?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 6.14-rc1
[linux.git] / net / core / sock.c
blobeae2ae70a2e03df370d8ef7750a7bb13cc3b8d8f
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88 #include <linux/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
94 #include <linux/in.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
121
122 #include <linux/uaccess.h>
123
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <linux/skbuff_ref.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <net/proto_memory.h>
132 #include <linux/net_tstamp.h>
133 #include <net/xfrm.h>
134 #include <linux/ipsec.h>
135 #include <net/cls_cgroup.h>
136 #include <net/netprio_cgroup.h>
137 #include <linux/sock_diag.h>
138
139 #include <linux/filter.h>
140 #include <net/sock_reuseport.h>
141 #include <net/bpf_sk_storage.h>
142
143 #include <trace/events/sock.h>
144
145 #include <net/tcp.h>
146 #include <net/busy_poll.h>
147 #include <net/phonet/phonet.h>
148
149 #include <linux/ethtool.h>
150
151 #include "dev.h"
152
153 static DEFINE_MUTEX(proto_list_mutex);
154 static LIST_HEAD(proto_list);
155
156 static void sock_def_write_space_wfree(struct sock *sk);
157 static void sock_def_write_space(struct sock *sk);
158
159 /**
160 * sk_ns_capable - General socket capability test
161 * @sk: Socket to use a capability on or through
162 * @user_ns: The user namespace of the capability to use
163 * @cap: The capability to use
164 *
165 * Test to see if the opener of the socket had when the socket was
166 * created and the current process has the capability @cap in the user
167 * namespace @user_ns.
168 */
169 bool sk_ns_capable(const struct sock *sk,
170 struct user_namespace *user_ns, int cap)
171 {
172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
173 ns_capable(user_ns, cap);
174 }
175 EXPORT_SYMBOL(sk_ns_capable);
176
177 /**
178 * sk_capable - Socket global capability test
179 * @sk: Socket to use a capability on or through
180 * @cap: The global capability to use
181 *
182 * Test to see if the opener of the socket had when the socket was
183 * created and the current process has the capability @cap in all user
184 * namespaces.
185 */
186 bool sk_capable(const struct sock *sk, int cap)
187 {
188 return sk_ns_capable(sk, &init_user_ns, cap);
189 }
190 EXPORT_SYMBOL(sk_capable);
191
192 /**
193 * sk_net_capable - Network namespace socket capability test
194 * @sk: Socket to use a capability on or through
195 * @cap: The capability to use
196 *
197 * Test to see if the opener of the socket had when the socket was created
198 * and the current process has the capability @cap over the network namespace
199 * the socket is a member of.
200 */
201 bool sk_net_capable(const struct sock *sk, int cap)
202 {
203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
204 }
205 EXPORT_SYMBOL(sk_net_capable);
206
207 /*
208 * Each address family might have different locking rules, so we have
209 * one slock key per address family and separate keys for internal and
210 * userspace sockets.
211 */
212 static struct lock_class_key af_family_keys[AF_MAX];
213 static struct lock_class_key af_family_kern_keys[AF_MAX];
214 static struct lock_class_key af_family_slock_keys[AF_MAX];
215 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
216
217 /*
218 * Make lock validator output more readable. (we pre-construct these
219 * strings build-time, so that runtime initialization of socket
220 * locks is fast):
221 */
222
223 #define _sock_locks(x) \
224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
233 x "27" , x "28" , x "AF_CAN" , \
234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 x "AF_MCTP" , \
240 x "AF_MAX"
241
242 static const char *const af_family_key_strings[AF_MAX+1] = {
243 _sock_locks("sk_lock-")
244 };
245 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("slock-")
247 };
248 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("clock-")
250 };
251
252 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
253 _sock_locks("k-sk_lock-")
254 };
255 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-slock-")
257 };
258 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
259 _sock_locks("k-clock-")
260 };
261 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 _sock_locks("rlock-")
263 };
264 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
265 _sock_locks("wlock-")
266 };
267 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
268 _sock_locks("elock-")
269 };
270
271 /*
272 * sk_callback_lock and sk queues locking rules are per-address-family,
273 * so split the lock classes by using a per-AF key:
274 */
275 static struct lock_class_key af_callback_keys[AF_MAX];
276 static struct lock_class_key af_rlock_keys[AF_MAX];
277 static struct lock_class_key af_wlock_keys[AF_MAX];
278 static struct lock_class_key af_elock_keys[AF_MAX];
279 static struct lock_class_key af_kern_callback_keys[AF_MAX];
280
281 /* Run time adjustable parameters. */
282 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
283 EXPORT_SYMBOL(sysctl_wmem_max);
284 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
285 EXPORT_SYMBOL(sysctl_rmem_max);
286 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
287 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
288
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 sk_set_prio_allowed(const struct sock *sk, int val)
458 {
459 return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) ||
460 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
461 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN));
462 }
463
464 static bool sock_needs_netstamp(const struct sock *sk)
465 {
466 switch (sk->sk_family) {
467 case AF_UNSPEC:
468 case AF_UNIX:
469 return false;
470 default:
471 return true;
472 }
473 }
474
475 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
476 {
477 if (sk->sk_flags & flags) {
478 sk->sk_flags &= ~flags;
479 if (sock_needs_netstamp(sk) &&
480 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
481 net_disable_timestamp();
482 }
483 }
484
485
486 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
487 {
488 unsigned long flags;
489 struct sk_buff_head *list = &sk->sk_receive_queue;
490
491 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
492 atomic_inc(&sk->sk_drops);
493 trace_sock_rcvqueue_full(sk, skb);
494 return -ENOMEM;
495 }
496
497 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
498 atomic_inc(&sk->sk_drops);
499 return -ENOBUFS;
500 }
501
502 skb->dev = NULL;
503 skb_set_owner_r(skb, sk);
504
505 /* we escape from rcu protected region, make sure we dont leak
506 * a norefcounted dst
507 */
508 skb_dst_force(skb);
509
510 spin_lock_irqsave(&list->lock, flags);
511 sock_skb_set_dropcount(sk, skb);
512 __skb_queue_tail(list, skb);
513 spin_unlock_irqrestore(&list->lock, flags);
514
515 if (!sock_flag(sk, SOCK_DEAD))
516 sk->sk_data_ready(sk);
517 return 0;
518 }
519 EXPORT_SYMBOL(__sock_queue_rcv_skb);
520
521 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
522 enum skb_drop_reason *reason)
523 {
524 enum skb_drop_reason drop_reason;
525 int err;
526
527 err = sk_filter(sk, skb);
528 if (err) {
529 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
530 goto out;
531 }
532 err = __sock_queue_rcv_skb(sk, skb);
533 switch (err) {
534 case -ENOMEM:
535 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
536 break;
537 case -ENOBUFS:
538 drop_reason = SKB_DROP_REASON_PROTO_MEM;
539 break;
540 default:
541 drop_reason = SKB_NOT_DROPPED_YET;
542 break;
543 }
544 out:
545 if (reason)
546 *reason = drop_reason;
547 return err;
548 }
549 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
550
551 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
552 const int nested, unsigned int trim_cap, bool refcounted)
553 {
554 int rc = NET_RX_SUCCESS;
555
556 if (sk_filter_trim_cap(sk, skb, trim_cap))
557 goto discard_and_relse;
558
559 skb->dev = NULL;
560
561 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
562 atomic_inc(&sk->sk_drops);
563 goto discard_and_relse;
564 }
565 if (nested)
566 bh_lock_sock_nested(sk);
567 else
568 bh_lock_sock(sk);
569 if (!sock_owned_by_user(sk)) {
570 /*
571 * trylock + unlock semantics:
572 */
573 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
574
575 rc = sk_backlog_rcv(sk, skb);
576
577 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
578 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
579 bh_unlock_sock(sk);
580 atomic_inc(&sk->sk_drops);
581 goto discard_and_relse;
582 }
583
584 bh_unlock_sock(sk);
585 out:
586 if (refcounted)
587 sock_put(sk);
588 return rc;
589 discard_and_relse:
590 kfree_skb(skb);
591 goto out;
592 }
593 EXPORT_SYMBOL(__sk_receive_skb);
594
595 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
596 u32));
597 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
598 u32));
599 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
600 {
601 struct dst_entry *dst = __sk_dst_get(sk);
602
603 if (dst && dst->obsolete &&
604 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
605 dst, cookie) == NULL) {
606 sk_tx_queue_clear(sk);
607 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
608 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
609 dst_release(dst);
610 return NULL;
611 }
612
613 return dst;
614 }
615 EXPORT_SYMBOL(__sk_dst_check);
616
617 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
618 {
619 struct dst_entry *dst = sk_dst_get(sk);
620
621 if (dst && dst->obsolete &&
622 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
623 dst, cookie) == NULL) {
624 sk_dst_reset(sk);
625 dst_release(dst);
626 return NULL;
627 }
628
629 return dst;
630 }
631 EXPORT_SYMBOL(sk_dst_check);
632
633 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
634 {
635 int ret = -ENOPROTOOPT;
636 #ifdef CONFIG_NETDEVICES
637 struct net *net = sock_net(sk);
638
639 /* Sorry... */
640 ret = -EPERM;
641 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
642 goto out;
643
644 ret = -EINVAL;
645 if (ifindex < 0)
646 goto out;
647
648 /* Paired with all READ_ONCE() done locklessly. */
649 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
650
651 if (sk->sk_prot->rehash)
652 sk->sk_prot->rehash(sk);
653 sk_dst_reset(sk);
654
655 ret = 0;
656
657 out:
658 #endif
659
660 return ret;
661 }
662
663 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
664 {
665 int ret;
666
667 if (lock_sk)
668 lock_sock(sk);
669 ret = sock_bindtoindex_locked(sk, ifindex);
670 if (lock_sk)
671 release_sock(sk);
672
673 return ret;
674 }
675 EXPORT_SYMBOL(sock_bindtoindex);
676
677 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
678 {
679 int ret = -ENOPROTOOPT;
680 #ifdef CONFIG_NETDEVICES
681 struct net *net = sock_net(sk);
682 char devname[IFNAMSIZ];
683 int index;
684
685 ret = -EINVAL;
686 if (optlen < 0)
687 goto out;
688
689 /* Bind this socket to a particular device like "eth0",
690 * as specified in the passed interface name. If the
691 * name is "" or the option length is zero the socket
692 * is not bound.
693 */
694 if (optlen > IFNAMSIZ - 1)
695 optlen = IFNAMSIZ - 1;
696 memset(devname, 0, sizeof(devname));
697
698 ret = -EFAULT;
699 if (copy_from_sockptr(devname, optval, optlen))
700 goto out;
701
702 index = 0;
703 if (devname[0] != '\0') {
704 struct net_device *dev;
705
706 rcu_read_lock();
707 dev = dev_get_by_name_rcu(net, devname);
708 if (dev)
709 index = dev->ifindex;
710 rcu_read_unlock();
711 ret = -ENODEV;
712 if (!dev)
713 goto out;
714 }
715
716 sockopt_lock_sock(sk);
717 ret = sock_bindtoindex_locked(sk, index);
718 sockopt_release_sock(sk);
719 out:
720 #endif
721
722 return ret;
723 }
724
725 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
726 sockptr_t optlen, int len)
727 {
728 int ret = -ENOPROTOOPT;
729 #ifdef CONFIG_NETDEVICES
730 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
731 struct net *net = sock_net(sk);
732 char devname[IFNAMSIZ];
733
734 if (bound_dev_if == 0) {
735 len = 0;
736 goto zero;
737 }
738
739 ret = -EINVAL;
740 if (len < IFNAMSIZ)
741 goto out;
742
743 ret = netdev_get_name(net, devname, bound_dev_if);
744 if (ret)
745 goto out;
746
747 len = strlen(devname) + 1;
748
749 ret = -EFAULT;
750 if (copy_to_sockptr(optval, devname, len))
751 goto out;
752
753 zero:
754 ret = -EFAULT;
755 if (copy_to_sockptr(optlen, &len, sizeof(int)))
756 goto out;
757
758 ret = 0;
759
760 out:
761 #endif
762
763 return ret;
764 }
765
766 bool sk_mc_loop(const struct sock *sk)
767 {
768 if (dev_recursion_level())
769 return false;
770 if (!sk)
771 return true;
772 /* IPV6_ADDRFORM can change sk->sk_family under us. */
773 switch (READ_ONCE(sk->sk_family)) {
774 case AF_INET:
775 return inet_test_bit(MC_LOOP, sk);
776 #if IS_ENABLED(CONFIG_IPV6)
777 case AF_INET6:
778 return inet6_test_bit(MC6_LOOP, sk);
779 #endif
780 }
781 WARN_ON_ONCE(1);
782 return true;
783 }
784 EXPORT_SYMBOL(sk_mc_loop);
785
786 void sock_set_reuseaddr(struct sock *sk)
787 {
788 lock_sock(sk);
789 sk->sk_reuse = SK_CAN_REUSE;
790 release_sock(sk);
791 }
792 EXPORT_SYMBOL(sock_set_reuseaddr);
793
794 void sock_set_reuseport(struct sock *sk)
795 {
796 lock_sock(sk);
797 sk->sk_reuseport = true;
798 release_sock(sk);
799 }
800 EXPORT_SYMBOL(sock_set_reuseport);
801
802 void sock_no_linger(struct sock *sk)
803 {
804 lock_sock(sk);
805 WRITE_ONCE(sk->sk_lingertime, 0);
806 sock_set_flag(sk, SOCK_LINGER);
807 release_sock(sk);
808 }
809 EXPORT_SYMBOL(sock_no_linger);
810
811 void sock_set_priority(struct sock *sk, u32 priority)
812 {
813 WRITE_ONCE(sk->sk_priority, priority);
814 }
815 EXPORT_SYMBOL(sock_set_priority);
816
817 void sock_set_sndtimeo(struct sock *sk, s64 secs)
818 {
819 lock_sock(sk);
820 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
821 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
822 else
823 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
824 release_sock(sk);
825 }
826 EXPORT_SYMBOL(sock_set_sndtimeo);
827
828 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
829 {
830 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
831 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
832 if (val) {
833 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
834 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
835 }
836 }
837
838 void sock_enable_timestamps(struct sock *sk)
839 {
840 lock_sock(sk);
841 __sock_set_timestamps(sk, true, false, true);
842 release_sock(sk);
843 }
844 EXPORT_SYMBOL(sock_enable_timestamps);
845
846 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
847 {
848 switch (optname) {
849 case SO_TIMESTAMP_OLD:
850 __sock_set_timestamps(sk, valbool, false, false);
851 break;
852 case SO_TIMESTAMP_NEW:
853 __sock_set_timestamps(sk, valbool, true, false);
854 break;
855 case SO_TIMESTAMPNS_OLD:
856 __sock_set_timestamps(sk, valbool, false, true);
857 break;
858 case SO_TIMESTAMPNS_NEW:
859 __sock_set_timestamps(sk, valbool, true, true);
860 break;
861 }
862 }
863
864 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
865 {
866 struct net *net = sock_net(sk);
867 struct net_device *dev = NULL;
868 bool match = false;
869 int *vclock_index;
870 int i, num;
871
872 if (sk->sk_bound_dev_if)
873 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
874
875 if (!dev) {
876 pr_err("%s: sock not bind to device\n", __func__);
877 return -EOPNOTSUPP;
878 }
879
880 num = ethtool_get_phc_vclocks(dev, &vclock_index);
881 dev_put(dev);
882
883 for (i = 0; i < num; i++) {
884 if (*(vclock_index + i) == phc_index) {
885 match = true;
886 break;
887 }
888 }
889
890 if (num > 0)
891 kfree(vclock_index);
892
893 if (!match)
894 return -EINVAL;
895
896 WRITE_ONCE(sk->sk_bind_phc, phc_index);
897
898 return 0;
899 }
900
901 int sock_set_timestamping(struct sock *sk, int optname,
902 struct so_timestamping timestamping)
903 {
904 int val = timestamping.flags;
905 int ret;
906
907 if (val & ~SOF_TIMESTAMPING_MASK)
908 return -EINVAL;
909
910 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
911 !(val & SOF_TIMESTAMPING_OPT_ID))
912 return -EINVAL;
913
914 if (val & SOF_TIMESTAMPING_OPT_ID &&
915 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
916 if (sk_is_tcp(sk)) {
917 if ((1 << sk->sk_state) &
918 (TCPF_CLOSE | TCPF_LISTEN))
919 return -EINVAL;
920 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
922 else
923 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
924 } else {
925 atomic_set(&sk->sk_tskey, 0);
926 }
927 }
928
929 if (val & SOF_TIMESTAMPING_OPT_STATS &&
930 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
931 return -EINVAL;
932
933 if (val & SOF_TIMESTAMPING_BIND_PHC) {
934 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
935 if (ret)
936 return ret;
937 }
938
939 WRITE_ONCE(sk->sk_tsflags, val);
940 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
941
942 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
943 sock_enable_timestamp(sk,
944 SOCK_TIMESTAMPING_RX_SOFTWARE);
945 else
946 sock_disable_timestamp(sk,
947 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
948 return 0;
949 }
950
951 void sock_set_keepalive(struct sock *sk)
952 {
953 lock_sock(sk);
954 if (sk->sk_prot->keepalive)
955 sk->sk_prot->keepalive(sk, true);
956 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
957 release_sock(sk);
958 }
959 EXPORT_SYMBOL(sock_set_keepalive);
960
961 static void __sock_set_rcvbuf(struct sock *sk, int val)
962 {
963 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
964 * as a negative value.
