]> git.ipfire.org Git - thirdparty/linux.git/blob - net/core/sock.c
projects / thirdparty / linux.git / blob
? search:


88bf810394a5521de83aaa7095543a208aa60573
[thirdparty/linux.git] / net / core / sock.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
94 #include <linux/in.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
121
122 #include <linux/uaccess.h>
123
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
136
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
139 #include <net/bpf_sk_storage.h>
140
141 #include <trace/events/sock.h>
142
143 #include <net/tcp.h>
144 #include <net/busy_poll.h>
145 #include <net/phonet/phonet.h>
146
147 #include <linux/ethtool.h>
148
149 #include "dev.h"
150
151 static DEFINE_MUTEX(proto_list_mutex);
152 static LIST_HEAD(proto_list);
153
154 static void sock_def_write_space_wfree(struct sock *sk);
155 static void sock_def_write_space(struct sock *sk);
156
157 /**
158 * sk_ns_capable - General socket capability test
159 * @sk: Socket to use a capability on or through
160 * @user_ns: The user namespace of the capability to use
161 * @cap: The capability to use
162 *
163 * Test to see if the opener of the socket had when the socket was
164 * created and the current process has the capability @cap in the user
165 * namespace @user_ns.
166 */
167 bool sk_ns_capable(const struct sock *sk,
168 struct user_namespace *user_ns, int cap)
169 {
170 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
171 ns_capable(user_ns, cap);
172 }
173 EXPORT_SYMBOL(sk_ns_capable);
174
175 /**
176 * sk_capable - Socket global capability test
177 * @sk: Socket to use a capability on or through
178 * @cap: The global capability to use
179 *
180 * Test to see if the opener of the socket had when the socket was
181 * created and the current process has the capability @cap in all user
182 * namespaces.
183 */
184 bool sk_capable(const struct sock *sk, int cap)
185 {
186 return sk_ns_capable(sk, &init_user_ns, cap);
187 }
188 EXPORT_SYMBOL(sk_capable);
189
190 /**
191 * sk_net_capable - Network namespace socket capability test
192 * @sk: Socket to use a capability on or through
193 * @cap: The capability to use
194 *
195 * Test to see if the opener of the socket had when the socket was created
196 * and the current process has the capability @cap over the network namespace
197 * the socket is a member of.
198 */
199 bool sk_net_capable(const struct sock *sk, int cap)
200 {
201 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202 }
203 EXPORT_SYMBOL(sk_net_capable);
204
205 /*
206 * Each address family might have different locking rules, so we have
207 * one slock key per address family and separate keys for internal and
208 * userspace sockets.
209 */
210 static struct lock_class_key af_family_keys[AF_MAX];
211 static struct lock_class_key af_family_kern_keys[AF_MAX];
212 static struct lock_class_key af_family_slock_keys[AF_MAX];
213 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
214
215 /*
216 * Make lock validator output more readable. (we pre-construct these
217 * strings build-time, so that runtime initialization of socket
218 * locks is fast):
219 */
220
221 #define _sock_locks(x) \
222 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
223 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
224 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
225 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
226 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
227 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
228 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
229 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
230 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
231 x "27" , x "28" , x "AF_CAN" , \
232 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
233 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
234 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
235 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
236 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
237 x "AF_MCTP" , \
238 x "AF_MAX"
239
240 static const char *const af_family_key_strings[AF_MAX+1] = {
241 _sock_locks("sk_lock-")
242 };
243 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("slock-")
245 };
246 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("clock-")
248 };
249
250 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
251 _sock_locks("k-sk_lock-")
252 };
253 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
254 _sock_locks("k-slock-")
255 };
256 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
257 _sock_locks("k-clock-")
258 };
259 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
260 _sock_locks("rlock-")
261 };
262 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
263 _sock_locks("wlock-")
264 };
265 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
266 _sock_locks("elock-")
267 };
268
269 /*
270 * sk_callback_lock and sk queues locking rules are per-address-family,
271 * so split the lock classes by using a per-AF key:
272 */
273 static struct lock_class_key af_callback_keys[AF_MAX];
274 static struct lock_class_key af_rlock_keys[AF_MAX];
275 static struct lock_class_key af_wlock_keys[AF_MAX];
276 static struct lock_class_key af_elock_keys[AF_MAX];
277 static struct lock_class_key af_kern_callback_keys[AF_MAX];
278
279 /* Run time adjustable parameters. */
280 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
281 EXPORT_SYMBOL(sysctl_wmem_max);
282 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
283 EXPORT_SYMBOL(sysctl_rmem_max);
284 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
285 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286
287 int sysctl_tstamp_allow_data __read_mostly = 1;
288
289 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
290 EXPORT_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 */
300 void 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 }
306 EXPORT_SYMBOL_GPL(sk_set_memalloc);
307
308 void 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 }
323 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324
325 int __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 }
342 EXPORT_SYMBOL(__sk_backlog_rcv);
343
344 void 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 }
358 EXPORT_SYMBOL(sk_error_report);
359
360 int 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 }
389 EXPORT_SYMBOL(sock_get_timeout);
390
391 int 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 }
422 EXPORT_SYMBOL(sock_copy_user_timeval);
423
424 static 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
457 static 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
468 static 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
479 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
480 {
481 unsigned long flags;
482 struct sk_buff_head *list = &sk->sk_receive_queue;
483
484 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
485 atomic_inc(&sk->sk_drops);
486 trace_sock_rcvqueue_full(sk, skb);
487 return -ENOMEM;
488 }
489
490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
491 atomic_inc(&sk->sk_drops);
492 return -ENOBUFS;
493 }
494
495 skb->dev = NULL;
496 skb_set_owner_r(skb, sk);
497
498 /* we escape from rcu protected region, make sure we dont leak
499 * a norefcounted dst
500 */
501 skb_dst_force(skb);
502
503 spin_lock_irqsave(&list->lock, flags);
504 sock_skb_set_dropcount(sk, skb);
505 __skb_queue_tail(list, skb);
506 spin_unlock_irqrestore(&list->lock, flags);
507
508 if (!sock_flag(sk, SOCK_DEAD))
509 sk->sk_data_ready(sk);
510 return 0;
511 }
512 EXPORT_SYMBOL(__sock_queue_rcv_skb);
513
514 int 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 }
537 out:
538 if (reason)
539 *reason = drop_reason;
540 return err;
541 }
542 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
543
544 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 const int nested, unsigned int trim_cap, bool refcounted)
546 {
547 int rc = NET_RX_SUCCESS;
548
549 if (sk_filter_trim_cap(sk, skb, trim_cap))
550 goto discard_and_relse;
551
552 skb->dev = NULL;
553
554 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
555 atomic_inc(&sk->sk_drops);
556 goto discard_and_relse;
557 }
558 if (nested)
559 bh_lock_sock_nested(sk);
560 else
561 bh_lock_sock(sk);
562 if (!sock_owned_by_user(sk)) {
563 /*
564 * trylock + unlock semantics:
565 */
566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
567
568 rc = sk_backlog_rcv(sk, skb);
569
570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
572 bh_unlock_sock(sk);
573 atomic_inc(&sk->sk_drops);
574 goto discard_and_relse;
575 }
576
577 bh_unlock_sock(sk);
578 out:
579 if (refcounted)
580 sock_put(sk);
581 return rc;
582 discard_and_relse:
583 kfree_skb(skb);
584 goto out;
585 }
586 EXPORT_SYMBOL(__sk_receive_skb);
587
588 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
589 u32));
590 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
591 u32));
592 struct 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 }
608 EXPORT_SYMBOL(__sk_dst_check);
609
610 struct 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 }
624 EXPORT_SYMBOL(sk_dst_check);
625
626 static 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
650 out:
651 #endif
652
653 return ret;
654 }
655
656 int 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 }
668 EXPORT_SYMBOL(sock_bindtoindex);
669
670 static 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);
712 out:
713 #endif
714
715 return ret;
716 }
717
718 static 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
746 zero:
747 ret = -EFAULT;
748 if (copy_to_sockptr(optlen, &len, sizeof(int)))
749 goto out;
750
751 ret = 0;
752
753 out:
754 #endif
755
756 return ret;
757 }
758
759 bool 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 }
777 EXPORT_SYMBOL(sk_mc_loop);
778
779 void sock_set_reuseaddr(struct sock *sk)
780 {
781 lock_sock(sk);
782 sk->sk_reuse = SK_CAN_REUSE;
783 release_sock(sk);
784 }
785 EXPORT_SYMBOL(sock_set_reuseaddr);
786
787 void sock_set_reuseport(struct sock *sk)
788 {
789 lock_sock(sk);
790 sk->sk_reuseport = true;
791 release_sock(sk);
792 }
793 EXPORT_SYMBOL(sock_set_reuseport);
794
795 void 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 }
802 EXPORT_SYMBOL(sock_no_linger);
803
804 void sock_set_priority(struct sock *sk, u32 priority)
805 {
806 WRITE_ONCE(sk->sk_priority, priority);
807 }
808 EXPORT_SYMBOL(sock_set_priority);
809
810 void 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 }
819 EXPORT_SYMBOL(sock_set_sndtimeo);
820
821 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
822 {
823 if (val) {
824 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
825 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
826 sock_set_flag(sk, SOCK_RCVTSTAMP);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
828 } else {
829 sock_reset_flag(sk, SOCK_RCVTSTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
831 }
832 }
833
834 void sock_enable_timestamps(struct sock *sk)
835 {
836 lock_sock(sk);
837 __sock_set_timestamps(sk, true, false, true);
838 release_sock(sk);
839 }
840 EXPORT_SYMBOL(sock_enable_timestamps);
841
842 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
843 {
844 switch (optname) {
845 case SO_TIMESTAMP_OLD:
846 __sock_set_timestamps(sk, valbool, false, false);
847 break;
848 case SO_TIMESTAMP_NEW:
849 __sock_set_timestamps(sk, valbool, true, false);
850 break;
851 case SO_TIMESTAMPNS_OLD:
852 __sock_set_timestamps(sk, valbool, false, true);
853 break;
854 case SO_TIMESTAMPNS_NEW:
855 __sock_set_timestamps(sk, valbool, true, true);
856 break;
857 }
858 }
859
860 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
861 {
862 struct net *net = sock_net(sk);
863 struct net_device *dev = NULL;
864 bool match = false;
865 int *vclock_index;
866 int i, num;
867
868 if (sk->sk_bound_dev_if)
869 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
870
871 if (!dev) {
872 pr_err("%s: sock not bind to device\n", __func__);
873 return -EOPNOTSUPP;
874 }
875
876 num = ethtool_get_phc_vclocks(dev, &vclock_index);
877 dev_put(dev);
878
879 for (i = 0; i < num; i++) {
880 if (*(vclock_index + i) == phc_index) {
881 match = true;
882 break;
883 }
884 }
885
886 if (num > 0)
887 kfree(vclock_index);
888
889 if (!match)
890 return -EINVAL;
891
892 WRITE_ONCE(sk->sk_bind_phc, phc_index);
893
894 return 0;
895 }
896
897 int sock_set_timestamping(struct sock *sk, int optname,
898 struct so_timestamping timestamping)
899 {
900 int val = timestamping.flags;
901 int ret;
902
903 if (val & ~SOF_TIMESTAMPING_MASK)
904 return -EINVAL;
905
906 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
907 !(val & SOF_TIMESTAMPING_OPT_ID))
908 return -EINVAL;
909
910 if (val & SOF_TIMESTAMPING_OPT_ID &&
911 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
912 if (sk_is_tcp(sk)) {
913 if ((1 << sk->sk_state) &
914 (TCPF_CLOSE | TCPF_LISTEN))
915 return -EINVAL;
916 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
918 else
919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
920 } else {
921 atomic_set(&sk->sk_tskey, 0);
922 }
923 }
924
925 if (val & SOF_TIMESTAMPING_OPT_STATS &&
926 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
927 return -EINVAL;
928
929 if (val & SOF_TIMESTAMPING_BIND_PHC) {
930 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
931 if (ret)
932 return ret;
933 }
934
935 WRITE_ONCE(sk->sk_tsflags, val);
936 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
937
938 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
939 sock_enable_timestamp(sk,
940 SOCK_TIMESTAMPING_RX_SOFTWARE);
941 else
942 sock_disable_timestamp(sk,
943 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
944 return 0;
945 }
946
947 void sock_set_keepalive(struct sock *sk)
948 {
949 lock_sock(sk);
950 if (sk->sk_prot->keepalive)
951 sk->sk_prot->keepalive(sk, true);
952 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
953 release_sock(sk);
954 }
955 EXPORT_SYMBOL(sock_set_keepalive);
956
957 static void __sock_set_rcvbuf(struct sock *sk, int val)
958 {
959 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
960 * as a negative value.
961 */
962 val = min_t(int, val, INT_MAX / 2);
963 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
964
965 /* We double it on the way in to account for "struct sk_buff" etc.
966 * overhead. Applications assume that the SO_RCVBUF setting they make
967 * will allow that much actual data to be received on that socket.
968 *
969 * Applications are unaware that "struct sk_buff" and other overheads
970 * allocate from the receive buffer during socket buffer allocation.
971 *
972 * And after considering the possible alternatives, returning the value
973 * we actually used in getsockopt is the most desirable behavior.
