Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88#include <asm/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/init.h>
111#include <linux/highmem.h>
112#include <linux/user_namespace.h>
113#include <linux/static_key.h>
114#include <linux/memcontrol.h>
115#include <linux/prefetch.h>
116#include <linux/compat.h>
117
118#include <linux/uaccess.h>
119
120#include <linux/netdevice.h>
121#include <net/protocol.h>
122#include <linux/skbuff.h>
123#include <net/net_namespace.h>
124#include <net/request_sock.h>
125#include <net/sock.h>
126#include <linux/net_tstamp.h>
127#include <net/xfrm.h>
128#include <linux/ipsec.h>
129#include <net/cls_cgroup.h>
130#include <net/netprio_cgroup.h>
131#include <linux/sock_diag.h>
132
133#include <linux/filter.h>
134#include <net/sock_reuseport.h>
135#include <net/bpf_sk_storage.h>
136
137#include <trace/events/sock.h>
138
139#include <net/tcp.h>
140#include <net/busy_poll.h>
141
142static DEFINE_MUTEX(proto_list_mutex);
143static LIST_HEAD(proto_list);
144
145static void sock_inuse_add(struct net *net, int val);
146
147/**
148 * sk_ns_capable - General socket capability test
149 * @sk: Socket to use a capability on or through
150 * @user_ns: The user namespace of the capability to use
151 * @cap: The capability to use
152 *
153 * Test to see if the opener of the socket had when the socket was
154 * created and the current process has the capability @cap in the user
155 * namespace @user_ns.
156 */
157bool sk_ns_capable(const struct sock *sk,
158 struct user_namespace *user_ns, int cap)
159{
160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
161 ns_capable(user_ns, cap);
162}
163EXPORT_SYMBOL(sk_ns_capable);
164
165/**
166 * sk_capable - Socket global capability test
167 * @sk: Socket to use a capability on or through
168 * @cap: The global capability to use
169 *
170 * Test to see if the opener of the socket had when the socket was
171 * created and the current process has the capability @cap in all user
172 * namespaces.
173 */
174bool sk_capable(const struct sock *sk, int cap)
175{
176 return sk_ns_capable(sk, &init_user_ns, cap);
177}
178EXPORT_SYMBOL(sk_capable);
179
180/**
181 * sk_net_capable - Network namespace socket capability test
182 * @sk: Socket to use a capability on or through
183 * @cap: The capability to use
184 *
185 * Test to see if the opener of the socket had when the socket was created
186 * and the current process has the capability @cap over the network namespace
187 * the socket is a member of.
188 */
189bool sk_net_capable(const struct sock *sk, int cap)
190{
191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
192}
193EXPORT_SYMBOL(sk_net_capable);
194
195/*
196 * Each address family might have different locking rules, so we have
197 * one slock key per address family and separate keys for internal and
198 * userspace sockets.
199 */
200static struct lock_class_key af_family_keys[AF_MAX];
201static struct lock_class_key af_family_kern_keys[AF_MAX];
202static struct lock_class_key af_family_slock_keys[AF_MAX];
203static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
204
205/*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
210
211#define _sock_locks(x) \
212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
221 x "27" , x "28" , x "AF_CAN" , \
222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
227 x "AF_MAX"
228
229static const char *const af_family_key_strings[AF_MAX+1] = {
230 _sock_locks("sk_lock-")
231};
232static const char *const af_family_slock_key_strings[AF_MAX+1] = {
233 _sock_locks("slock-")
234};
235static const char *const af_family_clock_key_strings[AF_MAX+1] = {
236 _sock_locks("clock-")
237};
238
239static const char *const af_family_kern_key_strings[AF_MAX+1] = {
240 _sock_locks("k-sk_lock-")
241};
242static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
243 _sock_locks("k-slock-")
244};
245static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
246 _sock_locks("k-clock-")
247};
248static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
249 _sock_locks("rlock-")
250};
251static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
252 _sock_locks("wlock-")
253};
254static const char *const af_family_elock_key_strings[AF_MAX+1] = {
255 _sock_locks("elock-")
256};
257
258/*
259 * sk_callback_lock and sk queues locking rules are per-address-family,
260 * so split the lock classes by using a per-AF key:
261 */
262static struct lock_class_key af_callback_keys[AF_MAX];
263static struct lock_class_key af_rlock_keys[AF_MAX];
264static struct lock_class_key af_wlock_keys[AF_MAX];
265static struct lock_class_key af_elock_keys[AF_MAX];
266static struct lock_class_key af_kern_callback_keys[AF_MAX];
267
268/* Run time adjustable parameters. */
269__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
270EXPORT_SYMBOL(sysctl_wmem_max);
271__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
272EXPORT_SYMBOL(sysctl_rmem_max);
273__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
274__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
275
276/* Maximal space eaten by iovec or ancillary data plus some space */
277int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
278EXPORT_SYMBOL(sysctl_optmem_max);
279
280int sysctl_tstamp_allow_data __read_mostly = 1;
281
282DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
283EXPORT_SYMBOL_GPL(memalloc_socks_key);
284
285/**
286 * sk_set_memalloc - sets %SOCK_MEMALLOC
287 * @sk: socket to set it on
288 *
289 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
290 * It's the responsibility of the admin to adjust min_free_kbytes
291 * to meet the requirements
292 */
293void sk_set_memalloc(struct sock *sk)
294{
295 sock_set_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation |= __GFP_MEMALLOC;
297 static_branch_inc(&memalloc_socks_key);
298}
299EXPORT_SYMBOL_GPL(sk_set_memalloc);
300
301void sk_clear_memalloc(struct sock *sk)
302{
303 sock_reset_flag(sk, SOCK_MEMALLOC);
304 sk->sk_allocation &= ~__GFP_MEMALLOC;
305 static_branch_dec(&memalloc_socks_key);
306
307 /*
308 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
309 * progress of swapping. SOCK_MEMALLOC may be cleared while
310 * it has rmem allocations due to the last swapfile being deactivated
311 * but there is a risk that the socket is unusable due to exceeding
312 * the rmem limits. Reclaim the reserves and obey rmem limits again.
313 */
314 sk_mem_reclaim(sk);
315}
316EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317
318int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
319{
320 int ret;
321 unsigned int noreclaim_flag;
322
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325
326 noreclaim_flag = memalloc_noreclaim_save();
327 ret = sk->sk_backlog_rcv(sk, skb);
328 memalloc_noreclaim_restore(noreclaim_flag);
329
330 return ret;
331}
332EXPORT_SYMBOL(__sk_backlog_rcv);
333
334static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
335{
336 struct __kernel_sock_timeval tv;
337
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
339 tv.tv_sec = 0;
340 tv.tv_usec = 0;
341 } else {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344 }
345
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
349 return sizeof(tv32);
350 }
351
352 if (old_timeval) {
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 return sizeof(old_tv);
358 }
359
360 *(struct __kernel_sock_timeval *)optval = tv;
361 return sizeof(tv);
362}
363
364static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
365 bool old_timeval)
366{
367 struct __kernel_sock_timeval tv;
368
369 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
370 struct old_timeval32 tv32;
371
372 if (optlen < sizeof(tv32))
373 return -EINVAL;
374
375 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
376 return -EFAULT;
377 tv.tv_sec = tv32.tv_sec;
378 tv.tv_usec = tv32.tv_usec;
379 } else if (old_timeval) {
380 struct __kernel_old_timeval old_tv;
381
382 if (optlen < sizeof(old_tv))
383 return -EINVAL;
384 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
385 return -EFAULT;
386 tv.tv_sec = old_tv.tv_sec;
387 tv.tv_usec = old_tv.tv_usec;
388 } else {
389 if (optlen < sizeof(tv))
390 return -EINVAL;
391 if (copy_from_sockptr(&tv, optval, sizeof(tv)))
392 return -EFAULT;
393 }
394 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
395 return -EDOM;
396
397 if (tv.tv_sec < 0) {
398 static int warned __read_mostly;
399
400 *timeo_p = 0;
401 if (warned < 10 && net_ratelimit()) {
402 warned++;
403 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
404 __func__, current->comm, task_pid_nr(current));
405 }
406 return 0;
407 }
408 *timeo_p = MAX_SCHEDULE_TIMEOUT;
409 if (tv.tv_sec == 0 && tv.tv_usec == 0)
410 return 0;
411 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
412 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
413 return 0;
414}
415
416static void sock_warn_obsolete_bsdism(const char *name)
417{
418 static int warned;
419 static char warncomm[TASK_COMM_LEN];
420 if (strcmp(warncomm, current->comm) && warned < 5) {
421 strcpy(warncomm, current->comm);
422 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
423 warncomm, name);
424 warned++;
425 }
426}
427
428static bool sock_needs_netstamp(const struct sock *sk)
429{
430 switch (sk->sk_family) {
431 case AF_UNSPEC:
432 case AF_UNIX:
433 return false;
434 default:
435 return true;
436 }
437}
438
439static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
440{
441 if (sk->sk_flags & flags) {
442 sk->sk_flags &= ~flags;
443 if (sock_needs_netstamp(sk) &&
444 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
445 net_disable_timestamp();
446 }
447}
448
449
450int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
451{
452 unsigned long flags;
453 struct sk_buff_head *list = &sk->sk_receive_queue;
454
455 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
456 atomic_inc(&sk->sk_drops);
457 trace_sock_rcvqueue_full(sk, skb);
458 return -ENOMEM;
459 }
460
461 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
462 atomic_inc(&sk->sk_drops);
463 return -ENOBUFS;
464 }
465
466 skb->dev = NULL;
467 skb_set_owner_r(skb, sk);
468
469 /* we escape from rcu protected region, make sure we dont leak
470 * a norefcounted dst
471 */
472 skb_dst_force(skb);
473
474 spin_lock_irqsave(&list->lock, flags);
475 sock_skb_set_dropcount(sk, skb);
476 __skb_queue_tail(list, skb);
477 spin_unlock_irqrestore(&list->lock, flags);
478
479 if (!sock_flag(sk, SOCK_DEAD))
480 sk->sk_data_ready(sk);
481 return 0;
482}
483EXPORT_SYMBOL(__sock_queue_rcv_skb);
484
485int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
486{
487 int err;
488
489 err = sk_filter(sk, skb);
490 if (err)
491 return err;
492
493 return __sock_queue_rcv_skb(sk, skb);
494}
495EXPORT_SYMBOL(sock_queue_rcv_skb);
496
497int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
498 const int nested, unsigned int trim_cap, bool refcounted)
499{
500 int rc = NET_RX_SUCCESS;
501
502 if (sk_filter_trim_cap(sk, skb, trim_cap))
503 goto discard_and_relse;
504
505 skb->dev = NULL;
506
507 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
508 atomic_inc(&sk->sk_drops);
509 goto discard_and_relse;
510 }
511 if (nested)
512 bh_lock_sock_nested(sk);
513 else
514 bh_lock_sock(sk);
515 if (!sock_owned_by_user(sk)) {
516 /*
517 * trylock + unlock semantics:
518 */
519 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
520
521 rc = sk_backlog_rcv(sk, skb);
522
523 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
524 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
525 bh_unlock_sock(sk);
526 atomic_inc(&sk->sk_drops);
527 goto discard_and_relse;
528 }
529
530 bh_unlock_sock(sk);
531out:
532 if (refcounted)
533 sock_put(sk);
534 return rc;
535discard_and_relse:
536 kfree_skb(skb);
537 goto out;
538}
539EXPORT_SYMBOL(__sk_receive_skb);
540
541struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
542{
543 struct dst_entry *dst = __sk_dst_get(sk);
544
545 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
546 sk_tx_queue_clear(sk);
547 sk->sk_dst_pending_confirm = 0;
548 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
549 dst_release(dst);
550 return NULL;
551 }
552
553 return dst;
554}
555EXPORT_SYMBOL(__sk_dst_check);
556
557struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
558{
559 struct dst_entry *dst = sk_dst_get(sk);
560
561 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
562 sk_dst_reset(sk);
563 dst_release(dst);
564 return NULL;
565 }
566
567 return dst;
568}
569EXPORT_SYMBOL(sk_dst_check);
570
571static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
572{
573 int ret = -ENOPROTOOPT;
574#ifdef CONFIG_NETDEVICES
575 struct net *net = sock_net(sk);
576
577 /* Sorry... */
578 ret = -EPERM;
579 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
580 goto out;
581
582 ret = -EINVAL;
583 if (ifindex < 0)
584 goto out;
585
586 sk->sk_bound_dev_if = ifindex;
587 if (sk->sk_prot->rehash)
588 sk->sk_prot->rehash(sk);
589 sk_dst_reset(sk);
590
591 ret = 0;
592
593out:
594#endif
595
596 return ret;
597}
598
599int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
600{
601 int ret;
602
603 if (lock_sk)
604 lock_sock(sk);
605 ret = sock_bindtoindex_locked(sk, ifindex);
606 if (lock_sk)
607 release_sock(sk);
608
609 return ret;
610}
611EXPORT_SYMBOL(sock_bindtoindex);
612
613static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
614{
615 int ret = -ENOPROTOOPT;
616#ifdef CONFIG_NETDEVICES
617 struct net *net = sock_net(sk);
618 char devname[IFNAMSIZ];
619 int index;
620
621 ret = -EINVAL;
622 if (optlen < 0)
623 goto out;
624
625 /* Bind this socket to a particular device like "eth0",
626 * as specified in the passed interface name. If the
627 * name is "" or the option length is zero the socket
628 * is not bound.
629 */
630 if (optlen > IFNAMSIZ - 1)
631 optlen = IFNAMSIZ - 1;
632 memset(devname, 0, sizeof(devname));
633
634 ret = -EFAULT;
635 if (copy_from_sockptr(devname, optval, optlen))
636 goto out;
637
638 index = 0;
639 if (devname[0] != '\0') {
640 struct net_device *dev;
641
642 rcu_read_lock();
643 dev = dev_get_by_name_rcu(net, devname);
644 if (dev)
645 index = dev->ifindex;
646 rcu_read_unlock();
647 ret = -ENODEV;
648 if (!dev)
649 goto out;
650 }
651
652 return sock_bindtoindex(sk, index, true);
653out:
654#endif
655
656 return ret;
657}
658
659static int sock_getbindtodevice(struct sock *sk, char __user *optval,
660 int __user *optlen, int len)
661{
662 int ret = -ENOPROTOOPT;
663#ifdef CONFIG_NETDEVICES
664 struct net *net = sock_net(sk);
665 char devname[IFNAMSIZ];
666
667 if (sk->sk_bound_dev_if == 0) {
668 len = 0;
669 goto zero;
670 }
671
672 ret = -EINVAL;
673 if (len < IFNAMSIZ)
674 goto out;
675
676 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
677 if (ret)
678 goto out;
679
680 len = strlen(devname) + 1;
681
682 ret = -EFAULT;
683 if (copy_to_user(optval, devname, len))
684 goto out;
685
686zero:
687 ret = -EFAULT;
688 if (put_user(len, optlen))
689 goto out;
690
691 ret = 0;
692
693out:
694#endif
695
696 return ret;
697}
698
699bool sk_mc_loop(struct sock *sk)
700{
701 if (dev_recursion_level())
702 return false;
703 if (!sk)
704 return true;
705 switch (sk->sk_family) {
706 case AF_INET:
707 return inet_sk(sk)->mc_loop;
708#if IS_ENABLED(CONFIG_IPV6)
709 case AF_INET6:
710 return inet6_sk(sk)->mc_loop;
711#endif
712 }
713 WARN_ON_ONCE(1);
714 return true;
715}
716EXPORT_SYMBOL(sk_mc_loop);
717
718void sock_set_reuseaddr(struct sock *sk)
719{
720 lock_sock(sk);
721 sk->sk_reuse = SK_CAN_REUSE;
722 release_sock(sk);
723}
724EXPORT_SYMBOL(sock_set_reuseaddr);
725
726void sock_set_reuseport(struct sock *sk)
727{
728 lock_sock(sk);
729 sk->sk_reuseport = true;
730 release_sock(sk);
731}
732EXPORT_SYMBOL(sock_set_reuseport);
733
734void sock_no_linger(struct sock *sk)
735{
736 lock_sock(sk);
737 sk->sk_lingertime = 0;
738 sock_set_flag(sk, SOCK_LINGER);
739 release_sock(sk);
740}
741EXPORT_SYMBOL(sock_no_linger);
742
743void sock_set_priority(struct sock *sk, u32 priority)
744{
745 lock_sock(sk);
746 sk->sk_priority = priority;
747 release_sock(sk);
748}
749EXPORT_SYMBOL(sock_set_priority);
750
751void sock_set_sndtimeo(struct sock *sk, s64 secs)
752{
753 lock_sock(sk);
754 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
755 sk->sk_sndtimeo = secs * HZ;
756 else
757 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
758 release_sock(sk);
759}
760EXPORT_SYMBOL(sock_set_sndtimeo);
761
762static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
763{
764 if (val) {
765 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
766 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
767 sock_set_flag(sk, SOCK_RCVTSTAMP);
768 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
769 } else {
770 sock_reset_flag(sk, SOCK_RCVTSTAMP);
771 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
772 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
773 }
774}
775
776void sock_enable_timestamps(struct sock *sk)
777{
778 lock_sock(sk);
779 __sock_set_timestamps(sk, true, false, true);
780 release_sock(sk);
781}
782EXPORT_SYMBOL(sock_enable_timestamps);
783
784void sock_set_keepalive(struct sock *sk)
785{
786 lock_sock(sk);
787 if (sk->sk_prot->keepalive)
788 sk->sk_prot->keepalive(sk, true);
789 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
790 release_sock(sk);
791}
792EXPORT_SYMBOL(sock_set_keepalive);
793
794static void __sock_set_rcvbuf(struct sock *sk, int val)
795{
796 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
797 * as a negative value.