965 */
966 val = min_t(int, val, INT_MAX / 2);
967 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
968
969 /* We double it on the way in to account for "struct sk_buff" etc.
970 * overhead. Applications assume that the SO_RCVBUF setting they make
971 * will allow that much actual data to be received on that socket.
972 *
973 * Applications are unaware that "struct sk_buff" and other overheads
974 * allocate from the receive buffer during socket buffer allocation.
975 *
976 * And after considering the possible alternatives, returning the value
977 * we actually used in getsockopt is the most desirable behavior.
978 */
979 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
980 }
981
982 void sock_set_rcvbuf(struct sock *sk, int val)
983 {
984 lock_sock(sk);
985 __sock_set_rcvbuf(sk, val);
986 release_sock(sk);
987 }
988 EXPORT_SYMBOL(sock_set_rcvbuf);
989
990 static void __sock_set_mark(struct sock *sk, u32 val)
991 {
992 if (val != sk->sk_mark) {
993 WRITE_ONCE(sk->sk_mark, val);
994 sk_dst_reset(sk);
995 }
996 }
997
998 void sock_set_mark(struct sock *sk, u32 val)
999 {
1000 lock_sock(sk);
1001 __sock_set_mark(sk, val);
1002 release_sock(sk);
1003 }
1004 EXPORT_SYMBOL(sock_set_mark);
1005
1006 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1007 {
1008 /* Round down bytes to multiple of pages */
1009 bytes = round_down(bytes, PAGE_SIZE);
1010
1011 WARN_ON(bytes > sk->sk_reserved_mem);
1012 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1013 sk_mem_reclaim(sk);
1014 }
1015
1016 static int sock_reserve_memory(struct sock *sk, int bytes)
1017 {
1018 long allocated;
1019 bool charged;
1020 int pages;
1021
1022 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1023 return -EOPNOTSUPP;
1024
1025 if (!bytes)
1026 return 0;
1027
1028 pages = sk_mem_pages(bytes);
1029
1030 /* pre-charge to memcg */
1031 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1032 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1033 if (!charged)
1034 return -ENOMEM;
1035
1036 /* pre-charge to forward_alloc */
1037 sk_memory_allocated_add(sk, pages);
1038 allocated = sk_memory_allocated(sk);
1039 /* If the system goes into memory pressure with this
1040 * precharge, give up and return error.
1041 */
1042 if (allocated > sk_prot_mem_limits(sk, 1)) {
1043 sk_memory_allocated_sub(sk, pages);
1044 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1045 return -ENOMEM;
1046 }
1047 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1048
1049 WRITE_ONCE(sk->sk_reserved_mem,
1050 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1051
1052 return 0;
1053 }
1054
1055 #ifdef CONFIG_PAGE_POOL
1056
1057 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1058 * in 1 syscall. The limit exists to limit the amount of memory the kernel
1059 * allocates to copy these tokens, and to prevent looping over the frags for
1060 * too long.
1061 */
1062 #define MAX_DONTNEED_TOKENS 128
1063 #define MAX_DONTNEED_FRAGS 1024
1064
1065 static noinline_for_stack int
1066 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1067 {
1068 unsigned int num_tokens, i, j, k, netmem_num = 0;
1069 struct dmabuf_token *tokens;
1070 int ret = 0, num_frags = 0;
1071 netmem_ref netmems[16];
1072
1073 if (!sk_is_tcp(sk))
1074 return -EBADF;
1075
1076 if (optlen % sizeof(*tokens) ||
1077 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1078 return -EINVAL;
1079
1080 num_tokens = optlen / sizeof(*tokens);
1081 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL);
1082 if (!tokens)
1083 return -ENOMEM;
1084
1085 if (copy_from_sockptr(tokens, optval, optlen)) {
1086 kvfree(tokens);
1087 return -EFAULT;
1088 }
1089
1090 xa_lock_bh(&sk->sk_user_frags);
1091 for (i = 0; i < num_tokens; i++) {
1092 for (j = 0; j < tokens[i].token_count; j++) {
1093 if (++num_frags > MAX_DONTNEED_FRAGS)
1094 goto frag_limit_reached;
1095
1096 netmem_ref netmem = (__force netmem_ref)__xa_erase(
1097 &sk->sk_user_frags, tokens[i].token_start + j);
1098
1099 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1100 continue;
1101
1102 netmems[netmem_num++] = netmem;
1103 if (netmem_num == ARRAY_SIZE(netmems)) {
1104 xa_unlock_bh(&sk->sk_user_frags);
1105 for (k = 0; k < netmem_num; k++)
1106 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1107 netmem_num = 0;
1108 xa_lock_bh(&sk->sk_user_frags);
1109 }
1110 ret++;
1111 }
1112 }
1113
1114 frag_limit_reached:
1115 xa_unlock_bh(&sk->sk_user_frags);
1116 for (k = 0; k < netmem_num; k++)
1117 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1118
1119 kvfree(tokens);
1120 return ret;
1121 }
1122 #endif
1123
1124 void sockopt_lock_sock(struct sock *sk)
1125 {
1126 /* When current->bpf_ctx is set, the setsockopt is called from
1127 * a bpf prog. bpf has ensured the sk lock has been
1128 * acquired before calling setsockopt().
1129 */
1130 if (has_current_bpf_ctx())
1131 return;
1132
1133 lock_sock(sk);
1134 }
1135 EXPORT_SYMBOL(sockopt_lock_sock);
1136
1137 void sockopt_release_sock(struct sock *sk)
1138 {
1139 if (has_current_bpf_ctx())
1140 return;
1141
1142 release_sock(sk);
1143 }
1144 EXPORT_SYMBOL(sockopt_release_sock);
1145
1146 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1147 {
1148 return has_current_bpf_ctx() || ns_capable(ns, cap);
1149 }
1150 EXPORT_SYMBOL(sockopt_ns_capable);
1151
1152 bool sockopt_capable(int cap)
1153 {
1154 return has_current_bpf_ctx() || capable(cap);
1155 }
1156 EXPORT_SYMBOL(sockopt_capable);
1157
1158 static int sockopt_validate_clockid(__kernel_clockid_t value)
1159 {
1160 switch (value) {
1161 case CLOCK_REALTIME:
1162 case CLOCK_MONOTONIC:
1163 case CLOCK_TAI:
1164 return 0;
1165 }
1166 return -EINVAL;
1167 }
1168
1169 /*
1170 * This is meant for all protocols to use and covers goings on
1171 * at the socket level. Everything here is generic.
1172 */
1173
1174 int sk_setsockopt(struct sock *sk, int level, int optname,
1175 sockptr_t optval, unsigned int optlen)
1176 {
1177 struct so_timestamping timestamping;
1178 struct socket *sock = sk->sk_socket;
1179 struct sock_txtime sk_txtime;
1180 int val;
1181 int valbool;
1182 struct linger ling;
1183 int ret = 0;
1184
1185 /*
1186 * Options without arguments
1187 */
1188
1189 if (optname == SO_BINDTODEVICE)
1190 return sock_setbindtodevice(sk, optval, optlen);
1191
1192 if (optlen < sizeof(int))
1193 return -EINVAL;
1194
1195 if (copy_from_sockptr(&val, optval, sizeof(val)))
1196 return -EFAULT;
1197
1198 valbool = val ? 1 : 0;
1199
1200 /* handle options which do not require locking the socket. */
1201 switch (optname) {
1202 case SO_PRIORITY:
1203 if (sk_set_prio_allowed(sk, val)) {
1204 sock_set_priority(sk, val);
1205 return 0;
1206 }
1207 return -EPERM;
1208 case SO_PASSSEC:
1209 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1210 return 0;
1211 case SO_PASSCRED:
1212 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1213 return 0;
1214 case SO_PASSPIDFD:
1215 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1216 return 0;
1217 case SO_TYPE:
1218 case SO_PROTOCOL:
1219 case SO_DOMAIN:
1220 case SO_ERROR:
1221 return -ENOPROTOOPT;
1222 #ifdef CONFIG_NET_RX_BUSY_POLL
1223 case SO_BUSY_POLL:
1224 if (val < 0)
1225 return -EINVAL;
1226 WRITE_ONCE(sk->sk_ll_usec, val);
1227 return 0;
1228 case SO_PREFER_BUSY_POLL:
1229 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1230 return -EPERM;
1231 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1232 return 0;
1233 case SO_BUSY_POLL_BUDGET:
1234 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1235 !sockopt_capable(CAP_NET_ADMIN))
1236 return -EPERM;
1237 if (val < 0 || val > U16_MAX)
1238 return -EINVAL;
1239 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1240 return 0;
1241 #endif
1242 case SO_MAX_PACING_RATE:
1243 {
1244 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1245 unsigned long pacing_rate;
1246
1247 if (sizeof(ulval) != sizeof(val) &&
1248 optlen >= sizeof(ulval) &&
1249 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1250 return -EFAULT;
1251 }
1252 if (ulval != ~0UL)
1253 cmpxchg(&sk->sk_pacing_status,
1254 SK_PACING_NONE,
1255 SK_PACING_NEEDED);
1256 /* Pairs with READ_ONCE() from sk_getsockopt() */
1257 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1258 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1259 if (ulval < pacing_rate)
1260 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1261 return 0;
1262 }
1263 case SO_TXREHASH:
1264 if (val < -1 || val > 1)
1265 return -EINVAL;
1266 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1267 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1268 /* Paired with READ_ONCE() in tcp_rtx_synack()
1269 * and sk_getsockopt().
1270 */
1271 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1272 return 0;
1273 case SO_PEEK_OFF:
1274 {
1275 int (*set_peek_off)(struct sock *sk, int val);
1276
1277 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1278 if (set_peek_off)
1279 ret = set_peek_off(sk, val);
1280 else
1281 ret = -EOPNOTSUPP;
1282 return ret;
1283 }
1284 #ifdef CONFIG_PAGE_POOL
1285 case SO_DEVMEM_DONTNEED:
1286 return sock_devmem_dontneed(sk, optval, optlen);
1287 #endif
1288 }
1289
1290 sockopt_lock_sock(sk);
1291
1292 switch (optname) {
1293 case SO_DEBUG:
1294 if (val && !sockopt_capable(CAP_NET_ADMIN))
1295 ret = -EACCES;
1296 else
1297 sock_valbool_flag(sk, SOCK_DBG, valbool);
1298 break;
1299 case SO_REUSEADDR:
1300 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1301 break;
1302 case SO_REUSEPORT:
1303 if (valbool && !sk_is_inet(sk))
1304 ret = -EOPNOTSUPP;
1305 else
1306 sk->sk_reuseport = valbool;
1307 break;
1308 case SO_DONTROUTE:
1309 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1310 sk_dst_reset(sk);
1311 break;
1312 case SO_BROADCAST:
1313 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1314 break;
1315 case SO_SNDBUF:
1316 /* Don't error on this BSD doesn't and if you think
1317 * about it this is right. Otherwise apps have to
1318 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1319 * are treated in BSD as hints
1320 */
1321 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1322 set_sndbuf:
1323 /* Ensure val * 2 fits into an int, to prevent max_t()
1324 * from treating it as a negative value.
1325 */
1326 val = min_t(int, val, INT_MAX / 2);
1327 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1328 WRITE_ONCE(sk->sk_sndbuf,
1329 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1330 /* Wake up sending tasks if we upped the value. */
1331 sk->sk_write_space(sk);
1332 break;
1333
1334 case SO_SNDBUFFORCE:
1335 if (!sockopt_capable(CAP_NET_ADMIN)) {
1336 ret = -EPERM;
1337 break;
1338 }
1339
1340 /* No negative values (to prevent underflow, as val will be
1341 * multiplied by 2).