974 */
975 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
976 }
977
978 void sock_set_rcvbuf(struct sock *sk, int val)
979 {
980 lock_sock(sk);
981 __sock_set_rcvbuf(sk, val);
982 release_sock(sk);
983 }
984 EXPORT_SYMBOL(sock_set_rcvbuf);
985
986 static void __sock_set_mark(struct sock *sk, u32 val)
987 {
988 if (val != sk->sk_mark) {
989 WRITE_ONCE(sk->sk_mark, val);
990 sk_dst_reset(sk);
991 }
992 }
993
994 void sock_set_mark(struct sock *sk, u32 val)
995 {
996 lock_sock(sk);
997 __sock_set_mark(sk, val);
998 release_sock(sk);
999 }
1000 EXPORT_SYMBOL(sock_set_mark);
1001
1002 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1003 {
1004 /* Round down bytes to multiple of pages */
1005 bytes = round_down(bytes, PAGE_SIZE);
1006
1007 WARN_ON(bytes > sk->sk_reserved_mem);
1008 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1009 sk_mem_reclaim(sk);
1010 }
1011
1012 static int sock_reserve_memory(struct sock *sk, int bytes)
1013 {
1014 long allocated;
1015 bool charged;
1016 int pages;
1017
1018 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1019 return -EOPNOTSUPP;
1020
1021 if (!bytes)
1022 return 0;
1023
1024 pages = sk_mem_pages(bytes);
1025
1026 /* pre-charge to memcg */
1027 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1028 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1029 if (!charged)
1030 return -ENOMEM;
1031
1032 /* pre-charge to forward_alloc */
1033 sk_memory_allocated_add(sk, pages);
1034 allocated = sk_memory_allocated(sk);
1035 /* If the system goes into memory pressure with this
1036 * precharge, give up and return error.
1037 */
1038 if (allocated > sk_prot_mem_limits(sk, 1)) {
1039 sk_memory_allocated_sub(sk, pages);
1040 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1041 return -ENOMEM;
1042 }
1043 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1044
1045 WRITE_ONCE(sk->sk_reserved_mem,
1046 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1047
1048 return 0;
1049 }
1050
1051 void sockopt_lock_sock(struct sock *sk)
1052 {
1053 /* When current->bpf_ctx is set, the setsockopt is called from
1054 * a bpf prog. bpf has ensured the sk lock has been
1055 * acquired before calling setsockopt().
1056 */
1057 if (has_current_bpf_ctx())
1058 return;
1059
1060 lock_sock(sk);
1061 }
1062 EXPORT_SYMBOL(sockopt_lock_sock);
1063
1064 void sockopt_release_sock(struct sock *sk)
1065 {
1066 if (has_current_bpf_ctx())
1067 return;
1068
1069 release_sock(sk);
1070 }
1071 EXPORT_SYMBOL(sockopt_release_sock);
1072
1073 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1074 {
1075 return has_current_bpf_ctx() || ns_capable(ns, cap);
1076 }
1077 EXPORT_SYMBOL(sockopt_ns_capable);
1078
1079 bool sockopt_capable(int cap)
1080 {
1081 return has_current_bpf_ctx() || capable(cap);
1082 }
1083 EXPORT_SYMBOL(sockopt_capable);
1084
1085 /*
1086 * This is meant for all protocols to use and covers goings on
1087 * at the socket level. Everything here is generic.
1088 */
1089
1090 int sk_setsockopt(struct sock *sk, int level, int optname,
1091 sockptr_t optval, unsigned int optlen)
1092 {
1093 struct so_timestamping timestamping;
1094 struct socket *sock = sk->sk_socket;
1095 struct sock_txtime sk_txtime;
1096 int val;
1097 int valbool;
1098 struct linger ling;
1099 int ret = 0;
1100
1101 /*
1102 * Options without arguments
1103 */
1104
1105 if (optname == SO_BINDTODEVICE)
1106 return sock_setbindtodevice(sk, optval, optlen);
1107
1108 if (optlen < sizeof(int))
1109 return -EINVAL;
1110
1111 if (copy_from_sockptr(&val, optval, sizeof(val)))
1112 return -EFAULT;
1113
1114 valbool = val ? 1 : 0;
1115
1116 /* handle options which do not require locking the socket. */
1117 switch (optname) {
1118 case SO_PRIORITY:
1119 if ((val >= 0 && val <= 6) ||
1120 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1121 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1122 sock_set_priority(sk, val);
1123 return 0;
1124 }
1125 return -EPERM;
1126 case SO_PASSSEC:
1127 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1128 return 0;
1129 case SO_PASSCRED:
1130 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1131 return 0;
1132 case SO_PASSPIDFD:
1133 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1134 return 0;
1135 case SO_TYPE:
1136 case SO_PROTOCOL:
1137 case SO_DOMAIN:
1138 case SO_ERROR:
1139 return -ENOPROTOOPT;
1140 #ifdef CONFIG_NET_RX_BUSY_POLL
1141 case SO_BUSY_POLL:
1142 if (val < 0)
1143 return -EINVAL;
1144 WRITE_ONCE(sk->sk_ll_usec, val);
1145 return 0;
1146 case SO_PREFER_BUSY_POLL:
1147 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1148 return -EPERM;
1149 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1150 return 0;
1151 case SO_BUSY_POLL_BUDGET:
1152 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1153 !sockopt_capable(CAP_NET_ADMIN))
1154 return -EPERM;
1155 if (val < 0 || val > U16_MAX)
1156 return -EINVAL;
1157 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1158 return 0;
1159 #endif
1160 case SO_MAX_PACING_RATE:
1161 {
1162 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1163 unsigned long pacing_rate;
1164
1165 if (sizeof(ulval) != sizeof(val) &&
1166 optlen >= sizeof(ulval) &&
1167 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1168 return -EFAULT;
1169 }
1170 if (ulval != ~0UL)
1171 cmpxchg(&sk->sk_pacing_status,
1172 SK_PACING_NONE,
1173 SK_PACING_NEEDED);
1174 /* Pairs with READ_ONCE() from sk_getsockopt() */
1175 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1176 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1177 if (ulval < pacing_rate)
1178 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1179 return 0;
1180 }
1181 case SO_TXREHASH:
1182 if (val < -1 || val > 1)
1183 return -EINVAL;
1184 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1185 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1186 /* Paired with READ_ONCE() in tcp_rtx_synack()
1187 * and sk_getsockopt().
1188 */
1189 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1190 return 0;
1191 }
1192
1193 sockopt_lock_sock(sk);
1194
1195 switch (optname) {
1196 case SO_DEBUG:
1197 if (val && !sockopt_capable(CAP_NET_ADMIN))
1198 ret = -EACCES;
1199 else
1200 sock_valbool_flag(sk, SOCK_DBG, valbool);
1201 break;
1202 case SO_REUSEADDR:
1203 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1204 break;
1205 case SO_REUSEPORT:
1206 sk->sk_reuseport = valbool;
1207 break;
1208 case SO_DONTROUTE:
1209 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1210 sk_dst_reset(sk);
1211 break;
1212 case SO_BROADCAST:
1213 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1214 break;
1215 case SO_SNDBUF:
1216 /* Don't error on this BSD doesn't and if you think
1217 * about it this is right. Otherwise apps have to
1218 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1219 * are treated in BSD as hints
1220 */
1221 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1222 set_sndbuf:
1223 /* Ensure val * 2 fits into an int, to prevent max_t()
1224 * from treating it as a negative value.
1225 */
1226 val = min_t(int, val, INT_MAX / 2);
1227 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1228 WRITE_ONCE(sk->sk_sndbuf,
1229 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1230 /* Wake up sending tasks if we upped the value. */
1231 sk->sk_write_space(sk);
1232 break;
1233
1234 case SO_SNDBUFFORCE:
1235 if (!sockopt_capable(CAP_NET_ADMIN)) {
1236 ret = -EPERM;
1237 break;
1238 }
1239
1240 /* No negative values (to prevent underflow, as val will be
1241 * multiplied by 2).
1242 */
1243 if (val < 0)
1244 val = 0;
1245 goto set_sndbuf;
1246
1247 case SO_RCVBUF:
1248 /* Don't error on this BSD doesn't and if you think
1249 * about it this is right. Otherwise apps have to
1250 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1251 * are treated in BSD as hints
1252 */
1253 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1254 break;
1255
1256 case SO_RCVBUFFORCE:
1257 if (!sockopt_capable(CAP_NET_ADMIN)) {
1258 ret = -EPERM;
1259 break;
1260 }
1261
1262 /* No negative values (to prevent underflow, as val will be
1263 * multiplied by 2).
1264 */
1265 __sock_set_rcvbuf(sk, max(val, 0));
1266 break;
1267
1268 case SO_KEEPALIVE:
1269 if (sk->sk_prot->keepalive)
1270 sk->sk_prot->keepalive(sk, valbool);
1271 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1272 break;
1273
1274 case SO_OOBINLINE:
1275 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1276 break;
1277
1278 case SO_NO_CHECK:
1279 sk->sk_no_check_tx = valbool;
1280 break;
1281
1282 case SO_LINGER:
1283 if (optlen < sizeof(ling)) {
1284 ret = -EINVAL; /* 1003.1g */
1285 break;
1286 }
1287 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1288 ret = -EFAULT;
1289 break;
1290 }
1291 if (!ling.l_onoff) {
1292 sock_reset_flag(sk, SOCK_LINGER);
1293 } else {
1294 unsigned long t_sec = ling.l_linger;
1295
1296 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1297 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1298 else
1299 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1300 sock_set_flag(sk, SOCK_LINGER);
1301 }
1302 break;
1303
1304 case SO_BSDCOMPAT:
1305 break;
1306
1307 case SO_TIMESTAMP_OLD:
1308 case SO_TIMESTAMP_NEW:
1309 case SO_TIMESTAMPNS_OLD:
1310 case SO_TIMESTAMPNS_NEW:
1311 sock_set_timestamp(sk, optname, valbool);
1312 break;
1313
1314 case SO_TIMESTAMPING_NEW:
1315 case SO_TIMESTAMPING_OLD:
1316 if (optlen == sizeof(timestamping)) {
1317 if (copy_from_sockptr(&timestamping, optval,
1318 sizeof(timestamping))) {
1319 ret = -EFAULT;
1320 break;
1321 }
1322 } else {
1323 memset(&timestamping, 0, sizeof(timestamping));
1324 timestamping.flags = val;
1325 }
1326 ret = sock_set_timestamping(sk, optname, timestamping);
1327 break;
1328
1329 case SO_RCVLOWAT:
1330 {
1331 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1332
1333 if (val < 0)
1334 val = INT_MAX;
1335 if (sock)
1336 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1337 if (set_rcvlowat)
1338 ret = set_rcvlowat(sk, val);
1339 else
1340 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1341 break;
1342 }
1343 case SO_RCVTIMEO_OLD:
1344 case SO_RCVTIMEO_NEW:
1345 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1346 optlen, optname == SO_RCVTIMEO_OLD);
1347 break;
1348
1349 case SO_SNDTIMEO_OLD:
1350 case SO_SNDTIMEO_NEW:
1351 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1352 optlen, optname == SO_SNDTIMEO_OLD);
1353 break;
1354
1355 case SO_ATTACH_FILTER: {
1356 struct sock_fprog fprog;
1357
1358 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1359 if (!ret)
1360 ret = sk_attach_filter(&fprog, sk);
1361 break;
1362 }
1363 case SO_ATTACH_BPF:
1364 ret = -EINVAL;
1365 if (optlen == sizeof(u32)) {
1366 u32 ufd;
1367
1368 ret = -EFAULT;
1369 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1370 break;
1371
1372 ret = sk_attach_bpf(ufd, sk);
1373 }
1374 break;
1375
1376 case SO_ATTACH_REUSEPORT_CBPF: {
1377 struct sock_fprog fprog;
1378
1379 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1380 if (!ret)
1381 ret = sk_reuseport_attach_filter(&fprog, sk);
1382 break;
1383 }
1384 case SO_ATTACH_REUSEPORT_EBPF:
1385 ret = -EINVAL;
1386 if (optlen == sizeof(u32)) {
1387 u32 ufd;
1388
1389 ret = -EFAULT;
1390 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1391 break;
1392
1393 ret = sk_reuseport_attach_bpf(ufd, sk);
1394 }
1395 break;
1396
1397 case SO_DETACH_REUSEPORT_BPF:
1398 ret = reuseport_detach_prog(sk);
1399 break;
1400
1401 case SO_DETACH_FILTER:
1402 ret = sk_detach_filter(sk);
1403 break;
1404
1405 case SO_LOCK_FILTER:
1406 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1407 ret = -EPERM;
1408 else
1409 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1410 break;
1411
1412 case SO_MARK:
1413 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1414 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1415 ret = -EPERM;
1416 break;
1417 }
1418
1419 __sock_set_mark(sk, val);
1420 break;
1421 case SO_RCVMARK:
1422 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1423 break;
1424
1425 case SO_RXQ_OVFL:
1426 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1427 break;
1428
1429 case SO_WIFI_STATUS:
1430 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1431 break;
1432
1433 case SO_PEEK_OFF:
1434 {
1435 int (*set_peek_off)(struct sock *sk, int val);
1436
1437 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1438 if (set_peek_off)
1439 ret = set_peek_off(sk, val);
1440 else
1441 ret = -EOPNOTSUPP;
1442 break;
1443 }
1444
1445 case SO_NOFCS:
1446 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1447 break;
1448
1449 case SO_SELECT_ERR_QUEUE:
1450 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1451 break;
1452
1453
1454 case SO_INCOMING_CPU:
1455 reuseport_update_incoming_cpu(sk, val);
1456 break;
1457
1458 case SO_CNX_ADVICE:
1459 if (val == 1)
1460 dst_negative_advice(sk);
1461 break;
1462
1463 case SO_ZEROCOPY:
1464 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1465 if (!(sk_is_tcp(sk) ||
1466 (sk->sk_type == SOCK_DGRAM &&
1467 sk->sk_protocol == IPPROTO_UDP)))
1468 ret = -EOPNOTSUPP;
1469 } else if (sk->sk_family != PF_RDS) {
1470 ret = -EOPNOTSUPP;
1471 }
1472 if (!ret) {
1473 if (val < 0 || val > 1)
1474 ret = -EINVAL;
1475 else
1476 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1477 }
1478 break;
1479
1480 case SO_TXTIME:
1481 if (optlen != sizeof(struct sock_txtime)) {
1482 ret = -EINVAL;
1483 break;
1484 } else if (copy_from_sockptr(&sk_txtime, optval,
1485 sizeof(struct sock_txtime))) {
1486 ret = -EFAULT;
1487 break;
1488 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1489 ret = -EINVAL;
1490 break;
1491 }
1492 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1493 * scheduler has enough safe guards.