798 */
799 val = min_t(int, val, INT_MAX / 2);
800 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
801
802 /* We double it on the way in to account for "struct sk_buff" etc.
803 * overhead. Applications assume that the SO_RCVBUF setting they make
804 * will allow that much actual data to be received on that socket.
805 *
806 * Applications are unaware that "struct sk_buff" and other overheads
807 * allocate from the receive buffer during socket buffer allocation.
808 *
809 * And after considering the possible alternatives, returning the value
810 * we actually used in getsockopt is the most desirable behavior.
811 */
812 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
813}
814
815void sock_set_rcvbuf(struct sock *sk, int val)
816{
817 lock_sock(sk);
818 __sock_set_rcvbuf(sk, val);
819 release_sock(sk);
820}
821EXPORT_SYMBOL(sock_set_rcvbuf);
822
823void sock_set_mark(struct sock *sk, u32 val)
824{
825 lock_sock(sk);
826 sk->sk_mark = val;
827 release_sock(sk);
828}
829EXPORT_SYMBOL(sock_set_mark);
830
831/*
832 * This is meant for all protocols to use and covers goings on
833 * at the socket level. Everything here is generic.
834 */
835
836int sock_setsockopt(struct socket *sock, int level, int optname,
837 sockptr_t optval, unsigned int optlen)
838{
839 struct sock_txtime sk_txtime;
840 struct sock *sk = sock->sk;
841 int val;
842 int valbool;
843 struct linger ling;
844 int ret = 0;
845
846 /*
847 * Options without arguments
848 */
849
850 if (optname == SO_BINDTODEVICE)
851 return sock_setbindtodevice(sk, optval, optlen);
852
853 if (optlen < sizeof(int))
854 return -EINVAL;
855
856 if (copy_from_sockptr(&val, optval, sizeof(val)))
857 return -EFAULT;
858
859 valbool = val ? 1 : 0;
860
861 lock_sock(sk);
862
863 switch (optname) {
864 case SO_DEBUG:
865 if (val && !capable(CAP_NET_ADMIN))
866 ret = -EACCES;
867 else
868 sock_valbool_flag(sk, SOCK_DBG, valbool);
869 break;
870 case SO_REUSEADDR:
871 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
872 break;
873 case SO_REUSEPORT:
874 sk->sk_reuseport = valbool;
875 break;
876 case SO_TYPE:
877 case SO_PROTOCOL:
878 case SO_DOMAIN:
879 case SO_ERROR:
880 ret = -ENOPROTOOPT;
881 break;
882 case SO_DONTROUTE:
883 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
884 sk_dst_reset(sk);
885 break;
886 case SO_BROADCAST:
887 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
888 break;
889 case SO_SNDBUF:
890 /* Don't error on this BSD doesn't and if you think
891 * about it this is right. Otherwise apps have to
892 * play 'guess the biggest size' games. RCVBUF/SNDBUF
893 * are treated in BSD as hints
894 */
895 val = min_t(u32, val, sysctl_wmem_max);
896set_sndbuf:
897 /* Ensure val * 2 fits into an int, to prevent max_t()
898 * from treating it as a negative value.
899 */
900 val = min_t(int, val, INT_MAX / 2);
901 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
902 WRITE_ONCE(sk->sk_sndbuf,
903 max_t(int, val * 2, SOCK_MIN_SNDBUF));
904 /* Wake up sending tasks if we upped the value. */
905 sk->sk_write_space(sk);
906 break;
907
908 case SO_SNDBUFFORCE:
909 if (!capable(CAP_NET_ADMIN)) {
910 ret = -EPERM;
911 break;
912 }
913
914 /* No negative values (to prevent underflow, as val will be
915 * multiplied by 2).
916 */
917 if (val < 0)
918 val = 0;
919 goto set_sndbuf;
920
921 case SO_RCVBUF:
922 /* Don't error on this BSD doesn't and if you think
923 * about it this is right. Otherwise apps have to
924 * play 'guess the biggest size' games. RCVBUF/SNDBUF
925 * are treated in BSD as hints
926 */
927 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
928 break;
929
930 case SO_RCVBUFFORCE:
931 if (!capable(CAP_NET_ADMIN)) {
932 ret = -EPERM;
933 break;
934 }
935
936 /* No negative values (to prevent underflow, as val will be
937 * multiplied by 2).
938 */
939 __sock_set_rcvbuf(sk, max(val, 0));
940 break;
941
942 case SO_KEEPALIVE:
943 if (sk->sk_prot->keepalive)
944 sk->sk_prot->keepalive(sk, valbool);
945 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
946 break;
947
948 case SO_OOBINLINE:
949 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
950 break;
951
952 case SO_NO_CHECK:
953 sk->sk_no_check_tx = valbool;
954 break;
955
956 case SO_PRIORITY:
957 if ((val >= 0 && val <= 6) ||
958 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
959 sk->sk_priority = val;
960 else
961 ret = -EPERM;
962 break;
963
964 case SO_LINGER:
965 if (optlen < sizeof(ling)) {
966 ret = -EINVAL; /* 1003.1g */
967 break;
968 }
969 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
970 ret = -EFAULT;
971 break;
972 }
973 if (!ling.l_onoff)
974 sock_reset_flag(sk, SOCK_LINGER);
975 else {
976#if (BITS_PER_LONG == 32)
977 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
978 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
979 else
980#endif
981 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
982 sock_set_flag(sk, SOCK_LINGER);
983 }
984 break;
985
986 case SO_BSDCOMPAT:
987 sock_warn_obsolete_bsdism("setsockopt");
988 break;
989
990 case SO_PASSCRED:
991 if (valbool)
992 set_bit(SOCK_PASSCRED, &sock->flags);
993 else
994 clear_bit(SOCK_PASSCRED, &sock->flags);
995 break;
996
997 case SO_TIMESTAMP_OLD:
998 __sock_set_timestamps(sk, valbool, false, false);
999 break;
1000 case SO_TIMESTAMP_NEW:
1001 __sock_set_timestamps(sk, valbool, true, false);
1002 break;
1003 case SO_TIMESTAMPNS_OLD:
1004 __sock_set_timestamps(sk, valbool, false, true);
1005 break;
1006 case SO_TIMESTAMPNS_NEW:
1007 __sock_set_timestamps(sk, valbool, true, true);
1008 break;
1009 case SO_TIMESTAMPING_NEW:
1010 sock_set_flag(sk, SOCK_TSTAMP_NEW);
1011 fallthrough;
1012 case SO_TIMESTAMPING_OLD:
1013 if (val & ~SOF_TIMESTAMPING_MASK) {
1014 ret = -EINVAL;
1015 break;
1016 }
1017
1018 if (val & SOF_TIMESTAMPING_OPT_ID &&
1019 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
1020 if (sk->sk_protocol == IPPROTO_TCP &&
1021 sk->sk_type == SOCK_STREAM) {
1022 if ((1 << sk->sk_state) &
1023 (TCPF_CLOSE | TCPF_LISTEN)) {
1024 ret = -EINVAL;
1025 break;
1026 }
1027 sk->sk_tskey = tcp_sk(sk)->snd_una;
1028 } else {
1029 sk->sk_tskey = 0;
1030 }
1031 }
1032
1033 if (val & SOF_TIMESTAMPING_OPT_STATS &&
1034 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
1035 ret = -EINVAL;
1036 break;
1037 }
1038
1039 sk->sk_tsflags = val;
1040 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
1041 sock_enable_timestamp(sk,
1042 SOCK_TIMESTAMPING_RX_SOFTWARE);
1043 else {
1044 if (optname == SO_TIMESTAMPING_NEW)
1045 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
1046
1047 sock_disable_timestamp(sk,
1048 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
1049 }
1050 break;
1051
1052 case SO_RCVLOWAT:
1053 if (val < 0)
1054 val = INT_MAX;
1055 if (sock->ops->set_rcvlowat)
1056 ret = sock->ops->set_rcvlowat(sk, val);
1057 else
1058 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1059 break;
1060
1061 case SO_RCVTIMEO_OLD:
1062 case SO_RCVTIMEO_NEW:
1063 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1064 optlen, optname == SO_RCVTIMEO_OLD);
1065 break;
1066
1067 case SO_SNDTIMEO_OLD:
1068 case SO_SNDTIMEO_NEW:
1069 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1070 optlen, optname == SO_SNDTIMEO_OLD);
1071 break;
1072
1073 case SO_ATTACH_FILTER: {
1074 struct sock_fprog fprog;
1075
1076 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1077 if (!ret)
1078 ret = sk_attach_filter(&fprog, sk);
1079 break;
1080 }
1081 case SO_ATTACH_BPF:
1082 ret = -EINVAL;
1083 if (optlen == sizeof(u32)) {
1084 u32 ufd;
1085
1086 ret = -EFAULT;
1087 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1088 break;
1089
1090 ret = sk_attach_bpf(ufd, sk);
1091 }
1092 break;
1093
1094 case SO_ATTACH_REUSEPORT_CBPF: {
1095 struct sock_fprog fprog;
1096
1097 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1098 if (!ret)
1099 ret = sk_reuseport_attach_filter(&fprog, sk);
1100 break;
1101 }
1102 case SO_ATTACH_REUSEPORT_EBPF:
1103 ret = -EINVAL;
1104 if (optlen == sizeof(u32)) {
1105 u32 ufd;
1106
1107 ret = -EFAULT;
1108 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1109 break;
1110
1111 ret = sk_reuseport_attach_bpf(ufd, sk);
1112 }
1113 break;
1114
1115 case SO_DETACH_REUSEPORT_BPF:
1116 ret = reuseport_detach_prog(sk);
1117 break;
1118
1119 case SO_DETACH_FILTER:
1120 ret = sk_detach_filter(sk);
1121 break;
1122
1123 case SO_LOCK_FILTER:
1124 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1125 ret = -EPERM;
1126 else
1127 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1128 break;
1129
1130 case SO_PASSSEC:
1131 if (valbool)
1132 set_bit(SOCK_PASSSEC, &sock->flags);
1133 else
1134 clear_bit(SOCK_PASSSEC, &sock->flags);
1135 break;
1136 case SO_MARK:
1137 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1138 ret = -EPERM;
1139 } else if (val != sk->sk_mark) {
1140 sk->sk_mark = val;
1141 sk_dst_reset(sk);
1142 }
1143 break;
1144
1145 case SO_RXQ_OVFL:
1146 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1147 break;
1148
1149 case SO_WIFI_STATUS:
1150 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1151 break;
1152
1153 case SO_PEEK_OFF:
1154 if (sock->ops->set_peek_off)
1155 ret = sock->ops->set_peek_off(sk, val);
1156 else
1157 ret = -EOPNOTSUPP;
1158 break;
1159
1160 case SO_NOFCS:
1161 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1162 break;
1163
1164 case SO_SELECT_ERR_QUEUE:
1165 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1166 break;
1167
1168#ifdef CONFIG_NET_RX_BUSY_POLL
1169 case SO_BUSY_POLL:
1170 /* allow unprivileged users to decrease the value */
1171 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1172 ret = -EPERM;
1173 else {
1174 if (val < 0)
1175 ret = -EINVAL;
1176 else
1177 sk->sk_ll_usec = val;
1178 }
1179 break;
1180#endif
1181
1182 case SO_MAX_PACING_RATE:
1183 {
1184 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1185
1186 if (sizeof(ulval) != sizeof(val) &&
1187 optlen >= sizeof(ulval) &&
1188 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1189 ret = -EFAULT;
1190 break;
1191 }
1192 if (ulval != ~0UL)
1193 cmpxchg(&sk->sk_pacing_status,
1194 SK_PACING_NONE,
1195 SK_PACING_NEEDED);
1196 sk->sk_max_pacing_rate = ulval;
1197 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1198 break;
1199 }
1200 case SO_INCOMING_CPU:
1201 WRITE_ONCE(sk->sk_incoming_cpu, val);
1202 break;
1203
1204 case SO_CNX_ADVICE:
1205 if (val == 1)
1206 dst_negative_advice(sk);
1207 break;
1208
1209 case SO_ZEROCOPY:
1210 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1211 if (!((sk->sk_type == SOCK_STREAM &&
1212 sk->sk_protocol == IPPROTO_TCP) ||
1213 (sk->sk_type == SOCK_DGRAM &&
1214 sk->sk_protocol == IPPROTO_UDP)))
1215 ret = -ENOTSUPP;
1216 } else if (sk->sk_family != PF_RDS) {
1217 ret = -ENOTSUPP;
1218 }
1219 if (!ret) {
1220 if (val < 0 || val > 1)
1221 ret = -EINVAL;
1222 else
1223 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1224 }
1225 break;
1226
1227 case SO_TXTIME:
1228 if (optlen != sizeof(struct sock_txtime)) {
1229 ret = -EINVAL;
1230 break;
1231 } else if (copy_from_sockptr(&sk_txtime, optval,
1232 sizeof(struct sock_txtime))) {
1233 ret = -EFAULT;
1234 break;
1235 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1236 ret = -EINVAL;
1237 break;
1238 }
1239 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1240 * scheduler has enough safe guards.
