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3055 /* Force charge with __GFP_NOFAIL */
3056 if (memcg_charge && !charged) {
3057 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3058 gfp_memcg_charge() | __GFP_NOFAIL);
3059 }
3060 return 1;
3061 }
3062 }
3063
3064 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3065 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3066
3067 sk_memory_allocated_sub(sk, amt);
3068
3069 if (memcg_charge && charged)
3070 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3071
3072 return 0;
3073}
3074
3075/**
3076 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3077 * @sk: socket
3078 * @size: memory size to allocate
3079 * @kind: allocation type
3080 *
3081 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3082 * rmem allocation. This function assumes that protocols which have
3083 * memory_pressure use sk_wmem_queued as write buffer accounting.
3084 */
3085int __sk_mem_schedule(struct sock *sk, int size, int kind)
3086{
3087 int ret, amt = sk_mem_pages(size);
3088
3089 sk->sk_forward_alloc += amt << PAGE_SHIFT;
3090 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3091 if (!ret)
3092 sk->sk_forward_alloc -= amt << PAGE_SHIFT;
3093 return ret;
3094}
3095EXPORT_SYMBOL(__sk_mem_schedule);
3096
3097/**
3098 * __sk_mem_reduce_allocated - reclaim memory_allocated
3099 * @sk: socket
3100 * @amount: number of quanta
3101 *
3102 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3103 */
3104void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3105{
3106 sk_memory_allocated_sub(sk, amount);
3107
3108 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3109 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3110
3111 if (sk_under_memory_pressure(sk) &&
3112 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3113 sk_leave_memory_pressure(sk);
3114}
3115
3116/**
3117 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3118 * @sk: socket
3119 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3120 */
3121void __sk_mem_reclaim(struct sock *sk, int amount)
3122{
3123 amount >>= PAGE_SHIFT;
3124 sk->sk_forward_alloc -= amount << PAGE_SHIFT;
3125 __sk_mem_reduce_allocated(sk, amount);
3126}
3127EXPORT_SYMBOL(__sk_mem_reclaim);
3128
3129int sk_set_peek_off(struct sock *sk, int val)
3130{
3131 sk->sk_peek_off = val;
3132 return 0;
3133}
3134EXPORT_SYMBOL_GPL(sk_set_peek_off);
3135
3136/*
3137 * Set of default routines for initialising struct proto_ops when
3138 * the protocol does not support a particular function. In certain
3139 * cases where it makes no sense for a protocol to have a "do nothing"
3140 * function, some default processing is provided.
3141 */
3142
3143int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3144{
3145 return -EOPNOTSUPP;
3146}
3147EXPORT_SYMBOL(sock_no_bind);
3148
3149int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3150 int len, int flags)
3151{
3152 return -EOPNOTSUPP;
3153}
3154EXPORT_SYMBOL(sock_no_connect);
3155
3156int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3157{
3158 return -EOPNOTSUPP;
3159}
3160EXPORT_SYMBOL(sock_no_socketpair);
3161
3162int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3163 bool kern)
3164{
3165 return -EOPNOTSUPP;
3166}
3167EXPORT_SYMBOL(sock_no_accept);
3168
3169int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3170 int peer)
3171{
3172 return -EOPNOTSUPP;
3173}
3174EXPORT_SYMBOL(sock_no_getname);
3175
3176int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3177{
3178 return -EOPNOTSUPP;
3179}
3180EXPORT_SYMBOL(sock_no_ioctl);
3181
3182int sock_no_listen(struct socket *sock, int backlog)
3183{
3184 return -EOPNOTSUPP;
3185}
3186EXPORT_SYMBOL(sock_no_listen);
3187
3188int sock_no_shutdown(struct socket *sock, int how)
3189{
3190 return -EOPNOTSUPP;
3191}
3192EXPORT_SYMBOL(sock_no_shutdown);
3193
3194int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3195{
3196 return -EOPNOTSUPP;
3197}
3198EXPORT_SYMBOL(sock_no_sendmsg);
3199
3200int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3201{
3202 return -EOPNOTSUPP;
3203}
3204EXPORT_SYMBOL(sock_no_sendmsg_locked);
3205
3206int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3207 int flags)
3208{
3209 return -EOPNOTSUPP;
3210}
3211EXPORT_SYMBOL(sock_no_recvmsg);
3212
3213int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3214{
3215 /* Mirror missing mmap method error code */
3216 return -ENODEV;
3217}
3218EXPORT_SYMBOL(sock_no_mmap);
3219
3220/*
3221 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3222 * various sock-based usage counts.