1494 */
1495 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1496 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1497 ret = -EPERM;
1498 break;
1499 }
1500 sock_valbool_flag(sk, SOCK_TXTIME, true);
1501 sk->sk_clockid = sk_txtime.clockid;
1502 sk->sk_txtime_deadline_mode =
1503 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1504 sk->sk_txtime_report_errors =
1505 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1506 break;
1507
1508 case SO_BINDTOIFINDEX:
1509 ret = sock_bindtoindex_locked(sk, val);
1510 break;
1511
1512 case SO_BUF_LOCK:
1513 if (val & ~SOCK_BUF_LOCK_MASK) {
1514 ret = -EINVAL;
1515 break;
1516 }
1517 sk->sk_userlocks = val | (sk->sk_userlocks &
1518 ~SOCK_BUF_LOCK_MASK);
1519 break;
1520
1521 case SO_RESERVE_MEM:
1522 {
1523 int delta;
1524
1525 if (val < 0) {
1526 ret = -EINVAL;
1527 break;
1528 }
1529
1530 delta = val - sk->sk_reserved_mem;
1531 if (delta < 0)
1532 sock_release_reserved_memory(sk, -delta);
1533 else
1534 ret = sock_reserve_memory(sk, delta);
1535 break;
1536 }
1537
1538 default:
1539 ret = -ENOPROTOOPT;
1540 break;
1541 }
1542 sockopt_release_sock(sk);
1543 return ret;
1544 }
1545
1546 int sock_setsockopt(struct socket *sock, int level, int optname,
1547 sockptr_t optval, unsigned int optlen)
1548 {
1549 return sk_setsockopt(sock->sk, level, optname,
1550 optval, optlen);
1551 }
1552 EXPORT_SYMBOL(sock_setsockopt);
1553
1554 static const struct cred *sk_get_peer_cred(struct sock *sk)
1555 {
1556 const struct cred *cred;
1557
1558 spin_lock(&sk->sk_peer_lock);
1559 cred = get_cred(sk->sk_peer_cred);
1560 spin_unlock(&sk->sk_peer_lock);
1561
1562 return cred;
1563 }
1564
1565 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1566 struct ucred *ucred)
1567 {
1568 ucred->pid = pid_vnr(pid);
1569 ucred->uid = ucred->gid = -1;
1570 if (cred) {
1571 struct user_namespace *current_ns = current_user_ns();
1572
1573 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1574 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1575 }
1576 }
1577
1578 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1579 {
1580 struct user_namespace *user_ns = current_user_ns();
1581 int i;
1582
1583 for (i = 0; i < src->ngroups; i++) {
1584 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1585
1586 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1587 return -EFAULT;
1588 }
1589
1590 return 0;
1591 }
1592
1593 int sk_getsockopt(struct sock *sk, int level, int optname,
1594 sockptr_t optval, sockptr_t optlen)
1595 {
1596 struct socket *sock = sk->sk_socket;
1597
1598 union {
1599 int val;
1600 u64 val64;
1601 unsigned long ulval;
1602 struct linger ling;
1603 struct old_timeval32 tm32;
1604 struct __kernel_old_timeval tm;
1605 struct __kernel_sock_timeval stm;
1606 struct sock_txtime txtime;
1607 struct so_timestamping timestamping;
1608 } v;
1609
1610 int lv = sizeof(int);
1611 int len;
1612
1613 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1614 return -EFAULT;
1615 if (len < 0)
1616 return -EINVAL;
1617
1618 memset(&v, 0, sizeof(v));
1619
1620 switch (optname) {
1621 case SO_DEBUG:
1622 v.val = sock_flag(sk, SOCK_DBG);
1623 break;
1624
1625 case SO_DONTROUTE:
1626 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1627 break;
1628
1629 case SO_BROADCAST:
1630 v.val = sock_flag(sk, SOCK_BROADCAST);
1631 break;
1632
1633 case SO_SNDBUF:
1634 v.val = READ_ONCE(sk->sk_sndbuf);
1635 break;
1636
1637 case SO_RCVBUF:
1638 v.val = READ_ONCE(sk->sk_rcvbuf);
1639 break;
1640
1641 case SO_REUSEADDR:
1642 v.val = sk->sk_reuse;
1643 break;
1644
1645 case SO_REUSEPORT:
1646 v.val = sk->sk_reuseport;
1647 break;
1648
1649 case SO_KEEPALIVE:
1650 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1651 break;
1652
1653 case SO_TYPE:
1654 v.val = sk->sk_type;
1655 break;
1656
1657 case SO_PROTOCOL:
1658 v.val = sk->sk_protocol;
1659 break;
1660
1661 case SO_DOMAIN:
1662 v.val = sk->sk_family;
1663 break;
1664
1665 case SO_ERROR:
1666 v.val = -sock_error(sk);
1667 if (v.val == 0)
1668 v.val = xchg(&sk->sk_err_soft, 0);
1669 break;
1670
1671 case SO_OOBINLINE:
1672 v.val = sock_flag(sk, SOCK_URGINLINE);
1673 break;
1674
1675 case SO_NO_CHECK:
1676 v.val = sk->sk_no_check_tx;
1677 break;
1678
1679 case SO_PRIORITY:
1680 v.val = READ_ONCE(sk->sk_priority);
1681 break;
1682
1683 case SO_LINGER:
1684 lv = sizeof(v.ling);
1685 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1686 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1687 break;
1688
1689 case SO_BSDCOMPAT:
1690 break;
1691
1692 case SO_TIMESTAMP_OLD:
1693 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1694 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1695 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1696 break;
1697
1698 case SO_TIMESTAMPNS_OLD:
1699 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1700 break;
1701
1702 case SO_TIMESTAMP_NEW:
1703 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1704 break;
1705
1706 case SO_TIMESTAMPNS_NEW:
1707 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1708 break;
1709
1710 case SO_TIMESTAMPING_OLD:
1711 case SO_TIMESTAMPING_NEW:
1712 lv = sizeof(v.timestamping);
1713 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1714 * returning the flags when they were set through the same option.
1715 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1716 */
1717 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1718 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1719 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1720 }
1721 break;
1722
1723 case SO_RCVTIMEO_OLD:
1724 case SO_RCVTIMEO_NEW:
1725 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1726 SO_RCVTIMEO_OLD == optname);
1727 break;
1728
1729 case SO_SNDTIMEO_OLD:
1730 case SO_SNDTIMEO_NEW:
1731 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1732 SO_SNDTIMEO_OLD == optname);
1733 break;
1734
1735 case SO_RCVLOWAT:
1736 v.val = READ_ONCE(sk->sk_rcvlowat);
1737 break;
1738
1739 case SO_SNDLOWAT:
1740 v.val = 1;
1741 break;
1742
1743 case SO_PASSCRED:
1744 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1745 break;
1746
1747 case SO_PASSPIDFD:
1748 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1749 break;
1750
1751 case SO_PEERCRED:
1752 {
1753 struct ucred peercred;
1754 if (len > sizeof(peercred))
1755 len = sizeof(peercred);
1756
1757 spin_lock(&sk->sk_peer_lock);
1758 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1759 spin_unlock(&sk->sk_peer_lock);
1760
1761 if (copy_to_sockptr(optval, &peercred, len))
1762 return -EFAULT;
1763 goto lenout;
1764 }
1765
1766 case SO_PEERPIDFD:
1767 {
1768 struct pid *peer_pid;
1769 struct file *pidfd_file = NULL;
1770 int pidfd;
1771
1772 if (len > sizeof(pidfd))
1773 len = sizeof(pidfd);
1774
1775 spin_lock(&sk->sk_peer_lock);
1776 peer_pid = get_pid(sk->sk_peer_pid);
1777 spin_unlock(&sk->sk_peer_lock);
1778
1779 if (!peer_pid)
1780 return -ENODATA;
1781
1782 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1783 put_pid(peer_pid);
1784 if (pidfd < 0)
1785 return pidfd;
1786
1787 if (copy_to_sockptr(optval, &pidfd, len) ||
1788 copy_to_sockptr(optlen, &len, sizeof(int))) {
1789 put_unused_fd(pidfd);
1790 fput(pidfd_file);
1791
1792 return -EFAULT;
1793 }
1794
1795 fd_install(pidfd, pidfd_file);
1796 return 0;
1797 }
1798
1799 case SO_PEERGROUPS:
1800 {
1801 const struct cred *cred;
1802 int ret, n;
1803
1804 cred = sk_get_peer_cred(sk);
1805 if (!cred)
1806 return -ENODATA;
1807
1808 n = cred->group_info->ngroups;
1809 if (len < n * sizeof(gid_t)) {
1810 len = n * sizeof(gid_t);
1811 put_cred(cred);
1812 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1813 }
1814 len = n * sizeof(gid_t);
1815
1816 ret = groups_to_user(optval, cred->group_info);
1817 put_cred(cred);
1818 if (ret)
1819 return ret;
1820 goto lenout;
1821 }
1822
1823 case SO_PEERNAME:
1824 {
1825 struct sockaddr_storage address;
1826
1827 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1828 if (lv < 0)
1829 return -ENOTCONN;
1830 if (lv < len)
1831 return -EINVAL;
1832 if (copy_to_sockptr(optval, &address, len))
1833 return -EFAULT;
1834 goto lenout;
1835 }
1836
1837 /* Dubious BSD thing... Probably nobody even uses it, but
1838 * the UNIX standard wants it for whatever reason... -DaveM
1839 */
1840 case SO_ACCEPTCONN:
1841 v.val = sk->sk_state == TCP_LISTEN;
1842 break;
1843
1844 case SO_PASSSEC:
1845 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1846 break;
1847
1848 case SO_PEERSEC:
1849 return security_socket_getpeersec_stream(sock,
1850 optval, optlen, len);
1851
1852 case SO_MARK:
1853 v.val = READ_ONCE(sk->sk_mark);
1854 break;
1855
1856 case SO_RCVMARK:
1857 v.val = sock_flag(sk, SOCK_RCVMARK);
1858 break;
1859
1860 case SO_RXQ_OVFL:
1861 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1862 break;
1863
1864 case SO_WIFI_STATUS:
1865 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1866 break;
1867
1868 case SO_PEEK_OFF:
1869 if (!READ_ONCE(sock->ops)->set_peek_off)
1870 return -EOPNOTSUPP;
1871
1872 v.val = READ_ONCE(sk->sk_peek_off);
1873 break;
1874 case SO_NOFCS:
1875 v.val = sock_flag(sk, SOCK_NOFCS);
1876 break;
1877
1878 case SO_BINDTODEVICE:
1879 return sock_getbindtodevice(sk, optval, optlen, len);
1880
1881 case SO_GET_FILTER:
1882 len = sk_get_filter(sk, optval, len);
1883 if (len < 0)
1884 return len;
1885
1886 goto lenout;
1887
1888 case SO_LOCK_FILTER:
1889 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1890 break;
1891
1892 case SO_BPF_EXTENSIONS:
1893 v.val = bpf_tell_extensions();
1894 break;
1895
1896 case SO_SELECT_ERR_QUEUE:
1897 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1898 break;
1899
1900 #ifdef CONFIG_NET_RX_BUSY_POLL
1901 case SO_BUSY_POLL:
1902 v.val = READ_ONCE(sk->sk_ll_usec);
1903 break;
1904 case SO_PREFER_BUSY_POLL:
1905 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1906 break;
1907 #endif
1908
1909 case SO_MAX_PACING_RATE:
1910 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1911 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1912 lv = sizeof(v.ulval);
1913 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1914 } else {
1915 /* 32bit version */
1916 v.val = min_t(unsigned long, ~0U,
1917 READ_ONCE(sk->sk_max_pacing_rate));
1918 }
1919 break;
1920
1921 case SO_INCOMING_CPU:
1922 v.val = READ_ONCE(sk->sk_incoming_cpu);
1923 break;
1924
1925 case SO_MEMINFO:
1926 {
1927 u32 meminfo[SK_MEMINFO_VARS];
1928
1929 sk_get_meminfo(sk, meminfo);
1930
1931 len = min_t(unsigned int, len, sizeof(meminfo));
1932 if (copy_to_sockptr(optval, &meminfo, len))
1933 return -EFAULT;
1934
1935 goto lenout;
1936 }
1937
1938 #ifdef CONFIG_NET_RX_BUSY_POLL
1939 case SO_INCOMING_NAPI_ID:
1940 v.val = READ_ONCE(sk->sk_napi_id);
1941
1942 /* aggregate non-NAPI IDs down to 0 */
1943 if (v.val < MIN_NAPI_ID)
1944 v.val = 0;
1945
1946 break;
1947 #endif
1948
1949 case SO_COOKIE:
1950 lv = sizeof(u64);
1951 if (len < lv)
1952 return -EINVAL;
1953 v.val64 = sock_gen_cookie(sk);
1954 break;
1955
1956 case SO_ZEROCOPY:
1957 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1958 break;
1959
1960 case SO_TXTIME:
1961 lv = sizeof(v.txtime);
1962 v.txtime.clockid = sk->sk_clockid;
1963 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1964 SOF_TXTIME_DEADLINE_MODE : 0;
1965 v.txtime.flags |= sk->sk_txtime_report_errors ?