1342 */
1343 if (val < 0)
1344 val = 0;
1345 goto set_sndbuf;
1346
1347 case SO_RCVBUF:
1348 /* Don't error on this BSD doesn't and if you think
1349 * about it this is right. Otherwise apps have to
1350 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1351 * are treated in BSD as hints
1352 */
1353 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1354 break;
1355
1356 case SO_RCVBUFFORCE:
1357 if (!sockopt_capable(CAP_NET_ADMIN)) {
1358 ret = -EPERM;
1359 break;
1360 }
1361
1362 /* No negative values (to prevent underflow, as val will be
1363 * multiplied by 2).
1364 */
1365 __sock_set_rcvbuf(sk, max(val, 0));
1366 break;
1367
1368 case SO_KEEPALIVE:
1369 if (sk->sk_prot->keepalive)
1370 sk->sk_prot->keepalive(sk, valbool);
1371 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1372 break;
1373
1374 case SO_OOBINLINE:
1375 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1376 break;
1377
1378 case SO_NO_CHECK:
1379 sk->sk_no_check_tx = valbool;
1380 break;
1381
1382 case SO_LINGER:
1383 if (optlen < sizeof(ling)) {
1384 ret = -EINVAL; /* 1003.1g */
1385 break;
1386 }
1387 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1388 ret = -EFAULT;
1389 break;
1390 }
1391 if (!ling.l_onoff) {
1392 sock_reset_flag(sk, SOCK_LINGER);
1393 } else {
1394 unsigned long t_sec = ling.l_linger;
1395
1396 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1397 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1398 else
1399 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1400 sock_set_flag(sk, SOCK_LINGER);
1401 }
1402 break;
1403
1404 case SO_BSDCOMPAT:
1405 break;
1406
1407 case SO_TIMESTAMP_OLD:
1408 case SO_TIMESTAMP_NEW:
1409 case SO_TIMESTAMPNS_OLD:
1410 case SO_TIMESTAMPNS_NEW:
1411 sock_set_timestamp(sk, optname, valbool);
1412 break;
1413
1414 case SO_TIMESTAMPING_NEW:
1415 case SO_TIMESTAMPING_OLD:
1416 if (optlen == sizeof(timestamping)) {
1417 if (copy_from_sockptr(&timestamping, optval,
1418 sizeof(timestamping))) {
1419 ret = -EFAULT;
1420 break;
1421 }
1422 } else {
1423 memset(&timestamping, 0, sizeof(timestamping));
1424 timestamping.flags = val;
1425 }
1426 ret = sock_set_timestamping(sk, optname, timestamping);
1427 break;
1428
1429 case SO_RCVLOWAT:
1430 {
1431 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1432
1433 if (val < 0)
1434 val = INT_MAX;
1435 if (sock)
1436 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1437 if (set_rcvlowat)
1438 ret = set_rcvlowat(sk, val);
1439 else
1440 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1441 break;
1442 }
1443 case SO_RCVTIMEO_OLD:
1444 case SO_RCVTIMEO_NEW:
1445 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1446 optlen, optname == SO_RCVTIMEO_OLD);
1447 break;
1448
1449 case SO_SNDTIMEO_OLD:
1450 case SO_SNDTIMEO_NEW:
1451 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1452 optlen, optname == SO_SNDTIMEO_OLD);
1453 break;
1454
1455 case SO_ATTACH_FILTER: {
1456 struct sock_fprog fprog;
1457
1458 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1459 if (!ret)
1460 ret = sk_attach_filter(&fprog, sk);
1461 break;
1462 }
1463 case SO_ATTACH_BPF:
1464 ret = -EINVAL;
1465 if (optlen == sizeof(u32)) {
1466 u32 ufd;
1467
1468 ret = -EFAULT;
1469 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1470 break;
1471
1472 ret = sk_attach_bpf(ufd, sk);
1473 }
1474 break;
1475
1476 case SO_ATTACH_REUSEPORT_CBPF: {
1477 struct sock_fprog fprog;
1478
1479 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1480 if (!ret)
1481 ret = sk_reuseport_attach_filter(&fprog, sk);
1482 break;
1483 }
1484 case SO_ATTACH_REUSEPORT_EBPF:
1485 ret = -EINVAL;
1486 if (optlen == sizeof(u32)) {
1487 u32 ufd;
1488
1489 ret = -EFAULT;
1490 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1491 break;
1492
1493 ret = sk_reuseport_attach_bpf(ufd, sk);
1494 }
1495 break;
1496
1497 case SO_DETACH_REUSEPORT_BPF:
1498 ret = reuseport_detach_prog(sk);
1499 break;
1500
1501 case SO_DETACH_FILTER:
1502 ret = sk_detach_filter(sk);
1503 break;
1504
1505 case SO_LOCK_FILTER:
1506 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1507 ret = -EPERM;
1508 else
1509 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1510 break;
1511
1512 case SO_MARK:
1513 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1514 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1515 ret = -EPERM;
1516 break;
1517 }
1518
1519 __sock_set_mark(sk, val);
1520 break;
1521 case SO_RCVMARK:
1522 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1523 break;
1524
1525 case SO_RCVPRIORITY:
1526 sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool);
1527 break;
1528
1529 case SO_RXQ_OVFL:
1530 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1531 break;
1532
1533 case SO_WIFI_STATUS:
1534 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1535 break;
1536
1537 case SO_NOFCS:
1538 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1539 break;
1540
1541 case SO_SELECT_ERR_QUEUE:
1542 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1543 break;
1544
1545
1546 case SO_INCOMING_CPU:
1547 reuseport_update_incoming_cpu(sk, val);
1548 break;
1549
1550 case SO_CNX_ADVICE:
1551 if (val == 1)
1552 dst_negative_advice(sk);
1553 break;
1554
1555 case SO_ZEROCOPY:
1556 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1557 if (!(sk_is_tcp(sk) ||
1558 (sk->sk_type == SOCK_DGRAM &&
1559 sk->sk_protocol == IPPROTO_UDP)))
1560 ret = -EOPNOTSUPP;
1561 } else if (sk->sk_family != PF_RDS) {
1562 ret = -EOPNOTSUPP;
1563 }
1564 if (!ret) {
1565 if (val < 0 || val > 1)
1566 ret = -EINVAL;
1567 else
1568 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1569 }
1570 break;
1571
1572 case SO_TXTIME:
1573 if (optlen != sizeof(struct sock_txtime)) {
1574 ret = -EINVAL;
1575 break;
1576 } else if (copy_from_sockptr(&sk_txtime, optval,
1577 sizeof(struct sock_txtime))) {
1578 ret = -EFAULT;
1579 break;
1580 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1581 ret = -EINVAL;
1582 break;
1583 }
1584 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1585 * scheduler has enough safe guards.
1586 */
1587 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1588 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1589 ret = -EPERM;
1590 break;
1591 }
1592
1593 ret = sockopt_validate_clockid(sk_txtime.clockid);
1594 if (ret)
1595 break;
1596
1597 sock_valbool_flag(sk, SOCK_TXTIME, true);
1598 sk->sk_clockid = sk_txtime.clockid;
1599 sk->sk_txtime_deadline_mode =
1600 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1601 sk->sk_txtime_report_errors =
1602 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1603 break;
1604
1605 case SO_BINDTOIFINDEX:
1606 ret = sock_bindtoindex_locked(sk, val);
1607 break;
1608
1609 case SO_BUF_LOCK:
1610 if (val & ~SOCK_BUF_LOCK_MASK) {
1611 ret = -EINVAL;
1612 break;
1613 }
1614 sk->sk_userlocks = val | (sk->sk_userlocks &
1615 ~SOCK_BUF_LOCK_MASK);
1616 break;
1617
1618 case SO_RESERVE_MEM:
1619 {
1620 int delta;
1621
1622 if (val < 0) {
1623 ret = -EINVAL;
1624 break;
1625 }
1626
1627 delta = val - sk->sk_reserved_mem;
1628 if (delta < 0)
1629 sock_release_reserved_memory(sk, -delta);
1630 else
1631 ret = sock_reserve_memory(sk, delta);
1632 break;
1633 }
1634
1635 default:
1636 ret = -ENOPROTOOPT;
1637 break;
1638 }
1639 sockopt_release_sock(sk);
1640 return ret;
1641 }
1642
1643 int sock_setsockopt(struct socket *sock, int level, int optname,
1644 sockptr_t optval, unsigned int optlen)
1645 {
1646 return sk_setsockopt(sock->sk, level, optname,
1647 optval, optlen);
1648 }
1649 EXPORT_SYMBOL(sock_setsockopt);
1650
1651 static const struct cred *sk_get_peer_cred(struct sock *sk)
1652 {
1653 const struct cred *cred;
1654
1655 spin_lock(&sk->sk_peer_lock);
1656 cred = get_cred(sk->sk_peer_cred);
1657 spin_unlock(&sk->sk_peer_lock);
1658
1659 return cred;
1660 }
1661
1662 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1663 struct ucred *ucred)
1664 {
1665 ucred->pid = pid_vnr(pid);
1666 ucred->uid = ucred->gid = -1;
1667 if (cred) {
1668 struct user_namespace *current_ns = current_user_ns();
1669
1670 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1671 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1672 }
1673 }
1674
1675 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1676 {
1677 struct user_namespace *user_ns = current_user_ns();
1678 int i;
1679
1680 for (i = 0; i < src->ngroups; i++) {
1681 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1682
1683 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1684 return -EFAULT;
1685 }
1686
1687 return 0;
1688 }
1689
1690 int sk_getsockopt(struct sock *sk, int level, int optname,
1691 sockptr_t optval, sockptr_t optlen)
1692 {
1693 struct socket *sock = sk->sk_socket;
1694
1695 union {
1696 int val;
1697 u64 val64;
1698 unsigned long ulval;
1699 struct linger ling;
1700 struct old_timeval32 tm32;
1701 struct __kernel_old_timeval tm;
1702 struct __kernel_sock_timeval stm;
1703 struct sock_txtime txtime;
1704 struct so_timestamping timestamping;
1705 } v;
1706
1707 int lv = sizeof(int);
1708 int len;
1709
1710 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1711 return -EFAULT;
1712 if (len < 0)
1713 return -EINVAL;
1714
1715 memset(&v, 0, sizeof(v));
1716
1717 switch (optname) {
1718 case SO_DEBUG:
1719 v.val = sock_flag(sk, SOCK_DBG);
1720 break;
1721
1722 case SO_DONTROUTE:
1723 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1724 break;
1725
1726 case SO_BROADCAST:
1727 v.val = sock_flag(sk, SOCK_BROADCAST);
1728 break;
1729
1730 case SO_SNDBUF:
1731 v.val = READ_ONCE(sk->sk_sndbuf);
1732 break;
1733
1734 case SO_RCVBUF:
1735 v.val = READ_ONCE(sk->sk_rcvbuf);
1736 break;
1737
1738 case SO_REUSEADDR:
1739 v.val = sk->sk_reuse;
1740 break;
1741
1742 case SO_REUSEPORT:
1743 v.val = sk->sk_reuseport;
1744 break;
1745
1746 case SO_KEEPALIVE:
1747 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1748 break;
1749
1750 case SO_TYPE:
1751 v.val = sk->sk_type;
1752 break;
1753
1754 case SO_PROTOCOL:
1755 v.val = sk->sk_protocol;
1756 break;
1757
1758 case SO_DOMAIN:
1759 v.val = sk->sk_family;
1760 break;
1761
1762 case SO_ERROR:
1763 v.val = -sock_error(sk);
1764 if (v.val == 0)
1765 v.val = xchg(&sk->sk_err_soft, 0);
1766 break;
1767
1768 case SO_OOBINLINE:
1769 v.val = sock_flag(sk, SOCK_URGINLINE);
1770 break;
1771
1772 case SO_NO_CHECK:
1773 v.val = sk->sk_no_check_tx;
1774 break;
1775
1776 case SO_PRIORITY:
1777 v.val = READ_ONCE(sk->sk_priority);
1778 break;
1779
1780 case SO_LINGER:
1781 lv = sizeof(v.ling);
1782 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1783 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1784 break;
1785
1786 case SO_BSDCOMPAT:
1787 break;
1788
1789 case SO_TIMESTAMP_OLD:
1790 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1791 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1792 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1793 break;
1794
1795 case SO_TIMESTAMPNS_OLD:
1796 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1797 break;
1798
1799 case SO_TIMESTAMP_NEW:
1800 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1801 break;
1802
1803 case SO_TIMESTAMPNS_NEW:
1804 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1805 break;
1806
1807 case SO_TIMESTAMPING_OLD:
1808 case SO_TIMESTAMPING_NEW:
1809 lv = sizeof(v.timestamping);
1810 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1811 * returning the flags when they were set through the same option.