1241 */
1242 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1243 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1244 ret = -EPERM;
1245 break;
1246 }
1247 sock_valbool_flag(sk, SOCK_TXTIME, true);
1248 sk->sk_clockid = sk_txtime.clockid;
1249 sk->sk_txtime_deadline_mode =
1250 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1251 sk->sk_txtime_report_errors =
1252 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1253 break;
1254
1255 case SO_BINDTOIFINDEX:
1256 ret = sock_bindtoindex_locked(sk, val);
1257 break;
1258
1259 default:
1260 ret = -ENOPROTOOPT;
1261 break;
1262 }
1263 release_sock(sk);
1264 return ret;
1265}
1266EXPORT_SYMBOL(sock_setsockopt);
1267
1268
1269static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1270 struct ucred *ucred)
1271{
1272 ucred->pid = pid_vnr(pid);
1273 ucred->uid = ucred->gid = -1;
1274 if (cred) {
1275 struct user_namespace *current_ns = current_user_ns();
1276
1277 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1278 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1279 }
1280}
1281
1282static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1283{
1284 struct user_namespace *user_ns = current_user_ns();
1285 int i;
1286
1287 for (i = 0; i < src->ngroups; i++)
1288 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1289 return -EFAULT;
1290
1291 return 0;
1292}
1293
1294int sock_getsockopt(struct socket *sock, int level, int optname,
1295 char __user *optval, int __user *optlen)
1296{
1297 struct sock *sk = sock->sk;
1298
1299 union {
1300 int val;
1301 u64 val64;
1302 unsigned long ulval;
1303 struct linger ling;
1304 struct old_timeval32 tm32;
1305 struct __kernel_old_timeval tm;
1306 struct __kernel_sock_timeval stm;
1307 struct sock_txtime txtime;
1308 } v;
1309
1310 int lv = sizeof(int);
1311 int len;
1312
1313 if (get_user(len, optlen))
1314 return -EFAULT;
1315 if (len < 0)
1316 return -EINVAL;
1317
1318 memset(&v, 0, sizeof(v));
1319
1320 switch (optname) {
1321 case SO_DEBUG:
1322 v.val = sock_flag(sk, SOCK_DBG);
1323 break;
1324
1325 case SO_DONTROUTE:
1326 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1327 break;
1328
1329 case SO_BROADCAST:
1330 v.val = sock_flag(sk, SOCK_BROADCAST);
1331 break;
1332
1333 case SO_SNDBUF:
1334 v.val = sk->sk_sndbuf;
1335 break;
1336
1337 case SO_RCVBUF:
1338 v.val = sk->sk_rcvbuf;
1339 break;
1340
1341 case SO_REUSEADDR:
1342 v.val = sk->sk_reuse;
1343 break;
1344
1345 case SO_REUSEPORT:
1346 v.val = sk->sk_reuseport;
1347 break;
1348
1349 case SO_KEEPALIVE:
1350 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1351 break;
1352
1353 case SO_TYPE:
1354 v.val = sk->sk_type;
1355 break;
1356
1357 case SO_PROTOCOL:
1358 v.val = sk->sk_protocol;
1359 break;
1360
1361 case SO_DOMAIN:
1362 v.val = sk->sk_family;
1363 break;
1364
1365 case SO_ERROR:
1366 v.val = -sock_error(sk);
1367 if (v.val == 0)
1368 v.val = xchg(&sk->sk_err_soft, 0);
1369 break;
1370
1371 case SO_OOBINLINE:
1372 v.val = sock_flag(sk, SOCK_URGINLINE);
1373 break;
1374
1375 case SO_NO_CHECK:
1376 v.val = sk->sk_no_check_tx;
1377 break;
1378
1379 case SO_PRIORITY:
1380 v.val = sk->sk_priority;
1381 break;
1382
1383 case SO_LINGER:
1384 lv = sizeof(v.ling);
1385 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1386 v.ling.l_linger = sk->sk_lingertime / HZ;
1387 break;
1388
1389 case SO_BSDCOMPAT:
1390 sock_warn_obsolete_bsdism("getsockopt");
1391 break;
1392
1393 case SO_TIMESTAMP_OLD:
1394 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1395 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1396 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1397 break;
1398
1399 case SO_TIMESTAMPNS_OLD:
1400 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1401 break;
1402
1403 case SO_TIMESTAMP_NEW:
1404 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1405 break;
1406
1407 case SO_TIMESTAMPNS_NEW:
1408 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1409 break;
1410
1411 case SO_TIMESTAMPING_OLD:
1412 v.val = sk->sk_tsflags;
1413 break;
1414
1415 case SO_RCVTIMEO_OLD:
1416 case SO_RCVTIMEO_NEW:
1417 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1418 break;
1419
1420 case SO_SNDTIMEO_OLD:
1421 case SO_SNDTIMEO_NEW:
1422 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1423 break;
1424
1425 case SO_RCVLOWAT:
1426 v.val = sk->sk_rcvlowat;
1427 break;
1428
1429 case SO_SNDLOWAT:
1430 v.val = 1;
1431 break;
1432
1433 case SO_PASSCRED:
1434 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1435 break;
1436
1437 case SO_PEERCRED:
1438 {
1439 struct ucred peercred;
1440 if (len > sizeof(peercred))
1441 len = sizeof(peercred);
1442 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1443 if (copy_to_user(optval, &peercred, len))
1444 return -EFAULT;
1445 goto lenout;
1446 }
1447
1448 case SO_PEERGROUPS:
1449 {
1450 int ret, n;
1451
1452 if (!sk->sk_peer_cred)
1453 return -ENODATA;
1454
1455 n = sk->sk_peer_cred->group_info->ngroups;
1456 if (len < n * sizeof(gid_t)) {
1457 len = n * sizeof(gid_t);
1458 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1459 }
1460 len = n * sizeof(gid_t);
1461
1462 ret = groups_to_user((gid_t __user *)optval,
1463 sk->sk_peer_cred->group_info);
1464 if (ret)
1465 return ret;
1466 goto lenout;
1467 }
1468
1469 case SO_PEERNAME:
1470 {
1471 char address[128];
1472
1473 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1474 if (lv < 0)
1475 return -ENOTCONN;
1476 if (lv < len)
1477 return -EINVAL;
1478 if (copy_to_user(optval, address, len))
1479 return -EFAULT;
1480 goto lenout;
1481 }
1482
1483 /* Dubious BSD thing... Probably nobody even uses it, but
1484 * the UNIX standard wants it for whatever reason... -DaveM
1485 */
1486 case SO_ACCEPTCONN:
1487 v.val = sk->sk_state == TCP_LISTEN;
1488 break;
1489
1490 case SO_PASSSEC:
1491 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1492 break;
1493
1494 case SO_PEERSEC:
1495 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1496
1497 case SO_MARK:
1498 v.val = sk->sk_mark;
1499 break;
1500
1501 case SO_RXQ_OVFL:
1502 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1503 break;
1504
1505 case SO_WIFI_STATUS:
1506 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1507 break;
1508
1509 case SO_PEEK_OFF:
1510 if (!sock->ops->set_peek_off)
1511 return -EOPNOTSUPP;
1512
1513 v.val = sk->sk_peek_off;
1514 break;
1515 case SO_NOFCS:
1516 v.val = sock_flag(sk, SOCK_NOFCS);
1517 break;
1518
1519 case SO_BINDTODEVICE:
1520 return sock_getbindtodevice(sk, optval, optlen, len);
1521
1522 case SO_GET_FILTER:
1523 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1524 if (len < 0)
1525 return len;
1526
1527 goto lenout;
1528
1529 case SO_LOCK_FILTER:
1530 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1531 break;
1532
1533 case SO_BPF_EXTENSIONS:
1534 v.val = bpf_tell_extensions();
1535 break;
1536
1537 case SO_SELECT_ERR_QUEUE:
1538 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1539 break;
1540
1541#ifdef CONFIG_NET_RX_BUSY_POLL
1542 case SO_BUSY_POLL:
1543 v.val = sk->sk_ll_usec;
1544 break;
1545#endif
1546
1547 case SO_MAX_PACING_RATE:
1548 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1549 lv = sizeof(v.ulval);
1550 v.ulval = sk->sk_max_pacing_rate;
1551 } else {
1552 /* 32bit version */
1553 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1554 }
1555 break;
1556
1557 case SO_INCOMING_CPU:
1558 v.val = READ_ONCE(sk->sk_incoming_cpu);
1559 break;
1560
1561 case SO_MEMINFO:
1562 {
1563 u32 meminfo[SK_MEMINFO_VARS];
1564
1565 sk_get_meminfo(sk, meminfo);
1566
1567 len = min_t(unsigned int, len, sizeof(meminfo));
1568 if (copy_to_user(optval, &meminfo, len))
1569 return -EFAULT;
1570
1571 goto lenout;
1572 }
1573
1574#ifdef CONFIG_NET_RX_BUSY_POLL
1575 case SO_INCOMING_NAPI_ID:
1576 v.val = READ_ONCE(sk->sk_napi_id);
1577
1578 /* aggregate non-NAPI IDs down to 0 */
1579 if (v.val < MIN_NAPI_ID)
1580 v.val = 0;
1581
1582 break;
1583#endif
1584
1585 case SO_COOKIE:
1586 lv = sizeof(u64);
1587 if (len < lv)
1588 return -EINVAL;
1589 v.val64 = sock_gen_cookie(sk);
1590 break;
1591
1592 case SO_ZEROCOPY:
1593 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1594 break;
1595
1596 case SO_TXTIME:
1597 lv = sizeof(v.txtime);
1598 v.txtime.clockid = sk->sk_clockid;
1599 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1600 SOF_TXTIME_DEADLINE_MODE : 0;
1601 v.txtime.flags |= sk->sk_txtime_report_errors ?
1602 SOF_TXTIME_REPORT_ERRORS : 0;
1603 break;
1604
1605 case SO_BINDTOIFINDEX:
1606 v.val = sk->sk_bound_dev_if;
1607 break;
1608
1609 default:
1610 /* We implement the SO_SNDLOWAT etc to not be settable
1611 * (1003.1g 7).
1612 */
1613 return -ENOPROTOOPT;
1614 }
1615
1616 if (len > lv)
1617 len = lv;
1618 if (copy_to_user(optval, &v, len))
1619 return -EFAULT;
1620lenout:
1621 if (put_user(len, optlen))
1622 return -EFAULT;
1623 return 0;
1624}
1625
1626/*
1627 * Initialize an sk_lock.
1628 *
1629 * (We also register the sk_lock with the lock validator.)
1630 */
1631static inline void sock_lock_init(struct sock *sk)
1632{
1633 if (sk->sk_kern_sock)
1634 sock_lock_init_class_and_name(
1635 sk,
1636 af_family_kern_slock_key_strings[sk->sk_family],
1637 af_family_kern_slock_keys + sk->sk_family,
1638 af_family_kern_key_strings[sk->sk_family],
1639 af_family_kern_keys + sk->sk_family);
1640 else
1641 sock_lock_init_class_and_name(
1642 sk,
1643 af_family_slock_key_strings[sk->sk_family],
1644 af_family_slock_keys + sk->sk_family,
1645 af_family_key_strings[sk->sk_family],
1646 af_family_keys + sk->sk_family);
1647}
1648
1649/*
1650 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1651 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1652 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1653 */
1654static void sock_copy(struct sock *nsk, const struct sock *osk)
1655{
1656 const struct proto *prot = READ_ONCE(osk->sk_prot);
1657#ifdef CONFIG_SECURITY_NETWORK
1658 void *sptr = nsk->sk_security;
1659#endif
1660 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1661
1662 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1663 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1664
1665#ifdef CONFIG_SECURITY_NETWORK
1666 nsk->sk_security = sptr;
1667 security_sk_clone(osk, nsk);
1668#endif
1669}
1670
1671static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1672 int family)
1673{
1674 struct sock *sk;
1675 struct kmem_cache *slab;
1676
1677 slab = prot->slab;
1678 if (slab != NULL) {
1679 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1680 if (!sk)
1681 return sk;
1682 if (want_init_on_alloc(priority))
1683 sk_prot_clear_nulls(sk, prot->obj_size);
1684 } else
1685 sk = kmalloc(prot->obj_size, priority);
1686
1687 if (sk != NULL) {
1688 if (security_sk_alloc(sk, family, priority))
1689 goto out_free;
1690
1691 if (!try_module_get(prot->owner))
1692 goto out_free_sec;
1693 sk_tx_queue_clear(sk);
1694 }
1695
1696 return sk;
1697
1698out_free_sec:
1699 security_sk_free(sk);
1700out_free:
1701 if (slab != NULL)
1702 kmem_cache_free(slab, sk);
1703 else
1704 kfree(sk);
1705 return NULL;
1706}
1707
1708static void sk_prot_free(struct proto *prot, struct sock *sk)
1709{
1710 struct kmem_cache *slab;
1711 struct module *owner;
1712
1713 owner = prot->owner;
1714 slab = prot->slab;
1715
1716 cgroup_sk_free(&sk->sk_cgrp_data);
1717 mem_cgroup_sk_free(sk);
1718 security_sk_free(sk);
1719 if (slab != NULL)
1720 kmem_cache_free(slab, sk);
1721 else
1722 kfree(sk);
1723 module_put(owner);
1724}
1725
1726/**
1727 * sk_alloc - All socket objects are allocated here
1728 * @net: the applicable net namespace
1729 * @family: protocol family
1730 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1731 * @prot: struct proto associated with this new sock instance
1732 * @kern: is this to be a kernel socket?
1733 */
1734struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1735 struct proto *prot, int kern)
1736{
1737 struct sock *sk;
1738
1739 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1740 if (sk) {
1741 sk->sk_family = family;
1742 /*
1743 * See comment in struct sock definition to understand
1744 * why we need sk_prot_creator -acme
1745 */
1746 sk->sk_prot = sk->sk_prot_creator = prot;
1747 sk->sk_kern_sock = kern;
1748 sock_lock_init(sk);
1749 sk->sk_net_refcnt = kern ? 0 : 1;
1750 if (likely(sk->sk_net_refcnt)) {
1751 get_net(net);
1752 sock_inuse_add(net, 1);
1753 }
1754
1755 sock_net_set(sk, net);
1756 refcount_set(&sk->sk_wmem_alloc, 1);
1757
1758 mem_cgroup_sk_alloc(sk);
1759 cgroup_sk_alloc(&sk->sk_cgrp_data);
1760 sock_update_classid(&sk->sk_cgrp_data);
1761 sock_update_netprioidx(&sk->sk_cgrp_data);
1762 sk_tx_queue_clear(sk);
1763 }
1764
1765 return sk;
1766}
1767EXPORT_SYMBOL(sk_alloc);
1768
1769/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1770 * grace period. This is the case for UDP sockets and TCP listeners.
1771 */
1772static void __sk_destruct(struct rcu_head *head)
1773{
1774 struct sock *sk = container_of(head, struct sock, sk_rcu);
1775 struct sk_filter *filter;
1776
1777 if (sk->sk_destruct)
1778 sk->sk_destruct(sk);
1779
1780 filter = rcu_dereference_check(sk->sk_filter,
1781 refcount_read(&sk->sk_wmem_alloc) == 0);
1782 if (filter) {
1783 sk_filter_uncharge(sk, filter);
1784 RCU_INIT_POINTER(sk->sk_filter, NULL);
1785 }
1786
1787 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1788
1789#ifdef CONFIG_BPF_SYSCALL
1790 bpf_sk_storage_free(sk);
1791#endif
1792
1793 if (atomic_read(&sk->sk_omem_alloc))
1794 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1795 __func__, atomic_read(&sk->sk_omem_alloc));
1796
1797 if (sk->sk_frag.page) {
1798 put_page(sk->sk_frag.page);
1799 sk->sk_frag.page = NULL;
1800 }
1801
1802 if (sk->sk_peer_cred)
1803 put_cred(sk->sk_peer_cred);
1804 put_pid(sk->sk_peer_pid);
1805 if (likely(sk->sk_net_refcnt))
1806 put_net(sock_net(sk));
1807 sk_prot_free(sk->sk_prot_creator, sk);
1808}
1809
1810void sk_destruct(struct sock *sk)
1811{
1812 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1813
1814 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1815 reuseport_detach_sock(sk);
1816 use_call_rcu = true;
1817 }
1818
1819 if (use_call_rcu)
1820 call_rcu(&sk->sk_rcu, __sk_destruct);
1821 else
1822 __sk_destruct(&sk->sk_rcu);
1823}
1824
1825static void __sk_free(struct sock *sk)
1826{
1827 if (likely(sk->sk_net_refcnt))
1828 sock_inuse_add(sock_net(sk), -1);
1829
1830 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1831 sock_diag_broadcast_destroy(sk);
1832 else
1833 sk_destruct(sk);
1834}
1835
1836void sk_free(struct sock *sk)
1837{
1838 /*
1839 * We subtract one from sk_wmem_alloc and can know if
1840 * some packets are still in some tx queue.
1841 * If not null, sock_wfree() will call __sk_free(sk) later
1842 */
1843 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1844 __sk_free(sk);
1845}
1846EXPORT_SYMBOL(sk_free);
1847
1848static void sk_init_common(struct sock *sk)
1849{
1850 skb_queue_head_init(&sk->sk_receive_queue);
1851 skb_queue_head_init(&sk->sk_write_queue);
1852 skb_queue_head_init(&sk->sk_error_queue);
1853
1854 rwlock_init(&sk->sk_callback_lock);
1855 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1856 af_rlock_keys + sk->sk_family,
1857 af_family_rlock_key_strings[sk->sk_family]);
1858 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1859 af_wlock_keys + sk->sk_family,
1860 af_family_wlock_key_strings[sk->sk_family]);
1861 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1862 af_elock_keys + sk->sk_family,
1863 af_family_elock_key_strings[sk->sk_family]);
1864 lockdep_set_class_and_name(&sk->sk_callback_lock,
1865 af_callback_keys + sk->sk_family,
1866 af_family_clock_key_strings[sk->sk_family]);
1867}
1868
1869/**
1870 * sk_clone_lock - clone a socket, and lock its clone
1871 * @sk: the socket to clone
1872 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1873 *
1874 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1875 */
1876struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1877{
1878 struct proto *prot = READ_ONCE(sk->sk_prot);
1879 struct sock *newsk;
1880 bool is_charged = true;
1881
1882 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
1883 if (newsk != NULL) {
1884 struct sk_filter *filter;
1885
1886 sock_copy(newsk, sk);
1887
1888 newsk->sk_prot_creator = prot;
1889
1890 /* SANITY */
1891 if (likely(newsk->sk_net_refcnt))
1892 get_net(sock_net(newsk));
1893 sk_node_init(&newsk->sk_node);
1894 sock_lock_init(newsk);
1895 bh_lock_sock(newsk);
1896 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1897 newsk->sk_backlog.len = 0;
1898
1899 atomic_set(&newsk->sk_rmem_alloc, 0);
1900 /*
1901 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1902 */
1903 refcount_set(&newsk->sk_wmem_alloc, 1);
1904 atomic_set(&newsk->sk_omem_alloc, 0);
1905 sk_init_common(newsk);
1906
1907 newsk->sk_dst_cache = NULL;
1908 newsk->sk_dst_pending_confirm = 0;
1909 newsk->sk_wmem_queued = 0;
1910 newsk->sk_forward_alloc = 0;
1911 atomic_set(&newsk->sk_drops, 0);
1912 newsk->sk_send_head = NULL;
1913 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1914 atomic_set(&newsk->sk_zckey, 0);
1915
1916 sock_reset_flag(newsk, SOCK_DONE);
1917
1918 /* sk->sk_memcg will be populated at accept() time */
1919 newsk->sk_memcg = NULL;
1920
1921 cgroup_sk_clone(&newsk->sk_cgrp_data);
1922
1923 rcu_read_lock();
1924 filter = rcu_dereference(sk->sk_filter);
1925 if (filter != NULL)
1926 /* though it's an empty new sock, the charging may fail
1927 * if sysctl_optmem_max was changed between creation of
1928 * original socket and cloning
1929 */
1930 is_charged = sk_filter_charge(newsk, filter);
1931 RCU_INIT_POINTER(newsk->sk_filter, filter);
1932 rcu_read_unlock();
1933
1934 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1935 /* We need to make sure that we don't uncharge the new
1936 * socket if we couldn't charge it in the first place
1937 * as otherwise we uncharge the parent's filter.
1938 */
1939 if (!is_charged)
1940 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1941 sk_free_unlock_clone(newsk);
1942 newsk = NULL;
1943 goto out;
1944 }
1945 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1946
1947 if (bpf_sk_storage_clone(sk, newsk)) {
1948 sk_free_unlock_clone(newsk);
1949 newsk = NULL;
1950 goto out;
1951 }
1952
1953 /* Clear sk_user_data if parent had the pointer tagged
1954 * as not suitable for copying when cloning.
1955 */
1956 if (sk_user_data_is_nocopy(newsk))
1957 newsk->sk_user_data = NULL;
1958
1959 newsk->sk_err = 0;
1960 newsk->sk_err_soft = 0;
1961 newsk->sk_priority = 0;
1962 newsk->sk_incoming_cpu = raw_smp_processor_id();
1963 if (likely(newsk->sk_net_refcnt))
1964 sock_inuse_add(sock_net(newsk), 1);
1965
1966 /*
1967 * Before updating sk_refcnt, we must commit prior changes to memory
1968 * (Documentation/RCU/rculist_nulls.rst for details)
1969 */
1970 smp_wmb();
1971 refcount_set(&newsk->sk_refcnt, 2);
1972
1973 /*
1974 * Increment the counter in the same struct proto as the master
1975 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1976 * is the same as sk->sk_prot->socks, as this field was copied
1977 * with memcpy).