1966 SOF_TXTIME_REPORT_ERRORS : 0;
1967 break;
1968
1969 case SO_BINDTOIFINDEX:
1970 v.val = READ_ONCE(sk->sk_bound_dev_if);
1971 break;
1972
1973 case SO_NETNS_COOKIE:
1974 lv = sizeof(u64);
1975 if (len != lv)
1976 return -EINVAL;
1977 v.val64 = sock_net(sk)->net_cookie;
1978 break;
1979
1980 case SO_BUF_LOCK:
1981 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1982 break;
1983
1984 case SO_RESERVE_MEM:
1985 v.val = READ_ONCE(sk->sk_reserved_mem);
1986 break;
1987
1988 case SO_TXREHASH:
1989 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1990 v.val = READ_ONCE(sk->sk_txrehash);
1991 break;
1992
1993 default:
1994 /* We implement the SO_SNDLOWAT etc to not be settable
1995 * (1003.1g 7).
1996 */
1997 return -ENOPROTOOPT;
1998 }
1999
2000 if (len > lv)
2001 len = lv;
2002 if (copy_to_sockptr(optval, &v, len))
2003 return -EFAULT;
2004 lenout:
2005 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2006 return -EFAULT;
2007 return 0;
2008 }
2009
2010 /*
2011 * Initialize an sk_lock.
2012 *
2013 * (We also register the sk_lock with the lock validator.)
2014 */
2015 static inline void sock_lock_init(struct sock *sk)
2016 {
2017 if (sk->sk_kern_sock)
2018 sock_lock_init_class_and_name(
2019 sk,
2020 af_family_kern_slock_key_strings[sk->sk_family],
2021 af_family_kern_slock_keys + sk->sk_family,
2022 af_family_kern_key_strings[sk->sk_family],
2023 af_family_kern_keys + sk->sk_family);
2024 else
2025 sock_lock_init_class_and_name(
2026 sk,
2027 af_family_slock_key_strings[sk->sk_family],
2028 af_family_slock_keys + sk->sk_family,
2029 af_family_key_strings[sk->sk_family],
2030 af_family_keys + sk->sk_family);
2031 }
2032
2033 /*
2034 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2035 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2036 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2037 */
2038 static void sock_copy(struct sock *nsk, const struct sock *osk)
2039 {
2040 const struct proto *prot = READ_ONCE(osk->sk_prot);
2041 #ifdef CONFIG_SECURITY_NETWORK
2042 void *sptr = nsk->sk_security;
2043 #endif
2044
2045 /* If we move sk_tx_queue_mapping out of the private section,
2046 * we must check if sk_tx_queue_clear() is called after
2047 * sock_copy() in sk_clone_lock().
2048 */
2049 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2050 offsetof(struct sock, sk_dontcopy_begin) ||
2051 offsetof(struct sock, sk_tx_queue_mapping) >=
2052 offsetof(struct sock, sk_dontcopy_end));
2053
2054 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2055
2056 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2057 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2058
2059 #ifdef CONFIG_SECURITY_NETWORK
2060 nsk->sk_security = sptr;
2061 security_sk_clone(osk, nsk);
2062 #endif
2063 }
2064
2065 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2066 int family)
2067 {
2068 struct sock *sk;
2069 struct kmem_cache *slab;
2070
2071 slab = prot->slab;
2072 if (slab != NULL) {
2073 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2074 if (!sk)
2075 return sk;
2076 if (want_init_on_alloc(priority))
2077 sk_prot_clear_nulls(sk, prot->obj_size);
2078 } else
2079 sk = kmalloc(prot->obj_size, priority);
2080
2081 if (sk != NULL) {
2082 if (security_sk_alloc(sk, family, priority))
2083 goto out_free;
2084
2085 if (!try_module_get(prot->owner))
2086 goto out_free_sec;
2087 }
2088
2089 return sk;
2090
2091 out_free_sec:
2092 security_sk_free(sk);
2093 out_free:
2094 if (slab != NULL)
2095 kmem_cache_free(slab, sk);
2096 else
2097 kfree(sk);
2098 return NULL;
2099 }
2100
2101 static void sk_prot_free(struct proto *prot, struct sock *sk)
2102 {
2103 struct kmem_cache *slab;
2104 struct module *owner;
2105
2106 owner = prot->owner;
2107 slab = prot->slab;
2108
2109 cgroup_sk_free(&sk->sk_cgrp_data);
2110 mem_cgroup_sk_free(sk);
2111 security_sk_free(sk);
2112 if (slab != NULL)
2113 kmem_cache_free(slab, sk);
2114 else
2115 kfree(sk);
2116 module_put(owner);
2117 }
2118
2119 /**
2120 * sk_alloc - All socket objects are allocated here
2121 * @net: the applicable net namespace
2122 * @family: protocol family
2123 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2124 * @prot: struct proto associated with this new sock instance
2125 * @kern: is this to be a kernel socket?
2126 */
2127 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2128 struct proto *prot, int kern)
2129 {
2130 struct sock *sk;
2131
2132 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2133 if (sk) {
2134 sk->sk_family = family;
2135 /*
2136 * See comment in struct sock definition to understand
2137 * why we need sk_prot_creator -acme
2138 */
2139 sk->sk_prot = sk->sk_prot_creator = prot;
2140 sk->sk_kern_sock = kern;
2141 sock_lock_init(sk);
2142 sk->sk_net_refcnt = kern ? 0 : 1;
2143 if (likely(sk->sk_net_refcnt)) {
2144 get_net_track(net, &sk->ns_tracker, priority);
2145 sock_inuse_add(net, 1);
2146 } else {
2147 __netns_tracker_alloc(net, &sk->ns_tracker,
2148 false, priority);
2149 }
2150
2151 sock_net_set(sk, net);
2152 refcount_set(&sk->sk_wmem_alloc, 1);
2153
2154 mem_cgroup_sk_alloc(sk);
2155 cgroup_sk_alloc(&sk->sk_cgrp_data);
2156 sock_update_classid(&sk->sk_cgrp_data);
2157 sock_update_netprioidx(&sk->sk_cgrp_data);
2158 sk_tx_queue_clear(sk);
2159 }
2160
2161 return sk;
2162 }
2163 EXPORT_SYMBOL(sk_alloc);
2164
2165 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2166 * grace period. This is the case for UDP sockets and TCP listeners.
2167 */
2168 static void __sk_destruct(struct rcu_head *head)
2169 {
2170 struct sock *sk = container_of(head, struct sock, sk_rcu);
2171 struct sk_filter *filter;
2172
2173 if (sk->sk_destruct)
2174 sk->sk_destruct(sk);
2175
2176 filter = rcu_dereference_check(sk->sk_filter,
2177 refcount_read(&sk->sk_wmem_alloc) == 0);
2178 if (filter) {
2179 sk_filter_uncharge(sk, filter);
2180 RCU_INIT_POINTER(sk->sk_filter, NULL);
2181 }
2182
2183 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2184
2185 #ifdef CONFIG_BPF_SYSCALL
2186 bpf_sk_storage_free(sk);
2187 #endif
2188
2189 if (atomic_read(&sk->sk_omem_alloc))
2190 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2191 __func__, atomic_read(&sk->sk_omem_alloc));
2192
2193 if (sk->sk_frag.page) {
2194 put_page(sk->sk_frag.page);
2195 sk->sk_frag.page = NULL;
2196 }
2197
2198 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2199 put_cred(sk->sk_peer_cred);
2200 put_pid(sk->sk_peer_pid);
2201
2202 if (likely(sk->sk_net_refcnt))
2203 put_net_track(sock_net(sk), &sk->ns_tracker);
2204 else
2205 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2206
2207 sk_prot_free(sk->sk_prot_creator, sk);
2208 }
2209
2210 void sk_destruct(struct sock *sk)
2211 {
2212 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2213
2214 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2215 reuseport_detach_sock(sk);
2216 use_call_rcu = true;
2217 }
2218
2219 if (use_call_rcu)
2220 call_rcu(&sk->sk_rcu, __sk_destruct);
2221 else
2222 __sk_destruct(&sk->sk_rcu);
2223 }
2224
2225 static void __sk_free(struct sock *sk)
2226 {
2227 if (likely(sk->sk_net_refcnt))
2228 sock_inuse_add(sock_net(sk), -1);
2229
2230 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2231 sock_diag_broadcast_destroy(sk);
2232 else
2233 sk_destruct(sk);
2234 }
2235
2236 void sk_free(struct sock *sk)
2237 {
2238 /*
2239 * We subtract one from sk_wmem_alloc and can know if
2240 * some packets are still in some tx queue.
2241 * If not null, sock_wfree() will call __sk_free(sk) later
2242 */
2243 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2244 __sk_free(sk);
2245 }
2246 EXPORT_SYMBOL(sk_free);
2247
2248 static void sk_init_common(struct sock *sk)
2249 {
2250 skb_queue_head_init(&sk->sk_receive_queue);
2251 skb_queue_head_init(&sk->sk_write_queue);
2252 skb_queue_head_init(&sk->sk_error_queue);
2253
2254 rwlock_init(&sk->sk_callback_lock);
2255 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2256 af_rlock_keys + sk->sk_family,
2257 af_family_rlock_key_strings[sk->sk_family]);
2258 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2259 af_wlock_keys + sk->sk_family,
2260 af_family_wlock_key_strings[sk->sk_family]);
2261 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2262 af_elock_keys + sk->sk_family,
2263 af_family_elock_key_strings[sk->sk_family]);
2264 lockdep_set_class_and_name(&sk->sk_callback_lock,
2265 af_callback_keys + sk->sk_family,
2266 af_family_clock_key_strings[sk->sk_family]);
2267 }
2268
2269 /**
2270 * sk_clone_lock - clone a socket, and lock its clone
2271 * @sk: the socket to clone
2272 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2273 *
2274 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2275 */
2276 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2277 {
2278 struct proto *prot = READ_ONCE(sk->sk_prot);
2279 struct sk_filter *filter;
2280 bool is_charged = true;
2281 struct sock *newsk;
2282
2283 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2284 if (!newsk)
2285 goto out;
2286
2287 sock_copy(newsk, sk);
2288
2289 newsk->sk_prot_creator = prot;
2290
2291 /* SANITY */
2292 if (likely(newsk->sk_net_refcnt)) {
2293 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2294 sock_inuse_add(sock_net(newsk), 1);
2295 } else {
2296 /* Kernel sockets are not elevating the struct net refcount.
2297 * Instead, use a tracker to more easily detect if a layer
2298 * is not properly dismantling its kernel sockets at netns
2299 * destroy time.
2300 */
2301 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2302 false, priority);
2303 }
2304 sk_node_init(&newsk->sk_node);
2305 sock_lock_init(newsk);
2306 bh_lock_sock(newsk);
2307 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2308 newsk->sk_backlog.len = 0;
2309
2310 atomic_set(&newsk->sk_rmem_alloc, 0);
2311
2312 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2313 refcount_set(&newsk->sk_wmem_alloc, 1);
2314
2315 atomic_set(&newsk->sk_omem_alloc, 0);
2316 sk_init_common(newsk);
2317
2318 newsk->sk_dst_cache = NULL;
2319 newsk->sk_dst_pending_confirm = 0;
2320 newsk->sk_wmem_queued = 0;
2321 newsk->sk_forward_alloc = 0;
2322 newsk->sk_reserved_mem = 0;
2323 atomic_set(&newsk->sk_drops, 0);
2324 newsk->sk_send_head = NULL;
2325 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2326 atomic_set(&newsk->sk_zckey, 0);
2327
2328 sock_reset_flag(newsk, SOCK_DONE);
2329
2330 /* sk->sk_memcg will be populated at accept() time */
2331 newsk->sk_memcg = NULL;
2332
2333 cgroup_sk_clone(&newsk->sk_cgrp_data);
2334
2335 rcu_read_lock();
2336 filter = rcu_dereference(sk->sk_filter);
2337 if (filter != NULL)
2338 /* though it's an empty new sock, the charging may fail
2339 * if sysctl_optmem_max was changed between creation of
2340 * original socket and cloning
2341 */
2342 is_charged = sk_filter_charge(newsk, filter);
2343 RCU_INIT_POINTER(newsk->sk_filter, filter);
2344 rcu_read_unlock();
2345
2346 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2347 /* We need to make sure that we don't uncharge the new
2348 * socket if we couldn't charge it in the first place
2349 * as otherwise we uncharge the parent's filter.
2350 */
2351 if (!is_charged)
2352 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2353 sk_free_unlock_clone(newsk);
2354 newsk = NULL;
2355 goto out;
2356 }
2357 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2358
2359 if (bpf_sk_storage_clone(sk, newsk)) {
2360 sk_free_unlock_clone(newsk);
2361 newsk = NULL;
2362 goto out;
2363 }
2364
2365 /* Clear sk_user_data if parent had the pointer tagged
2366 * as not suitable for copying when cloning.