3223 */
3224void __receive_sock(struct file *file)
3225{
3226 struct socket *sock;
3227
3228 sock = sock_from_file(file);
3229 if (sock) {
3230 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3231 sock_update_classid(&sock->sk->sk_cgrp_data);
3232 }
3233}
3234
3235ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3236{
3237 ssize_t res;
3238 struct msghdr msg = {.msg_flags = flags};
3239 struct kvec iov;
3240 char *kaddr = kmap(page);
3241 iov.iov_base = kaddr + offset;
3242 iov.iov_len = size;
3243 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3244 kunmap(page);
3245 return res;
3246}
3247EXPORT_SYMBOL(sock_no_sendpage);
3248
3249ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3250 int offset, size_t size, int flags)
3251{
3252 ssize_t res;
3253 struct msghdr msg = {.msg_flags = flags};
3254 struct kvec iov;
3255 char *kaddr = kmap(page);
3256
3257 iov.iov_base = kaddr + offset;
3258 iov.iov_len = size;
3259 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3260 kunmap(page);
3261 return res;
3262}
3263EXPORT_SYMBOL(sock_no_sendpage_locked);
3264
3265/*
3266 * Default Socket Callbacks
3267 */
3268
3269static void sock_def_wakeup(struct sock *sk)
3270{
3271 struct socket_wq *wq;
3272
3273 rcu_read_lock();
3274 wq = rcu_dereference(sk->sk_wq);
3275 if (skwq_has_sleeper(wq))
3276 wake_up_interruptible_all(&wq->wait);
3277 rcu_read_unlock();
3278}
3279
3280static void sock_def_error_report(struct sock *sk)
3281{
3282 struct socket_wq *wq;
3283
3284 rcu_read_lock();
3285 wq = rcu_dereference(sk->sk_wq);
3286 if (skwq_has_sleeper(wq))
3287 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3288 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3289 rcu_read_unlock();
3290}
3291
3292void sock_def_readable(struct sock *sk)
3293{
3294 struct socket_wq *wq;
3295
3296 rcu_read_lock();
3297 wq = rcu_dereference(sk->sk_wq);
3298 if (skwq_has_sleeper(wq))
3299 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3300 EPOLLRDNORM | EPOLLRDBAND);
3301 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3302 rcu_read_unlock();
3303}
3304
3305static void sock_def_write_space(struct sock *sk)
3306{
3307 struct socket_wq *wq;
3308
3309 rcu_read_lock();
3310
3311 /* Do not wake up a writer until he can make "significant"
3312 * progress. --DaveM
3313 */
3314 if (sock_writeable(sk)) {
3315 wq = rcu_dereference(sk->sk_wq);
3316 if (skwq_has_sleeper(wq))
3317 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3318 EPOLLWRNORM | EPOLLWRBAND);
3319
3320 /* Should agree with poll, otherwise some programs break */
3321 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3322 }
3323
3324 rcu_read_unlock();
3325}
3326
3327/* An optimised version of sock_def_write_space(), should only be called
3328 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3329 * ->sk_wmem_alloc.