1812 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1813 */
1814 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1815 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1816 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1817 }
1818 break;
1819
1820 case SO_RCVTIMEO_OLD:
1821 case SO_RCVTIMEO_NEW:
1822 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1823 SO_RCVTIMEO_OLD == optname);
1824 break;
1825
1826 case SO_SNDTIMEO_OLD:
1827 case SO_SNDTIMEO_NEW:
1828 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1829 SO_SNDTIMEO_OLD == optname);
1830 break;
1831
1832 case SO_RCVLOWAT:
1833 v.val = READ_ONCE(sk->sk_rcvlowat);
1834 break;
1835
1836 case SO_SNDLOWAT:
1837 v.val = 1;
1838 break;
1839
1840 case SO_PASSCRED:
1841 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1842 break;
1843
1844 case SO_PASSPIDFD:
1845 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1846 break;
1847
1848 case SO_PEERCRED:
1849 {
1850 struct ucred peercred;
1851 if (len > sizeof(peercred))
1852 len = sizeof(peercred);
1853
1854 spin_lock(&sk->sk_peer_lock);
1855 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1856 spin_unlock(&sk->sk_peer_lock);
1857
1858 if (copy_to_sockptr(optval, &peercred, len))
1859 return -EFAULT;
1860 goto lenout;
1861 }
1862
1863 case SO_PEERPIDFD:
1864 {
1865 struct pid *peer_pid;
1866 struct file *pidfd_file = NULL;
1867 int pidfd;
1868
1869 if (len > sizeof(pidfd))
1870 len = sizeof(pidfd);
1871
1872 spin_lock(&sk->sk_peer_lock);
1873 peer_pid = get_pid(sk->sk_peer_pid);
1874 spin_unlock(&sk->sk_peer_lock);
1875
1876 if (!peer_pid)
1877 return -ENODATA;
1878
1879 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1880 put_pid(peer_pid);
1881 if (pidfd < 0)
1882 return pidfd;
1883
1884 if (copy_to_sockptr(optval, &pidfd, len) ||
1885 copy_to_sockptr(optlen, &len, sizeof(int))) {
1886 put_unused_fd(pidfd);
1887 fput(pidfd_file);
1888
1889 return -EFAULT;
1890 }
1891
1892 fd_install(pidfd, pidfd_file);
1893 return 0;
1894 }
1895
1896 case SO_PEERGROUPS:
1897 {
1898 const struct cred *cred;
1899 int ret, n;
1900
1901 cred = sk_get_peer_cred(sk);
1902 if (!cred)
1903 return -ENODATA;
1904
1905 n = cred->group_info->ngroups;
1906 if (len < n * sizeof(gid_t)) {
1907 len = n * sizeof(gid_t);
1908 put_cred(cred);
1909 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1910 }
1911 len = n * sizeof(gid_t);
1912
1913 ret = groups_to_user(optval, cred->group_info);
1914 put_cred(cred);
1915 if (ret)
1916 return ret;
1917 goto lenout;
1918 }
1919
1920 case SO_PEERNAME:
1921 {
1922 struct sockaddr_storage address;
1923
1924 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1925 if (lv < 0)
1926 return -ENOTCONN;
1927 if (lv < len)
1928 return -EINVAL;
1929 if (copy_to_sockptr(optval, &address, len))
1930 return -EFAULT;
1931 goto lenout;
1932 }
1933
1934 /* Dubious BSD thing... Probably nobody even uses it, but
1935 * the UNIX standard wants it for whatever reason... -DaveM
1936 */
1937 case SO_ACCEPTCONN:
1938 v.val = sk->sk_state == TCP_LISTEN;
1939 break;
1940
1941 case SO_PASSSEC:
1942 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1943 break;
1944
1945 case SO_PEERSEC:
1946 return security_socket_getpeersec_stream(sock,
1947 optval, optlen, len);
1948
1949 case SO_MARK:
1950 v.val = READ_ONCE(sk->sk_mark);
1951 break;
1952
1953 case SO_RCVMARK:
1954 v.val = sock_flag(sk, SOCK_RCVMARK);
1955 break;
1956
1957 case SO_RCVPRIORITY:
1958 v.val = sock_flag(sk, SOCK_RCVPRIORITY);
1959 break;
1960
1961 case SO_RXQ_OVFL:
1962 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1963 break;
1964
1965 case SO_WIFI_STATUS:
1966 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1967 break;
1968
1969 case SO_PEEK_OFF:
1970 if (!READ_ONCE(sock->ops)->set_peek_off)
1971 return -EOPNOTSUPP;
1972
1973 v.val = READ_ONCE(sk->sk_peek_off);
1974 break;
1975 case SO_NOFCS:
1976 v.val = sock_flag(sk, SOCK_NOFCS);
1977 break;
1978
1979 case SO_BINDTODEVICE:
1980 return sock_getbindtodevice(sk, optval, optlen, len);
1981
1982 case SO_GET_FILTER:
1983 len = sk_get_filter(sk, optval, len);
1984 if (len < 0)
1985 return len;
1986
1987 goto lenout;
1988
1989 case SO_LOCK_FILTER:
1990 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1991 break;
1992
1993 case SO_BPF_EXTENSIONS:
1994 v.val = bpf_tell_extensions();
1995 break;
1996
1997 case SO_SELECT_ERR_QUEUE:
1998 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1999 break;
2000
2001 #ifdef CONFIG_NET_RX_BUSY_POLL
2002 case SO_BUSY_POLL:
2003 v.val = READ_ONCE(sk->sk_ll_usec);
2004 break;
2005 case SO_PREFER_BUSY_POLL:
2006 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
2007 break;
2008 #endif
2009
2010 case SO_MAX_PACING_RATE:
2011 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
2012 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2013 lv = sizeof(v.ulval);
2014 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2015 } else {
2016 /* 32bit version */
2017 v.val = min_t(unsigned long, ~0U,
2018 READ_ONCE(sk->sk_max_pacing_rate));
2019 }
2020 break;
2021
2022 case SO_INCOMING_CPU:
2023 v.val = READ_ONCE(sk->sk_incoming_cpu);
2024 break;
2025
2026 case SO_MEMINFO:
2027 {
2028 u32 meminfo[SK_MEMINFO_VARS];
2029
2030 sk_get_meminfo(sk, meminfo);
2031
2032 len = min_t(unsigned int, len, sizeof(meminfo));
2033 if (copy_to_sockptr(optval, &meminfo, len))
2034 return -EFAULT;
2035
2036 goto lenout;
2037 }
2038
2039 #ifdef CONFIG_NET_RX_BUSY_POLL
2040 case SO_INCOMING_NAPI_ID:
2041 v.val = READ_ONCE(sk->sk_napi_id);
2042
2043 /* aggregate non-NAPI IDs down to 0 */
2044 if (v.val < MIN_NAPI_ID)
2045 v.val = 0;
2046
2047 break;
2048 #endif
2049
2050 case SO_COOKIE:
2051 lv = sizeof(u64);
2052 if (len < lv)
2053 return -EINVAL;
2054 v.val64 = sock_gen_cookie(sk);
2055 break;
2056
2057 case SO_ZEROCOPY:
2058 v.val = sock_flag(sk, SOCK_ZEROCOPY);
2059 break;
2060
2061 case SO_TXTIME:
2062 lv = sizeof(v.txtime);
2063 v.txtime.clockid = sk->sk_clockid;
2064 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2065 SOF_TXTIME_DEADLINE_MODE : 0;
2066 v.txtime.flags |= sk->sk_txtime_report_errors ?
2067 SOF_TXTIME_REPORT_ERRORS : 0;
2068 break;
2069
2070 case SO_BINDTOIFINDEX:
2071 v.val = READ_ONCE(sk->sk_bound_dev_if);
2072 break;
2073
2074 case SO_NETNS_COOKIE:
2075 lv = sizeof(u64);
2076 if (len != lv)
2077 return -EINVAL;
2078 v.val64 = sock_net(sk)->net_cookie;
2079 break;
2080
2081 case SO_BUF_LOCK:
2082 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2083 break;
2084
2085 case SO_RESERVE_MEM:
2086 v.val = READ_ONCE(sk->sk_reserved_mem);
2087 break;
2088
2089 case SO_TXREHASH:
2090 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2091 v.val = READ_ONCE(sk->sk_txrehash);
2092 break;
2093
2094 default:
2095 /* We implement the SO_SNDLOWAT etc to not be settable
2096 * (1003.1g 7).
2097 */
2098 return -ENOPROTOOPT;
2099 }
2100
2101 if (len > lv)
2102 len = lv;
2103 if (copy_to_sockptr(optval, &v, len))
2104 return -EFAULT;
2105 lenout:
2106 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2107 return -EFAULT;
2108 return 0;
2109 }
2110
2111 /*
2112 * Initialize an sk_lock.
2113 *
2114 * (We also register the sk_lock with the lock validator.)
2115 */
2116 static inline void sock_lock_init(struct sock *sk)
2117 {
2118 if (sk->sk_kern_sock)
2119 sock_lock_init_class_and_name(
2120 sk,
2121 af_family_kern_slock_key_strings[sk->sk_family],
2122 af_family_kern_slock_keys + sk->sk_family,
2123 af_family_kern_key_strings[sk->sk_family],
2124 af_family_kern_keys + sk->sk_family);
2125 else
2126 sock_lock_init_class_and_name(
2127 sk,
2128 af_family_slock_key_strings[sk->sk_family],
2129 af_family_slock_keys + sk->sk_family,
2130 af_family_key_strings[sk->sk_family],
2131 af_family_keys + sk->sk_family);
2132 }
2133
2134 /*
2135 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2136 * even temporarily, because of RCU lookups. sk_node should also be left as is.
2137 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2138 */
2139 static void sock_copy(struct sock *nsk, const struct sock *osk)
2140 {
2141 const struct proto *prot = READ_ONCE(osk->sk_prot);
2142 #ifdef CONFIG_SECURITY_NETWORK
2143 void *sptr = nsk->sk_security;
2144 #endif
2145
2146 /* If we move sk_tx_queue_mapping out of the private section,
2147 * we must check if sk_tx_queue_clear() is called after
2148 * sock_copy() in sk_clone_lock().
2149 */
2150 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2151 offsetof(struct sock, sk_dontcopy_begin) ||
2152 offsetof(struct sock, sk_tx_queue_mapping) >=
2153 offsetof(struct sock, sk_dontcopy_end));
2154
2155 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2156
2157 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2158 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2159 /* alloc is larger than struct, see sk_prot_alloc() */);
2160
2161 #ifdef CONFIG_SECURITY_NETWORK
2162 nsk->sk_security = sptr;
2163 security_sk_clone(osk, nsk);
2164 #endif
2165 }
2166
2167 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2168 int family)
2169 {
2170 struct sock *sk;
2171 struct kmem_cache *slab;
2172
2173 slab = prot->slab;
2174 if (slab != NULL) {
2175 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2176 if (!sk)
2177 return sk;
2178 if (want_init_on_alloc(priority))
2179 sk_prot_clear_nulls(sk, prot->obj_size);
2180 } else
2181 sk = kmalloc(prot->obj_size, priority);
2182
2183 if (sk != NULL) {
2184 if (security_sk_alloc(sk, family, priority))
2185 goto out_free;
2186
2187 if (!try_module_get(prot->owner))
2188 goto out_free_sec;
2189 }
2190
2191 return sk;
2192
2193 out_free_sec:
2194 security_sk_free(sk);
2195 out_free:
2196 if (slab != NULL)
2197 kmem_cache_free(slab, sk);
2198 else
2199 kfree(sk);
2200 return NULL;
2201 }
2202
2203 static void sk_prot_free(struct proto *prot, struct sock *sk)
2204 {
2205 struct kmem_cache *slab;
2206 struct module *owner;
2207
2208 owner = prot->owner;
2209 slab = prot->slab;
2210
2211 cgroup_sk_free(&sk->sk_cgrp_data);
2212 mem_cgroup_sk_free(sk);
2213 security_sk_free(sk);
2214 if (slab != NULL)
2215 kmem_cache_free(slab, sk);
2216 else
2217 kfree(sk);
2218 module_put(owner);
2219 }
2220
2221 /**
2222 * sk_alloc - All socket objects are allocated here
2223 * @net: the applicable net namespace
2224 * @family: protocol family
2225 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2226 * @prot: struct proto associated with this new sock instance
2227 * @kern: is this to be a kernel socket?
2228 */
2229 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2230 struct proto *prot, int kern)
2231 {
2232 struct sock *sk;
2233
2234 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2235 if (sk) {
2236 sk->sk_family = family;
2237 /*
2238 * See comment in struct sock definition to understand
2239 * why we need sk_prot_creator -acme
2240 */
2241 sk->sk_prot = sk->sk_prot_creator = prot;
2242 sk->sk_kern_sock = kern;
2243 sock_lock_init(sk);
2244 sk->sk_net_refcnt = kern ? 0 : 1;
2245 if (likely(sk->sk_net_refcnt)) {
2246 get_net_track(net, &sk->ns_tracker, priority);
2247 sock_inuse_add(net, 1);
2248 } else {
2249 __netns_tracker_alloc(net, &sk->ns_tracker,
2250 false, priority);
2251 }
2252
2253 sock_net_set(sk, net);
2254 refcount_set(&sk->sk_wmem_alloc, 1);
2255
2256 mem_cgroup_sk_alloc(sk);
2257 cgroup_sk_alloc(&sk->sk_cgrp_data);
2258 sock_update_classid(&sk->sk_cgrp_data);
2259 sock_update_netprioidx(&sk->sk_cgrp_data);
2260 sk_tx_queue_clear(sk);
2261 }
2262
2263 return sk;
2264 }
2265 EXPORT_SYMBOL(sk_alloc);
2266
2267 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2268 * grace period. This is the case for UDP sockets and TCP listeners.
2269 */
2270 static void __sk_destruct(struct rcu_head *head)
2271 {
2272 struct sock *sk = container_of(head, struct sock, sk_rcu);
2273 struct sk_filter *filter;
2274
2275 if (sk->sk_destruct)
2276 sk->sk_destruct(sk);
2277
2278 filter = rcu_dereference_check(sk->sk_filter,
2279 refcount_read(&sk->sk_wmem_alloc) == 0);
2280 if (filter) {
2281 sk_filter_uncharge(sk, filter);
2282 RCU_INIT_POINTER(sk->sk_filter, NULL);
2283 }
2284
2285 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2286
2287 #ifdef CONFIG_BPF_SYSCALL
2288 bpf_sk_storage_free(sk);
2289 #endif
2290
2291 if (atomic_read(&sk->sk_omem_alloc))
2292 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2293 __func__, atomic_read(&sk->sk_omem_alloc));
2294
2295 if (sk->sk_frag.page) {
2296 put_page(sk->sk_frag.page);
2297 sk->sk_frag.page = NULL;
2298 }
2299
2300 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2301 put_cred(sk->sk_peer_cred);
2302 put_pid(sk->sk_peer_pid);
2303
2304 if (likely(sk->sk_net_refcnt))
2305 put_net_track(sock_net(sk), &sk->ns_tracker);
2306 else
2307 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2308
2309 sk_prot_free(sk->sk_prot_creator, sk);
2310 }
2311
2312 void sk_destruct(struct sock *sk)
2313 {
2314 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2315
2316 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2317 reuseport_detach_sock(sk);
2318 use_call_rcu = true;
2319 }
2320
2321 if (use_call_rcu)
2322 call_rcu(&sk->sk_rcu, __sk_destruct);
2323 else
2324 __sk_destruct(&sk->sk_rcu);
2325 }
2326
2327 static void __sk_free(struct sock *sk)
2328 {
2329 if (likely(sk->sk_net_refcnt))
2330 sock_inuse_add(sock_net(sk), -1);
2331
2332 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2333 sock_diag_broadcast_destroy(sk);
2334 else
2335 sk_destruct(sk);
2336 }
2337
2338 void sk_free(struct sock *sk)
2339 {
2340 /*
2341 * We subtract one from sk_wmem_alloc and can know if
2342 * some packets are still in some tx queue.