1978 *
1979 * This _changes_ the previous behaviour, where
1980 * tcp_create_openreq_child always was incrementing the
1981 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1982 * to be taken into account in all callers. -acme
1983 */
1984 sk_refcnt_debug_inc(newsk);
1985 sk_set_socket(newsk, NULL);
1986 sk_tx_queue_clear(newsk);
1987 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1988
1989 if (newsk->sk_prot->sockets_allocated)
1990 sk_sockets_allocated_inc(newsk);
1991
1992 if (sock_needs_netstamp(sk) &&
1993 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1994 net_enable_timestamp();
1995 }
1996out:
1997 return newsk;
1998}
1999EXPORT_SYMBOL_GPL(sk_clone_lock);
2000
2001void sk_free_unlock_clone(struct sock *sk)
2002{
2003 /* It is still raw copy of parent, so invalidate
2004 * destructor and make plain sk_free() */
2005 sk->sk_destruct = NULL;
2006 bh_unlock_sock(sk);
2007 sk_free(sk);
2008}
2009EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2010
2011void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2012{
2013 u32 max_segs = 1;
2014
2015 sk_dst_set(sk, dst);
2016 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
2017 if (sk->sk_route_caps & NETIF_F_GSO)
2018 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2019 sk->sk_route_caps &= ~sk->sk_route_nocaps;
2020 if (sk_can_gso(sk)) {
2021 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2022 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2023 } else {
2024 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2025 sk->sk_gso_max_size = dst->dev->gso_max_size;
2026 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
2027 }
2028 }
2029 sk->sk_gso_max_segs = max_segs;
2030}
2031EXPORT_SYMBOL_GPL(sk_setup_caps);
2032
2033/*
2034 * Simple resource managers for sockets.
2035 */
2036
2037
2038/*
2039 * Write buffer destructor automatically called from kfree_skb.
2040 */
2041void sock_wfree(struct sk_buff *skb)
2042{
2043 struct sock *sk = skb->sk;
2044 unsigned int len = skb->truesize;
2045
2046 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2047 /*
2048 * Keep a reference on sk_wmem_alloc, this will be released
2049 * after sk_write_space() call
2050 */
2051 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2052 sk->sk_write_space(sk);
2053 len = 1;
2054 }
2055 /*
2056 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2057 * could not do because of in-flight packets
2058 */
2059 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2060 __sk_free(sk);
2061}
2062EXPORT_SYMBOL(sock_wfree);
2063
2064/* This variant of sock_wfree() is used by TCP,
2065 * since it sets SOCK_USE_WRITE_QUEUE.
2066 */
2067void __sock_wfree(struct sk_buff *skb)
2068{
2069 struct sock *sk = skb->sk;
2070
2071 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2072 __sk_free(sk);
2073}
2074
2075void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2076{
2077 skb_orphan(skb);
2078 skb->sk = sk;
2079#ifdef CONFIG_INET
2080 if (unlikely(!sk_fullsock(sk))) {
2081 skb->destructor = sock_edemux;
2082 sock_hold(sk);
2083 return;
2084 }
2085#endif
2086 skb->destructor = sock_wfree;
2087 skb_set_hash_from_sk(skb, sk);
2088 /*
2089 * We used to take a refcount on sk, but following operation
2090 * is enough to guarantee sk_free() wont free this sock until
2091 * all in-flight packets are completed
2092 */
2093 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2094}
2095EXPORT_SYMBOL(skb_set_owner_w);
2096
2097static bool can_skb_orphan_partial(const struct sk_buff *skb)
2098{
2099#ifdef CONFIG_TLS_DEVICE
2100 /* Drivers depend on in-order delivery for crypto offload,
2101 * partial orphan breaks out-of-order-OK logic.
2102 */
2103 if (skb->decrypted)
2104 return false;
2105#endif
2106 return (skb->destructor == sock_wfree ||
2107 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2108}
2109
2110/* This helper is used by netem, as it can hold packets in its
2111 * delay queue. We want to allow the owner socket to send more
2112 * packets, as if they were already TX completed by a typical driver.
2113 * But we also want to keep skb->sk set because some packet schedulers
2114 * rely on it (sch_fq for example).
2115 */
2116void skb_orphan_partial(struct sk_buff *skb)
2117{
2118 if (skb_is_tcp_pure_ack(skb))
2119 return;
2120
2121 if (can_skb_orphan_partial(skb)) {
2122 struct sock *sk = skb->sk;
2123
2124 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2125 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2126 skb->destructor = sock_efree;
2127 }
2128 } else {
2129 skb_orphan(skb);
2130 }
2131}
2132EXPORT_SYMBOL(skb_orphan_partial);
2133
2134/*
2135 * Read buffer destructor automatically called from kfree_skb.
2136 */
2137void sock_rfree(struct sk_buff *skb)
2138{
2139 struct sock *sk = skb->sk;
2140 unsigned int len = skb->truesize;
2141
2142 atomic_sub(len, &sk->sk_rmem_alloc);
2143 sk_mem_uncharge(sk, len);
2144}
2145EXPORT_SYMBOL(sock_rfree);
2146
2147/*
2148 * Buffer destructor for skbs that are not used directly in read or write
2149 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2150 */
2151void sock_efree(struct sk_buff *skb)
2152{
2153 sock_put(skb->sk);
2154}
2155EXPORT_SYMBOL(sock_efree);
2156
2157/* Buffer destructor for prefetch/receive path where reference count may
2158 * not be held, e.g. for listen sockets.
2159 */
2160#ifdef CONFIG_INET
2161void sock_pfree(struct sk_buff *skb)
2162{
2163 if (sk_is_refcounted(skb->sk))
2164 sock_gen_put(skb->sk);
2165}
2166EXPORT_SYMBOL(sock_pfree);
2167#endif /* CONFIG_INET */
2168
2169kuid_t sock_i_uid(struct sock *sk)
2170{
2171 kuid_t uid;
2172
2173 read_lock_bh(&sk->sk_callback_lock);
2174 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2175 read_unlock_bh(&sk->sk_callback_lock);
2176 return uid;
2177}
2178EXPORT_SYMBOL(sock_i_uid);
2179
2180unsigned long sock_i_ino(struct sock *sk)
2181{
2182 unsigned long ino;
2183
2184 read_lock_bh(&sk->sk_callback_lock);
2185 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2186 read_unlock_bh(&sk->sk_callback_lock);
2187 return ino;
2188}
2189EXPORT_SYMBOL(sock_i_ino);
2190
2191/*
2192 * Allocate a skb from the socket's send buffer.
2193 */
2194struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2195 gfp_t priority)
2196{
2197 if (force ||
2198 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2199 struct sk_buff *skb = alloc_skb(size, priority);
2200
2201 if (skb) {
2202 skb_set_owner_w(skb, sk);
2203 return skb;
2204 }
2205 }
2206 return NULL;
2207}
2208EXPORT_SYMBOL(sock_wmalloc);
2209
2210static void sock_ofree(struct sk_buff *skb)
2211{
2212 struct sock *sk = skb->sk;
2213
2214 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2215}
2216
2217struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2218 gfp_t priority)
2219{
2220 struct sk_buff *skb;
2221
2222 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2223 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2224 sysctl_optmem_max)
2225 return NULL;
2226
2227 skb = alloc_skb(size, priority);
2228 if (!skb)
2229 return NULL;
2230
2231 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2232 skb->sk = sk;
2233 skb->destructor = sock_ofree;
2234 return skb;
2235}
2236
2237/*
2238 * Allocate a memory block from the socket's option memory buffer.
2239 */
2240void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2241{
2242 if ((unsigned int)size <= sysctl_optmem_max &&
2243 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2244 void *mem;
2245 /* First do the add, to avoid the race if kmalloc
2246 * might sleep.
2247 */
2248 atomic_add(size, &sk->sk_omem_alloc);
2249 mem = kmalloc(size, priority);
2250 if (mem)
2251 return mem;
2252 atomic_sub(size, &sk->sk_omem_alloc);
2253 }
2254 return NULL;
2255}
2256EXPORT_SYMBOL(sock_kmalloc);
2257
2258/* Free an option memory block. Note, we actually want the inline
2259 * here as this allows gcc to detect the nullify and fold away the
2260 * condition entirely.
2261 */
2262static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2263 const bool nullify)
2264{
2265 if (WARN_ON_ONCE(!mem))
2266 return;
2267 if (nullify)
2268 kfree_sensitive(mem);
2269 else
2270 kfree(mem);
2271 atomic_sub(size, &sk->sk_omem_alloc);
2272}
2273
2274void sock_kfree_s(struct sock *sk, void *mem, int size)
2275{
2276 __sock_kfree_s(sk, mem, size, false);
2277}
2278EXPORT_SYMBOL(sock_kfree_s);
2279
2280void sock_kzfree_s(struct sock *sk, void *mem, int size)
2281{
2282 __sock_kfree_s(sk, mem, size, true);
2283}
2284EXPORT_SYMBOL(sock_kzfree_s);
2285
2286/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2287 I think, these locks should be removed for datagram sockets.
2288 */
2289static long sock_wait_for_wmem(struct sock *sk, long timeo)
2290{
2291 DEFINE_WAIT(wait);
2292
2293 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2294 for (;;) {
2295 if (!timeo)
2296 break;
2297 if (signal_pending(current))
2298 break;
2299 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2300 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2301 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2302 break;
2303 if (sk->sk_shutdown & SEND_SHUTDOWN)
2304 break;
2305 if (sk->sk_err)
2306 break;
2307 timeo = schedule_timeout(timeo);
2308 }
2309 finish_wait(sk_sleep(sk), &wait);
2310 return timeo;
2311}
2312
2313
2314/*
2315 * Generic send/receive buffer handlers
2316 */
2317
2318struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2319 unsigned long data_len, int noblock,
2320 int *errcode, int max_page_order)
2321{
2322 struct sk_buff *skb;
2323 long timeo;
2324 int err;
2325
2326 timeo = sock_sndtimeo(sk, noblock);
2327 for (;;) {
2328 err = sock_error(sk);
2329 if (err != 0)
2330 goto failure;
2331
2332 err = -EPIPE;
2333 if (sk->sk_shutdown & SEND_SHUTDOWN)
2334 goto failure;
2335
2336 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2337 break;
2338
2339 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2340 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2341 err = -EAGAIN;
2342 if (!timeo)
2343 goto failure;
2344 if (signal_pending(current))
2345 goto interrupted;
2346 timeo = sock_wait_for_wmem(sk, timeo);
2347 }
2348 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2349 errcode, sk->sk_allocation);
2350 if (skb)
2351 skb_set_owner_w(skb, sk);
2352 return skb;
2353
2354interrupted:
2355 err = sock_intr_errno(timeo);
2356failure:
2357 *errcode = err;
2358 return NULL;
2359}
2360EXPORT_SYMBOL(sock_alloc_send_pskb);
2361
2362struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2363 int noblock, int *errcode)
2364{
2365 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2366}
2367EXPORT_SYMBOL(sock_alloc_send_skb);
2368
2369int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2370 struct sockcm_cookie *sockc)
2371{
2372 u32 tsflags;
2373
2374 switch (cmsg->cmsg_type) {
2375 case SO_MARK:
2376 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2377 return -EPERM;
2378 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2379 return -EINVAL;
2380 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2381 break;
2382 case SO_TIMESTAMPING_OLD:
2383 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2384 return -EINVAL;
2385
2386 tsflags = *(u32 *)CMSG_DATA(cmsg);
2387 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2388 return -EINVAL;
2389
2390 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2391 sockc->tsflags |= tsflags;
2392 break;
2393 case SCM_TXTIME:
2394 if (!sock_flag(sk, SOCK_TXTIME))
2395 return -EINVAL;
2396 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2397 return -EINVAL;
2398 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2399 break;
2400 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2401 case SCM_RIGHTS:
2402 case SCM_CREDENTIALS:
2403 break;
2404 default:
2405 return -EINVAL;
2406 }
2407 return 0;
2408}
2409EXPORT_SYMBOL(__sock_cmsg_send);
2410
2411int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2412 struct sockcm_cookie *sockc)
2413{
2414 struct cmsghdr *cmsg;
2415 int ret;
2416
2417 for_each_cmsghdr(cmsg, msg) {
2418 if (!CMSG_OK(msg, cmsg))
2419 return -EINVAL;
2420 if (cmsg->cmsg_level != SOL_SOCKET)
2421 continue;
2422 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2423 if (ret)
2424 return ret;
2425 }
2426 return 0;
2427}
2428EXPORT_SYMBOL(sock_cmsg_send);
2429
2430static void sk_enter_memory_pressure(struct sock *sk)
2431{
2432 if (!sk->sk_prot->enter_memory_pressure)
2433 return;
2434
2435 sk->sk_prot->enter_memory_pressure(sk);
2436}
2437
2438static void sk_leave_memory_pressure(struct sock *sk)
2439{
2440 if (sk->sk_prot->leave_memory_pressure) {
2441 sk->sk_prot->leave_memory_pressure(sk);
2442 } else {
2443 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2444
2445 if (memory_pressure && READ_ONCE(*memory_pressure))
2446 WRITE_ONCE(*memory_pressure, 0);
2447 }
2448}
2449
2450#define SKB_FRAG_PAGE_ORDER get_order(32768)
2451DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2452
2453/**
2454 * skb_page_frag_refill - check that a page_frag contains enough room
2455 * @sz: minimum size of the fragment we want to get
2456 * @pfrag: pointer to page_frag
2457 * @gfp: priority for memory allocation
2458 *
2459 * Note: While this allocator tries to use high order pages, there is
2460 * no guarantee that allocations succeed. Therefore, @sz MUST be
2461 * less or equal than PAGE_SIZE.
2462 */
2463bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2464{
2465 if (pfrag->page) {
2466 if (page_ref_count(pfrag->page) == 1) {
2467 pfrag->offset = 0;
2468 return true;
2469 }
2470 if (pfrag->offset + sz <= pfrag->size)
2471 return true;
2472 put_page(pfrag->page);
2473 }
2474
2475 pfrag->offset = 0;
2476 if (SKB_FRAG_PAGE_ORDER &&
2477 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2478 /* Avoid direct reclaim but allow kswapd to wake */
2479 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2480 __GFP_COMP | __GFP_NOWARN |
2481 __GFP_NORETRY,
2482 SKB_FRAG_PAGE_ORDER);
2483 if (likely(pfrag->page)) {
2484 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2485 return true;
2486 }
2487 }
2488 pfrag->page = alloc_page(gfp);
2489 if (likely(pfrag->page)) {
2490 pfrag->size = PAGE_SIZE;
2491 return true;
2492 }
2493 return false;
2494}
2495EXPORT_SYMBOL(skb_page_frag_refill);
2496
2497bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2498{
2499 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2500 return true;
2501
2502 sk_enter_memory_pressure(sk);
2503 sk_stream_moderate_sndbuf(sk);
2504 return false;
2505}
2506EXPORT_SYMBOL(sk_page_frag_refill);
2507
2508static void __lock_sock(struct sock *sk)
2509 __releases(&sk->sk_lock.slock)
2510 __acquires(&sk->sk_lock.slock)
2511{
2512 DEFINE_WAIT(wait);
2513
2514 for (;;) {
2515 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2516 TASK_UNINTERRUPTIBLE);
2517 spin_unlock_bh(&sk->sk_lock.slock);
2518 schedule();
2519 spin_lock_bh(&sk->sk_lock.slock);
2520 if (!sock_owned_by_user(sk))
2521 break;
2522 }
2523 finish_wait(&sk->sk_lock.wq, &wait);
2524}
2525
2526void __release_sock(struct sock *sk)
2527 __releases(&sk->sk_lock.slock)
2528 __acquires(&sk->sk_lock.slock)
2529{
2530 struct sk_buff *skb, *next;
2531
2532 while ((skb = sk->sk_backlog.head) != NULL) {
2533 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2534
2535 spin_unlock_bh(&sk->sk_lock.slock);
2536
2537 do {
2538 next = skb->next;
2539 prefetch(next);
2540 WARN_ON_ONCE(skb_dst_is_noref(skb));
2541 skb_mark_not_on_list(skb);
2542 sk_backlog_rcv(sk, skb);
2543
2544 cond_resched();
2545
2546 skb = next;
2547 } while (skb != NULL);
2548
2549 spin_lock_bh(&sk->sk_lock.slock);
2550 }
2551
2552 /*
2553 * Doing the zeroing here guarantee we can not loop forever
2554 * while a wild producer attempts to flood us.
2555 */
2556 sk->sk_backlog.len = 0;
2557}
2558
2559void __sk_flush_backlog(struct sock *sk)
2560{
2561 spin_lock_bh(&sk->sk_lock.slock);
2562 __release_sock(sk);
2563 spin_unlock_bh(&sk->sk_lock.slock);
2564}
2565
2566/**
2567 * sk_wait_data - wait for data to arrive at sk_receive_queue
2568 * @sk: sock to wait on
2569 * @timeo: for how long
2570 * @skb: last skb seen on sk_receive_queue
2571 *
2572 * Now socket state including sk->sk_err is changed only under lock,
2573 * hence we may omit checks after joining wait queue.
2574 * We check receive queue before schedule() only as optimization;
2575 * it is very likely that release_sock() added new data.