3330 */
3331static void sock_def_write_space_wfree(struct sock *sk)
3332{
3333 /* Do not wake up a writer until he can make "significant"
3334 * progress. --DaveM
3335 */
3336 if (sock_writeable(sk)) {
3337 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3338
3339 /* rely on refcount_sub from sock_wfree() */
3340 smp_mb__after_atomic();
3341 if (wq && waitqueue_active(&wq->wait))
3342 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3343 EPOLLWRNORM | EPOLLWRBAND);
3344
3345 /* Should agree with poll, otherwise some programs break */
3346 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3347 }
3348}
3349
3350static void sock_def_destruct(struct sock *sk)
3351{
3352}
3353
3354void sk_send_sigurg(struct sock *sk)
3355{
3356 if (sk->sk_socket && sk->sk_socket->file)
3357 if (send_sigurg(&sk->sk_socket->file->f_owner))
3358 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3359}
3360EXPORT_SYMBOL(sk_send_sigurg);
3361
3362void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3363 unsigned long expires)
3364{
3365 if (!mod_timer(timer, expires))
3366 sock_hold(sk);
3367}
3368EXPORT_SYMBOL(sk_reset_timer);
3369
3370void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3371{
3372 if (del_timer(timer))
3373 __sock_put(sk);
3374}
3375EXPORT_SYMBOL(sk_stop_timer);
3376
3377void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3378{
3379 if (del_timer_sync(timer))
3380 __sock_put(sk);
3381}
3382EXPORT_SYMBOL(sk_stop_timer_sync);
3383
3384void sock_init_data(struct socket *sock, struct sock *sk)
3385{
3386 sk_init_common(sk);
3387 sk->sk_send_head = NULL;
3388
3389 timer_setup(&sk->sk_timer, NULL, 0);
3390
3391 sk->sk_allocation = GFP_KERNEL;
3392 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3393 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3394 sk->sk_state = TCP_CLOSE;
3395 sk->sk_use_task_frag = true;
3396 sk_set_socket(sk, sock);
3397
3398 sock_set_flag(sk, SOCK_ZAPPED);
3399
3400 if (sock) {
3401 sk->sk_type = sock->type;
3402 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3403 sock->sk = sk;
3404 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3405 } else {
3406 RCU_INIT_POINTER(sk->sk_wq, NULL);
3407 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3408 }
3409
3410 rwlock_init(&sk->sk_callback_lock);
3411 if (sk->sk_kern_sock)
3412 lockdep_set_class_and_name(
3413 &sk->sk_callback_lock,
3414 af_kern_callback_keys + sk->sk_family,
3415 af_family_kern_clock_key_strings[sk->sk_family]);
3416 else
3417 lockdep_set_class_and_name(
3418 &sk->sk_callback_lock,
3419 af_callback_keys + sk->sk_family,
3420 af_family_clock_key_strings[sk->sk_family]);
3421
3422 sk->sk_state_change = sock_def_wakeup;
3423 sk->sk_data_ready = sock_def_readable;
3424 sk->sk_write_space = sock_def_write_space;
3425 sk->sk_error_report = sock_def_error_report;
3426 sk->sk_destruct = sock_def_destruct;
3427
3428 sk->sk_frag.page = NULL;
3429 sk->sk_frag.offset = 0;
3430 sk->sk_peek_off = -1;
3431
3432 sk->sk_peer_pid = NULL;
3433 sk->sk_peer_cred = NULL;
3434 spin_lock_init(&sk->sk_peer_lock);
3435
3436 sk->sk_write_pending = 0;
3437 sk->sk_rcvlowat = 1;
3438 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3439 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3440
3441 sk->sk_stamp = SK_DEFAULT_STAMP;
3442#if BITS_PER_LONG==32
3443 seqlock_init(&sk->sk_stamp_seq);
3444#endif
3445 atomic_set(&sk->sk_zckey, 0);
3446
3447#ifdef CONFIG_NET_RX_BUSY_POLL
3448 sk->sk_napi_id = 0;
3449 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3450#endif