2343 * If not null, sock_wfree() will call __sk_free(sk) later
2344 */
2345 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2346 __sk_free(sk);
2347 }
2348 EXPORT_SYMBOL(sk_free);
2349
2350 static void sk_init_common(struct sock *sk)
2351 {
2352 skb_queue_head_init(&sk->sk_receive_queue);
2353 skb_queue_head_init(&sk->sk_write_queue);
2354 skb_queue_head_init(&sk->sk_error_queue);
2355
2356 rwlock_init(&sk->sk_callback_lock);
2357 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2358 af_rlock_keys + sk->sk_family,
2359 af_family_rlock_key_strings[sk->sk_family]);
2360 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2361 af_wlock_keys + sk->sk_family,
2362 af_family_wlock_key_strings[sk->sk_family]);
2363 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2364 af_elock_keys + sk->sk_family,
2365 af_family_elock_key_strings[sk->sk_family]);
2366 if (sk->sk_kern_sock)
2367 lockdep_set_class_and_name(&sk->sk_callback_lock,
2368 af_kern_callback_keys + sk->sk_family,
2369 af_family_kern_clock_key_strings[sk->sk_family]);
2370 else
2371 lockdep_set_class_and_name(&sk->sk_callback_lock,
2372 af_callback_keys + sk->sk_family,
2373 af_family_clock_key_strings[sk->sk_family]);
2374 }
2375
2376 /**
2377 * sk_clone_lock - clone a socket, and lock its clone
2378 * @sk: the socket to clone
2379 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2380 *
2381 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2382 */
2383 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2384 {
2385 struct proto *prot = READ_ONCE(sk->sk_prot);
2386 struct sk_filter *filter;
2387 bool is_charged = true;
2388 struct sock *newsk;
2389
2390 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2391 if (!newsk)
2392 goto out;
2393
2394 sock_copy(newsk, sk);
2395
2396 newsk->sk_prot_creator = prot;
2397
2398 /* SANITY */
2399 if (likely(newsk->sk_net_refcnt)) {
2400 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2401 sock_inuse_add(sock_net(newsk), 1);
2402 } else {
2403 /* Kernel sockets are not elevating the struct net refcount.
2404 * Instead, use a tracker to more easily detect if a layer
2405 * is not properly dismantling its kernel sockets at netns
2406 * destroy time.
2407 */
2408 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2409 false, priority);
2410 }
2411 sk_node_init(&newsk->sk_node);
2412 sock_lock_init(newsk);
2413 bh_lock_sock(newsk);
2414 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2415 newsk->sk_backlog.len = 0;
2416
2417 atomic_set(&newsk->sk_rmem_alloc, 0);
2418
2419 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2420 refcount_set(&newsk->sk_wmem_alloc, 1);
2421
2422 atomic_set(&newsk->sk_omem_alloc, 0);
2423 sk_init_common(newsk);
2424
2425 newsk->sk_dst_cache = NULL;
2426 newsk->sk_dst_pending_confirm = 0;
2427 newsk->sk_wmem_queued = 0;
2428 newsk->sk_forward_alloc = 0;
2429 newsk->sk_reserved_mem = 0;
2430 atomic_set(&newsk->sk_drops, 0);
2431 newsk->sk_send_head = NULL;
2432 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2433 atomic_set(&newsk->sk_zckey, 0);
2434
2435 sock_reset_flag(newsk, SOCK_DONE);
2436
2437 /* sk->sk_memcg will be populated at accept() time */
2438 newsk->sk_memcg = NULL;
2439
2440 cgroup_sk_clone(&newsk->sk_cgrp_data);
2441
2442 rcu_read_lock();
2443 filter = rcu_dereference(sk->sk_filter);
2444 if (filter != NULL)
2445 /* though it's an empty new sock, the charging may fail
2446 * if sysctl_optmem_max was changed between creation of
2447 * original socket and cloning
2448 */
2449 is_charged = sk_filter_charge(newsk, filter);
2450 RCU_INIT_POINTER(newsk->sk_filter, filter);
2451 rcu_read_unlock();
2452
2453 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2454 /* We need to make sure that we don't uncharge the new
2455 * socket if we couldn't charge it in the first place
2456 * as otherwise we uncharge the parent's filter.
2457 */
2458 if (!is_charged)
2459 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2460 sk_free_unlock_clone(newsk);
2461 newsk = NULL;
2462 goto out;
2463 }
2464 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2465
2466 if (bpf_sk_storage_clone(sk, newsk)) {
2467 sk_free_unlock_clone(newsk);
2468 newsk = NULL;
2469 goto out;
2470 }
2471
2472 /* Clear sk_user_data if parent had the pointer tagged
2473 * as not suitable for copying when cloning.
2474 */
2475 if (sk_user_data_is_nocopy(newsk))
2476 newsk->sk_user_data = NULL;
2477
2478 newsk->sk_err = 0;
2479 newsk->sk_err_soft = 0;
2480 newsk->sk_priority = 0;
2481 newsk->sk_incoming_cpu = raw_smp_processor_id();
2482
2483 /* Before updating sk_refcnt, we must commit prior changes to memory
2484 * (Documentation/RCU/rculist_nulls.rst for details)
2485 */
2486 smp_wmb();
2487 refcount_set(&newsk->sk_refcnt, 2);
2488
2489 sk_set_socket(newsk, NULL);
2490 sk_tx_queue_clear(newsk);
2491 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2492
2493 if (newsk->sk_prot->sockets_allocated)
2494 sk_sockets_allocated_inc(newsk);
2495
2496 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2497 net_enable_timestamp();
2498 out:
2499 return newsk;
2500 }
2501 EXPORT_SYMBOL_GPL(sk_clone_lock);
2502
2503 void sk_free_unlock_clone(struct sock *sk)
2504 {
2505 /* It is still raw copy of parent, so invalidate
2506 * destructor and make plain sk_free() */
2507 sk->sk_destruct = NULL;
2508 bh_unlock_sock(sk);
2509 sk_free(sk);
2510 }
2511 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2512
2513 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2514 {
2515 bool is_ipv6 = false;
2516 u32 max_size;
2517
2518 #if IS_ENABLED(CONFIG_IPV6)
2519 is_ipv6 = (sk->sk_family == AF_INET6 &&
2520 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2521 #endif
2522 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2523 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2524 READ_ONCE(dst->dev->gso_ipv4_max_size);
2525 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2526 max_size = GSO_LEGACY_MAX_SIZE;
2527
2528 return max_size - (MAX_TCP_HEADER + 1);
2529 }
2530
2531 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2532 {
2533 u32 max_segs = 1;
2534
2535 sk->sk_route_caps = dst->dev->features;
2536 if (sk_is_tcp(sk))
2537 sk->sk_route_caps |= NETIF_F_GSO;
2538 if (sk->sk_route_caps & NETIF_F_GSO)
2539 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2540 if (unlikely(sk->sk_gso_disabled))
2541 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2542 if (sk_can_gso(sk)) {
2543 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2544 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2545 } else {
2546 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2547 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2548 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2549 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2550 }
2551 }
2552 sk->sk_gso_max_segs = max_segs;
2553 sk_dst_set(sk, dst);
2554 }
2555 EXPORT_SYMBOL_GPL(sk_setup_caps);
2556
2557 /*
2558 * Simple resource managers for sockets.
2559 */
2560
2561
2562 /*
2563 * Write buffer destructor automatically called from kfree_skb.
2564 */
2565 void sock_wfree(struct sk_buff *skb)
2566 {
2567 struct sock *sk = skb->sk;
2568 unsigned int len = skb->truesize;
2569 bool free;
2570
2571 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2572 if (sock_flag(sk, SOCK_RCU_FREE) &&
2573 sk->sk_write_space == sock_def_write_space) {
2574 rcu_read_lock();
2575 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2576 sock_def_write_space_wfree(sk);
2577 rcu_read_unlock();
2578 if (unlikely(free))
2579 __sk_free(sk);
2580 return;
2581 }
2582
2583 /*
2584 * Keep a reference on sk_wmem_alloc, this will be released
2585 * after sk_write_space() call
2586 */
2587 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2588 sk->sk_write_space(sk);
2589 len = 1;
2590 }
2591 /*
2592 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2593 * could not do because of in-flight packets
2594 */
2595 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2596 __sk_free(sk);
2597 }
2598 EXPORT_SYMBOL(sock_wfree);
2599
2600 /* This variant of sock_wfree() is used by TCP,
2601 * since it sets SOCK_USE_WRITE_QUEUE.
2602 */
2603 void __sock_wfree(struct sk_buff *skb)
2604 {
2605 struct sock *sk = skb->sk;
2606
2607 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2608 __sk_free(sk);
2609 }
2610
2611 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2612 {
2613 skb_orphan(skb);
2614 #ifdef CONFIG_INET
2615 if (unlikely(!sk_fullsock(sk)))
2616 return skb_set_owner_edemux(skb, sk);
2617 #endif
2618 skb->sk = sk;
2619 skb->destructor = sock_wfree;
2620 skb_set_hash_from_sk(skb, sk);
2621 /*
2622 * We used to take a refcount on sk, but following operation
2623 * is enough to guarantee sk_free() won't free this sock until
2624 * all in-flight packets are completed
2625 */
2626 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2627 }
2628 EXPORT_SYMBOL(skb_set_owner_w);
2629
2630 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2631 {
2632 /* Drivers depend on in-order delivery for crypto offload,
2633 * partial orphan breaks out-of-order-OK logic.
2634 */
2635 if (skb_is_decrypted(skb))
2636 return false;
2637
2638 return (skb->destructor == sock_wfree ||
2639 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2640 }
2641
2642 /* This helper is used by netem, as it can hold packets in its
2643 * delay queue. We want to allow the owner socket to send more
2644 * packets, as if they were already TX completed by a typical driver.
2645 * But we also want to keep skb->sk set because some packet schedulers
2646 * rely on it (sch_fq for example).
2647 */
2648 void skb_orphan_partial(struct sk_buff *skb)
2649 {
2650 if (skb_is_tcp_pure_ack(skb))
2651 return;
2652
2653 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2654 return;
2655
2656 skb_orphan(skb);
2657 }
2658 EXPORT_SYMBOL(skb_orphan_partial);
2659
2660 /*
2661 * Read buffer destructor automatically called from kfree_skb.
2662 */
2663 void sock_rfree(struct sk_buff *skb)
2664 {
2665 struct sock *sk = skb->sk;
2666 unsigned int len = skb->truesize;
2667
2668 atomic_sub(len, &sk->sk_rmem_alloc);
2669 sk_mem_uncharge(sk, len);
2670 }
2671 EXPORT_SYMBOL(sock_rfree);
2672
2673 /*
2674 * Buffer destructor for skbs that are not used directly in read or write
2675 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2676 */
2677 void sock_efree(struct sk_buff *skb)
2678 {
2679 sock_put(skb->sk);
2680 }
2681 EXPORT_SYMBOL(sock_efree);
2682
2683 /* Buffer destructor for prefetch/receive path where reference count may
2684 * not be held, e.g. for listen sockets.
2685 */
2686 #ifdef CONFIG_INET
2687 void sock_pfree(struct sk_buff *skb)
2688 {
2689 struct sock *sk = skb->sk;
2690
2691 if (!sk_is_refcounted(sk))
2692 return;
2693
2694 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2695 inet_reqsk(sk)->rsk_listener = NULL;
2696 reqsk_free(inet_reqsk(sk));
2697 return;
2698 }
2699
2700 sock_gen_put(sk);
2701 }
2702 EXPORT_SYMBOL(sock_pfree);
2703 #endif /* CONFIG_INET */
2704
2705 kuid_t sock_i_uid(struct sock *sk)
2706 {
2707 kuid_t uid;
2708
2709 read_lock_bh(&sk->sk_callback_lock);
2710 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2711 read_unlock_bh(&sk->sk_callback_lock);
2712 return uid;
2713 }
2714 EXPORT_SYMBOL(sock_i_uid);
2715
2716 unsigned long __sock_i_ino(struct sock *sk)
2717 {
2718 unsigned long ino;
2719
2720 read_lock(&sk->sk_callback_lock);
2721 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2722 read_unlock(&sk->sk_callback_lock);
2723 return ino;
2724 }
2725 EXPORT_SYMBOL(__sock_i_ino);
2726
2727 unsigned long sock_i_ino(struct sock *sk)
2728 {
2729 unsigned long ino;
2730
2731 local_bh_disable();
2732 ino = __sock_i_ino(sk);
2733 local_bh_enable();
2734 return ino;
2735 }
2736 EXPORT_SYMBOL(sock_i_ino);
2737
2738 /*
2739 * Allocate a skb from the socket's send buffer.
2740 */
2741 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2742 gfp_t priority)
2743 {
2744 if (force ||
2745 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2746 struct sk_buff *skb = alloc_skb(size, priority);
2747
2748 if (skb) {
2749 skb_set_owner_w(skb, sk);
2750 return skb;
2751 }
2752 }
2753 return NULL;
2754 }
2755 EXPORT_SYMBOL(sock_wmalloc);
2756
2757 static void sock_ofree(struct sk_buff *skb)
2758 {
2759 struct sock *sk = skb->sk;
2760
2761 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2762 }
2763
2764 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2765 gfp_t priority)
2766 {
2767 struct sk_buff *skb;
2768
2769 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2770 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2771 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2772 return NULL;
2773
2774 skb = alloc_skb(size, priority);
2775 if (!skb)
2776 return NULL;
2777
2778 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2779 skb->sk = sk;
2780 skb->destructor = sock_ofree;
2781 return skb;
2782 }
2783
2784 /*
2785 * Allocate a memory block from the socket's option memory buffer.
2786 */
2787 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2788 {
2789 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2790
2791 if ((unsigned int)size <= optmem_max &&
2792 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2793 void *mem;
2794 /* First do the add, to avoid the race if kmalloc
2795 * might sleep.
2796 */
2797 atomic_add(size, &sk->sk_omem_alloc);
2798 mem = kmalloc(size, priority);
2799 if (mem)
2800 return mem;
2801 atomic_sub(size, &sk->sk_omem_alloc);
2802 }
2803 return NULL;
2804 }
2805 EXPORT_SYMBOL(sock_kmalloc);
2806
2807 /* Free an option memory block. Note, we actually want the inline
2808 * here as this allows gcc to detect the nullify and fold away the
2809 * condition entirely.
2810 */
2811 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2812 const bool nullify)
2813 {
2814 if (WARN_ON_ONCE(!mem))
2815 return;
2816 if (nullify)
2817 kfree_sensitive(mem);
2818 else
2819 kfree(mem);
2820 atomic_sub(size, &sk->sk_omem_alloc);
2821 }
2822
2823 void sock_kfree_s(struct sock *sk, void *mem, int size)
2824 {
2825 __sock_kfree_s(sk, mem, size, false);
2826 }
2827 EXPORT_SYMBOL(sock_kfree_s);
2828
2829 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2830 {
2831 __sock_kfree_s(sk, mem, size, true);
2832 }
2833 EXPORT_SYMBOL(sock_kzfree_s);
2834
2835 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2836 I think, these locks should be removed for datagram sockets.