2576 */
2577int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2578{
2579 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2580 int rc;
2581
2582 add_wait_queue(sk_sleep(sk), &wait);
2583 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2584 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2585 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2586 remove_wait_queue(sk_sleep(sk), &wait);
2587 return rc;
2588}
2589EXPORT_SYMBOL(sk_wait_data);
2590
2591/**
2592 * __sk_mem_raise_allocated - increase memory_allocated
2593 * @sk: socket
2594 * @size: memory size to allocate
2595 * @amt: pages to allocate
2596 * @kind: allocation type
2597 *
2598 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2599 */
2600int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2601{
2602 struct proto *prot = sk->sk_prot;
2603 long allocated = sk_memory_allocated_add(sk, amt);
2604 bool charged = true;
2605
2606 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2607 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2608 goto suppress_allocation;
2609
2610 /* Under limit. */
2611 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2612 sk_leave_memory_pressure(sk);
2613 return 1;
2614 }
2615
2616 /* Under pressure. */
2617 if (allocated > sk_prot_mem_limits(sk, 1))
2618 sk_enter_memory_pressure(sk);
2619
2620 /* Over hard limit. */
2621 if (allocated > sk_prot_mem_limits(sk, 2))
2622 goto suppress_allocation;
2623
2624 /* guarantee minimum buffer size under pressure */
2625 if (kind == SK_MEM_RECV) {
2626 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2627 return 1;
2628
2629 } else { /* SK_MEM_SEND */
2630 int wmem0 = sk_get_wmem0(sk, prot);
2631
2632 if (sk->sk_type == SOCK_STREAM) {
2633 if (sk->sk_wmem_queued < wmem0)
2634 return 1;
2635 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2636 return 1;
2637 }
2638 }
2639
2640 if (sk_has_memory_pressure(sk)) {
2641 u64 alloc;
2642
2643 if (!sk_under_memory_pressure(sk))
2644 return 1;
2645 alloc = sk_sockets_allocated_read_positive(sk);
2646 if (sk_prot_mem_limits(sk, 2) > alloc *
2647 sk_mem_pages(sk->sk_wmem_queued +
2648 atomic_read(&sk->sk_rmem_alloc) +
2649 sk->sk_forward_alloc))
2650 return 1;
2651 }
2652
2653suppress_allocation:
2654
2655 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2656 sk_stream_moderate_sndbuf(sk);
2657
2658 /* Fail only if socket is _under_ its sndbuf.
2659 * In this case we cannot block, so that we have to fail.
2660 */
2661 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2662 return 1;
2663 }
2664
2665 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2666 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2667
2668 sk_memory_allocated_sub(sk, amt);
2669
2670 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2671 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2672
2673 return 0;
2674}
2675EXPORT_SYMBOL(__sk_mem_raise_allocated);
2676
2677/**
2678 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2679 * @sk: socket
2680 * @size: memory size to allocate
2681 * @kind: allocation type
2682 *
2683 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2684 * rmem allocation. This function assumes that protocols which have
2685 * memory_pressure use sk_wmem_queued as write buffer accounting.
2686 */
2687int __sk_mem_schedule(struct sock *sk, int size, int kind)
2688{
2689 int ret, amt = sk_mem_pages(size);
2690
2691 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2692 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2693 if (!ret)
2694 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2695 return ret;
2696}
2697EXPORT_SYMBOL(__sk_mem_schedule);
2698
2699/**
2700 * __sk_mem_reduce_allocated - reclaim memory_allocated
2701 * @sk: socket
2702 * @amount: number of quanta
2703 *
2704 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2705 */
2706void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2707{
2708 sk_memory_allocated_sub(sk, amount);
2709
2710 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2711 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2712
2713 if (sk_under_memory_pressure(sk) &&
2714 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2715 sk_leave_memory_pressure(sk);
2716}
2717EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2718
2719/**
2720 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2721 * @sk: socket
2722 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2723 */
2724void __sk_mem_reclaim(struct sock *sk, int amount)
2725{
2726 amount >>= SK_MEM_QUANTUM_SHIFT;
2727 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2728 __sk_mem_reduce_allocated(sk, amount);
2729}
2730EXPORT_SYMBOL(__sk_mem_reclaim);
2731
2732int sk_set_peek_off(struct sock *sk, int val)
2733{
2734 sk->sk_peek_off = val;
2735 return 0;
2736}
2737EXPORT_SYMBOL_GPL(sk_set_peek_off);
2738
2739/*
2740 * Set of default routines for initialising struct proto_ops when
2741 * the protocol does not support a particular function. In certain
2742 * cases where it makes no sense for a protocol to have a "do nothing"
2743 * function, some default processing is provided.
2744 */
2745
2746int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2747{
2748 return -EOPNOTSUPP;
2749}
2750EXPORT_SYMBOL(sock_no_bind);
2751
2752int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2753 int len, int flags)
2754{
2755 return -EOPNOTSUPP;
2756}
2757EXPORT_SYMBOL(sock_no_connect);
2758
2759int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2760{
2761 return -EOPNOTSUPP;
2762}
2763EXPORT_SYMBOL(sock_no_socketpair);
2764
2765int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2766 bool kern)
2767{
2768 return -EOPNOTSUPP;
2769}
2770EXPORT_SYMBOL(sock_no_accept);
2771
2772int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2773 int peer)
2774{
2775 return -EOPNOTSUPP;
2776}
2777EXPORT_SYMBOL(sock_no_getname);
2778
2779int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2780{
2781 return -EOPNOTSUPP;
2782}
2783EXPORT_SYMBOL(sock_no_ioctl);
2784
2785int sock_no_listen(struct socket *sock, int backlog)
2786{
2787 return -EOPNOTSUPP;
2788}
2789EXPORT_SYMBOL(sock_no_listen);
2790
2791int sock_no_shutdown(struct socket *sock, int how)
2792{
2793 return -EOPNOTSUPP;
2794}
2795EXPORT_SYMBOL(sock_no_shutdown);
2796
2797int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2798{
2799 return -EOPNOTSUPP;
2800}
2801EXPORT_SYMBOL(sock_no_sendmsg);
2802
2803int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2804{
2805 return -EOPNOTSUPP;
2806}
2807EXPORT_SYMBOL(sock_no_sendmsg_locked);
2808
2809int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2810 int flags)
2811{
2812 return -EOPNOTSUPP;
2813}
2814EXPORT_SYMBOL(sock_no_recvmsg);
2815
2816int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2817{
2818 /* Mirror missing mmap method error code */
2819 return -ENODEV;
2820}
2821EXPORT_SYMBOL(sock_no_mmap);
2822
2823/*
2824 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2825 * various sock-based usage counts.
2826 */
2827void __receive_sock(struct file *file)
2828{
2829 struct socket *sock;
2830 int error;
2831
2832 /*
2833 * The resulting value of "error" is ignored here since we only
2834 * need to take action when the file is a socket and testing
2835 * "sock" for NULL is sufficient.
2836 */
2837 sock = sock_from_file(file, &error);
2838 if (sock) {
2839 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2840 sock_update_classid(&sock->sk->sk_cgrp_data);
2841 }
2842}
2843
2844ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2845{
2846 ssize_t res;
2847 struct msghdr msg = {.msg_flags = flags};
2848 struct kvec iov;
2849 char *kaddr = kmap(page);
2850 iov.iov_base = kaddr + offset;
2851 iov.iov_len = size;
2852 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2853 kunmap(page);
2854 return res;
2855}
2856EXPORT_SYMBOL(sock_no_sendpage);
2857
2858ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2859 int offset, size_t size, int flags)
2860{
2861 ssize_t res;
2862 struct msghdr msg = {.msg_flags = flags};
2863 struct kvec iov;
2864 char *kaddr = kmap(page);
2865
2866 iov.iov_base = kaddr + offset;
2867 iov.iov_len = size;
2868 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2869 kunmap(page);
2870 return res;
2871}
2872EXPORT_SYMBOL(sock_no_sendpage_locked);
2873
2874/*
2875 * Default Socket Callbacks
2876 */
2877
2878static void sock_def_wakeup(struct sock *sk)
2879{
2880 struct socket_wq *wq;
2881
2882 rcu_read_lock();
2883 wq = rcu_dereference(sk->sk_wq);
2884 if (skwq_has_sleeper(wq))
2885 wake_up_interruptible_all(&wq->wait);
2886 rcu_read_unlock();
2887}
2888
2889static void sock_def_error_report(struct sock *sk)
2890{
2891 struct socket_wq *wq;
2892
2893 rcu_read_lock();
2894 wq = rcu_dereference(sk->sk_wq);
2895 if (skwq_has_sleeper(wq))
2896 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2897 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2898 rcu_read_unlock();
2899}
2900
2901void sock_def_readable(struct sock *sk)
2902{
2903 struct socket_wq *wq;
2904
2905 rcu_read_lock();
2906 wq = rcu_dereference(sk->sk_wq);
2907 if (skwq_has_sleeper(wq))
2908 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2909 EPOLLRDNORM | EPOLLRDBAND);
2910 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2911 rcu_read_unlock();
2912}
2913
2914static void sock_def_write_space(struct sock *sk)
2915{
2916 struct socket_wq *wq;
2917
2918 rcu_read_lock();
2919
2920 /* Do not wake up a writer until he can make "significant"
2921 * progress. --DaveM
2922 */
2923 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2924 wq = rcu_dereference(sk->sk_wq);
2925 if (skwq_has_sleeper(wq))
2926 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2927 EPOLLWRNORM | EPOLLWRBAND);
2928
2929 /* Should agree with poll, otherwise some programs break */
2930 if (sock_writeable(sk))
2931 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2932 }
2933
2934 rcu_read_unlock();
2935}
2936
2937static void sock_def_destruct(struct sock *sk)
2938{
2939}
2940
2941void sk_send_sigurg(struct sock *sk)
2942{
2943 if (sk->sk_socket && sk->sk_socket->file)
2944 if (send_sigurg(&sk->sk_socket->file->f_owner))
2945 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2946}
2947EXPORT_SYMBOL(sk_send_sigurg);
2948
2949void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2950 unsigned long expires)
2951{
2952 if (!mod_timer(timer, expires))
2953 sock_hold(sk);
2954}
2955EXPORT_SYMBOL(sk_reset_timer);
2956
2957void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2958{
2959 if (del_timer(timer))
2960 __sock_put(sk);
2961}
2962EXPORT_SYMBOL(sk_stop_timer);
2963
2964void sock_init_data(struct socket *sock, struct sock *sk)
2965{
2966 sk_init_common(sk);
2967 sk->sk_send_head = NULL;
2968
2969 timer_setup(&sk->sk_timer, NULL, 0);
2970
2971 sk->sk_allocation = GFP_KERNEL;
2972 sk->sk_rcvbuf = sysctl_rmem_default;
2973 sk->sk_sndbuf = sysctl_wmem_default;
2974 sk->sk_state = TCP_CLOSE;
2975 sk_set_socket(sk, sock);
2976
2977 sock_set_flag(sk, SOCK_ZAPPED);
2978
2979 if (sock) {
2980 sk->sk_type = sock->type;
2981 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2982 sock->sk = sk;
2983 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2984 } else {
2985 RCU_INIT_POINTER(sk->sk_wq, NULL);
2986 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2987 }
2988
2989 rwlock_init(&sk->sk_callback_lock);
2990 if (sk->sk_kern_sock)
2991 lockdep_set_class_and_name(
2992 &sk->sk_callback_lock,
2993 af_kern_callback_keys + sk->sk_family,
2994 af_family_kern_clock_key_strings[sk->sk_family]);
2995 else
2996 lockdep_set_class_and_name(
2997 &sk->sk_callback_lock,
2998 af_callback_keys + sk->sk_family,
2999 af_family_clock_key_strings[sk->sk_family]);
3000
3001 sk->sk_state_change = sock_def_wakeup;
3002 sk->sk_data_ready = sock_def_readable;
3003 sk->sk_write_space = sock_def_write_space;
3004 sk->sk_error_report = sock_def_error_report;
3005 sk->sk_destruct = sock_def_destruct;
3006
3007 sk->sk_frag.page = NULL;
3008 sk->sk_frag.offset = 0;
3009 sk->sk_peek_off = -1;
3010
3011 sk->sk_peer_pid = NULL;
3012 sk->sk_peer_cred = NULL;
3013 sk->sk_write_pending = 0;
3014 sk->sk_rcvlowat = 1;
3015 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3016 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3017
3018 sk->sk_stamp = SK_DEFAULT_STAMP;
3019#if BITS_PER_LONG==32
3020 seqlock_init(&sk->sk_stamp_seq);
3021#endif
3022 atomic_set(&sk->sk_zckey, 0);
3023
3024#ifdef CONFIG_NET_RX_BUSY_POLL
3025 sk->sk_napi_id = 0;
3026 sk->sk_ll_usec = sysctl_net_busy_read;
3027#endif
3028
3029 sk->sk_max_pacing_rate = ~0UL;
3030 sk->sk_pacing_rate = ~0UL;
3031 WRITE_ONCE(sk->sk_pacing_shift, 10);
3032 sk->sk_incoming_cpu = -1;
3033
3034 sk_rx_queue_clear(sk);
3035 /*
3036 * Before updating sk_refcnt, we must commit prior changes to memory
3037 * (Documentation/RCU/rculist_nulls.rst for details)
3038 */
3039 smp_wmb();
3040 refcount_set(&sk->sk_refcnt, 1);
3041 atomic_set(&sk->sk_drops, 0);
3042}
3043EXPORT_SYMBOL(sock_init_data);
3044
3045void lock_sock_nested(struct sock *sk, int subclass)
3046{
3047 might_sleep();
3048 spin_lock_bh(&sk->sk_lock.slock);
3049 if (sk->sk_lock.owned)
3050 __lock_sock(sk);
3051 sk->sk_lock.owned = 1;
3052 spin_unlock(&sk->sk_lock.slock);
3053 /*
3054 * The sk_lock has mutex_lock() semantics here:
3055 */
3056 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3057 local_bh_enable();
3058}
3059EXPORT_SYMBOL(lock_sock_nested);
3060
3061void release_sock(struct sock *sk)
3062{
3063 spin_lock_bh(&sk->sk_lock.slock);
3064 if (sk->sk_backlog.tail)
3065 __release_sock(sk);
3066
3067 /* Warning : release_cb() might need to release sk ownership,
3068 * ie call sock_release_ownership(sk) before us.
3069 */
3070 if (sk->sk_prot->release_cb)
3071 sk->sk_prot->release_cb(sk);
3072
3073 sock_release_ownership(sk);
3074 if (waitqueue_active(&sk->sk_lock.wq))
3075 wake_up(&sk->sk_lock.wq);
3076 spin_unlock_bh(&sk->sk_lock.slock);
3077}
3078EXPORT_SYMBOL(release_sock);
3079
3080/**
3081 * lock_sock_fast - fast version of lock_sock
3082 * @sk: socket
3083 *
3084 * This version should be used for very small section, where process wont block
3085 * return false if fast path is taken:
3086 *
3087 * sk_lock.slock locked, owned = 0, BH disabled
3088 *
3089 * return true if slow path is taken:
3090 *
3091 * sk_lock.slock unlocked, owned = 1, BH enabled
3092 */
3093bool lock_sock_fast(struct sock *sk)
3094{
3095 might_sleep();
3096 spin_lock_bh(&sk->sk_lock.slock);
3097
3098 if (!sk->sk_lock.owned)
3099 /*
3100 * Note : We must disable BH
3101 */
3102 return false;
3103
3104 __lock_sock(sk);
3105 sk->sk_lock.owned = 1;
3106 spin_unlock(&sk->sk_lock.slock);
3107 /*
3108 * The sk_lock has mutex_lock() semantics here:
3109 */
3110 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3111 local_bh_enable();
3112 return true;
3113}
3114EXPORT_SYMBOL(lock_sock_fast);
3115
3116int sock_gettstamp(struct socket *sock, void __user *userstamp,
3117 bool timeval, bool time32)
3118{
3119 struct sock *sk = sock->sk;
3120 struct timespec64 ts;
3121
3122 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3123 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3124 if (ts.tv_sec == -1)
3125 return -ENOENT;
3126 if (ts.tv_sec == 0) {
3127 ktime_t kt = ktime_get_real();
3128 sock_write_timestamp(sk, kt);
3129 ts = ktime_to_timespec64(kt);
3130 }
3131
3132 if (timeval)
3133 ts.tv_nsec /= 1000;
3134
3135#ifdef CONFIG_COMPAT_32BIT_TIME
3136 if (time32)
3137 return put_old_timespec32(&ts, userstamp);
3138#endif
3139#ifdef CONFIG_SPARC64
3140 /* beware of padding in sparc64 timeval */
3141 if (timeval && !in_compat_syscall()) {
3142 struct __kernel_old_timeval __user tv = {
3143 .tv_sec = ts.tv_sec,
3144 .tv_usec = ts.tv_nsec,
3145 };
3146 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3147 return -EFAULT;
3148 return 0;
3149 }
3150#endif
3151 return put_timespec64(&ts, userstamp);
3152}
3153EXPORT_SYMBOL(sock_gettstamp);
3154
3155void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3156{
3157 if (!sock_flag(sk, flag)) {
3158 unsigned long previous_flags = sk->sk_flags;
3159
3160 sock_set_flag(sk, flag);
3161 /*
3162 * we just set one of the two flags which require net
3163 * time stamping, but time stamping might have been on
3164 * already because of the other one
3165 */
3166 if (sock_needs_netstamp(sk) &&
3167 !(previous_flags & SK_FLAGS_TIMESTAMP))
3168 net_enable_timestamp();
3169 }
3170}
3171
3172int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3173 int level, int type)
3174{
3175 struct sock_exterr_skb *serr;
3176 struct sk_buff *skb;
3177 int copied, err;
3178
3179 err = -EAGAIN;
3180 skb = sock_dequeue_err_skb(sk);
3181 if (skb == NULL)
3182 goto out;
3183
3184 copied = skb->len;
3185 if (copied > len) {
3186 msg->msg_flags |= MSG_TRUNC;
3187 copied = len;
3188 }
3189 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3190 if (err)
3191 goto out_free_skb;
3192
3193 sock_recv_timestamp(msg, sk, skb);
3194
3195 serr = SKB_EXT_ERR(skb);
3196 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3197
3198 msg->msg_flags |= MSG_ERRQUEUE;
3199 err = copied;
3200
3201out_free_skb:
3202 kfree_skb(skb);
3203out:
3204 return err;
3205}
3206EXPORT_SYMBOL(sock_recv_errqueue);
3207
3208/*
3209 * Get a socket option on an socket.