3451
3452 sk->sk_max_pacing_rate = ~0UL;
3453 sk->sk_pacing_rate = ~0UL;
3454 WRITE_ONCE(sk->sk_pacing_shift, 10);
3455 sk->sk_incoming_cpu = -1;
3456
3457 sk_rx_queue_clear(sk);
3458 /*
3459 * Before updating sk_refcnt, we must commit prior changes to memory
3460 * (Documentation/RCU/rculist_nulls.rst for details)
3461 */
3462 smp_wmb();
3463 refcount_set(&sk->sk_refcnt, 1);
3464 atomic_set(&sk->sk_drops, 0);
3465}
3466EXPORT_SYMBOL(sock_init_data);
3467
3468void lock_sock_nested(struct sock *sk, int subclass)
3469{
3470 /* The sk_lock has mutex_lock() semantics here. */
3471 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3472
3473 might_sleep();
3474 spin_lock_bh(&sk->sk_lock.slock);
3475 if (sock_owned_by_user_nocheck(sk))
3476 __lock_sock(sk);
3477 sk->sk_lock.owned = 1;
3478 spin_unlock_bh(&sk->sk_lock.slock);
3479}
3480EXPORT_SYMBOL(lock_sock_nested);
3481
3482void release_sock(struct sock *sk)
3483{
3484 spin_lock_bh(&sk->sk_lock.slock);
3485 if (sk->sk_backlog.tail)
3486 __release_sock(sk);
3487
3488 /* Warning : release_cb() might need to release sk ownership,
3489 * ie call sock_release_ownership(sk) before us.
3490 */
3491 if (sk->sk_prot->release_cb)
3492 sk->sk_prot->release_cb(sk);
3493
3494 sock_release_ownership(sk);
3495 if (waitqueue_active(&sk->sk_lock.wq))
3496 wake_up(&sk->sk_lock.wq);
3497 spin_unlock_bh(&sk->sk_lock.slock);
3498}
3499EXPORT_SYMBOL(release_sock);
3500
3501bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3502{
3503 might_sleep();
3504 spin_lock_bh(&sk->sk_lock.slock);
3505
3506 if (!sock_owned_by_user_nocheck(sk)) {
3507 /*
3508 * Fast path return with bottom halves disabled and
3509 * sock::sk_lock.slock held.
3510 *
3511 * The 'mutex' is not contended and holding
3512 * sock::sk_lock.slock prevents all other lockers to
3513 * proceed so the corresponding unlock_sock_fast() can
3514 * avoid the slow path of release_sock() completely and
3515 * just release slock.
3516 *
3517 * From a semantical POV this is equivalent to 'acquiring'
3518 * the 'mutex', hence the corresponding lockdep
3519 * mutex_release() has to happen in the fast path of
3520 * unlock_sock_fast().
3521 */
3522 return false;
3523 }
3524
3525 __lock_sock(sk);
3526 sk->sk_lock.owned = 1;
3527 __acquire(&sk->sk_lock.slock);
3528 spin_unlock_bh(&sk->sk_lock.slock);
3529 return true;
3530}
3531EXPORT_SYMBOL(__lock_sock_fast);
3532
3533int sock_gettstamp(struct socket *sock, void __user *userstamp,
3534 bool timeval, bool time32)
3535{
3536 struct sock *sk = sock->sk;
3537 struct timespec64 ts;
3538
3539 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3540 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3541 if (ts.tv_sec == -1)
3542 return -ENOENT;
3543 if (ts.tv_sec == 0) {
3544 ktime_t kt = ktime_get_real();
3545 sock_write_timestamp(sk, kt);
3546 ts = ktime_to_timespec64(kt);
3547 }
3548
3549 if (timeval)
3550 ts.tv_nsec /= 1000;
3551
3552#ifdef CONFIG_COMPAT_32BIT_TIME
3553 if (time32)
3554 return put_old_timespec32(&ts, userstamp);
3555#endif
3556#ifdef CONFIG_SPARC64
3557 /* beware of padding in sparc64 timeval */
3558 if (timeval && !in_compat_syscall()) {
3559 struct __kernel_old_timeval __user tv = {
3560 .tv_sec = ts.tv_sec,
3561 .tv_usec = ts.tv_nsec,
3562 };
3563 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3564 return -EFAULT;
3565 return 0;
3566 }
3567#endif
3568 return put_timespec64(&ts, userstamp);
3569}
3570EXPORT_SYMBOL(sock_gettstamp);
3571
3572void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3573{