2367 */
2368 if (sk_user_data_is_nocopy(newsk))
2369 newsk->sk_user_data = NULL;
2370
2371 newsk->sk_err = 0;
2372 newsk->sk_err_soft = 0;
2373 newsk->sk_priority = 0;
2374 newsk->sk_incoming_cpu = raw_smp_processor_id();
2375
2376 /* Before updating sk_refcnt, we must commit prior changes to memory
2377 * (Documentation/RCU/rculist_nulls.rst for details)
2378 */
2379 smp_wmb();
2380 refcount_set(&newsk->sk_refcnt, 2);
2381
2382 sk_set_socket(newsk, NULL);
2383 sk_tx_queue_clear(newsk);
2384 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2385
2386 if (newsk->sk_prot->sockets_allocated)
2387 sk_sockets_allocated_inc(newsk);
2388
2389 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2390 net_enable_timestamp();
2391 out:
2392 return newsk;
2393 }
2394 EXPORT_SYMBOL_GPL(sk_clone_lock);
2395
2396 void sk_free_unlock_clone(struct sock *sk)
2397 {
2398 /* It is still raw copy of parent, so invalidate
2399 * destructor and make plain sk_free() */
2400 sk->sk_destruct = NULL;
2401 bh_unlock_sock(sk);
2402 sk_free(sk);
2403 }
2404 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2405
2406 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2407 {
2408 bool is_ipv6 = false;
2409 u32 max_size;
2410
2411 #if IS_ENABLED(CONFIG_IPV6)
2412 is_ipv6 = (sk->sk_family == AF_INET6 &&
2413 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2414 #endif
2415 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2416 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2417 READ_ONCE(dst->dev->gso_ipv4_max_size);
2418 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2419 max_size = GSO_LEGACY_MAX_SIZE;
2420
2421 return max_size - (MAX_TCP_HEADER + 1);
2422 }
2423
2424 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2425 {
2426 u32 max_segs = 1;
2427
2428 sk->sk_route_caps = dst->dev->features;
2429 if (sk_is_tcp(sk))
2430 sk->sk_route_caps |= NETIF_F_GSO;
2431 if (sk->sk_route_caps & NETIF_F_GSO)
2432 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2433 if (unlikely(sk->sk_gso_disabled))
2434 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2435 if (sk_can_gso(sk)) {
2436 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2437 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2438 } else {
2439 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2440 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2441 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2442 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2443 }
2444 }
2445 sk->sk_gso_max_segs = max_segs;
2446 sk_dst_set(sk, dst);
2447 }
2448 EXPORT_SYMBOL_GPL(sk_setup_caps);
2449
2450 /*
2451 * Simple resource managers for sockets.
2452 */
2453
2454
2455 /*
2456 * Write buffer destructor automatically called from kfree_skb.
2457 */
2458 void sock_wfree(struct sk_buff *skb)
2459 {
2460 struct sock *sk = skb->sk;
2461 unsigned int len = skb->truesize;
2462 bool free;
2463
2464 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2465 if (sock_flag(sk, SOCK_RCU_FREE) &&
2466 sk->sk_write_space == sock_def_write_space) {
2467 rcu_read_lock();
2468 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2469 sock_def_write_space_wfree(sk);
2470 rcu_read_unlock();
2471 if (unlikely(free))
2472 __sk_free(sk);
2473 return;
2474 }
2475
2476 /*
2477 * Keep a reference on sk_wmem_alloc, this will be released
2478 * after sk_write_space() call
2479 */
2480 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2481 sk->sk_write_space(sk);
2482 len = 1;
2483 }
2484 /*
2485 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2486 * could not do because of in-flight packets
2487 */
2488 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2489 __sk_free(sk);
2490 }
2491 EXPORT_SYMBOL(sock_wfree);
2492
2493 /* This variant of sock_wfree() is used by TCP,
2494 * since it sets SOCK_USE_WRITE_QUEUE.
2495 */
2496 void __sock_wfree(struct sk_buff *skb)
2497 {
2498 struct sock *sk = skb->sk;
2499
2500 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2501 __sk_free(sk);
2502 }
2503
2504 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2505 {
2506 skb_orphan(skb);
2507 skb->sk = sk;
2508 #ifdef CONFIG_INET
2509 if (unlikely(!sk_fullsock(sk))) {
2510 skb->destructor = sock_edemux;
2511 sock_hold(sk);
2512 return;
2513 }
2514 #endif
2515 skb->destructor = sock_wfree;
2516 skb_set_hash_from_sk(skb, sk);
2517 /*
2518 * We used to take a refcount on sk, but following operation
2519 * is enough to guarantee sk_free() wont free this sock until
2520 * all in-flight packets are completed
2521 */
2522 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2523 }
2524 EXPORT_SYMBOL(skb_set_owner_w);
2525
2526 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2527 {
2528 #ifdef CONFIG_TLS_DEVICE
2529 /* Drivers depend on in-order delivery for crypto offload,
2530 * partial orphan breaks out-of-order-OK logic.
2531 */
2532 if (skb->decrypted)
2533 return false;
2534 #endif
2535 return (skb->destructor == sock_wfree ||
2536 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2537 }
2538
2539 /* This helper is used by netem, as it can hold packets in its
2540 * delay queue. We want to allow the owner socket to send more
2541 * packets, as if they were already TX completed by a typical driver.
2542 * But we also want to keep skb->sk set because some packet schedulers
2543 * rely on it (sch_fq for example).
2544 */
2545 void skb_orphan_partial(struct sk_buff *skb)
2546 {
2547 if (skb_is_tcp_pure_ack(skb))
2548 return;
2549
2550 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2551 return;
2552
2553 skb_orphan(skb);
2554 }
2555 EXPORT_SYMBOL(skb_orphan_partial);
2556
2557 /*
2558 * Read buffer destructor automatically called from kfree_skb.
2559 */
2560 void sock_rfree(struct sk_buff *skb)
2561 {
2562 struct sock *sk = skb->sk;
2563 unsigned int len = skb->truesize;
2564
2565 atomic_sub(len, &sk->sk_rmem_alloc);
2566 sk_mem_uncharge(sk, len);
2567 }
2568 EXPORT_SYMBOL(sock_rfree);
2569
2570 /*
2571 * Buffer destructor for skbs that are not used directly in read or write
2572 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2573 */
2574 void sock_efree(struct sk_buff *skb)
2575 {
2576 sock_put(skb->sk);
2577 }
2578 EXPORT_SYMBOL(sock_efree);
2579
2580 /* Buffer destructor for prefetch/receive path where reference count may
2581 * not be held, e.g. for listen sockets.
2582 */
2583 #ifdef CONFIG_INET
2584 void sock_pfree(struct sk_buff *skb)
2585 {
2586 struct sock *sk = skb->sk;
2587
2588 if (!sk_is_refcounted(sk))
2589 return;
2590
2591 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2592 inet_reqsk(sk)->rsk_listener = NULL;
2593 reqsk_free(inet_reqsk(sk));
2594 return;
2595 }
2596
2597 sock_gen_put(sk);
2598 }
2599 EXPORT_SYMBOL(sock_pfree);
2600 #endif /* CONFIG_INET */
2601
2602 kuid_t sock_i_uid(struct sock *sk)
2603 {
2604 kuid_t uid;
2605
2606 read_lock_bh(&sk->sk_callback_lock);
2607 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2608 read_unlock_bh(&sk->sk_callback_lock);
2609 return uid;
2610 }
2611 EXPORT_SYMBOL(sock_i_uid);
2612
2613 unsigned long __sock_i_ino(struct sock *sk)
2614 {
2615 unsigned long ino;
2616
2617 read_lock(&sk->sk_callback_lock);
2618 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2619 read_unlock(&sk->sk_callback_lock);
2620 return ino;
2621 }
2622 EXPORT_SYMBOL(__sock_i_ino);
2623
2624 unsigned long sock_i_ino(struct sock *sk)
2625 {
2626 unsigned long ino;
2627
2628 local_bh_disable();
2629 ino = __sock_i_ino(sk);
2630 local_bh_enable();
2631 return ino;
2632 }
2633 EXPORT_SYMBOL(sock_i_ino);
2634
2635 /*
2636 * Allocate a skb from the socket's send buffer.
2637 */
2638 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2639 gfp_t priority)
2640 {
2641 if (force ||
2642 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2643 struct sk_buff *skb = alloc_skb(size, priority);
2644
2645 if (skb) {
2646 skb_set_owner_w(skb, sk);
2647 return skb;
2648 }
2649 }
2650 return NULL;
2651 }
2652 EXPORT_SYMBOL(sock_wmalloc);
2653
2654 static void sock_ofree(struct sk_buff *skb)
2655 {
2656 struct sock *sk = skb->sk;
2657
2658 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2659 }
2660
2661 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2662 gfp_t priority)
2663 {
2664 struct sk_buff *skb;
2665
2666 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2667 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2668 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2669 return NULL;
2670
2671 skb = alloc_skb(size, priority);
2672 if (!skb)
2673 return NULL;
2674
2675 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2676 skb->sk = sk;
2677 skb->destructor = sock_ofree;
2678 return skb;
2679 }
2680
2681 /*
2682 * Allocate a memory block from the socket's option memory buffer.
2683 */
2684 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2685 {
2686 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2687
2688 if ((unsigned int)size <= optmem_max &&
2689 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2690 void *mem;
2691 /* First do the add, to avoid the race if kmalloc
2692 * might sleep.
2693 */
2694 atomic_add(size, &sk->sk_omem_alloc);
2695 mem = kmalloc(size, priority);
2696 if (mem)
2697 return mem;
2698 atomic_sub(size, &sk->sk_omem_alloc);
2699 }
2700 return NULL;
2701 }
2702 EXPORT_SYMBOL(sock_kmalloc);
2703
2704 /* Free an option memory block. Note, we actually want the inline
2705 * here as this allows gcc to detect the nullify and fold away the
2706 * condition entirely.
2707 */
2708 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2709 const bool nullify)
2710 {
2711 if (WARN_ON_ONCE(!mem))
2712 return;
2713 if (nullify)
2714 kfree_sensitive(mem);
2715 else
2716 kfree(mem);
2717 atomic_sub(size, &sk->sk_omem_alloc);
2718 }
2719
2720 void sock_kfree_s(struct sock *sk, void *mem, int size)
2721 {
2722 __sock_kfree_s(sk, mem, size, false);
2723 }
2724 EXPORT_SYMBOL(sock_kfree_s);
2725
2726 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2727 {
2728 __sock_kfree_s(sk, mem, size, true);
2729 }
2730 EXPORT_SYMBOL(sock_kzfree_s);
2731
2732 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2733 I think, these locks should be removed for datagram sockets.
2734 */
2735 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2736 {
2737 DEFINE_WAIT(wait);
2738
2739 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2740 for (;;) {
2741 if (!timeo)
2742 break;
2743 if (signal_pending(current))
2744 break;
2745 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2746 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2747 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2748 break;
2749 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2750 break;
2751 if (READ_ONCE(sk->sk_err))
2752 break;
2753 timeo = schedule_timeout(timeo);
2754 }
2755 finish_wait(sk_sleep(sk), &wait);
2756 return timeo;
2757 }
2758
2759
2760 /*
2761 * Generic send/receive buffer handlers
2762 */
2763
2764 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2765 unsigned long data_len, int noblock,
2766 int *errcode, int max_page_order)
2767 {
2768 struct sk_buff *skb;
2769 long timeo;
2770 int err;
2771
2772 timeo = sock_sndtimeo(sk, noblock);
2773 for (;;) {
2774 err = sock_error(sk);
2775 if (err != 0)
2776 goto failure;
2777
2778 err = -EPIPE;
2779 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2780 goto failure;
2781
2782 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2783 break;
2784
2785 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2786 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2787 err = -EAGAIN;
2788 if (!timeo)
2789 goto failure;
2790 if (signal_pending(current))
2791 goto interrupted;
2792 timeo = sock_wait_for_wmem(sk, timeo);
2793 }
2794 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2795 errcode, sk->sk_allocation);
2796 if (skb)
2797 skb_set_owner_w(skb, sk);
2798 return skb;
2799
2800 interrupted:
2801 err = sock_intr_errno(timeo);
2802 failure:
2803 *errcode = err;
2804 return NULL;
2805 }
2806 EXPORT_SYMBOL(sock_alloc_send_pskb);
2807
2808 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2809 struct sockcm_cookie *sockc)
2810 {
2811 u32 tsflags;
2812
2813 switch (cmsg->cmsg_type) {
2814 case SO_MARK:
2815 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2816 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2817 return -EPERM;
2818 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2819 return -EINVAL;
2820 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2821 break;
2822 case SO_TIMESTAMPING_OLD:
2823 case SO_TIMESTAMPING_NEW:
2824 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2825 return -EINVAL;
2826
2827 tsflags = *(u32 *)CMSG_DATA(cmsg);
2828 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2829 return -EINVAL;
2830
2831 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2832 sockc->tsflags |= tsflags;
2833 break;
2834 case SCM_TXTIME:
2835 if (!sock_flag(sk, SOCK_TXTIME))
2836 return -EINVAL;
2837 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2838 return -EINVAL;
2839 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2840 break;
2841 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2842 case SCM_RIGHTS:
2843 case SCM_CREDENTIALS:
2844 break;
2845 default:
2846 return -EINVAL;
2847 }
2848 return 0;
2849 }
2850 EXPORT_SYMBOL(__sock_cmsg_send);
2851
2852 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2853 struct sockcm_cookie *sockc)
2854 {
2855 struct cmsghdr *cmsg;
2856 int ret;
2857
2858 for_each_cmsghdr(cmsg, msg) {
2859 if (!CMSG_OK(msg, cmsg))
2860 return -EINVAL;
2861 if (cmsg->cmsg_level != SOL_SOCKET)
2862 continue;
2863 ret = __sock_cmsg_send(sk, cmsg, sockc);
2864 if (ret)
2865 return ret;
2866 }
2867 return 0;
2868 }
2869 EXPORT_SYMBOL(sock_cmsg_send);
2870
2871 static void sk_enter_memory_pressure(struct sock *sk)
2872 {
2873 if (!sk->sk_prot->enter_memory_pressure)
2874 return;
2875
2876 sk->sk_prot->enter_memory_pressure(sk);
2877 }
2878
2879 static void sk_leave_memory_pressure(struct sock *sk)
2880 {
2881 if (sk->sk_prot->leave_memory_pressure) {
2882 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2883 tcp_leave_memory_pressure, sk);
2884 } else {
2885 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2886
2887 if (memory_pressure && READ_ONCE(*memory_pressure))
2888 WRITE_ONCE(*memory_pressure, 0);
2889 }
2890 }
2891
2892 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2893
2894 /**
2895 * skb_page_frag_refill - check that a page_frag contains enough room
2896 * @sz: minimum size of the fragment we want to get
2897 * @pfrag: pointer to page_frag
2898 * @gfp: priority for memory allocation
2899 *
2900 * Note: While this allocator tries to use high order pages, there is
2901 * no guarantee that allocations succeed. Therefore, @sz MUST be
2902 * less or equal than PAGE_SIZE.