3210 *
3211 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3212 * asynchronous errors should be reported by getsockopt. We assume
3213 * this means if you specify SO_ERROR (otherwise whats the point of it).
3214 */
3215int sock_common_getsockopt(struct socket *sock, int level, int optname,
3216 char __user *optval, int __user *optlen)
3217{
3218 struct sock *sk = sock->sk;
3219
3220 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3221}
3222EXPORT_SYMBOL(sock_common_getsockopt);
3223
3224int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3225 int flags)
3226{
3227 struct sock *sk = sock->sk;
3228 int addr_len = 0;
3229 int err;
3230
3231 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3232 flags & ~MSG_DONTWAIT, &addr_len);
3233 if (err >= 0)
3234 msg->msg_namelen = addr_len;
3235 return err;
3236}
3237EXPORT_SYMBOL(sock_common_recvmsg);
3238
3239/*
3240 * Set socket options on an inet socket.
3241 */
3242int sock_common_setsockopt(struct socket *sock, int level, int optname,
3243 sockptr_t optval, unsigned int optlen)
3244{
3245 struct sock *sk = sock->sk;
3246
3247 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3248}
3249EXPORT_SYMBOL(sock_common_setsockopt);
3250
3251void sk_common_release(struct sock *sk)
3252{
3253 if (sk->sk_prot->destroy)
3254 sk->sk_prot->destroy(sk);
3255
3256 /*
3257 * Observation: when sk_common_release is called, processes have
3258 * no access to socket. But net still has.
3259 * Step one, detach it from networking:
3260 *
3261 * A. Remove from hash tables.
3262 */
3263
3264 sk->sk_prot->unhash(sk);
3265
3266 /*
3267 * In this point socket cannot receive new packets, but it is possible
3268 * that some packets are in flight because some CPU runs receiver and
3269 * did hash table lookup before we unhashed socket. They will achieve
3270 * receive queue and will be purged by socket destructor.
3271 *
3272 * Also we still have packets pending on receive queue and probably,
3273 * our own packets waiting in device queues. sock_destroy will drain
3274 * receive queue, but transmitted packets will delay socket destruction
3275 * until the last reference will be released.
3276 */
3277
3278 sock_orphan(sk);
3279
3280 xfrm_sk_free_policy(sk);
3281
3282 sk_refcnt_debug_release(sk);
3283
3284 sock_put(sk);
3285}
3286EXPORT_SYMBOL(sk_common_release);
3287
3288void sk_get_meminfo(const struct sock *sk, u32 *mem)
3289{
3290 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3291
3292 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3293 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3294 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3295 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3296 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3297 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3298 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3299 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3300 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3301}
3302
3303#ifdef CONFIG_PROC_FS
3304#define PROTO_INUSE_NR 64 /* should be enough for the first time */
3305struct prot_inuse {
3306 int val[PROTO_INUSE_NR];
3307};
3308
3309static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3310
3311void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3312{
3313 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3314}
3315EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3316
3317int sock_prot_inuse_get(struct net *net, struct proto *prot)
3318{
3319 int cpu, idx = prot->inuse_idx;
3320 int res = 0;
3321
3322 for_each_possible_cpu(cpu)
3323 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3324
3325 return res >= 0 ? res : 0;
3326}
3327EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3328
3329static void sock_inuse_add(struct net *net, int val)
3330{
3331 this_cpu_add(*net->core.sock_inuse, val);
3332}
3333
3334int sock_inuse_get(struct net *net)
3335{
3336 int cpu, res = 0;
3337
3338 for_each_possible_cpu(cpu)
3339 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3340
3341 return res;
3342}
3343
3344EXPORT_SYMBOL_GPL(sock_inuse_get);
3345
3346static int __net_init sock_inuse_init_net(struct net *net)
3347{
3348 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3349 if (net->core.prot_inuse == NULL)
3350 return -ENOMEM;
3351
3352 net->core.sock_inuse = alloc_percpu(int);
3353 if (net->core.sock_inuse == NULL)
3354 goto out;
3355
3356 return 0;
3357
3358out:
3359 free_percpu(net->core.prot_inuse);
3360 return -ENOMEM;
3361}
3362
3363static void __net_exit sock_inuse_exit_net(struct net *net)
3364{
3365 free_percpu(net->core.prot_inuse);
3366 free_percpu(net->core.sock_inuse);
3367}
3368
3369static struct pernet_operations net_inuse_ops = {
3370 .init = sock_inuse_init_net,
3371 .exit = sock_inuse_exit_net,
3372};
3373
3374static __init int net_inuse_init(void)
3375{
3376 if (register_pernet_subsys(&net_inuse_ops))
3377 panic("Cannot initialize net inuse counters");
3378
3379 return 0;
3380}
3381
3382core_initcall(net_inuse_init);
3383
3384static int assign_proto_idx(struct proto *prot)
3385{
3386 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3387
3388 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3389 pr_err("PROTO_INUSE_NR exhausted\n");
3390 return -ENOSPC;
3391 }
3392
3393 set_bit(prot->inuse_idx, proto_inuse_idx);
3394 return 0;
3395}
3396
3397static void release_proto_idx(struct proto *prot)
3398{
3399 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3400 clear_bit(prot->inuse_idx, proto_inuse_idx);
3401}
3402#else
3403static inline int assign_proto_idx(struct proto *prot)
3404{
3405 return 0;
3406}
3407
3408static inline void release_proto_idx(struct proto *prot)
3409{
3410}
3411
3412static void sock_inuse_add(struct net *net, int val)
3413{
3414}
3415#endif
3416
3417static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3418{
3419 if (!twsk_prot)
3420 return;
3421 kfree(twsk_prot->twsk_slab_name);
3422 twsk_prot->twsk_slab_name = NULL;
3423 kmem_cache_destroy(twsk_prot->twsk_slab);
3424 twsk_prot->twsk_slab = NULL;
3425}
3426
3427static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3428{
3429 if (!rsk_prot)
3430 return;
3431 kfree(rsk_prot->slab_name);
3432 rsk_prot->slab_name = NULL;
3433 kmem_cache_destroy(rsk_prot->slab);
3434 rsk_prot->slab = NULL;
3435}
3436
3437static int req_prot_init(const struct proto *prot)
3438{
3439 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3440
3441 if (!rsk_prot)
3442 return 0;
3443
3444 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3445 prot->name);
3446 if (!rsk_prot->slab_name)
3447 return -ENOMEM;
3448
3449 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3450 rsk_prot->obj_size, 0,
3451 SLAB_ACCOUNT | prot->slab_flags,
3452 NULL);
3453
3454 if (!rsk_prot->slab) {
3455 pr_crit("%s: Can't create request sock SLAB cache!\n",
3456 prot->name);
3457 return -ENOMEM;
3458 }
3459 return 0;
3460}
3461
3462int proto_register(struct proto *prot, int alloc_slab)
3463{
3464 int ret = -ENOBUFS;
3465
3466 if (alloc_slab) {
3467 prot->slab = kmem_cache_create_usercopy(prot->name,
3468 prot->obj_size, 0,
3469 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3470 prot->slab_flags,
3471 prot->useroffset, prot->usersize,
3472 NULL);
3473
3474 if (prot->slab == NULL) {
3475 pr_crit("%s: Can't create sock SLAB cache!\n",
3476 prot->name);
3477 goto out;
3478 }
3479
3480 if (req_prot_init(prot))
3481 goto out_free_request_sock_slab;
3482
3483 if (prot->twsk_prot != NULL) {
3484 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3485
3486 if (prot->twsk_prot->twsk_slab_name == NULL)
3487 goto out_free_request_sock_slab;
3488
3489 prot->twsk_prot->twsk_slab =
3490 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3491 prot->twsk_prot->twsk_obj_size,
3492 0,
3493 SLAB_ACCOUNT |
3494 prot->slab_flags,
3495 NULL);
3496 if (prot->twsk_prot->twsk_slab == NULL)
3497 goto out_free_timewait_sock_slab;
3498 }
3499 }
3500
3501 mutex_lock(&proto_list_mutex);
3502 ret = assign_proto_idx(prot);
3503 if (ret) {
3504 mutex_unlock(&proto_list_mutex);
3505 goto out_free_timewait_sock_slab;
3506 }
3507 list_add(&prot->node, &proto_list);
3508 mutex_unlock(&proto_list_mutex);
3509 return ret;
3510
3511out_free_timewait_sock_slab:
3512 if (alloc_slab && prot->twsk_prot)
3513 tw_prot_cleanup(prot->twsk_prot);
3514out_free_request_sock_slab:
3515 if (alloc_slab) {
3516 req_prot_cleanup(prot->rsk_prot);
3517
3518 kmem_cache_destroy(prot->slab);
3519 prot->slab = NULL;
3520 }
3521out:
3522 return ret;
3523}
3524EXPORT_SYMBOL(proto_register);
3525
3526void proto_unregister(struct proto *prot)
3527{
3528 mutex_lock(&proto_list_mutex);
3529 release_proto_idx(prot);
3530 list_del(&prot->node);
3531 mutex_unlock(&proto_list_mutex);
3532
3533 kmem_cache_destroy(prot->slab);
3534 prot->slab = NULL;
3535
3536 req_prot_cleanup(prot->rsk_prot);
3537 tw_prot_cleanup(prot->twsk_prot);
3538}
3539EXPORT_SYMBOL(proto_unregister);
3540
3541int sock_load_diag_module(int family, int protocol)
3542{
3543 if (!protocol) {
3544 if (!sock_is_registered(family))
3545 return -ENOENT;
3546
3547 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3548 NETLINK_SOCK_DIAG, family);
3549 }
3550
3551#ifdef CONFIG_INET
3552 if (family == AF_INET &&
3553 protocol != IPPROTO_RAW &&
3554 protocol < MAX_INET_PROTOS &&
3555 !rcu_access_pointer(inet_protos[protocol]))
3556 return -ENOENT;
3557#endif
3558
3559 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3560 NETLINK_SOCK_DIAG, family, protocol);
3561}
3562EXPORT_SYMBOL(sock_load_diag_module);
3563
3564#ifdef CONFIG_PROC_FS
3565static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3566 __acquires(proto_list_mutex)
3567{
3568 mutex_lock(&proto_list_mutex);
3569 return seq_list_start_head(&proto_list, *pos);
3570}
3571
3572static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3573{
3574 return seq_list_next(v, &proto_list, pos);
3575}
3576
3577static void proto_seq_stop(struct seq_file *seq, void *v)
3578 __releases(proto_list_mutex)
3579{
3580 mutex_unlock(&proto_list_mutex);
3581}
3582
3583static char proto_method_implemented(const void *method)
3584{
3585 return method == NULL ? 'n' : 'y';
3586}
3587static long sock_prot_memory_allocated(struct proto *proto)
3588{
3589 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3590}
3591
3592static const char *sock_prot_memory_pressure(struct proto *proto)
3593{
3594 return proto->memory_pressure != NULL ?
3595 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3596}
3597
3598static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3599{
3600
3601 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3602 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3603 proto->name,
3604 proto->obj_size,
3605 sock_prot_inuse_get(seq_file_net(seq), proto),
3606 sock_prot_memory_allocated(proto),
3607 sock_prot_memory_pressure(proto),
3608 proto->max_header,
3609 proto->slab == NULL ? "no" : "yes",
3610 module_name(proto->owner),
3611 proto_method_implemented(proto->close),
3612 proto_method_implemented(proto->connect),
3613 proto_method_implemented(proto->disconnect),
3614 proto_method_implemented(proto->accept),
3615 proto_method_implemented(proto->ioctl),
3616 proto_method_implemented(proto->init),
3617 proto_method_implemented(proto->destroy),
3618 proto_method_implemented(proto->shutdown),
3619 proto_method_implemented(proto->setsockopt),
3620 proto_method_implemented(proto->getsockopt),
3621 proto_method_implemented(proto->sendmsg),
3622 proto_method_implemented(proto->recvmsg),
3623 proto_method_implemented(proto->sendpage),
3624 proto_method_implemented(proto->bind),
3625 proto_method_implemented(proto->backlog_rcv),
3626 proto_method_implemented(proto->hash),
3627 proto_method_implemented(proto->unhash),
3628 proto_method_implemented(proto->get_port),
3629 proto_method_implemented(proto->enter_memory_pressure));
3630}
3631
3632static int proto_seq_show(struct seq_file *seq, void *v)
3633{
3634 if (v == &proto_list)
3635 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3636 "protocol",
3637 "size",
3638 "sockets",
3639 "memory",
3640 "press",
3641 "maxhdr",
3642 "slab",
3643 "module",
3644 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3645 else
3646 proto_seq_printf(seq, list_entry(v, struct proto, node));
3647 return 0;
3648}
3649
3650static const struct seq_operations proto_seq_ops = {
3651 .start = proto_seq_start,
3652 .next = proto_seq_next,
3653 .stop = proto_seq_stop,
3654 .show = proto_seq_show,
3655};
3656
3657static __net_init int proto_init_net(struct net *net)
3658{
3659 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3660 sizeof(struct seq_net_private)))
3661 return -ENOMEM;
3662
3663 return 0;
3664}
3665
3666static __net_exit void proto_exit_net(struct net *net)
3667{
3668 remove_proc_entry("protocols", net->proc_net);
3669}
3670
3671
3672static __net_initdata struct pernet_operations proto_net_ops = {
3673 .init = proto_init_net,
3674 .exit = proto_exit_net,
3675};
3676
3677static int __init proto_init(void)
3678{
3679 return register_pernet_subsys(&proto_net_ops);
3680}
3681
3682subsys_initcall(proto_init);
3683
3684#endif /* PROC_FS */
3685
3686#ifdef CONFIG_NET_RX_BUSY_POLL
3687bool sk_busy_loop_end(void *p, unsigned long start_time)
3688{
3689 struct sock *sk = p;
3690
3691 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3692 sk_busy_loop_timeout(sk, start_time);
3693}
3694EXPORT_SYMBOL(sk_busy_loop_end);
3695#endif /* CONFIG_NET_RX_BUSY_POLL */
3696
3697int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
3698{
3699 if (!sk->sk_prot->bind_add)
3700 return -EOPNOTSUPP;
3701 return sk->sk_prot->bind_add(sk, addr, addr_len);
3702}
3703EXPORT_SYMBOL(sock_bind_add);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88#include <asm/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/init.h>
111#include <linux/highmem.h>
112#include <linux/user_namespace.h>
113#include <linux/static_key.h>
114#include <linux/memcontrol.h>
115#include <linux/prefetch.h>
116#include <linux/compat.h>
117
118#include <linux/uaccess.h>
119
120#include <linux/netdevice.h>
121#include <net/protocol.h>
122#include <linux/skbuff.h>
123#include <net/net_namespace.h>
124#include <net/request_sock.h>
125#include <net/sock.h>
126#include <linux/net_tstamp.h>
127#include <net/xfrm.h>
128#include <linux/ipsec.h>
129#include <net/cls_cgroup.h>
130#include <net/netprio_cgroup.h>
131#include <linux/sock_diag.h>
132
133#include <linux/filter.h>
134#include <net/sock_reuseport.h>
135#include <net/bpf_sk_storage.h>
136
137#include <trace/events/sock.h>
138
139#include <net/tcp.h>
140#include <net/busy_poll.h>
141
142#include <linux/ethtool.h>
143
144#include "dev.h"
145
146static DEFINE_MUTEX(proto_list_mutex);
147static LIST_HEAD(proto_list);
148
149static void sock_def_write_space_wfree(struct sock *sk);
150static void sock_def_write_space(struct sock *sk);
151
152/**
153 * sk_ns_capable - General socket capability test
154 * @sk: Socket to use a capability on or through
155 * @user_ns: The user namespace of the capability to use
156 * @cap: The capability to use
157 *
158 * Test to see if the opener of the socket had when the socket was
159 * created and the current process has the capability @cap in the user
160 * namespace @user_ns.
161 */
162bool sk_ns_capable(const struct sock *sk,
163 struct user_namespace *user_ns, int cap)
164{
165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
166 ns_capable(user_ns, cap);
167}
168EXPORT_SYMBOL(sk_ns_capable);
169
170/**
171 * sk_capable - Socket global capability test
172 * @sk: Socket to use a capability on or through
173 * @cap: The global capability to use
174 *
175 * Test to see if the opener of the socket had when the socket was
176 * created and the current process has the capability @cap in all user
177 * namespaces.