3574 if (!sock_flag(sk, flag)) {
3575 unsigned long previous_flags = sk->sk_flags;
3576
3577 sock_set_flag(sk, flag);
3578 /*
3579 * we just set one of the two flags which require net
3580 * time stamping, but time stamping might have been on
3581 * already because of the other one
3582 */
3583 if (sock_needs_netstamp(sk) &&
3584 !(previous_flags & SK_FLAGS_TIMESTAMP))
3585 net_enable_timestamp();
3586 }
3587}
3588
3589int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3590 int level, int type)
3591{
3592 struct sock_exterr_skb *serr;
3593 struct sk_buff *skb;
3594 int copied, err;
3595
3596 err = -EAGAIN;
3597 skb = sock_dequeue_err_skb(sk);
3598 if (skb == NULL)
3599 goto out;
3600
3601 copied = skb->len;
3602 if (copied > len) {
3603 msg->msg_flags |= MSG_TRUNC;
3604 copied = len;
3605 }
3606 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3607 if (err)
3608 goto out_free_skb;
3609
3610 sock_recv_timestamp(msg, sk, skb);
3611
3612 serr = SKB_EXT_ERR(skb);
3613 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3614
3615 msg->msg_flags |= MSG_ERRQUEUE;
3616 err = copied;
3617
3618out_free_skb:
3619 kfree_skb(skb);
3620out:
3621 return err;
3622}
3623EXPORT_SYMBOL(sock_recv_errqueue);
3624
3625/*
3626 * Get a socket option on an socket.
3627 *
3628 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3629 * asynchronous errors should be reported by getsockopt. We assume
3630 * this means if you specify SO_ERROR (otherwise whats the point of it).
3631 */
3632int sock_common_getsockopt(struct socket *sock, int level, int optname,
3633 char __user *optval, int __user *optlen)
3634{
3635 struct sock *sk = sock->sk;
3636
3637 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3638 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3639}
3640EXPORT_SYMBOL(sock_common_getsockopt);
3641
3642int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3643 int flags)
3644{
3645 struct sock *sk = sock->sk;
3646 int addr_len = 0;
3647 int err;
3648
3649 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3650 if (err >= 0)
3651 msg->msg_namelen = addr_len;
3652 return err;
3653}
3654EXPORT_SYMBOL(sock_common_recvmsg);
3655
3656/*
3657 * Set socket options on an inet socket.
3658 */
3659int sock_common_setsockopt(struct socket *sock, int level, int optname,
3660 sockptr_t optval, unsigned int optlen)
3661{
3662 struct sock *sk = sock->sk;
3663
3664 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3665 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3666}
3667EXPORT_SYMBOL(sock_common_setsockopt);
3668
3669void sk_common_release(struct sock *sk)
3670{
3671 if (sk->sk_prot->destroy)
3672 sk->sk_prot->destroy(sk);
3673
3674 /*
3675 * Observation: when sk_common_release is called, processes have
3676 * no access to socket. But net still has.
3677 * Step one, detach it from networking:
3678 *
3679 * A. Remove from hash tables.
3680 */
3681
3682 sk->sk_prot->unhash(sk);
3683
3684 /*
3685 * In this point socket cannot receive new packets, but it is possible
3686 * that some packets are in flight because some CPU runs receiver and
3687 * did hash table lookup before we unhashed socket. They will achieve
3688 * receive queue and will be purged by socket destructor.
3689 *
3690 * Also we still have packets pending on receive queue and probably,
3691 * our own packets waiting in device queues. sock_destroy will drain
3692 * receive queue, but transmitted packets will delay socket destruction
3693 * until the last reference will be released.