2903 */
2904 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2905 {
2906 if (pfrag->page) {
2907 if (page_ref_count(pfrag->page) == 1) {
2908 pfrag->offset = 0;
2909 return true;
2910 }
2911 if (pfrag->offset + sz <= pfrag->size)
2912 return true;
2913 put_page(pfrag->page);
2914 }
2915
2916 pfrag->offset = 0;
2917 if (SKB_FRAG_PAGE_ORDER &&
2918 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2919 /* Avoid direct reclaim but allow kswapd to wake */
2920 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2921 __GFP_COMP | __GFP_NOWARN |
2922 __GFP_NORETRY,
2923 SKB_FRAG_PAGE_ORDER);
2924 if (likely(pfrag->page)) {
2925 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2926 return true;
2927 }
2928 }
2929 pfrag->page = alloc_page(gfp);
2930 if (likely(pfrag->page)) {
2931 pfrag->size = PAGE_SIZE;
2932 return true;
2933 }
2934 return false;
2935 }
2936 EXPORT_SYMBOL(skb_page_frag_refill);
2937
2938 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2939 {
2940 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2941 return true;
2942
2943 sk_enter_memory_pressure(sk);
2944 sk_stream_moderate_sndbuf(sk);
2945 return false;
2946 }
2947 EXPORT_SYMBOL(sk_page_frag_refill);
2948
2949 void __lock_sock(struct sock *sk)
2950 __releases(&sk->sk_lock.slock)
2951 __acquires(&sk->sk_lock.slock)
2952 {
2953 DEFINE_WAIT(wait);
2954
2955 for (;;) {
2956 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2957 TASK_UNINTERRUPTIBLE);
2958 spin_unlock_bh(&sk->sk_lock.slock);
2959 schedule();
2960 spin_lock_bh(&sk->sk_lock.slock);
2961 if (!sock_owned_by_user(sk))
2962 break;
2963 }
2964 finish_wait(&sk->sk_lock.wq, &wait);
2965 }
2966
2967 void __release_sock(struct sock *sk)
2968 __releases(&sk->sk_lock.slock)
2969 __acquires(&sk->sk_lock.slock)
2970 {
2971 struct sk_buff *skb, *next;
2972
2973 while ((skb = sk->sk_backlog.head) != NULL) {
2974 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2975
2976 spin_unlock_bh(&sk->sk_lock.slock);
2977
2978 do {
2979 next = skb->next;
2980 prefetch(next);
2981 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2982 skb_mark_not_on_list(skb);
2983 sk_backlog_rcv(sk, skb);
2984
2985 cond_resched();
2986
2987 skb = next;
2988 } while (skb != NULL);
2989
2990 spin_lock_bh(&sk->sk_lock.slock);
2991 }
2992
2993 /*
2994 * Doing the zeroing here guarantee we can not loop forever
2995 * while a wild producer attempts to flood us.
2996 */
2997 sk->sk_backlog.len = 0;
2998 }
2999
3000 void __sk_flush_backlog(struct sock *sk)
3001 {
3002 spin_lock_bh(&sk->sk_lock.slock);
3003 __release_sock(sk);
3004
3005 if (sk->sk_prot->release_cb)
3006 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3007 tcp_release_cb, sk);
3008
3009 spin_unlock_bh(&sk->sk_lock.slock);
3010 }
3011 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3012
3013 /**
3014 * sk_wait_data - wait for data to arrive at sk_receive_queue
3015 * @sk: sock to wait on
3016 * @timeo: for how long
3017 * @skb: last skb seen on sk_receive_queue
3018 *
3019 * Now socket state including sk->sk_err is changed only under lock,
3020 * hence we may omit checks after joining wait queue.
3021 * We check receive queue before schedule() only as optimization;
3022 * it is very likely that release_sock() added new data.
3023 */
3024 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3025 {
3026 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3027 int rc;
3028
3029 add_wait_queue(sk_sleep(sk), &wait);
3030 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3031 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3032 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3033 remove_wait_queue(sk_sleep(sk), &wait);
3034 return rc;
3035 }
3036 EXPORT_SYMBOL(sk_wait_data);
3037
3038 /**
3039 * __sk_mem_raise_allocated - increase memory_allocated
3040 * @sk: socket
3041 * @size: memory size to allocate
3042 * @amt: pages to allocate
3043 * @kind: allocation type
3044 *
3045 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3046 *
3047 * Unlike the globally shared limits among the sockets under same protocol,
3048 * consuming the budget of a memcg won't have direct effect on other ones.
3049 * So be optimistic about memcg's tolerance, and leave the callers to decide
3050 * whether or not to raise allocated through sk_under_memory_pressure() or
3051 * its variants.
3052 */
3053 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3054 {
3055 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3056 struct proto *prot = sk->sk_prot;
3057 bool charged = false;
3058 long allocated;
3059
3060 sk_memory_allocated_add(sk, amt);
3061 allocated = sk_memory_allocated(sk);
3062
3063 if (memcg) {
3064 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3065 goto suppress_allocation;
3066 charged = true;
3067 }
3068
3069 /* Under limit. */
3070 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3071 sk_leave_memory_pressure(sk);
3072 return 1;
3073 }
3074
3075 /* Under pressure. */
3076 if (allocated > sk_prot_mem_limits(sk, 1))
3077 sk_enter_memory_pressure(sk);
3078
3079 /* Over hard limit. */
3080 if (allocated > sk_prot_mem_limits(sk, 2))
3081 goto suppress_allocation;
3082
3083 /* Guarantee minimum buffer size under pressure (either global
3084 * or memcg) to make sure features described in RFC 7323 (TCP
3085 * Extensions for High Performance) work properly.
3086 *
3087 * This rule does NOT stand when exceeds global or memcg's hard
3088 * limit, or else a DoS attack can be taken place by spawning
3089 * lots of sockets whose usage are under minimum buffer size.
3090 */
3091 if (kind == SK_MEM_RECV) {
3092 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3093 return 1;
3094
3095 } else { /* SK_MEM_SEND */
3096 int wmem0 = sk_get_wmem0(sk, prot);
3097
3098 if (sk->sk_type == SOCK_STREAM) {
3099 if (sk->sk_wmem_queued < wmem0)
3100 return 1;
3101 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3102 return 1;
3103 }
3104 }
3105
3106 if (sk_has_memory_pressure(sk)) {
3107 u64 alloc;
3108
3109 /* The following 'average' heuristic is within the
3110 * scope of global accounting, so it only makes
3111 * sense for global memory pressure.
3112 */
3113 if (!sk_under_global_memory_pressure(sk))
3114 return 1;
3115
3116 /* Try to be fair among all the sockets under global
3117 * pressure by allowing the ones that below average
3118 * usage to raise.
3119 */
3120 alloc = sk_sockets_allocated_read_positive(sk);
3121 if (sk_prot_mem_limits(sk, 2) > alloc *
3122 sk_mem_pages(sk->sk_wmem_queued +
3123 atomic_read(&sk->sk_rmem_alloc) +
3124 sk->sk_forward_alloc))
3125 return 1;
3126 }
3127
3128 suppress_allocation:
3129
3130 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3131 sk_stream_moderate_sndbuf(sk);
3132
3133 /* Fail only if socket is _under_ its sndbuf.
3134 * In this case we cannot block, so that we have to fail.
3135 */
3136 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3137 /* Force charge with __GFP_NOFAIL */
3138 if (memcg && !charged) {
3139 mem_cgroup_charge_skmem(memcg, amt,
3140 gfp_memcg_charge() | __GFP_NOFAIL);
3141 }
3142 return 1;
3143 }
3144 }
3145
3146 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3147 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3148
3149 sk_memory_allocated_sub(sk, amt);
3150
3151 if (charged)
3152 mem_cgroup_uncharge_skmem(memcg, amt);
3153
3154 return 0;
3155 }
3156
3157 /**
3158 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3159 * @sk: socket
3160 * @size: memory size to allocate
3161 * @kind: allocation type
3162 *
3163 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3164 * rmem allocation. This function assumes that protocols which have
3165 * memory_pressure use sk_wmem_queued as write buffer accounting.
3166 */
3167 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3168 {
3169 int ret, amt = sk_mem_pages(size);
3170
3171 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3172 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3173 if (!ret)
3174 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3175 return ret;
3176 }
3177 EXPORT_SYMBOL(__sk_mem_schedule);
3178
3179 /**
3180 * __sk_mem_reduce_allocated - reclaim memory_allocated
3181 * @sk: socket
3182 * @amount: number of quanta
3183 *
3184 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3185 */
3186 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3187 {
3188 sk_memory_allocated_sub(sk, amount);
3189
3190 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3191 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3192
3193 if (sk_under_global_memory_pressure(sk) &&
3194 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3195 sk_leave_memory_pressure(sk);
3196 }
3197
3198 /**
3199 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3200 * @sk: socket
3201 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3202 */
3203 void __sk_mem_reclaim(struct sock *sk, int amount)
3204 {
3205 amount >>= PAGE_SHIFT;
3206 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3207 __sk_mem_reduce_allocated(sk, amount);
3208 }
3209 EXPORT_SYMBOL(__sk_mem_reclaim);
3210
3211 int sk_set_peek_off(struct sock *sk, int val)
3212 {
3213 WRITE_ONCE(sk->sk_peek_off, val);
3214 return 0;
3215 }
3216 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3217
3218 /*
3219 * Set of default routines for initialising struct proto_ops when
3220 * the protocol does not support a particular function. In certain
3221 * cases where it makes no sense for a protocol to have a "do nothing"
3222 * function, some default processing is provided.
3223 */
3224
3225 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3226 {
3227 return -EOPNOTSUPP;
3228 }
3229 EXPORT_SYMBOL(sock_no_bind);
3230
3231 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3232 int len, int flags)
3233 {
3234 return -EOPNOTSUPP;
3235 }
3236 EXPORT_SYMBOL(sock_no_connect);
3237
3238 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3239 {
3240 return -EOPNOTSUPP;
3241 }
3242 EXPORT_SYMBOL(sock_no_socketpair);
3243
3244 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3245 bool kern)
3246 {
3247 return -EOPNOTSUPP;
3248 }
3249 EXPORT_SYMBOL(sock_no_accept);
3250
3251 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3252 int peer)
3253 {
3254 return -EOPNOTSUPP;
3255 }
3256 EXPORT_SYMBOL(sock_no_getname);
3257
3258 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3259 {
3260 return -EOPNOTSUPP;
3261 }
3262 EXPORT_SYMBOL(sock_no_ioctl);
3263
3264 int sock_no_listen(struct socket *sock, int backlog)
3265 {
3266 return -EOPNOTSUPP;
3267 }
3268 EXPORT_SYMBOL(sock_no_listen);
3269
3270 int sock_no_shutdown(struct socket *sock, int how)
3271 {
3272 return -EOPNOTSUPP;
3273 }
3274 EXPORT_SYMBOL(sock_no_shutdown);
3275
3276 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3277 {
3278 return -EOPNOTSUPP;
3279 }
3280 EXPORT_SYMBOL(sock_no_sendmsg);
3281
3282 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3283 {
3284 return -EOPNOTSUPP;
3285 }
3286 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3287
3288 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3289 int flags)
3290 {
3291 return -EOPNOTSUPP;
3292 }
3293 EXPORT_SYMBOL(sock_no_recvmsg);
3294
3295 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3296 {
3297 /* Mirror missing mmap method error code */
3298 return -ENODEV;
3299 }
3300 EXPORT_SYMBOL(sock_no_mmap);
3301
3302 /*
3303 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3304 * various sock-based usage counts.