178 */
179bool sk_capable(const struct sock *sk, int cap)
180{
181 return sk_ns_capable(sk, &init_user_ns, cap);
182}
183EXPORT_SYMBOL(sk_capable);
184
185/**
186 * sk_net_capable - Network namespace socket capability test
187 * @sk: Socket to use a capability on or through
188 * @cap: The capability to use
189 *
190 * Test to see if the opener of the socket had when the socket was created
191 * and the current process has the capability @cap over the network namespace
192 * the socket is a member of.
193 */
194bool sk_net_capable(const struct sock *sk, int cap)
195{
196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
197}
198EXPORT_SYMBOL(sk_net_capable);
199
200/*
201 * Each address family might have different locking rules, so we have
202 * one slock key per address family and separate keys for internal and
203 * userspace sockets.
204 */
205static struct lock_class_key af_family_keys[AF_MAX];
206static struct lock_class_key af_family_kern_keys[AF_MAX];
207static struct lock_class_key af_family_slock_keys[AF_MAX];
208static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
209
210/*
211 * Make lock validator output more readable. (we pre-construct these
212 * strings build-time, so that runtime initialization of socket
213 * locks is fast):
214 */
215
216#define _sock_locks(x) \
217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
226 x "27" , x "28" , x "AF_CAN" , \
227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
232 x "AF_MCTP" , \
233 x "AF_MAX"
234
235static const char *const af_family_key_strings[AF_MAX+1] = {
236 _sock_locks("sk_lock-")
237};
238static const char *const af_family_slock_key_strings[AF_MAX+1] = {
239 _sock_locks("slock-")
240};
241static const char *const af_family_clock_key_strings[AF_MAX+1] = {
242 _sock_locks("clock-")
243};
244
245static const char *const af_family_kern_key_strings[AF_MAX+1] = {
246 _sock_locks("k-sk_lock-")
247};
248static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
249 _sock_locks("k-slock-")
250};
251static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
252 _sock_locks("k-clock-")
253};
254static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
255 _sock_locks("rlock-")
256};
257static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
258 _sock_locks("wlock-")
259};
260static const char *const af_family_elock_key_strings[AF_MAX+1] = {
261 _sock_locks("elock-")
262};
263
264/*
265 * sk_callback_lock and sk queues locking rules are per-address-family,
266 * so split the lock classes by using a per-AF key:
267 */
268static struct lock_class_key af_callback_keys[AF_MAX];
269static struct lock_class_key af_rlock_keys[AF_MAX];
270static struct lock_class_key af_wlock_keys[AF_MAX];
271static struct lock_class_key af_elock_keys[AF_MAX];
272static struct lock_class_key af_kern_callback_keys[AF_MAX];
273
274/* Run time adjustable parameters. */
275__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
276EXPORT_SYMBOL(sysctl_wmem_max);
277__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
278EXPORT_SYMBOL(sysctl_rmem_max);
279__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
280__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
281
282/* Maximal space eaten by iovec or ancillary data plus some space */
283int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
284EXPORT_SYMBOL(sysctl_optmem_max);
285
286int sysctl_tstamp_allow_data __read_mostly = 1;
287
288DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
289EXPORT_SYMBOL_GPL(memalloc_socks_key);
290
291/**
292 * sk_set_memalloc - sets %SOCK_MEMALLOC
293 * @sk: socket to set it on
294 *
295 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
296 * It's the responsibility of the admin to adjust min_free_kbytes
297 * to meet the requirements
298 */
299void sk_set_memalloc(struct sock *sk)
300{
301 sock_set_flag(sk, SOCK_MEMALLOC);
302 sk->sk_allocation |= __GFP_MEMALLOC;
303 static_branch_inc(&memalloc_socks_key);
304}
305EXPORT_SYMBOL_GPL(sk_set_memalloc);
306
307void sk_clear_memalloc(struct sock *sk)
308{
309 sock_reset_flag(sk, SOCK_MEMALLOC);
310 sk->sk_allocation &= ~__GFP_MEMALLOC;
311 static_branch_dec(&memalloc_socks_key);
312
313 /*
314 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
315 * progress of swapping. SOCK_MEMALLOC may be cleared while
316 * it has rmem allocations due to the last swapfile being deactivated
317 * but there is a risk that the socket is unusable due to exceeding
318 * the rmem limits. Reclaim the reserves and obey rmem limits again.
319 */
320 sk_mem_reclaim(sk);
321}
322EXPORT_SYMBOL_GPL(sk_clear_memalloc);
323
324int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
325{
326 int ret;
327 unsigned int noreclaim_flag;
328
329 /* these should have been dropped before queueing */
330 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
331
332 noreclaim_flag = memalloc_noreclaim_save();
333 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
334 tcp_v6_do_rcv,
335 tcp_v4_do_rcv,
336 sk, skb);
337 memalloc_noreclaim_restore(noreclaim_flag);
338
339 return ret;
340}
341EXPORT_SYMBOL(__sk_backlog_rcv);
342
343void sk_error_report(struct sock *sk)
344{
345 sk->sk_error_report(sk);
346
347 switch (sk->sk_family) {
348 case AF_INET:
349 fallthrough;
350 case AF_INET6:
351 trace_inet_sk_error_report(sk);
352 break;
353 default:
354 break;
355 }
356}
357EXPORT_SYMBOL(sk_error_report);
358
359int sock_get_timeout(long timeo, void *optval, bool old_timeval)
360{
361 struct __kernel_sock_timeval tv;
362
363 if (timeo == MAX_SCHEDULE_TIMEOUT) {
364 tv.tv_sec = 0;
365 tv.tv_usec = 0;
366 } else {
367 tv.tv_sec = timeo / HZ;
368 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
369 }
370
371 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
372 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
373 *(struct old_timeval32 *)optval = tv32;
374 return sizeof(tv32);
375 }
376
377 if (old_timeval) {
378 struct __kernel_old_timeval old_tv;
379 old_tv.tv_sec = tv.tv_sec;
380 old_tv.tv_usec = tv.tv_usec;
381 *(struct __kernel_old_timeval *)optval = old_tv;
382 return sizeof(old_tv);
383 }
384
385 *(struct __kernel_sock_timeval *)optval = tv;
386 return sizeof(tv);
387}
388EXPORT_SYMBOL(sock_get_timeout);
389
390int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
391 sockptr_t optval, int optlen, bool old_timeval)
392{
393 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
394 struct old_timeval32 tv32;
395
396 if (optlen < sizeof(tv32))
397 return -EINVAL;
398
399 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
400 return -EFAULT;
401 tv->tv_sec = tv32.tv_sec;
402 tv->tv_usec = tv32.tv_usec;
403 } else if (old_timeval) {
404 struct __kernel_old_timeval old_tv;
405
406 if (optlen < sizeof(old_tv))
407 return -EINVAL;
408 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
409 return -EFAULT;
410 tv->tv_sec = old_tv.tv_sec;
411 tv->tv_usec = old_tv.tv_usec;
412 } else {
413 if (optlen < sizeof(*tv))
414 return -EINVAL;
415 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
416 return -EFAULT;
417 }
418
419 return 0;
420}
421EXPORT_SYMBOL(sock_copy_user_timeval);
422
423static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
424 bool old_timeval)
425{
426 struct __kernel_sock_timeval tv;
427 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
428
429 if (err)
430 return err;
431
432 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
433 return -EDOM;
434
435 if (tv.tv_sec < 0) {
436 static int warned __read_mostly;
437
438 *timeo_p = 0;
439 if (warned < 10 && net_ratelimit()) {
440 warned++;
441 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
442 __func__, current->comm, task_pid_nr(current));
443 }
444 return 0;
445 }
446 *timeo_p = MAX_SCHEDULE_TIMEOUT;
447 if (tv.tv_sec == 0 && tv.tv_usec == 0)
448 return 0;
449 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
450 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
451 return 0;
452}
453
454static bool sock_needs_netstamp(const struct sock *sk)
455{
456 switch (sk->sk_family) {
457 case AF_UNSPEC:
458 case AF_UNIX:
459 return false;
460 default:
461 return true;
462 }
463}
464
465static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
466{
467 if (sk->sk_flags & flags) {
468 sk->sk_flags &= ~flags;
469 if (sock_needs_netstamp(sk) &&
470 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
471 net_disable_timestamp();
472 }
473}
474
475
476int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
477{
478 unsigned long flags;
479 struct sk_buff_head *list = &sk->sk_receive_queue;
480
481 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
482 atomic_inc(&sk->sk_drops);
483 trace_sock_rcvqueue_full(sk, skb);
484 return -ENOMEM;
485 }
486
487 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
488 atomic_inc(&sk->sk_drops);
489 return -ENOBUFS;
490 }
491
492 skb->dev = NULL;
493 skb_set_owner_r(skb, sk);
494
495 /* we escape from rcu protected region, make sure we dont leak
496 * a norefcounted dst
497 */
498 skb_dst_force(skb);
499
500 spin_lock_irqsave(&list->lock, flags);
501 sock_skb_set_dropcount(sk, skb);
502 __skb_queue_tail(list, skb);
503 spin_unlock_irqrestore(&list->lock, flags);
504
505 if (!sock_flag(sk, SOCK_DEAD))
506 sk->sk_data_ready(sk);
507 return 0;
508}
509EXPORT_SYMBOL(__sock_queue_rcv_skb);
510
511int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
512 enum skb_drop_reason *reason)
513{
514 enum skb_drop_reason drop_reason;
515 int err;
516
517 err = sk_filter(sk, skb);
518 if (err) {
519 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
520 goto out;
521 }
522 err = __sock_queue_rcv_skb(sk, skb);
523 switch (err) {
524 case -ENOMEM:
525 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
526 break;
527 case -ENOBUFS:
528 drop_reason = SKB_DROP_REASON_PROTO_MEM;
529 break;
530 default:
531 drop_reason = SKB_NOT_DROPPED_YET;
532 break;
533 }
534out:
535 if (reason)
536 *reason = drop_reason;
537 return err;
538}
539EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
540
541int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
542 const int nested, unsigned int trim_cap, bool refcounted)
543{
544 int rc = NET_RX_SUCCESS;
545
546 if (sk_filter_trim_cap(sk, skb, trim_cap))
547 goto discard_and_relse;
548
549 skb->dev = NULL;
550
551 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
552 atomic_inc(&sk->sk_drops);
553 goto discard_and_relse;
554 }
555 if (nested)
556 bh_lock_sock_nested(sk);
557 else
558 bh_lock_sock(sk);
559 if (!sock_owned_by_user(sk)) {
560 /*
561 * trylock + unlock semantics:
562 */
563 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
564
565 rc = sk_backlog_rcv(sk, skb);
566
567 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
568 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
569 bh_unlock_sock(sk);
570 atomic_inc(&sk->sk_drops);
571 goto discard_and_relse;
572 }
573
574 bh_unlock_sock(sk);
575out:
576 if (refcounted)
577 sock_put(sk);
578 return rc;
579discard_and_relse:
580 kfree_skb(skb);
581 goto out;
582}
583EXPORT_SYMBOL(__sk_receive_skb);
584
585INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
586 u32));
587INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
588 u32));
589struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
590{
591 struct dst_entry *dst = __sk_dst_get(sk);
592
593 if (dst && dst->obsolete &&
594 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
595 dst, cookie) == NULL) {
596 sk_tx_queue_clear(sk);
597 sk->sk_dst_pending_confirm = 0;
598 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
599 dst_release(dst);
600 return NULL;
601 }
602
603 return dst;
604}
605EXPORT_SYMBOL(__sk_dst_check);
606
607struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
608{
609 struct dst_entry *dst = sk_dst_get(sk);
610
611 if (dst && dst->obsolete &&
612 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
613 dst, cookie) == NULL) {
614 sk_dst_reset(sk);
615 dst_release(dst);
616 return NULL;
617 }
618
619 return dst;
620}
621EXPORT_SYMBOL(sk_dst_check);
622
623static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
624{
625 int ret = -ENOPROTOOPT;
626#ifdef CONFIG_NETDEVICES
627 struct net *net = sock_net(sk);
628
629 /* Sorry... */
630 ret = -EPERM;
631 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
632 goto out;
633
634 ret = -EINVAL;
635 if (ifindex < 0)
636 goto out;
637
638 /* Paired with all READ_ONCE() done locklessly. */
639 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
640
641 if (sk->sk_prot->rehash)
642 sk->sk_prot->rehash(sk);
643 sk_dst_reset(sk);
644
645 ret = 0;
646
647out:
648#endif
649
650 return ret;
651}
652
653int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
654{
655 int ret;
656
657 if (lock_sk)
658 lock_sock(sk);
659 ret = sock_bindtoindex_locked(sk, ifindex);
660 if (lock_sk)
661 release_sock(sk);
662
663 return ret;
664}
665EXPORT_SYMBOL(sock_bindtoindex);
666
667static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
668{
669 int ret = -ENOPROTOOPT;
670#ifdef CONFIG_NETDEVICES
671 struct net *net = sock_net(sk);
672 char devname[IFNAMSIZ];
673 int index;
674
675 ret = -EINVAL;
676 if (optlen < 0)
677 goto out;
678
679 /* Bind this socket to a particular device like "eth0",
680 * as specified in the passed interface name. If the
681 * name is "" or the option length is zero the socket
682 * is not bound.
683 */
684 if (optlen > IFNAMSIZ - 1)
685 optlen = IFNAMSIZ - 1;
686 memset(devname, 0, sizeof(devname));
687
688 ret = -EFAULT;
689 if (copy_from_sockptr(devname, optval, optlen))
690 goto out;
691
692 index = 0;
693 if (devname[0] != '\0') {
694 struct net_device *dev;
695
696 rcu_read_lock();
697 dev = dev_get_by_name_rcu(net, devname);
698 if (dev)
699 index = dev->ifindex;
700 rcu_read_unlock();
701 ret = -ENODEV;
702 if (!dev)
703 goto out;
704 }
705
706 sockopt_lock_sock(sk);
707 ret = sock_bindtoindex_locked(sk, index);
708 sockopt_release_sock(sk);
709out:
710#endif
711
712 return ret;
713}
714
715static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
716 sockptr_t optlen, int len)
717{
718 int ret = -ENOPROTOOPT;
719#ifdef CONFIG_NETDEVICES
720 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
721 struct net *net = sock_net(sk);
722 char devname[IFNAMSIZ];
723
724 if (bound_dev_if == 0) {
725 len = 0;
726 goto zero;
727 }
728
729 ret = -EINVAL;
730 if (len < IFNAMSIZ)
731 goto out;
732
733 ret = netdev_get_name(net, devname, bound_dev_if);
734 if (ret)
735 goto out;
736
737 len = strlen(devname) + 1;
738
739 ret = -EFAULT;
740 if (copy_to_sockptr(optval, devname, len))
741 goto out;
742
743zero:
744 ret = -EFAULT;
745 if (copy_to_sockptr(optlen, &len, sizeof(int)))
746 goto out;
747
748 ret = 0;
749
750out:
751#endif
752
753 return ret;
754}
755
756bool sk_mc_loop(struct sock *sk)
757{
758 if (dev_recursion_level())
759 return false;
760 if (!sk)
761 return true;
762 switch (sk->sk_family) {
763 case AF_INET:
764 return inet_sk(sk)->mc_loop;
765#if IS_ENABLED(CONFIG_IPV6)
766 case AF_INET6:
767 return inet6_sk(sk)->mc_loop;
768#endif
769 }
770 WARN_ON_ONCE(1);
771 return true;
772}
773EXPORT_SYMBOL(sk_mc_loop);
774
775void sock_set_reuseaddr(struct sock *sk)
776{
777 lock_sock(sk);
778 sk->sk_reuse = SK_CAN_REUSE;
779 release_sock(sk);
780}
781EXPORT_SYMBOL(sock_set_reuseaddr);
782
783void sock_set_reuseport(struct sock *sk)
784{
785 lock_sock(sk);
786 sk->sk_reuseport = true;
787 release_sock(sk);
788}
789EXPORT_SYMBOL(sock_set_reuseport);
790
791void sock_no_linger(struct sock *sk)
792{
793 lock_sock(sk);
794 sk->sk_lingertime = 0;
795 sock_set_flag(sk, SOCK_LINGER);
796 release_sock(sk);
797}
798EXPORT_SYMBOL(sock_no_linger);
799
800void sock_set_priority(struct sock *sk, u32 priority)
801{
802 lock_sock(sk);
803 sk->sk_priority = priority;
804 release_sock(sk);
805}
806EXPORT_SYMBOL(sock_set_priority);
807
808void sock_set_sndtimeo(struct sock *sk, s64 secs)
809{
810 lock_sock(sk);
811 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
812 sk->sk_sndtimeo = secs * HZ;
813 else
814 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
815 release_sock(sk);
816}
817EXPORT_SYMBOL(sock_set_sndtimeo);
818
819static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
820{
821 if (val) {
822 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
823 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
824 sock_set_flag(sk, SOCK_RCVTSTAMP);
825 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
826 } else {
827 sock_reset_flag(sk, SOCK_RCVTSTAMP);
828 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
829 }
830}
831
832void sock_enable_timestamps(struct sock *sk)
833{
834 lock_sock(sk);
835 __sock_set_timestamps(sk, true, false, true);
836 release_sock(sk);
837}
838EXPORT_SYMBOL(sock_enable_timestamps);
839
840void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
841{
842 switch (optname) {
843 case SO_TIMESTAMP_OLD:
844 __sock_set_timestamps(sk, valbool, false, false);
845 break;
846 case SO_TIMESTAMP_NEW:
847 __sock_set_timestamps(sk, valbool, true, false);
848 break;
849 case SO_TIMESTAMPNS_OLD:
850 __sock_set_timestamps(sk, valbool, false, true);
851 break;
852 case SO_TIMESTAMPNS_NEW:
853 __sock_set_timestamps(sk, valbool, true, true);
854 break;
855 }
856}
857
858static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
859{
860 struct net *net = sock_net(sk);
861 struct net_device *dev = NULL;
862 bool match = false;
863 int *vclock_index;
864 int i, num;
865
866 if (sk->sk_bound_dev_if)
867 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
868
869 if (!dev) {
870 pr_err("%s: sock not bind to device\n", __func__);
871 return -EOPNOTSUPP;
872 }
873
874 num = ethtool_get_phc_vclocks(dev, &vclock_index);
875 dev_put(dev);
876
877 for (i = 0; i < num; i++) {
878 if (*(vclock_index + i) == phc_index) {
879 match = true;
880 break;
881 }
882 }
883
884 if (num > 0)
885 kfree(vclock_index);
886
887 if (!match)
888 return -EINVAL;
889
890 sk->sk_bind_phc = phc_index;
891
892 return 0;
893}
894
895int sock_set_timestamping(struct sock *sk, int optname,
896 struct so_timestamping timestamping)
897{
898 int val = timestamping.flags;
899 int ret;
900
901 if (val & ~SOF_TIMESTAMPING_MASK)
902 return -EINVAL;
903
904 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
905 !(val & SOF_TIMESTAMPING_OPT_ID))
906 return -EINVAL;
907
908 if (val & SOF_TIMESTAMPING_OPT_ID &&
909 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
910 if (sk_is_tcp(sk)) {
911 if ((1 << sk->sk_state) &
912 (TCPF_CLOSE | TCPF_LISTEN))
913 return -EINVAL;
914 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
915 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
916 else
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
918 } else {
919 atomic_set(&sk->sk_tskey, 0);
920 }
921 }
922
923 if (val & SOF_TIMESTAMPING_OPT_STATS &&
924 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
925 return -EINVAL;
926
927 if (val & SOF_TIMESTAMPING_BIND_PHC) {
928 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
929 if (ret)
930 return ret;
931 }
932
933 sk->sk_tsflags = val;
934 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
935
936 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
937 sock_enable_timestamp(sk,
938 SOCK_TIMESTAMPING_RX_SOFTWARE);
939 else
940 sock_disable_timestamp(sk,
941 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
942 return 0;
943}
944
945void sock_set_keepalive(struct sock *sk)
946{
947 lock_sock(sk);
948 if (sk->sk_prot->keepalive)
949 sk->sk_prot->keepalive(sk, true);
950 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
951 release_sock(sk);
952}
953EXPORT_SYMBOL(sock_set_keepalive);
954
955static void __sock_set_rcvbuf(struct sock *sk, int val)
956{
957 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
958 * as a negative value.