3694 */
3695
3696 sock_orphan(sk);
3697
3698 xfrm_sk_free_policy(sk);
3699
3700 sk_refcnt_debug_release(sk);
3701
3702 sock_put(sk);
3703}
3704EXPORT_SYMBOL(sk_common_release);
3705
3706void sk_get_meminfo(const struct sock *sk, u32 *mem)
3707{
3708 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3709
3710 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3711 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3712 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3713 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3714 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3715 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3716 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3717 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3718 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3719}
3720
3721#ifdef CONFIG_PROC_FS
3722static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3723
3724int sock_prot_inuse_get(struct net *net, struct proto *prot)
3725{
3726 int cpu, idx = prot->inuse_idx;
3727 int res = 0;
3728
3729 for_each_possible_cpu(cpu)
3730 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3731
3732 return res >= 0 ? res : 0;
3733}
3734EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3735
3736int sock_inuse_get(struct net *net)
3737{
3738 int cpu, res = 0;
3739
3740 for_each_possible_cpu(cpu)
3741 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3742
3743 return res;
3744}
3745
3746EXPORT_SYMBOL_GPL(sock_inuse_get);
3747
3748static int __net_init sock_inuse_init_net(struct net *net)
3749{
3750 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3751 if (net->core.prot_inuse == NULL)
3752 return -ENOMEM;
3753 return 0;
3754}
3755
3756static void __net_exit sock_inuse_exit_net(struct net *net)
3757{
3758 free_percpu(net->core.prot_inuse);
3759}
3760
3761static struct pernet_operations net_inuse_ops = {
3762 .init = sock_inuse_init_net,
3763 .exit = sock_inuse_exit_net,
3764};
3765
3766static __init int net_inuse_init(void)
3767{
3768 if (register_pernet_subsys(&net_inuse_ops))
3769 panic("Cannot initialize net inuse counters");
3770
3771 return 0;
3772}
3773
3774core_initcall(net_inuse_init);
3775
3776static int assign_proto_idx(struct proto *prot)
3777{
3778 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3779
3780 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3781 pr_err("PROTO_INUSE_NR exhausted\n");
3782 return -ENOSPC;
3783 }
3784
3785 set_bit(prot->inuse_idx, proto_inuse_idx);
3786 return 0;
3787}
3788
3789static void release_proto_idx(struct proto *prot)
3790{
3791 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3792 clear_bit(prot->inuse_idx, proto_inuse_idx);
3793}
3794#else
3795static inline int assign_proto_idx(struct proto *prot)
3796{
3797 return 0;
3798}
3799
3800static inline void release_proto_idx(struct proto *prot)
3801{
3802}
3803
3804#endif
3805
3806static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3807{
3808 if (!twsk_prot)
3809 return;
3810 kfree(twsk_prot->twsk_slab_name);
3811 twsk_prot->twsk_slab_name = NULL;
3812 kmem_cache_destroy(twsk_prot->twsk_slab);
3813 twsk_prot->twsk_slab = NULL;
3814}
3815
3816static int tw_prot_init(const struct proto *prot)
3817{
3818 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3819
3820 if (!twsk_prot)
3821 return 0;
3822
3823 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3824 prot->name);
3825 if (!twsk_prot->twsk_slab_name)
3826 return -ENOMEM;
3827
3828 twsk_prot->twsk_slab =
3829 kmem_cache_create(twsk_prot->twsk_slab_name,
3830 twsk_prot->twsk_obj_size, 0,
3831 SLAB_ACCOUNT | prot->slab_flags,
3832 NULL);
3833 if (!twsk_prot->twsk_slab) {
3834 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3835 prot->name);
3836 return -ENOMEM;
3837 }
3838
3839 return 0;
3840}
3841
3842static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3843{
3844 if (!rsk_prot)
3845 return;
3846 kfree(rsk_prot->slab_name);
3847 rsk_prot->slab_name = NULL;
3848 kmem_cache_destroy(rsk_prot->slab);