3305 */
3306 void __receive_sock(struct file *file)
3307 {
3308 struct socket *sock;
3309
3310 sock = sock_from_file(file);
3311 if (sock) {
3312 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3313 sock_update_classid(&sock->sk->sk_cgrp_data);
3314 }
3315 }
3316
3317 /*
3318 * Default Socket Callbacks
3319 */
3320
3321 static void sock_def_wakeup(struct sock *sk)
3322 {
3323 struct socket_wq *wq;
3324
3325 rcu_read_lock();
3326 wq = rcu_dereference(sk->sk_wq);
3327 if (skwq_has_sleeper(wq))
3328 wake_up_interruptible_all(&wq->wait);
3329 rcu_read_unlock();
3330 }
3331
3332 static void sock_def_error_report(struct sock *sk)
3333 {
3334 struct socket_wq *wq;
3335
3336 rcu_read_lock();
3337 wq = rcu_dereference(sk->sk_wq);
3338 if (skwq_has_sleeper(wq))
3339 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3340 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3341 rcu_read_unlock();
3342 }
3343
3344 void sock_def_readable(struct sock *sk)
3345 {
3346 struct socket_wq *wq;
3347
3348 trace_sk_data_ready(sk);
3349
3350 rcu_read_lock();
3351 wq = rcu_dereference(sk->sk_wq);
3352 if (skwq_has_sleeper(wq))
3353 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3354 EPOLLRDNORM | EPOLLRDBAND);
3355 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3356 rcu_read_unlock();
3357 }
3358
3359 static void sock_def_write_space(struct sock *sk)
3360 {
3361 struct socket_wq *wq;
3362
3363 rcu_read_lock();
3364
3365 /* Do not wake up a writer until he can make "significant"
3366 * progress. --DaveM
3367 */
3368 if (sock_writeable(sk)) {
3369 wq = rcu_dereference(sk->sk_wq);
3370 if (skwq_has_sleeper(wq))
3371 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3372 EPOLLWRNORM | EPOLLWRBAND);
3373
3374 /* Should agree with poll, otherwise some programs break */
3375 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3376 }
3377
3378 rcu_read_unlock();
3379 }
3380
3381 /* An optimised version of sock_def_write_space(), should only be called
3382 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3383 * ->sk_wmem_alloc.
3384 */
3385 static void sock_def_write_space_wfree(struct sock *sk)
3386 {
3387 /* Do not wake up a writer until he can make "significant"
3388 * progress. --DaveM
3389 */
3390 if (sock_writeable(sk)) {
3391 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3392
3393 /* rely on refcount_sub from sock_wfree() */
3394 smp_mb__after_atomic();
3395 if (wq && waitqueue_active(&wq->wait))
3396 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3397 EPOLLWRNORM | EPOLLWRBAND);
3398
3399 /* Should agree with poll, otherwise some programs break */
3400 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3401 }
3402 }
3403
3404 static void sock_def_destruct(struct sock *sk)
3405 {
3406 }
3407
3408 void sk_send_sigurg(struct sock *sk)
3409 {
3410 if (sk->sk_socket && sk->sk_socket->file)
3411 if (send_sigurg(&sk->sk_socket->file->f_owner))
3412 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3413 }
3414 EXPORT_SYMBOL(sk_send_sigurg);
3415
3416 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3417 unsigned long expires)
3418 {
3419 if (!mod_timer(timer, expires))
3420 sock_hold(sk);
3421 }
3422 EXPORT_SYMBOL(sk_reset_timer);
3423
3424 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3425 {
3426 if (del_timer(timer))
3427 __sock_put(sk);
3428 }
3429 EXPORT_SYMBOL(sk_stop_timer);
3430
3431 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3432 {
3433 if (del_timer_sync(timer))
3434 __sock_put(sk);
3435 }
3436 EXPORT_SYMBOL(sk_stop_timer_sync);
3437
3438 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3439 {
3440 sk_init_common(sk);
3441 sk->sk_send_head = NULL;
3442
3443 timer_setup(&sk->sk_timer, NULL, 0);
3444
3445 sk->sk_allocation = GFP_KERNEL;
3446 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3447 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3448 sk->sk_state = TCP_CLOSE;
3449 sk->sk_use_task_frag = true;
3450 sk_set_socket(sk, sock);
3451
3452 sock_set_flag(sk, SOCK_ZAPPED);
3453
3454 if (sock) {
3455 sk->sk_type = sock->type;
3456 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3457 sock->sk = sk;
3458 } else {
3459 RCU_INIT_POINTER(sk->sk_wq, NULL);
3460 }
3461 sk->sk_uid = uid;
3462
3463 rwlock_init(&sk->sk_callback_lock);
3464 if (sk->sk_kern_sock)
3465 lockdep_set_class_and_name(
3466 &sk->sk_callback_lock,
3467 af_kern_callback_keys + sk->sk_family,
3468 af_family_kern_clock_key_strings[sk->sk_family]);
3469 else
3470 lockdep_set_class_and_name(
3471 &sk->sk_callback_lock,
3472 af_callback_keys + sk->sk_family,
3473 af_family_clock_key_strings[sk->sk_family]);
3474
3475 sk->sk_state_change = sock_def_wakeup;
3476 sk->sk_data_ready = sock_def_readable;
3477 sk->sk_write_space = sock_def_write_space;
3478 sk->sk_error_report = sock_def_error_report;
3479 sk->sk_destruct = sock_def_destruct;
3480
3481 sk->sk_frag.page = NULL;
3482 sk->sk_frag.offset = 0;
3483 sk->sk_peek_off = -1;
3484
3485 sk->sk_peer_pid = NULL;
3486 sk->sk_peer_cred = NULL;
3487 spin_lock_init(&sk->sk_peer_lock);
3488
3489 sk->sk_write_pending = 0;
3490 sk->sk_rcvlowat = 1;
3491 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3492 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3493
3494 sk->sk_stamp = SK_DEFAULT_STAMP;
3495 #if BITS_PER_LONG==32
3496 seqlock_init(&sk->sk_stamp_seq);
3497 #endif
3498 atomic_set(&sk->sk_zckey, 0);
3499
3500 #ifdef CONFIG_NET_RX_BUSY_POLL
3501 sk->sk_napi_id = 0;
3502 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3503 #endif
3504
3505 sk->sk_max_pacing_rate = ~0UL;
3506 sk->sk_pacing_rate = ~0UL;
3507 WRITE_ONCE(sk->sk_pacing_shift, 10);
3508 sk->sk_incoming_cpu = -1;
3509
3510 sk_rx_queue_clear(sk);
3511 /*
3512 * Before updating sk_refcnt, we must commit prior changes to memory
3513 * (Documentation/RCU/rculist_nulls.rst for details)
3514 */
3515 smp_wmb();
3516 refcount_set(&sk->sk_refcnt, 1);
3517 atomic_set(&sk->sk_drops, 0);
3518 }
3519 EXPORT_SYMBOL(sock_init_data_uid);
3520
3521 void sock_init_data(struct socket *sock, struct sock *sk)
3522 {
3523 kuid_t uid = sock ?
3524 SOCK_INODE(sock)->i_uid :
3525 make_kuid(sock_net(sk)->user_ns, 0);
3526
3527 sock_init_data_uid(sock, sk, uid);
3528 }
3529 EXPORT_SYMBOL(sock_init_data);
3530
3531 void lock_sock_nested(struct sock *sk, int subclass)
3532 {
3533 /* The sk_lock has mutex_lock() semantics here. */
3534 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3535
3536 might_sleep();
3537 spin_lock_bh(&sk->sk_lock.slock);
3538 if (sock_owned_by_user_nocheck(sk))
3539 __lock_sock(sk);
3540 sk->sk_lock.owned = 1;
3541 spin_unlock_bh(&sk->sk_lock.slock);
3542 }
3543 EXPORT_SYMBOL(lock_sock_nested);
3544
3545 void release_sock(struct sock *sk)
3546 {
3547 spin_lock_bh(&sk->sk_lock.slock);
3548 if (sk->sk_backlog.tail)
3549 __release_sock(sk);
3550
3551 if (sk->sk_prot->release_cb)
3552 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3553 tcp_release_cb, sk);
3554
3555 sock_release_ownership(sk);
3556 if (waitqueue_active(&sk->sk_lock.wq))
3557 wake_up(&sk->sk_lock.wq);
3558 spin_unlock_bh(&sk->sk_lock.slock);
3559 }
3560 EXPORT_SYMBOL(release_sock);
3561
3562 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3563 {
3564 might_sleep();
3565 spin_lock_bh(&sk->sk_lock.slock);
3566
3567 if (!sock_owned_by_user_nocheck(sk)) {
3568 /*
3569 * Fast path return with bottom halves disabled and
3570 * sock::sk_lock.slock held.
3571 *
3572 * The 'mutex' is not contended and holding
3573 * sock::sk_lock.slock prevents all other lockers to
3574 * proceed so the corresponding unlock_sock_fast() can
3575 * avoid the slow path of release_sock() completely and
3576 * just release slock.
3577 *
3578 * From a semantical POV this is equivalent to 'acquiring'
3579 * the 'mutex', hence the corresponding lockdep
3580 * mutex_release() has to happen in the fast path of
3581 * unlock_sock_fast().
3582 */
3583 return false;
3584 }
3585
3586 __lock_sock(sk);
3587 sk->sk_lock.owned = 1;
3588 __acquire(&sk->sk_lock.slock);
3589 spin_unlock_bh(&sk->sk_lock.slock);
3590 return true;
3591 }
3592 EXPORT_SYMBOL(__lock_sock_fast);
3593
3594 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3595 bool timeval, bool time32)
3596 {
3597 struct sock *sk = sock->sk;
3598 struct timespec64 ts;
3599
3600 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3601 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3602 if (ts.tv_sec == -1)
3603 return -ENOENT;
3604 if (ts.tv_sec == 0) {
3605 ktime_t kt = ktime_get_real();
3606 sock_write_timestamp(sk, kt);
3607 ts = ktime_to_timespec64(kt);
3608 }
3609
3610 if (timeval)
3611 ts.tv_nsec /= 1000;
3612
3613 #ifdef CONFIG_COMPAT_32BIT_TIME
3614 if (time32)
3615 return put_old_timespec32(&ts, userstamp);
3616 #endif
3617 #ifdef CONFIG_SPARC64
3618 /* beware of padding in sparc64 timeval */
3619 if (timeval && !in_compat_syscall()) {
3620 struct __kernel_old_timeval __user tv = {
3621 .tv_sec = ts.tv_sec,
3622 .tv_usec = ts.tv_nsec,
3623 };
3624 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3625 return -EFAULT;
3626 return 0;
3627 }
3628 #endif
3629 return put_timespec64(&ts, userstamp);
3630 }
3631 EXPORT_SYMBOL(sock_gettstamp);
3632
3633 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3634 {
3635 if (!sock_flag(sk, flag)) {
3636 unsigned long previous_flags = sk->sk_flags;
3637
3638 sock_set_flag(sk, flag);
3639 /*
3640 * we just set one of the two flags which require net
3641 * time stamping, but time stamping might have been on
3642 * already because of the other one
3643 */
3644 if (sock_needs_netstamp(sk) &&
3645 !(previous_flags & SK_FLAGS_TIMESTAMP))
3646 net_enable_timestamp();
3647 }
3648 }
3649
3650 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3651 int level, int type)
3652 {
3653 struct sock_exterr_skb *serr;
3654 struct sk_buff *skb;
3655 int copied, err;
3656
3657 err = -EAGAIN;
3658 skb = sock_dequeue_err_skb(sk);
3659 if (skb == NULL)
3660 goto out;
3661
3662 copied = skb->len;
3663 if (copied > len) {
3664 msg->msg_flags |= MSG_TRUNC;
3665 copied = len;
3666 }
3667 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3668 if (err)
3669 goto out_free_skb;
3670
3671 sock_recv_timestamp(msg, sk, skb);
3672
3673 serr = SKB_EXT_ERR(skb);
3674 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3675
3676 msg->msg_flags |= MSG_ERRQUEUE;
3677 err = copied;
3678
3679 out_free_skb:
3680 kfree_skb(skb);
3681 out:
3682 return err;
3683 }
3684 EXPORT_SYMBOL(sock_recv_errqueue);
3685
3686 /*
3687 * Get a socket option on an socket.
3688 *
3689 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3690 * asynchronous errors should be reported by getsockopt. We assume
3691 * this means if you specify SO_ERROR (otherwise whats the point of it).
3692 */
3693 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3694 char __user *optval, int __user *optlen)
3695 {
3696 struct sock *sk = sock->sk;
3697
3698 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3699 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3700 }
3701 EXPORT_SYMBOL(sock_common_getsockopt);
3702
3703 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3704 int flags)
3705 {
3706 struct sock *sk = sock->sk;
3707 int addr_len = 0;
3708 int err;
3709
3710 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3711 if (err >= 0)
3712 msg->msg_namelen = addr_len;
3713 return err;
3714 }
3715 EXPORT_SYMBOL(sock_common_recvmsg);
3716
3717 /*
3718 * Set socket options on an inet socket.
3719 */
3720 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3721 sockptr_t optval, unsigned int optlen)
3722 {
3723 struct sock *sk = sock->sk;
3724
3725 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3726 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3727 }
3728 EXPORT_SYMBOL(sock_common_setsockopt);
3729
3730 void sk_common_release(struct sock *sk)
3731 {
3732 if (sk->sk_prot->destroy)
3733 sk->sk_prot->destroy(sk);
3734
3735 /*
3736 * Observation: when sk_common_release is called, processes have
3737 * no access to socket. But net still has.
3738 * Step one, detach it from networking:
3739 *
3740 * A. Remove from hash tables.
3741 */
3742
3743 sk->sk_prot->unhash(sk);
3744
3745 /*
3746 * In this point socket cannot receive new packets, but it is possible
3747 * that some packets are in flight because some CPU runs receiver and
3748 * did hash table lookup before we unhashed socket. They will achieve
3749 * receive queue and will be purged by socket destructor.
3750 *
3751 * Also we still have packets pending on receive queue and probably,
3752 * our own packets waiting in device queues. sock_destroy will drain
3753 * receive queue, but transmitted packets will delay socket destruction
3754 * until the last reference will be released.