2837 */
2838 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2839 {
2840 DEFINE_WAIT(wait);
2841
2842 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2843 for (;;) {
2844 if (!timeo)
2845 break;
2846 if (signal_pending(current))
2847 break;
2848 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2849 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2850 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2851 break;
2852 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2853 break;
2854 if (READ_ONCE(sk->sk_err))
2855 break;
2856 timeo = schedule_timeout(timeo);
2857 }
2858 finish_wait(sk_sleep(sk), &wait);
2859 return timeo;
2860 }
2861
2862
2863 /*
2864 * Generic send/receive buffer handlers
2865 */
2866
2867 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2868 unsigned long data_len, int noblock,
2869 int *errcode, int max_page_order)
2870 {
2871 struct sk_buff *skb;
2872 long timeo;
2873 int err;
2874
2875 timeo = sock_sndtimeo(sk, noblock);
2876 for (;;) {
2877 err = sock_error(sk);
2878 if (err != 0)
2879 goto failure;
2880
2881 err = -EPIPE;
2882 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2883 goto failure;
2884
2885 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2886 break;
2887
2888 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2889 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2890 err = -EAGAIN;
2891 if (!timeo)
2892 goto failure;
2893 if (signal_pending(current))
2894 goto interrupted;
2895 timeo = sock_wait_for_wmem(sk, timeo);
2896 }
2897 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2898 errcode, sk->sk_allocation);
2899 if (skb)
2900 skb_set_owner_w(skb, sk);
2901 return skb;
2902
2903 interrupted:
2904 err = sock_intr_errno(timeo);
2905 failure:
2906 *errcode = err;
2907 return NULL;
2908 }
2909 EXPORT_SYMBOL(sock_alloc_send_pskb);
2910
2911 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2912 struct sockcm_cookie *sockc)
2913 {
2914 u32 tsflags;
2915
2916 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
2917
2918 switch (cmsg->cmsg_type) {
2919 case SO_MARK:
2920 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2921 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2922 return -EPERM;
2923 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2924 return -EINVAL;
2925 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2926 break;
2927 case SO_TIMESTAMPING_OLD:
2928 case SO_TIMESTAMPING_NEW:
2929 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2930 return -EINVAL;
2931
2932 tsflags = *(u32 *)CMSG_DATA(cmsg);
2933 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2934 return -EINVAL;
2935
2936 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2937 sockc->tsflags |= tsflags;
2938 break;
2939 case SCM_TXTIME:
2940 if (!sock_flag(sk, SOCK_TXTIME))
2941 return -EINVAL;
2942 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2943 return -EINVAL;
2944 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2945 break;
2946 case SCM_TS_OPT_ID:
2947 if (sk_is_tcp(sk))
2948 return -EINVAL;
2949 tsflags = READ_ONCE(sk->sk_tsflags);
2950 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
2951 return -EINVAL;
2952 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2953 return -EINVAL;
2954 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
2955 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
2956 break;
2957 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2958 case SCM_RIGHTS:
2959 case SCM_CREDENTIALS:
2960 break;
2961 case SO_PRIORITY:
2962 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2963 return -EINVAL;
2964 if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg)))
2965 return -EPERM;
2966 sockc->priority = *(u32 *)CMSG_DATA(cmsg);
2967 break;
2968 default:
2969 return -EINVAL;
2970 }
2971 return 0;
2972 }
2973 EXPORT_SYMBOL(__sock_cmsg_send);
2974
2975 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2976 struct sockcm_cookie *sockc)
2977 {
2978 struct cmsghdr *cmsg;
2979 int ret;
2980
2981 for_each_cmsghdr(cmsg, msg) {
2982 if (!CMSG_OK(msg, cmsg))
2983 return -EINVAL;
2984 if (cmsg->cmsg_level != SOL_SOCKET)
2985 continue;
2986 ret = __sock_cmsg_send(sk, cmsg, sockc);
2987 if (ret)
2988 return ret;
2989 }
2990 return 0;
2991 }
2992 EXPORT_SYMBOL(sock_cmsg_send);
2993
2994 static void sk_enter_memory_pressure(struct sock *sk)
2995 {
2996 if (!sk->sk_prot->enter_memory_pressure)
2997 return;
2998
2999 sk->sk_prot->enter_memory_pressure(sk);
3000 }
3001
3002 static void sk_leave_memory_pressure(struct sock *sk)
3003 {
3004 if (sk->sk_prot->leave_memory_pressure) {
3005 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3006 tcp_leave_memory_pressure, sk);
3007 } else {
3008 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3009
3010 if (memory_pressure && READ_ONCE(*memory_pressure))
3011 WRITE_ONCE(*memory_pressure, 0);
3012 }
3013 }
3014
3015 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3016
3017 /**
3018 * skb_page_frag_refill - check that a page_frag contains enough room
3019 * @sz: minimum size of the fragment we want to get
3020 * @pfrag: pointer to page_frag
3021 * @gfp: priority for memory allocation
3022 *
3023 * Note: While this allocator tries to use high order pages, there is
3024 * no guarantee that allocations succeed. Therefore, @sz MUST be
3025 * less or equal than PAGE_SIZE.
3026 */
3027 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3028 {
3029 if (pfrag->page) {
3030 if (page_ref_count(pfrag->page) == 1) {
3031 pfrag->offset = 0;
3032 return true;
3033 }
3034 if (pfrag->offset + sz <= pfrag->size)
3035 return true;
3036 put_page(pfrag->page);
3037 }
3038
3039 pfrag->offset = 0;
3040 if (SKB_FRAG_PAGE_ORDER &&
3041 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3042 /* Avoid direct reclaim but allow kswapd to wake */
3043 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3044 __GFP_COMP | __GFP_NOWARN |
3045 __GFP_NORETRY,
3046 SKB_FRAG_PAGE_ORDER);
3047 if (likely(pfrag->page)) {
3048 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3049 return true;
3050 }
3051 }
3052 pfrag->page = alloc_page(gfp);
3053 if (likely(pfrag->page)) {
3054 pfrag->size = PAGE_SIZE;
3055 return true;
3056 }
3057 return false;
3058 }
3059 EXPORT_SYMBOL(skb_page_frag_refill);
3060
3061 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3062 {
3063 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3064 return true;
3065
3066 sk_enter_memory_pressure(sk);
3067 sk_stream_moderate_sndbuf(sk);
3068 return false;
3069 }
3070 EXPORT_SYMBOL(sk_page_frag_refill);
3071
3072 void __lock_sock(struct sock *sk)
3073 __releases(&sk->sk_lock.slock)
3074 __acquires(&sk->sk_lock.slock)
3075 {
3076 DEFINE_WAIT(wait);
3077
3078 for (;;) {
3079 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3080 TASK_UNINTERRUPTIBLE);
3081 spin_unlock_bh(&sk->sk_lock.slock);
3082 schedule();
3083 spin_lock_bh(&sk->sk_lock.slock);
3084 if (!sock_owned_by_user(sk))
3085 break;
3086 }
3087 finish_wait(&sk->sk_lock.wq, &wait);
3088 }
3089
3090 void __release_sock(struct sock *sk)
3091 __releases(&sk->sk_lock.slock)
3092 __acquires(&sk->sk_lock.slock)
3093 {
3094 struct sk_buff *skb, *next;
3095
3096 while ((skb = sk->sk_backlog.head) != NULL) {
3097 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3098
3099 spin_unlock_bh(&sk->sk_lock.slock);
3100
3101 do {
3102 next = skb->next;
3103 prefetch(next);
3104 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3105 skb_mark_not_on_list(skb);
3106 sk_backlog_rcv(sk, skb);
3107
3108 cond_resched();
3109
3110 skb = next;
3111 } while (skb != NULL);
3112
3113 spin_lock_bh(&sk->sk_lock.slock);
3114 }
3115
3116 /*
3117 * Doing the zeroing here guarantee we can not loop forever
3118 * while a wild producer attempts to flood us.
3119 */
3120 sk->sk_backlog.len = 0;
3121 }
3122
3123 void __sk_flush_backlog(struct sock *sk)
3124 {
3125 spin_lock_bh(&sk->sk_lock.slock);
3126 __release_sock(sk);
3127
3128 if (sk->sk_prot->release_cb)
3129 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3130 tcp_release_cb, sk);
3131
3132 spin_unlock_bh(&sk->sk_lock.slock);
3133 }
3134 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3135
3136 /**
3137 * sk_wait_data - wait for data to arrive at sk_receive_queue
3138 * @sk: sock to wait on
3139 * @timeo: for how long
3140 * @skb: last skb seen on sk_receive_queue
3141 *
3142 * Now socket state including sk->sk_err is changed only under lock,
3143 * hence we may omit checks after joining wait queue.
3144 * We check receive queue before schedule() only as optimization;
3145 * it is very likely that release_sock() added new data.
3146 */
3147 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3148 {
3149 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3150 int rc;
3151
3152 add_wait_queue(sk_sleep(sk), &wait);
3153 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3154 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3155 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3156 remove_wait_queue(sk_sleep(sk), &wait);
3157 return rc;
3158 }
3159 EXPORT_SYMBOL(sk_wait_data);
3160
3161 /**
3162 * __sk_mem_raise_allocated - increase memory_allocated
3163 * @sk: socket
3164 * @size: memory size to allocate
3165 * @amt: pages to allocate
3166 * @kind: allocation type
3167 *
3168 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3169 *
3170 * Unlike the globally shared limits among the sockets under same protocol,
3171 * consuming the budget of a memcg won't have direct effect on other ones.
3172 * So be optimistic about memcg's tolerance, and leave the callers to decide
3173 * whether or not to raise allocated through sk_under_memory_pressure() or
3174 * its variants.
3175 */
3176 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3177 {
3178 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3179 struct proto *prot = sk->sk_prot;
3180 bool charged = false;
3181 long allocated;
3182
3183 sk_memory_allocated_add(sk, amt);
3184 allocated = sk_memory_allocated(sk);
3185
3186 if (memcg) {
3187 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3188 goto suppress_allocation;
3189 charged = true;
3190 }
3191
3192 /* Under limit. */
3193 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3194 sk_leave_memory_pressure(sk);
3195 return 1;
3196 }
3197
3198 /* Under pressure. */
3199 if (allocated > sk_prot_mem_limits(sk, 1))
3200 sk_enter_memory_pressure(sk);
3201
3202 /* Over hard limit. */
3203 if (allocated > sk_prot_mem_limits(sk, 2))
3204 goto suppress_allocation;
3205
3206 /* Guarantee minimum buffer size under pressure (either global
3207 * or memcg) to make sure features described in RFC 7323 (TCP
3208 * Extensions for High Performance) work properly.
3209 *
3210 * This rule does NOT stand when exceeds global or memcg's hard
3211 * limit, or else a DoS attack can be taken place by spawning
3212 * lots of sockets whose usage are under minimum buffer size.
3213 */
3214 if (kind == SK_MEM_RECV) {
3215 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3216 return 1;
3217
3218 } else { /* SK_MEM_SEND */
3219 int wmem0 = sk_get_wmem0(sk, prot);
3220
3221 if (sk->sk_type == SOCK_STREAM) {
3222 if (sk->sk_wmem_queued < wmem0)
3223 return 1;
3224 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3225 return 1;
3226 }
3227 }
3228
3229 if (sk_has_memory_pressure(sk)) {
3230 u64 alloc;
3231
3232 /* The following 'average' heuristic is within the
3233 * scope of global accounting, so it only makes
3234 * sense for global memory pressure.
3235 */
3236 if (!sk_under_global_memory_pressure(sk))
3237 return 1;
3238
3239 /* Try to be fair among all the sockets under global
3240 * pressure by allowing the ones that below average
3241 * usage to raise.
3242 */
3243 alloc = sk_sockets_allocated_read_positive(sk);
3244 if (sk_prot_mem_limits(sk, 2) > alloc *
3245 sk_mem_pages(sk->sk_wmem_queued +
3246 atomic_read(&sk->sk_rmem_alloc) +
3247 sk->sk_forward_alloc))
3248 return 1;
3249 }
3250
3251 suppress_allocation:
3252
3253 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3254 sk_stream_moderate_sndbuf(sk);
3255
3256 /* Fail only if socket is _under_ its sndbuf.
3257 * In this case we cannot block, so that we have to fail.
3258 */
3259 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3260 /* Force charge with __GFP_NOFAIL */
3261 if (memcg && !charged) {
3262 mem_cgroup_charge_skmem(memcg, amt,
3263 gfp_memcg_charge() | __GFP_NOFAIL);
3264 }
3265 return 1;
3266 }
3267 }
3268
3269 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3270 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3271
3272 sk_memory_allocated_sub(sk, amt);
3273
3274 if (charged)
3275 mem_cgroup_uncharge_skmem(memcg, amt);
3276
3277 return 0;
3278 }
3279
3280 /**
3281 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3282 * @sk: socket
3283 * @size: memory size to allocate
3284 * @kind: allocation type
3285 *
3286 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3287 * rmem allocation. This function assumes that protocols which have
3288 * memory_pressure use sk_wmem_queued as write buffer accounting.
3289 */
3290 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3291 {
3292 int ret, amt = sk_mem_pages(size);
3293
3294 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3295 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3296 if (!ret)
3297 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3298 return ret;
3299 }
3300 EXPORT_SYMBOL(__sk_mem_schedule);
3301
3302 /**
3303 * __sk_mem_reduce_allocated - reclaim memory_allocated
3304 * @sk: socket
3305 * @amount: number of quanta
3306 *
3307 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3308 */
3309 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3310 {
3311 sk_memory_allocated_sub(sk, amount);
3312
3313 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3314 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3315
3316 if (sk_under_global_memory_pressure(sk) &&
3317 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3318 sk_leave_memory_pressure(sk);
3319 }
3320
3321 /**
3322 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3323 * @sk: socket
3324 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3325 */
3326 void __sk_mem_reclaim(struct sock *sk, int amount)
3327 {
3328 amount >>= PAGE_SHIFT;
3329 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3330 __sk_mem_reduce_allocated(sk, amount);
3331 }
3332 EXPORT_SYMBOL(__sk_mem_reclaim);
3333
3334 int sk_set_peek_off(struct sock *sk, int val)
3335 {
3336 WRITE_ONCE(sk->sk_peek_off, val);
3337 return 0;
3338 }
3339 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3340
3341 /*
3342 * Set of default routines for initialising struct proto_ops when
3343 * the protocol does not support a particular function. In certain
3344 * cases where it makes no sense for a protocol to have a "do nothing"
3345 * function, some default processing is provided.