3849 rsk_prot->slab = NULL;
3850}
3851
3852static int req_prot_init(const struct proto *prot)
3853{
3854 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3855
3856 if (!rsk_prot)
3857 return 0;
3858
3859 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3860 prot->name);
3861 if (!rsk_prot->slab_name)
3862 return -ENOMEM;
3863
3864 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3865 rsk_prot->obj_size, 0,
3866 SLAB_ACCOUNT | prot->slab_flags,
3867 NULL);
3868
3869 if (!rsk_prot->slab) {
3870 pr_crit("%s: Can't create request sock SLAB cache!\n",
3871 prot->name);
3872 return -ENOMEM;
3873 }
3874 return 0;
3875}
3876
3877int proto_register(struct proto *prot, int alloc_slab)
3878{
3879 int ret = -ENOBUFS;
3880
3881 if (prot->memory_allocated && !prot->sysctl_mem) {
3882 pr_err("%s: missing sysctl_mem\n", prot->name);
3883 return -EINVAL;
3884 }
3885 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3886 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3887 return -EINVAL;
3888 }
3889 if (alloc_slab) {
3890 prot->slab = kmem_cache_create_usercopy(prot->name,
3891 prot->obj_size, 0,
3892 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3893 prot->slab_flags,
3894 prot->useroffset, prot->usersize,
3895 NULL);
3896
3897 if (prot->slab == NULL) {
3898 pr_crit("%s: Can't create sock SLAB cache!\n",
3899 prot->name);
3900 goto out;
3901 }
3902
3903 if (req_prot_init(prot))
3904 goto out_free_request_sock_slab;
3905
3906 if (tw_prot_init(prot))
3907 goto out_free_timewait_sock_slab;
3908 }
3909
3910 mutex_lock(&proto_list_mutex);
3911 ret = assign_proto_idx(prot);
3912 if (ret) {
3913 mutex_unlock(&proto_list_mutex);
3914 goto out_free_timewait_sock_slab;
3915 }
3916 list_add(&prot->node, &proto_list);
3917 mutex_unlock(&proto_list_mutex);
3918 return ret;
3919
3920out_free_timewait_sock_slab:
3921 if (alloc_slab)
3922 tw_prot_cleanup(prot->twsk_prot);
3923out_free_request_sock_slab:
3924 if (alloc_slab) {
3925 req_prot_cleanup(prot->rsk_prot);
3926
3927 kmem_cache_destroy(prot->slab);
3928 prot->slab = NULL;
3929 }
3930out:
3931 return ret;
3932}
3933EXPORT_SYMBOL(proto_register);
3934
3935void proto_unregister(struct proto *prot)
3936{
3937 mutex_lock(&proto_list_mutex);
3938 release_proto_idx(prot);
3939 list_del(&prot->node);
3940 mutex_unlock(&proto_list_mutex);
3941
3942 kmem_cache_destroy(prot->slab);
3943 prot->slab = NULL;
3944
3945 req_prot_cleanup(prot->rsk_prot);
3946 tw_prot_cleanup(prot->twsk_prot);
3947}
3948EXPORT_SYMBOL(proto_unregister);
3949
3950int sock_load_diag_module(int family, int protocol)
3951{
3952 if (!protocol) {
3953 if (!sock_is_registered(family))
3954 return -ENOENT;
3955
3956 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3957 NETLINK_SOCK_DIAG, family);
3958 }
3959
3960#ifdef CONFIG_INET
3961 if (family == AF_INET &&
3962 protocol != IPPROTO_RAW &&
3963 protocol < MAX_INET_PROTOS &&
3964 !rcu_access_pointer(inet_protos[protocol]))
3965 return -ENOENT;
3966#endif
3967
3968 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3969 NETLINK_SOCK_DIAG, family, protocol);
3970}
3971EXPORT_SYMBOL(sock_load_diag_module);
3972
3973#ifdef CONFIG_PROC_FS
3974static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3975 __acquires(proto_list_mutex)
3976{
3977 mutex_lock(&proto_list_mutex);
3978 return seq_list_start_head(&proto_list, *pos);
3979}
3980
3981static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3982{
3983 return seq_list_next(v, &proto_list, pos);
3984}
3985
3986static void proto_seq_stop(struct seq_file *seq, void *v)
3987 __releases(proto_list_mutex)
3988{
3989 mutex_unlock(&proto_list_mutex);
3990}
3991
3992static char proto_method_implemented(const void *method)
3993{
3994 return method == NULL ? 'n' : 'y';
3995}
3996static long sock_prot_memory_allocated(struct proto *proto)
3997{
3998 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3999}
4000
4001static const char *sock_prot_memory_pressure(struct proto *proto)
4002{
4003 return proto->memory_pressure != NULL ?