3755 */
3756
3757 sock_orphan(sk);
3758
3759 xfrm_sk_free_policy(sk);
3760
3761 sock_put(sk);
3762 }
3763 EXPORT_SYMBOL(sk_common_release);
3764
3765 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3766 {
3767 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3768
3769 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3770 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3771 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3772 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3773 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3774 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3775 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3776 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3777 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3778 }
3779
3780 #ifdef CONFIG_PROC_FS
3781 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3782
3783 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3784 {
3785 int cpu, idx = prot->inuse_idx;
3786 int res = 0;
3787
3788 for_each_possible_cpu(cpu)
3789 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3790
3791 return res >= 0 ? res : 0;
3792 }
3793 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3794
3795 int sock_inuse_get(struct net *net)
3796 {
3797 int cpu, res = 0;
3798
3799 for_each_possible_cpu(cpu)
3800 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3801
3802 return res;
3803 }
3804
3805 EXPORT_SYMBOL_GPL(sock_inuse_get);
3806
3807 static int __net_init sock_inuse_init_net(struct net *net)
3808 {
3809 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3810 if (net->core.prot_inuse == NULL)
3811 return -ENOMEM;
3812 return 0;
3813 }
3814
3815 static void __net_exit sock_inuse_exit_net(struct net *net)
3816 {
3817 free_percpu(net->core.prot_inuse);
3818 }
3819
3820 static struct pernet_operations net_inuse_ops = {
3821 .init = sock_inuse_init_net,
3822 .exit = sock_inuse_exit_net,
3823 };
3824
3825 static __init int net_inuse_init(void)
3826 {
3827 if (register_pernet_subsys(&net_inuse_ops))
3828 panic("Cannot initialize net inuse counters");
3829
3830 return 0;
3831 }
3832
3833 core_initcall(net_inuse_init);
3834
3835 static int assign_proto_idx(struct proto *prot)
3836 {
3837 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3838
3839 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3840 pr_err("PROTO_INUSE_NR exhausted\n");
3841 return -ENOSPC;
3842 }
3843
3844 set_bit(prot->inuse_idx, proto_inuse_idx);
3845 return 0;
3846 }
3847
3848 static void release_proto_idx(struct proto *prot)
3849 {
3850 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3851 clear_bit(prot->inuse_idx, proto_inuse_idx);
3852 }
3853 #else
3854 static inline int assign_proto_idx(struct proto *prot)
3855 {
3856 return 0;
3857 }
3858
3859 static inline void release_proto_idx(struct proto *prot)
3860 {
3861 }
3862
3863 #endif
3864
3865 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3866 {
3867 if (!twsk_prot)
3868 return;
3869 kfree(twsk_prot->twsk_slab_name);
3870 twsk_prot->twsk_slab_name = NULL;
3871 kmem_cache_destroy(twsk_prot->twsk_slab);
3872 twsk_prot->twsk_slab = NULL;
3873 }
3874
3875 static int tw_prot_init(const struct proto *prot)
3876 {
3877 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3878
3879 if (!twsk_prot)
3880 return 0;
3881
3882 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3883 prot->name);
3884 if (!twsk_prot->twsk_slab_name)
3885 return -ENOMEM;
3886
3887 twsk_prot->twsk_slab =
3888 kmem_cache_create(twsk_prot->twsk_slab_name,
3889 twsk_prot->twsk_obj_size, 0,
3890 SLAB_ACCOUNT | prot->slab_flags,
3891 NULL);
3892 if (!twsk_prot->twsk_slab) {
3893 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3894 prot->name);
3895 return -ENOMEM;
3896 }
3897
3898 return 0;
3899 }
3900
3901 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3902 {
3903 if (!rsk_prot)
3904 return;
3905 kfree(rsk_prot->slab_name);
3906 rsk_prot->slab_name = NULL;
3907 kmem_cache_destroy(rsk_prot->slab);
3908 rsk_prot->slab = NULL;
3909 }
3910
3911 static int req_prot_init(const struct proto *prot)
3912 {
3913 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3914
3915 if (!rsk_prot)
3916 return 0;
3917
3918 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3919 prot->name);
3920 if (!rsk_prot->slab_name)
3921 return -ENOMEM;
3922
3923 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3924 rsk_prot->obj_size, 0,
3925 SLAB_ACCOUNT | prot->slab_flags,
3926 NULL);
3927
3928 if (!rsk_prot->slab) {
3929 pr_crit("%s: Can't create request sock SLAB cache!\n",
3930 prot->name);
3931 return -ENOMEM;
3932 }
3933 return 0;
3934 }
3935
3936 int proto_register(struct proto *prot, int alloc_slab)
3937 {
3938 int ret = -ENOBUFS;
3939
3940 if (prot->memory_allocated && !prot->sysctl_mem) {
3941 pr_err("%s: missing sysctl_mem\n", prot->name);
3942 return -EINVAL;
3943 }
3944 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3945 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3946 return -EINVAL;
3947 }
3948 if (alloc_slab) {
3949 prot->slab = kmem_cache_create_usercopy(prot->name,
3950 prot->obj_size, 0,
3951 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3952 prot->slab_flags,
3953 prot->useroffset, prot->usersize,
3954 NULL);
3955
3956 if (prot->slab == NULL) {
3957 pr_crit("%s: Can't create sock SLAB cache!\n",
3958 prot->name);
3959 goto out;
3960 }
3961
3962 if (req_prot_init(prot))
3963 goto out_free_request_sock_slab;
3964
3965 if (tw_prot_init(prot))
3966 goto out_free_timewait_sock_slab;
3967 }
3968
3969 mutex_lock(&proto_list_mutex);
3970 ret = assign_proto_idx(prot);
3971 if (ret) {
3972 mutex_unlock(&proto_list_mutex);
3973 goto out_free_timewait_sock_slab;
3974 }
3975 list_add(&prot->node, &proto_list);
3976 mutex_unlock(&proto_list_mutex);
3977 return ret;
3978
3979 out_free_timewait_sock_slab:
3980 if (alloc_slab)
3981 tw_prot_cleanup(prot->twsk_prot);
3982 out_free_request_sock_slab:
3983 if (alloc_slab) {
3984 req_prot_cleanup(prot->rsk_prot);
3985
3986 kmem_cache_destroy(prot->slab);
3987 prot->slab = NULL;
3988 }
3989 out:
3990 return ret;
3991 }
3992 EXPORT_SYMBOL(proto_register);
3993
3994 void proto_unregister(struct proto *prot)
3995 {
3996 mutex_lock(&proto_list_mutex);
3997 release_proto_idx(prot);
3998 list_del(&prot->node);
3999 mutex_unlock(&proto_list_mutex);
4000
4001 kmem_cache_destroy(prot->slab);
4002 prot->slab = NULL;
4003
4004 req_prot_cleanup(prot->rsk_prot);
4005 tw_prot_cleanup(prot->twsk_prot);
4006 }
4007 EXPORT_SYMBOL(proto_unregister);
4008
4009 int sock_load_diag_module(int family, int protocol)
4010 {
4011 if (!protocol) {
4012 if (!sock_is_registered(family))
4013 return -ENOENT;
4014
4015 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4016 NETLINK_SOCK_DIAG, family);
4017 }
4018
4019 #ifdef CONFIG_INET
4020 if (family == AF_INET &&
4021 protocol != IPPROTO_RAW &&
4022 protocol < MAX_INET_PROTOS &&
4023 !rcu_access_pointer(inet_protos[protocol]))
4024 return -ENOENT;
4025 #endif
4026
4027 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4028 NETLINK_SOCK_DIAG, family, protocol);
4029 }
4030 EXPORT_SYMBOL(sock_load_diag_module);
4031
4032 #ifdef CONFIG_PROC_FS
4033 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4034 __acquires(proto_list_mutex)
4035 {
4036 mutex_lock(&proto_list_mutex);
4037 return seq_list_start_head(&proto_list, *pos);
4038 }
4039
4040 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4041 {
4042 return seq_list_next(v, &proto_list, pos);
4043 }
4044
4045 static void proto_seq_stop(struct seq_file *seq, void *v)
4046 __releases(proto_list_mutex)
4047 {
4048 mutex_unlock(&proto_list_mutex);
4049 }
4050
4051 static char proto_method_implemented(const void *method)
4052 {
4053 return method == NULL ? 'n' : 'y';
4054 }
4055 static long sock_prot_memory_allocated(struct proto *proto)
4056 {
4057 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4058 }
4059
4060 static const char *sock_prot_memory_pressure(struct proto *proto)
4061 {
4062 return proto->memory_pressure != NULL ?
4063 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4064 }
4065
4066 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4067 {
4068
4069 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4070 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4071 proto->name,
4072 proto->obj_size,
4073 sock_prot_inuse_get(seq_file_net(seq), proto),
4074 sock_prot_memory_allocated(proto),
4075 sock_prot_memory_pressure(proto),
4076 proto->max_header,
4077 proto->slab == NULL ? "no" : "yes",
4078 module_name(proto->owner),
4079 proto_method_implemented(proto->close),
4080 proto_method_implemented(proto->connect),
4081 proto_method_implemented(proto->disconnect),
4082 proto_method_implemented(proto->accept),
4083 proto_method_implemented(proto->ioctl),
4084 proto_method_implemented(proto->init),
4085 proto_method_implemented(proto->destroy),
4086 proto_method_implemented(proto->shutdown),
4087 proto_method_implemented(proto->setsockopt),
4088 proto_method_implemented(proto->getsockopt),
4089 proto_method_implemented(proto->sendmsg),
4090 proto_method_implemented(proto->recvmsg),
4091 proto_method_implemented(proto->bind),
4092 proto_method_implemented(proto->backlog_rcv),
4093 proto_method_implemented(proto->hash),
4094 proto_method_implemented(proto->unhash),
4095 proto_method_implemented(proto->get_port),
4096 proto_method_implemented(proto->enter_memory_pressure));
4097 }
4098
4099 static int proto_seq_show(struct seq_file *seq, void *v)
4100 {
4101 if (v == &proto_list)
4102 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4103 "protocol",
4104 "size",
4105 "sockets",
4106 "memory",
4107 "press",
4108 "maxhdr",
4109 "slab",
4110 "module",
4111 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4112 else
4113 proto_seq_printf(seq, list_entry(v, struct proto, node));
4114 return 0;
4115 }
4116
4117 static const struct seq_operations proto_seq_ops = {
4118 .start = proto_seq_start,
4119 .next = proto_seq_next,
4120 .stop = proto_seq_stop,
4121 .show = proto_seq_show,
4122 };
4123
4124 static __net_init int proto_init_net(struct net *net)
4125 {
4126 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4127 sizeof(struct seq_net_private)))
4128 return -ENOMEM;
4129
4130 return 0;
4131 }
4132
4133 static __net_exit void proto_exit_net(struct net *net)
4134 {
4135 remove_proc_entry("protocols", net->proc_net);
4136 }
4137
4138
4139 static __net_initdata struct pernet_operations proto_net_ops = {
4140 .init = proto_init_net,
4141 .exit = proto_exit_net,
4142 };
4143
4144 static int __init proto_init(void)
4145 {
4146 return register_pernet_subsys(&proto_net_ops);
4147 }
4148
4149 subsys_initcall(proto_init);
4150
4151 #endif /* PROC_FS */
4152
4153 #ifdef CONFIG_NET_RX_BUSY_POLL
4154 bool sk_busy_loop_end(void *p, unsigned long start_time)
4155 {
4156 struct sock *sk = p;
4157
4158 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4159 return true;
4160
4161 if (sk_is_udp(sk) &&
4162 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4163 return true;
4164
4165 return sk_busy_loop_timeout(sk, start_time);
4166 }
4167 EXPORT_SYMBOL(sk_busy_loop_end);
4168 #endif /* CONFIG_NET_RX_BUSY_POLL */
4169
4170 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4171 {
4172 if (!sk->sk_prot->bind_add)
4173 return -EOPNOTSUPP;
4174 return sk->sk_prot->bind_add(sk, addr, addr_len);
4175 }
4176 EXPORT_SYMBOL(sock_bind_add);
4177
4178 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4179 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4180 void __user *arg, void *karg, size_t size)
4181 {
4182 int ret;
4183
4184 if (copy_from_user(karg, arg, size))
4185 return -EFAULT;
4186
4187 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4188 if (ret)
4189 return ret;
4190
4191 if (copy_to_user(arg, karg, size))
4192 return -EFAULT;
4193
4194 return 0;
4195 }
4196 EXPORT_SYMBOL(sock_ioctl_inout);
4197
4198 /* This is the most common ioctl prep function, where the result (4 bytes) is
4199 * copied back to userspace if the ioctl() returns successfully. No input is
4200 * copied from userspace as input argument.
4201 */
4202 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4203 {
4204 int ret, karg = 0;
4205
4206 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4207 if (ret)
4208 return ret;
4209
4210 return put_user(karg, (int __user *)arg);
4211 }
4212
4213 /* A wrapper around sock ioctls, which copies the data from userspace
4214 * (depending on the protocol/ioctl), and copies back the result to userspace.
4215 * The main motivation for this function is to pass kernel memory to the
4216 * protocol ioctl callbacks, instead of userspace memory.
4217 */
4218 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4219 {
4220 int rc = 1;
4221
4222 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4223 rc = ipmr_sk_ioctl(sk, cmd, arg);
4224 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4225 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4226 else if (sk_is_phonet(sk))
4227 rc = phonet_sk_ioctl(sk, cmd, arg);
4228
4229 /* If ioctl was processed, returns its value */
4230 if (rc <= 0)
4231 return rc;
4232
4233 /* Otherwise call the default handler */
4234 return sock_ioctl_out(sk, cmd, arg);
4235 }
4236 EXPORT_SYMBOL(sk_ioctl);
A mirror of Linus' kernel repository