3346 */
3347
3348 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3349 {
3350 return -EOPNOTSUPP;
3351 }
3352 EXPORT_SYMBOL(sock_no_bind);
3353
3354 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3355 int len, int flags)
3356 {
3357 return -EOPNOTSUPP;
3358 }
3359 EXPORT_SYMBOL(sock_no_connect);
3360
3361 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3362 {
3363 return -EOPNOTSUPP;
3364 }
3365 EXPORT_SYMBOL(sock_no_socketpair);
3366
3367 int sock_no_accept(struct socket *sock, struct socket *newsock,
3368 struct proto_accept_arg *arg)
3369 {
3370 return -EOPNOTSUPP;
3371 }
3372 EXPORT_SYMBOL(sock_no_accept);
3373
3374 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3375 int peer)
3376 {
3377 return -EOPNOTSUPP;
3378 }
3379 EXPORT_SYMBOL(sock_no_getname);
3380
3381 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3382 {
3383 return -EOPNOTSUPP;
3384 }
3385 EXPORT_SYMBOL(sock_no_ioctl);
3386
3387 int sock_no_listen(struct socket *sock, int backlog)
3388 {
3389 return -EOPNOTSUPP;
3390 }
3391 EXPORT_SYMBOL(sock_no_listen);
3392
3393 int sock_no_shutdown(struct socket *sock, int how)
3394 {
3395 return -EOPNOTSUPP;
3396 }
3397 EXPORT_SYMBOL(sock_no_shutdown);
3398
3399 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3400 {
3401 return -EOPNOTSUPP;
3402 }
3403 EXPORT_SYMBOL(sock_no_sendmsg);
3404
3405 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3406 {
3407 return -EOPNOTSUPP;
3408 }
3409 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3410
3411 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3412 int flags)
3413 {
3414 return -EOPNOTSUPP;
3415 }
3416 EXPORT_SYMBOL(sock_no_recvmsg);
3417
3418 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3419 {
3420 /* Mirror missing mmap method error code */
3421 return -ENODEV;
3422 }
3423 EXPORT_SYMBOL(sock_no_mmap);
3424
3425 /*
3426 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3427 * various sock-based usage counts.
3428 */
3429 void __receive_sock(struct file *file)
3430 {
3431 struct socket *sock;
3432
3433 sock = sock_from_file(file);
3434 if (sock) {
3435 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3436 sock_update_classid(&sock->sk->sk_cgrp_data);
3437 }
3438 }
3439
3440 /*
3441 * Default Socket Callbacks
3442 */
3443
3444 static void sock_def_wakeup(struct sock *sk)
3445 {
3446 struct socket_wq *wq;
3447
3448 rcu_read_lock();
3449 wq = rcu_dereference(sk->sk_wq);
3450 if (skwq_has_sleeper(wq))
3451 wake_up_interruptible_all(&wq->wait);
3452 rcu_read_unlock();
3453 }
3454
3455 static void sock_def_error_report(struct sock *sk)
3456 {
3457 struct socket_wq *wq;
3458
3459 rcu_read_lock();
3460 wq = rcu_dereference(sk->sk_wq);
3461 if (skwq_has_sleeper(wq))
3462 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3463 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3464 rcu_read_unlock();
3465 }
3466
3467 void sock_def_readable(struct sock *sk)
3468 {
3469 struct socket_wq *wq;
3470
3471 trace_sk_data_ready(sk);
3472
3473 rcu_read_lock();
3474 wq = rcu_dereference(sk->sk_wq);
3475 if (skwq_has_sleeper(wq))
3476 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3477 EPOLLRDNORM | EPOLLRDBAND);
3478 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3479 rcu_read_unlock();
3480 }
3481
3482 static void sock_def_write_space(struct sock *sk)
3483 {
3484 struct socket_wq *wq;
3485
3486 rcu_read_lock();
3487
3488 /* Do not wake up a writer until he can make "significant"
3489 * progress. --DaveM
3490 */
3491 if (sock_writeable(sk)) {
3492 wq = rcu_dereference(sk->sk_wq);
3493 if (skwq_has_sleeper(wq))
3494 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3495 EPOLLWRNORM | EPOLLWRBAND);
3496
3497 /* Should agree with poll, otherwise some programs break */
3498 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3499 }
3500
3501 rcu_read_unlock();
3502 }
3503
3504 /* An optimised version of sock_def_write_space(), should only be called
3505 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3506 * ->sk_wmem_alloc.
3507 */
3508 static void sock_def_write_space_wfree(struct sock *sk)
3509 {
3510 /* Do not wake up a writer until he can make "significant"
3511 * progress. --DaveM
3512 */
3513 if (sock_writeable(sk)) {
3514 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3515
3516 /* rely on refcount_sub from sock_wfree() */
3517 smp_mb__after_atomic();
3518 if (wq && waitqueue_active(&wq->wait))
3519 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3520 EPOLLWRNORM | EPOLLWRBAND);
3521
3522 /* Should agree with poll, otherwise some programs break */
3523 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3524 }
3525 }
3526
3527 static void sock_def_destruct(struct sock *sk)
3528 {
3529 }
3530
3531 void sk_send_sigurg(struct sock *sk)
3532 {
3533 if (sk->sk_socket && sk->sk_socket->file)
3534 if (send_sigurg(sk->sk_socket->file))
3535 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3536 }
3537 EXPORT_SYMBOL(sk_send_sigurg);
3538
3539 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3540 unsigned long expires)
3541 {
3542 if (!mod_timer(timer, expires))
3543 sock_hold(sk);
3544 }
3545 EXPORT_SYMBOL(sk_reset_timer);
3546
3547 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3548 {
3549 if (del_timer(timer))
3550 __sock_put(sk);
3551 }
3552 EXPORT_SYMBOL(sk_stop_timer);
3553
3554 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3555 {
3556 if (del_timer_sync(timer))
3557 __sock_put(sk);
3558 }
3559 EXPORT_SYMBOL(sk_stop_timer_sync);
3560
3561 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3562 {
3563 sk_init_common(sk);
3564 sk->sk_send_head = NULL;
3565
3566 timer_setup(&sk->sk_timer, NULL, 0);
3567
3568 sk->sk_allocation = GFP_KERNEL;
3569 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3570 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3571 sk->sk_state = TCP_CLOSE;
3572 sk->sk_use_task_frag = true;
3573 sk_set_socket(sk, sock);
3574
3575 sock_set_flag(sk, SOCK_ZAPPED);
3576
3577 if (sock) {
3578 sk->sk_type = sock->type;
3579 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3580 sock->sk = sk;
3581 } else {
3582 RCU_INIT_POINTER(sk->sk_wq, NULL);
3583 }
3584 sk->sk_uid = uid;
3585
3586 sk->sk_state_change = sock_def_wakeup;
3587 sk->sk_data_ready = sock_def_readable;
3588 sk->sk_write_space = sock_def_write_space;
3589 sk->sk_error_report = sock_def_error_report;
3590 sk->sk_destruct = sock_def_destruct;
3591
3592 sk->sk_frag.page = NULL;
3593 sk->sk_frag.offset = 0;
3594 sk->sk_peek_off = -1;
3595
3596 sk->sk_peer_pid = NULL;
3597 sk->sk_peer_cred = NULL;
3598 spin_lock_init(&sk->sk_peer_lock);
3599
3600 sk->sk_write_pending = 0;
3601 sk->sk_rcvlowat = 1;
3602 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3603 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3604
3605 sk->sk_stamp = SK_DEFAULT_STAMP;
3606 #if BITS_PER_LONG==32
3607 seqlock_init(&sk->sk_stamp_seq);
3608 #endif
3609 atomic_set(&sk->sk_zckey, 0);
3610
3611 #ifdef CONFIG_NET_RX_BUSY_POLL
3612 sk->sk_napi_id = 0;
3613 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3614 #endif
3615
3616 sk->sk_max_pacing_rate = ~0UL;
3617 sk->sk_pacing_rate = ~0UL;
3618 WRITE_ONCE(sk->sk_pacing_shift, 10);
3619 sk->sk_incoming_cpu = -1;
3620
3621 sk_rx_queue_clear(sk);
3622 /*
3623 * Before updating sk_refcnt, we must commit prior changes to memory
3624 * (Documentation/RCU/rculist_nulls.rst for details)
3625 */
3626 smp_wmb();
3627 refcount_set(&sk->sk_refcnt, 1);
3628 atomic_set(&sk->sk_drops, 0);
3629 }
3630 EXPORT_SYMBOL(sock_init_data_uid);
3631
3632 void sock_init_data(struct socket *sock, struct sock *sk)
3633 {
3634 kuid_t uid = sock ?
3635 SOCK_INODE(sock)->i_uid :
3636 make_kuid(sock_net(sk)->user_ns, 0);
3637
3638 sock_init_data_uid(sock, sk, uid);
3639 }
3640 EXPORT_SYMBOL(sock_init_data);
3641
3642 void lock_sock_nested(struct sock *sk, int subclass)
3643 {
3644 /* The sk_lock has mutex_lock() semantics here. */
3645 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3646
3647 might_sleep();
3648 spin_lock_bh(&sk->sk_lock.slock);
3649 if (sock_owned_by_user_nocheck(sk))
3650 __lock_sock(sk);
3651 sk->sk_lock.owned = 1;
3652 spin_unlock_bh(&sk->sk_lock.slock);
3653 }
3654 EXPORT_SYMBOL(lock_sock_nested);
3655
3656 void release_sock(struct sock *sk)
3657 {
3658 spin_lock_bh(&sk->sk_lock.slock);
3659 if (sk->sk_backlog.tail)
3660 __release_sock(sk);
3661
3662 if (sk->sk_prot->release_cb)
3663 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3664 tcp_release_cb, sk);
3665
3666 sock_release_ownership(sk);
3667 if (waitqueue_active(&sk->sk_lock.wq))
3668 wake_up(&sk->sk_lock.wq);
3669 spin_unlock_bh(&sk->sk_lock.slock);
3670 }
3671 EXPORT_SYMBOL(release_sock);
3672
3673 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3674 {
3675 might_sleep();
3676 spin_lock_bh(&sk->sk_lock.slock);
3677
3678 if (!sock_owned_by_user_nocheck(sk)) {
3679 /*
3680 * Fast path return with bottom halves disabled and
3681 * sock::sk_lock.slock held.
3682 *
3683 * The 'mutex' is not contended and holding
3684 * sock::sk_lock.slock prevents all other lockers to
3685 * proceed so the corresponding unlock_sock_fast() can
3686 * avoid the slow path of release_sock() completely and
3687 * just release slock.
3688 *
3689 * From a semantical POV this is equivalent to 'acquiring'
3690 * the 'mutex', hence the corresponding lockdep
3691 * mutex_release() has to happen in the fast path of
3692 * unlock_sock_fast().
3693 */
3694 return false;
3695 }
3696
3697 __lock_sock(sk);
3698 sk->sk_lock.owned = 1;
3699 __acquire(&sk->sk_lock.slock);
3700 spin_unlock_bh(&sk->sk_lock.slock);
3701 return true;
3702 }
3703 EXPORT_SYMBOL(__lock_sock_fast);
3704
3705 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3706 bool timeval, bool time32)
3707 {
3708 struct sock *sk = sock->sk;
3709 struct timespec64 ts;
3710
3711 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3712 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3713 if (ts.tv_sec == -1)
3714 return -ENOENT;
3715 if (ts.tv_sec == 0) {
3716 ktime_t kt = ktime_get_real();
3717 sock_write_timestamp(sk, kt);
3718 ts = ktime_to_timespec64(kt);
3719 }
3720
3721 if (timeval)
3722 ts.tv_nsec /= 1000;
3723
3724 #ifdef CONFIG_COMPAT_32BIT_TIME
3725 if (time32)
3726 return put_old_timespec32(&ts, userstamp);
3727 #endif
3728 #ifdef CONFIG_SPARC64
3729 /* beware of padding in sparc64 timeval */
3730 if (timeval && !in_compat_syscall()) {
3731 struct __kernel_old_timeval __user tv = {
3732 .tv_sec = ts.tv_sec,
3733 .tv_usec = ts.tv_nsec,
3734 };
3735 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3736 return -EFAULT;
3737 return 0;
3738 }
3739 #endif
3740 return put_timespec64(&ts, userstamp);
3741 }
3742 EXPORT_SYMBOL(sock_gettstamp);
3743
3744 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3745 {
3746 if (!sock_flag(sk, flag)) {
3747 unsigned long previous_flags = sk->sk_flags;
3748
3749 sock_set_flag(sk, flag);
3750 /*
3751 * we just set one of the two flags which require net
3752 * time stamping, but time stamping might have been on
3753 * already because of the other one
3754 */
3755 if (sock_needs_netstamp(sk) &&
3756 !(previous_flags & SK_FLAGS_TIMESTAMP))
3757 net_enable_timestamp();
3758 }
3759 }
3760
3761 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3762 int level, int type)
3763 {
3764 struct sock_exterr_skb *serr;
3765 struct sk_buff *skb;
3766 int copied, err;
3767
3768 err = -EAGAIN;
3769 skb = sock_dequeue_err_skb(sk);
3770 if (skb == NULL)
3771 goto out;
3772
3773 copied = skb->len;
3774 if (copied > len) {
3775 msg->msg_flags |= MSG_TRUNC;
3776 copied = len;
3777 }
3778 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3779 if (err)
3780 goto out_free_skb;
3781
3782 sock_recv_timestamp(msg, sk, skb);
3783
3784 serr = SKB_EXT_ERR(skb);
3785 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3786
3787 msg->msg_flags |= MSG_ERRQUEUE;
3788 err = copied;
3789
3790 out_free_skb:
3791 kfree_skb(skb);
3792 out:
3793 return err;
3794 }
3795 EXPORT_SYMBOL(sock_recv_errqueue);
3796
3797 /*
3798 * Get a socket option on an socket.
3799 *
3800 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3801 * asynchronous errors should be reported by getsockopt. We assume
3802 * this means if you specify SO_ERROR (otherwise what is the point of it).
3803 */
3804 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3805 char __user *optval, int __user *optlen)
3806 {
3807 struct sock *sk = sock->sk;
3808
3809 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3810 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3811 }
3812 EXPORT_SYMBOL(sock_common_getsockopt);
3813
3814 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3815 int flags)
3816 {
3817 struct sock *sk = sock->sk;
3818 int addr_len = 0;
3819 int err;
3820
3821 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3822 if (err >= 0)
3823 msg->msg_namelen = addr_len;
3824 return err;
3825 }
3826 EXPORT_SYMBOL(sock_common_recvmsg);
3827
3828 /*
3829 * Set socket options on an inet socket.
3830 */
3831 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3832 sockptr_t optval, unsigned int optlen)
3833 {
3834 struct sock *sk = sock->sk;
3835
3836 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3837 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3838 }
3839 EXPORT_SYMBOL(sock_common_setsockopt);
3840
3841 void sk_common_release(struct sock *sk)
3842 {
3843 if (sk->sk_prot->destroy)
3844 sk->sk_prot->destroy(sk);
3845
3846 /*
3847 * Observation: when sk_common_release is called, processes have
3848 * no access to socket. But net still has.
3849 * Step one, detach it from networking:
3850 *
3851 * A. Remove from hash tables.