4004 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4005}
4006
4007static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4008{
4009
4010 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4011 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4012 proto->name,
4013 proto->obj_size,
4014 sock_prot_inuse_get(seq_file_net(seq), proto),
4015 sock_prot_memory_allocated(proto),
4016 sock_prot_memory_pressure(proto),
4017 proto->max_header,
4018 proto->slab == NULL ? "no" : "yes",
4019 module_name(proto->owner),
4020 proto_method_implemented(proto->close),
4021 proto_method_implemented(proto->connect),
4022 proto_method_implemented(proto->disconnect),
4023 proto_method_implemented(proto->accept),
4024 proto_method_implemented(proto->ioctl),
4025 proto_method_implemented(proto->init),
4026 proto_method_implemented(proto->destroy),
4027 proto_method_implemented(proto->shutdown),
4028 proto_method_implemented(proto->setsockopt),
4029 proto_method_implemented(proto->getsockopt),
4030 proto_method_implemented(proto->sendmsg),
4031 proto_method_implemented(proto->recvmsg),
4032 proto_method_implemented(proto->sendpage),
4033 proto_method_implemented(proto->bind),
4034 proto_method_implemented(proto->backlog_rcv),
4035 proto_method_implemented(proto->hash),
4036 proto_method_implemented(proto->unhash),
4037 proto_method_implemented(proto->get_port),
4038 proto_method_implemented(proto->enter_memory_pressure));
4039}
4040
4041static int proto_seq_show(struct seq_file *seq, void *v)
4042{
4043 if (v == &proto_list)
4044 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4045 "protocol",
4046 "size",
4047 "sockets",
4048 "memory",
4049 "press",
4050 "maxhdr",
4051 "slab",
4052 "module",
4053 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
4054 else
4055 proto_seq_printf(seq, list_entry(v, struct proto, node));
4056 return 0;
4057}
4058
4059static const struct seq_operations proto_seq_ops = {
4060 .start = proto_seq_start,
4061 .next = proto_seq_next,
4062 .stop = proto_seq_stop,
4063 .show = proto_seq_show,
4064};
4065
4066static __net_init int proto_init_net(struct net *net)
4067{
4068 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4069 sizeof(struct seq_net_private)))
4070 return -ENOMEM;
4071
4072 return 0;
4073}
4074
4075static __net_exit void proto_exit_net(struct net *net)
4076{
4077 remove_proc_entry("protocols", net->proc_net);
4078}
4079
4080
4081static __net_initdata struct pernet_operations proto_net_ops = {
4082 .init = proto_init_net,
4083 .exit = proto_exit_net,
4084};
4085
4086static int __init proto_init(void)
4087{
4088 return register_pernet_subsys(&proto_net_ops);
4089}
4090
4091subsys_initcall(proto_init);
4092
4093#endif /* PROC_FS */
4094
4095#ifdef CONFIG_NET_RX_BUSY_POLL
4096bool sk_busy_loop_end(void *p, unsigned long start_time)
4097{
4098 struct sock *sk = p;
4099
4100 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4101 sk_busy_loop_timeout(sk, start_time);
4102}
4103EXPORT_SYMBOL(sk_busy_loop_end);
4104#endif /* CONFIG_NET_RX_BUSY_POLL */
4105
4106int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4107{
4108 if (!sk->sk_prot->bind_add)
4109 return -EOPNOTSUPP;
4110 return sk->sk_prot->bind_add(sk, addr, addr_len);
4111}
4112EXPORT_SYMBOL(sock_bind_add);