3852 */
3853
3854 sk->sk_prot->unhash(sk);
3855
3856 /*
3857 * In this point socket cannot receive new packets, but it is possible
3858 * that some packets are in flight because some CPU runs receiver and
3859 * did hash table lookup before we unhashed socket. They will achieve
3860 * receive queue and will be purged by socket destructor.
3861 *
3862 * Also we still have packets pending on receive queue and probably,
3863 * our own packets waiting in device queues. sock_destroy will drain
3864 * receive queue, but transmitted packets will delay socket destruction
3865 * until the last reference will be released.
3866 */
3867
3868 sock_orphan(sk);
3869
3870 xfrm_sk_free_policy(sk);
3871
3872 sock_put(sk);
3873 }
3874 EXPORT_SYMBOL(sk_common_release);
3875
3876 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3877 {
3878 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3879
3880 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3881 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3882 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3883 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3884 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3885 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3886 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3887 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3888 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3889 }
3890
3891 #ifdef CONFIG_PROC_FS
3892 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3893
3894 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3895 {
3896 int cpu, idx = prot->inuse_idx;
3897 int res = 0;
3898
3899 for_each_possible_cpu(cpu)
3900 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3901
3902 return res >= 0 ? res : 0;
3903 }
3904 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3905
3906 int sock_inuse_get(struct net *net)
3907 {
3908 int cpu, res = 0;
3909
3910 for_each_possible_cpu(cpu)
3911 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3912
3913 return res;
3914 }
3915
3916 EXPORT_SYMBOL_GPL(sock_inuse_get);
3917
3918 static int __net_init sock_inuse_init_net(struct net *net)
3919 {
3920 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3921 if (net->core.prot_inuse == NULL)
3922 return -ENOMEM;
3923 return 0;
3924 }
3925
3926 static void __net_exit sock_inuse_exit_net(struct net *net)
3927 {
3928 free_percpu(net->core.prot_inuse);
3929 }
3930
3931 static struct pernet_operations net_inuse_ops = {
3932 .init = sock_inuse_init_net,
3933 .exit = sock_inuse_exit_net,
3934 };
3935
3936 static __init int net_inuse_init(void)
3937 {
3938 if (register_pernet_subsys(&net_inuse_ops))
3939 panic("Cannot initialize net inuse counters");
3940
3941 return 0;
3942 }
3943
3944 core_initcall(net_inuse_init);
3945
3946 static int assign_proto_idx(struct proto *prot)
3947 {
3948 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3949
3950 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3951 pr_err("PROTO_INUSE_NR exhausted\n");
3952 return -ENOSPC;
3953 }
3954
3955 set_bit(prot->inuse_idx, proto_inuse_idx);
3956 return 0;
3957 }
3958
3959 static void release_proto_idx(struct proto *prot)
3960 {
3961 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3962 clear_bit(prot->inuse_idx, proto_inuse_idx);
3963 }
3964 #else
3965 static inline int assign_proto_idx(struct proto *prot)
3966 {
3967 return 0;
3968 }
3969
3970 static inline void release_proto_idx(struct proto *prot)
3971 {
3972 }
3973
3974 #endif
3975
3976 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3977 {
3978 if (!twsk_prot)
3979 return;
3980 kfree(twsk_prot->twsk_slab_name);
3981 twsk_prot->twsk_slab_name = NULL;
3982 kmem_cache_destroy(twsk_prot->twsk_slab);
3983 twsk_prot->twsk_slab = NULL;
3984 }
3985
3986 static int tw_prot_init(const struct proto *prot)
3987 {
3988 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3989
3990 if (!twsk_prot)
3991 return 0;
3992
3993 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3994 prot->name);
3995 if (!twsk_prot->twsk_slab_name)
3996 return -ENOMEM;
3997
3998 twsk_prot->twsk_slab =
3999 kmem_cache_create(twsk_prot->twsk_slab_name,
4000 twsk_prot->twsk_obj_size, 0,
4001 SLAB_ACCOUNT | prot->slab_flags,
4002 NULL);
4003 if (!twsk_prot->twsk_slab) {
4004 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4005 prot->name);
4006 return -ENOMEM;
4007 }
4008
4009 return 0;
4010 }
4011
4012 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4013 {
4014 if (!rsk_prot)
4015 return;
4016 kfree(rsk_prot->slab_name);
4017 rsk_prot->slab_name = NULL;
4018 kmem_cache_destroy(rsk_prot->slab);
4019 rsk_prot->slab = NULL;
4020 }
4021
4022 static int req_prot_init(const struct proto *prot)
4023 {
4024 struct request_sock_ops *rsk_prot = prot->rsk_prot;
4025
4026 if (!rsk_prot)
4027 return 0;
4028
4029 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4030 prot->name);
4031 if (!rsk_prot->slab_name)
4032 return -ENOMEM;
4033
4034 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4035 rsk_prot->obj_size, 0,
4036 SLAB_ACCOUNT | prot->slab_flags,
4037 NULL);
4038
4039 if (!rsk_prot->slab) {
4040 pr_crit("%s: Can't create request sock SLAB cache!\n",
4041 prot->name);
4042 return -ENOMEM;
4043 }
4044 return 0;
4045 }
4046
4047 int proto_register(struct proto *prot, int alloc_slab)
4048 {
4049 int ret = -ENOBUFS;
4050
4051 if (prot->memory_allocated && !prot->sysctl_mem) {
4052 pr_err("%s: missing sysctl_mem\n", prot->name);
4053 return -EINVAL;
4054 }
4055 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4056 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4057 return -EINVAL;
4058 }
4059 if (alloc_slab) {
4060 prot->slab = kmem_cache_create_usercopy(prot->name,
4061 prot->obj_size, 0,
4062 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4063 prot->slab_flags,
4064 prot->useroffset, prot->usersize,
4065 NULL);
4066
4067 if (prot->slab == NULL) {
4068 pr_crit("%s: Can't create sock SLAB cache!\n",
4069 prot->name);
4070 goto out;
4071 }
4072
4073 if (req_prot_init(prot))
4074 goto out_free_request_sock_slab;
4075
4076 if (tw_prot_init(prot))
4077 goto out_free_timewait_sock_slab;
4078 }
4079
4080 mutex_lock(&proto_list_mutex);
4081 ret = assign_proto_idx(prot);
4082 if (ret) {
4083 mutex_unlock(&proto_list_mutex);
4084 goto out_free_timewait_sock_slab;
4085 }
4086 list_add(&prot->node, &proto_list);
4087 mutex_unlock(&proto_list_mutex);
4088 return ret;
4089
4090 out_free_timewait_sock_slab:
4091 if (alloc_slab)
4092 tw_prot_cleanup(prot->twsk_prot);
4093 out_free_request_sock_slab:
4094 if (alloc_slab) {
4095 req_prot_cleanup(prot->rsk_prot);
4096
4097 kmem_cache_destroy(prot->slab);
4098 prot->slab = NULL;
4099 }
4100 out:
4101 return ret;
4102 }
4103 EXPORT_SYMBOL(proto_register);
4104
4105 void proto_unregister(struct proto *prot)
4106 {
4107 mutex_lock(&proto_list_mutex);
4108 release_proto_idx(prot);
4109 list_del(&prot->node);
4110 mutex_unlock(&proto_list_mutex);
4111
4112 kmem_cache_destroy(prot->slab);
4113 prot->slab = NULL;
4114
4115 req_prot_cleanup(prot->rsk_prot);
4116 tw_prot_cleanup(prot->twsk_prot);
4117 }
4118 EXPORT_SYMBOL(proto_unregister);
4119
4120 int sock_load_diag_module(int family, int protocol)
4121 {
4122 if (!protocol) {
4123 if (!sock_is_registered(family))
4124 return -ENOENT;
4125
4126 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4127 NETLINK_SOCK_DIAG, family);
4128 }
4129
4130 #ifdef CONFIG_INET
4131 if (family == AF_INET &&
4132 protocol != IPPROTO_RAW &&
4133 protocol < MAX_INET_PROTOS &&
4134 !rcu_access_pointer(inet_protos[protocol]))
4135 return -ENOENT;
4136 #endif
4137
4138 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4139 NETLINK_SOCK_DIAG, family, protocol);
4140 }
4141 EXPORT_SYMBOL(sock_load_diag_module);
4142
4143 #ifdef CONFIG_PROC_FS
4144 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4145 __acquires(proto_list_mutex)
4146 {
4147 mutex_lock(&proto_list_mutex);
4148 return seq_list_start_head(&proto_list, *pos);
4149 }
4150
4151 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4152 {
4153 return seq_list_next(v, &proto_list, pos);
4154 }
4155
4156 static void proto_seq_stop(struct seq_file *seq, void *v)
4157 __releases(proto_list_mutex)
4158 {
4159 mutex_unlock(&proto_list_mutex);
4160 }
4161
4162 static char proto_method_implemented(const void *method)
4163 {
4164 return method == NULL ? 'n' : 'y';
4165 }
4166 static long sock_prot_memory_allocated(struct proto *proto)
4167 {
4168 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4169 }
4170
4171 static const char *sock_prot_memory_pressure(struct proto *proto)
4172 {
4173 return proto->memory_pressure != NULL ?
4174 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4175 }
4176
4177 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4178 {
4179
4180 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4181 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4182 proto->name,
4183 proto->obj_size,
4184 sock_prot_inuse_get(seq_file_net(seq), proto),
4185 sock_prot_memory_allocated(proto),
4186 sock_prot_memory_pressure(proto),
4187 proto->max_header,
4188 proto->slab == NULL ? "no" : "yes",
4189 module_name(proto->owner),
4190 proto_method_implemented(proto->close),
4191 proto_method_implemented(proto->connect),
4192 proto_method_implemented(proto->disconnect),
4193 proto_method_implemented(proto->accept),
4194 proto_method_implemented(proto->ioctl),
4195 proto_method_implemented(proto->init),
4196 proto_method_implemented(proto->destroy),
4197 proto_method_implemented(proto->shutdown),
4198 proto_method_implemented(proto->setsockopt),
4199 proto_method_implemented(proto->getsockopt),
4200 proto_method_implemented(proto->sendmsg),
4201 proto_method_implemented(proto->recvmsg),
4202 proto_method_implemented(proto->bind),
4203 proto_method_implemented(proto->backlog_rcv),
4204 proto_method_implemented(proto->hash),
4205 proto_method_implemented(proto->unhash),
4206 proto_method_implemented(proto->get_port),
4207 proto_method_implemented(proto->enter_memory_pressure));
4208 }
4209
4210 static int proto_seq_show(struct seq_file *seq, void *v)
4211 {
4212 if (v == &proto_list)
4213 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4214 "protocol",
4215 "size",
4216 "sockets",
4217 "memory",
4218 "press",
4219 "maxhdr",
4220 "slab",
4221 "module",
4222 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4223 else
4224 proto_seq_printf(seq, list_entry(v, struct proto, node));
4225 return 0;
4226 }
4227
4228 static const struct seq_operations proto_seq_ops = {
4229 .start = proto_seq_start,
4230 .next = proto_seq_next,
4231 .stop = proto_seq_stop,
4232 .show = proto_seq_show,
4233 };
4234
4235 static __net_init int proto_init_net(struct net *net)
4236 {
4237 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4238 sizeof(struct seq_net_private)))
4239 return -ENOMEM;
4240
4241 return 0;
4242 }
4243
4244 static __net_exit void proto_exit_net(struct net *net)
4245 {
4246 remove_proc_entry("protocols", net->proc_net);
4247 }
4248
4249
4250 static __net_initdata struct pernet_operations proto_net_ops = {
4251 .init = proto_init_net,
4252 .exit = proto_exit_net,
4253 };
4254
4255 static int __init proto_init(void)
4256 {
4257 return register_pernet_subsys(&proto_net_ops);
4258 }
4259
4260 subsys_initcall(proto_init);
4261
4262 #endif /* PROC_FS */
4263
4264 #ifdef CONFIG_NET_RX_BUSY_POLL
4265 bool sk_busy_loop_end(void *p, unsigned long start_time)
4266 {
4267 struct sock *sk = p;
4268
4269 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4270 return true;
4271
4272 if (sk_is_udp(sk) &&
4273 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4274 return true;
4275
4276 return sk_busy_loop_timeout(sk, start_time);
4277 }
4278 EXPORT_SYMBOL(sk_busy_loop_end);
4279 #endif /* CONFIG_NET_RX_BUSY_POLL */
4280
4281 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4282 {
4283 if (!sk->sk_prot->bind_add)
4284 return -EOPNOTSUPP;
4285 return sk->sk_prot->bind_add(sk, addr, addr_len);
4286 }
4287 EXPORT_SYMBOL(sock_bind_add);
4288
4289 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4290 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4291 void __user *arg, void *karg, size_t size)
4292 {
4293 int ret;
4294
4295 if (copy_from_user(karg, arg, size))
4296 return -EFAULT;
4297
4298 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4299 if (ret)
4300 return ret;
4301
4302 if (copy_to_user(arg, karg, size))
4303 return -EFAULT;
4304
4305 return 0;
4306 }
4307 EXPORT_SYMBOL(sock_ioctl_inout);
4308
4309 /* This is the most common ioctl prep function, where the result (4 bytes) is
4310 * copied back to userspace if the ioctl() returns successfully. No input is
4311 * copied from userspace as input argument.
4312 */
4313 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4314 {
4315 int ret, karg = 0;
4316
4317 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4318 if (ret)
4319 return ret;
4320
4321 return put_user(karg, (int __user *)arg);
4322 }
4323
4324 /* A wrapper around sock ioctls, which copies the data from userspace
4325 * (depending on the protocol/ioctl), and copies back the result to userspace.
4326 * The main motivation for this function is to pass kernel memory to the
4327 * protocol ioctl callbacks, instead of userspace memory.
4328 */
4329 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4330 {
4331 int rc = 1;
4332
4333 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4334 rc = ipmr_sk_ioctl(sk, cmd, arg);
4335 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4336 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4337 else if (sk_is_phonet(sk))
4338 rc = phonet_sk_ioctl(sk, cmd, arg);
4339
4340 /* If ioctl was processed, returns its value */
4341 if (rc <= 0)
4342 return rc;
4343
4344 /* Otherwise call the default handler */
4345 return sock_ioctl_out(sk, cmd, arg);
4346 }
4347 EXPORT_SYMBOL(sk_ioctl);
4348
4349 static int __init sock_struct_check(void)
4350 {
4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4353 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4356
4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4363 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4366
4367 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4370
4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4372 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4375
4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4385 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4386 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4387 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4388 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4389 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4390 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4391 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4392
4393 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4394 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4395 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4396 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4397 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4398 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4399 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4400 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4401 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4402 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4403 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4404 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4405 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4406 return 0;
4407 }
4408
4409 core_initcall(sock_struct_check);
Linux Kernel