959 */
960 val = min_t(int, val, INT_MAX / 2);
961 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
962
963 /* We double it on the way in to account for "struct sk_buff" etc.
964 * overhead. Applications assume that the SO_RCVBUF setting they make
965 * will allow that much actual data to be received on that socket.
966 *
967 * Applications are unaware that "struct sk_buff" and other overheads
968 * allocate from the receive buffer during socket buffer allocation.
969 *
970 * And after considering the possible alternatives, returning the value
971 * we actually used in getsockopt is the most desirable behavior.
972 */
973 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
974}
975
976void sock_set_rcvbuf(struct sock *sk, int val)
977{
978 lock_sock(sk);
979 __sock_set_rcvbuf(sk, val);
980 release_sock(sk);
981}
982EXPORT_SYMBOL(sock_set_rcvbuf);
983
984static void __sock_set_mark(struct sock *sk, u32 val)
985{
986 if (val != sk->sk_mark) {
987 sk->sk_mark = val;
988 sk_dst_reset(sk);
989 }
990}
991
992void sock_set_mark(struct sock *sk, u32 val)
993{
994 lock_sock(sk);
995 __sock_set_mark(sk, val);
996 release_sock(sk);
997}
998EXPORT_SYMBOL(sock_set_mark);
999
1000static void sock_release_reserved_memory(struct sock *sk, int bytes)
1001{
1002 /* Round down bytes to multiple of pages */
1003 bytes = round_down(bytes, PAGE_SIZE);
1004
1005 WARN_ON(bytes > sk->sk_reserved_mem);
1006 sk->sk_reserved_mem -= bytes;
1007 sk_mem_reclaim(sk);
1008}
1009
1010static int sock_reserve_memory(struct sock *sk, int bytes)
1011{
1012 long allocated;
1013 bool charged;
1014 int pages;
1015
1016 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1017 return -EOPNOTSUPP;
1018
1019 if (!bytes)
1020 return 0;
1021
1022 pages = sk_mem_pages(bytes);
1023
1024 /* pre-charge to memcg */
1025 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1026 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1027 if (!charged)
1028 return -ENOMEM;
1029
1030 /* pre-charge to forward_alloc */
1031 sk_memory_allocated_add(sk, pages);
1032 allocated = sk_memory_allocated(sk);
1033 /* If the system goes into memory pressure with this
1034 * precharge, give up and return error.
1035 */
1036 if (allocated > sk_prot_mem_limits(sk, 1)) {
1037 sk_memory_allocated_sub(sk, pages);
1038 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1039 return -ENOMEM;
1040 }
1041 sk->sk_forward_alloc += pages << PAGE_SHIFT;
1042
1043 sk->sk_reserved_mem += pages << PAGE_SHIFT;
1044
1045 return 0;
1046}
1047
1048void sockopt_lock_sock(struct sock *sk)
1049{
1050 /* When current->bpf_ctx is set, the setsockopt is called from
1051 * a bpf prog. bpf has ensured the sk lock has been
1052 * acquired before calling setsockopt().
1053 */
1054 if (has_current_bpf_ctx())
1055 return;
1056
1057 lock_sock(sk);
1058}
1059EXPORT_SYMBOL(sockopt_lock_sock);
1060
1061void sockopt_release_sock(struct sock *sk)
1062{
1063 if (has_current_bpf_ctx())
1064 return;
1065
1066 release_sock(sk);
1067}
1068EXPORT_SYMBOL(sockopt_release_sock);
1069
1070bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1071{
1072 return has_current_bpf_ctx() || ns_capable(ns, cap);
1073}
1074EXPORT_SYMBOL(sockopt_ns_capable);
1075
1076bool sockopt_capable(int cap)
1077{
1078 return has_current_bpf_ctx() || capable(cap);
1079}
1080EXPORT_SYMBOL(sockopt_capable);
1081
1082/*
1083 * This is meant for all protocols to use and covers goings on
1084 * at the socket level. Everything here is generic.
1085 */
1086
1087int sk_setsockopt(struct sock *sk, int level, int optname,
1088 sockptr_t optval, unsigned int optlen)
1089{
1090 struct so_timestamping timestamping;
1091 struct socket *sock = sk->sk_socket;
1092 struct sock_txtime sk_txtime;
1093 int val;
1094 int valbool;
1095 struct linger ling;
1096 int ret = 0;
1097
1098 /*
1099 * Options without arguments
1100 */
1101
1102 if (optname == SO_BINDTODEVICE)
1103 return sock_setbindtodevice(sk, optval, optlen);
1104
1105 if (optlen < sizeof(int))
1106 return -EINVAL;
1107
1108 if (copy_from_sockptr(&val, optval, sizeof(val)))
1109 return -EFAULT;
1110
1111 valbool = val ? 1 : 0;
1112
1113 sockopt_lock_sock(sk);
1114
1115 switch (optname) {
1116 case SO_DEBUG:
1117 if (val && !sockopt_capable(CAP_NET_ADMIN))
1118 ret = -EACCES;
1119 else
1120 sock_valbool_flag(sk, SOCK_DBG, valbool);
1121 break;
1122 case SO_REUSEADDR:
1123 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1124 break;
1125 case SO_REUSEPORT:
1126 sk->sk_reuseport = valbool;
1127 break;
1128 case SO_TYPE:
1129 case SO_PROTOCOL:
1130 case SO_DOMAIN:
1131 case SO_ERROR:
1132 ret = -ENOPROTOOPT;
1133 break;
1134 case SO_DONTROUTE:
1135 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1136 sk_dst_reset(sk);
1137 break;
1138 case SO_BROADCAST:
1139 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1140 break;
1141 case SO_SNDBUF:
1142 /* Don't error on this BSD doesn't and if you think
1143 * about it this is right. Otherwise apps have to
1144 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1145 * are treated in BSD as hints
1146 */
1147 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1148set_sndbuf:
1149 /* Ensure val * 2 fits into an int, to prevent max_t()
1150 * from treating it as a negative value.
1151 */
1152 val = min_t(int, val, INT_MAX / 2);
1153 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1154 WRITE_ONCE(sk->sk_sndbuf,
1155 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1156 /* Wake up sending tasks if we upped the value. */
1157 sk->sk_write_space(sk);
1158 break;
1159
1160 case SO_SNDBUFFORCE:
1161 if (!sockopt_capable(CAP_NET_ADMIN)) {
1162 ret = -EPERM;
1163 break;
1164 }
1165
1166 /* No negative values (to prevent underflow, as val will be
1167 * multiplied by 2).
1168 */
1169 if (val < 0)
1170 val = 0;
1171 goto set_sndbuf;
1172
1173 case SO_RCVBUF:
1174 /* Don't error on this BSD doesn't and if you think
1175 * about it this is right. Otherwise apps have to
1176 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1177 * are treated in BSD as hints
1178 */
1179 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1180 break;
1181
1182 case SO_RCVBUFFORCE:
1183 if (!sockopt_capable(CAP_NET_ADMIN)) {
1184 ret = -EPERM;
1185 break;
1186 }
1187
1188 /* No negative values (to prevent underflow, as val will be
1189 * multiplied by 2).
1190 */
1191 __sock_set_rcvbuf(sk, max(val, 0));
1192 break;
1193
1194 case SO_KEEPALIVE:
1195 if (sk->sk_prot->keepalive)
1196 sk->sk_prot->keepalive(sk, valbool);
1197 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1198 break;
1199
1200 case SO_OOBINLINE:
1201 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1202 break;
1203
1204 case SO_NO_CHECK:
1205 sk->sk_no_check_tx = valbool;
1206 break;
1207
1208 case SO_PRIORITY:
1209 if ((val >= 0 && val <= 6) ||
1210 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1211 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1212 sk->sk_priority = val;
1213 else
1214 ret = -EPERM;
1215 break;
1216
1217 case SO_LINGER:
1218 if (optlen < sizeof(ling)) {
1219 ret = -EINVAL; /* 1003.1g */
1220 break;
1221 }
1222 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1223 ret = -EFAULT;
1224 break;
1225 }
1226 if (!ling.l_onoff)
1227 sock_reset_flag(sk, SOCK_LINGER);
1228 else {
1229#if (BITS_PER_LONG == 32)
1230 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
1231 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1232 else
1233#endif
1234 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
1235 sock_set_flag(sk, SOCK_LINGER);
1236 }
1237 break;
1238
1239 case SO_BSDCOMPAT:
1240 break;
1241
1242 case SO_PASSCRED:
1243 if (valbool)
1244 set_bit(SOCK_PASSCRED, &sock->flags);
1245 else
1246 clear_bit(SOCK_PASSCRED, &sock->flags);
1247 break;
1248
1249 case SO_TIMESTAMP_OLD:
1250 case SO_TIMESTAMP_NEW:
1251 case SO_TIMESTAMPNS_OLD:
1252 case SO_TIMESTAMPNS_NEW:
1253 sock_set_timestamp(sk, optname, valbool);
1254 break;
1255
1256 case SO_TIMESTAMPING_NEW:
1257 case SO_TIMESTAMPING_OLD:
1258 if (optlen == sizeof(timestamping)) {
1259 if (copy_from_sockptr(×tamping, optval,
1260 sizeof(timestamping))) {
1261 ret = -EFAULT;
1262 break;
1263 }
1264 } else {
1265 memset(×tamping, 0, sizeof(timestamping));
1266 timestamping.flags = val;
1267 }
1268 ret = sock_set_timestamping(sk, optname, timestamping);
1269 break;
1270
1271 case SO_RCVLOWAT:
1272 if (val < 0)
1273 val = INT_MAX;
1274 if (sock && sock->ops->set_rcvlowat)
1275 ret = sock->ops->set_rcvlowat(sk, val);
1276 else
1277 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1278 break;
1279
1280 case SO_RCVTIMEO_OLD:
1281 case SO_RCVTIMEO_NEW:
1282 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1283 optlen, optname == SO_RCVTIMEO_OLD);
1284 break;
1285
1286 case SO_SNDTIMEO_OLD:
1287 case SO_SNDTIMEO_NEW:
1288 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1289 optlen, optname == SO_SNDTIMEO_OLD);
1290 break;
1291
1292 case SO_ATTACH_FILTER: {
1293 struct sock_fprog fprog;
1294
1295 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1296 if (!ret)
1297 ret = sk_attach_filter(&fprog, sk);
1298 break;
1299 }
1300 case SO_ATTACH_BPF:
1301 ret = -EINVAL;
1302 if (optlen == sizeof(u32)) {
1303 u32 ufd;
1304
1305 ret = -EFAULT;
1306 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1307 break;
1308
1309 ret = sk_attach_bpf(ufd, sk);
1310 }
1311 break;
1312
1313 case SO_ATTACH_REUSEPORT_CBPF: {
1314 struct sock_fprog fprog;
1315
1316 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1317 if (!ret)
1318 ret = sk_reuseport_attach_filter(&fprog, sk);
1319 break;
1320 }
1321 case SO_ATTACH_REUSEPORT_EBPF:
1322 ret = -EINVAL;
1323 if (optlen == sizeof(u32)) {
1324 u32 ufd;
1325
1326 ret = -EFAULT;
1327 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1328 break;
1329
1330 ret = sk_reuseport_attach_bpf(ufd, sk);
1331 }
1332 break;
1333
1334 case SO_DETACH_REUSEPORT_BPF:
1335 ret = reuseport_detach_prog(sk);
1336 break;
1337
1338 case SO_DETACH_FILTER:
1339 ret = sk_detach_filter(sk);
1340 break;
1341
1342 case SO_LOCK_FILTER:
1343 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1344 ret = -EPERM;
1345 else
1346 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1347 break;
1348
1349 case SO_PASSSEC:
1350 if (valbool)
1351 set_bit(SOCK_PASSSEC, &sock->flags);
1352 else
1353 clear_bit(SOCK_PASSSEC, &sock->flags);
1354 break;
1355 case SO_MARK:
1356 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1357 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1358 ret = -EPERM;
1359 break;
1360 }
1361
1362 __sock_set_mark(sk, val);
1363 break;
1364 case SO_RCVMARK:
1365 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1366 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1367 ret = -EPERM;
1368 break;
1369 }
1370
1371 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1372 break;
1373
1374 case SO_RXQ_OVFL:
1375 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1376 break;
1377
1378 case SO_WIFI_STATUS:
1379 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1380 break;
1381
1382 case SO_PEEK_OFF:
1383 if (sock->ops->set_peek_off)
1384 ret = sock->ops->set_peek_off(sk, val);
1385 else
1386 ret = -EOPNOTSUPP;
1387 break;
1388
1389 case SO_NOFCS:
1390 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1391 break;
1392
1393 case SO_SELECT_ERR_QUEUE:
1394 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1395 break;
1396
1397#ifdef CONFIG_NET_RX_BUSY_POLL
1398 case SO_BUSY_POLL:
1399 /* allow unprivileged users to decrease the value */
1400 if ((val > sk->sk_ll_usec) && !sockopt_capable(CAP_NET_ADMIN))
1401 ret = -EPERM;
1402 else {
1403 if (val < 0)
1404 ret = -EINVAL;
1405 else
1406 WRITE_ONCE(sk->sk_ll_usec, val);
1407 }
1408 break;
1409 case SO_PREFER_BUSY_POLL:
1410 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1411 ret = -EPERM;
1412 else
1413 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1414 break;
1415 case SO_BUSY_POLL_BUDGET:
1416 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) {
1417 ret = -EPERM;
1418 } else {
1419 if (val < 0 || val > U16_MAX)
1420 ret = -EINVAL;
1421 else
1422 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1423 }
1424 break;
1425#endif
1426
1427 case SO_MAX_PACING_RATE:
1428 {
1429 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1430
1431 if (sizeof(ulval) != sizeof(val) &&
1432 optlen >= sizeof(ulval) &&
1433 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1434 ret = -EFAULT;
1435 break;
1436 }
1437 if (ulval != ~0UL)
1438 cmpxchg(&sk->sk_pacing_status,
1439 SK_PACING_NONE,
1440 SK_PACING_NEEDED);
1441 sk->sk_max_pacing_rate = ulval;
1442 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1443 break;
1444 }
1445 case SO_INCOMING_CPU:
1446 reuseport_update_incoming_cpu(sk, val);
1447 break;
1448
1449 case SO_CNX_ADVICE:
1450 if (val == 1)
1451 dst_negative_advice(sk);
1452 break;
1453
1454 case SO_ZEROCOPY:
1455 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1456 if (!(sk_is_tcp(sk) ||
1457 (sk->sk_type == SOCK_DGRAM &&
1458 sk->sk_protocol == IPPROTO_UDP)))
1459 ret = -EOPNOTSUPP;
1460 } else if (sk->sk_family != PF_RDS) {
1461 ret = -EOPNOTSUPP;
1462 }
1463 if (!ret) {
1464 if (val < 0 || val > 1)
1465 ret = -EINVAL;
1466 else
1467 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1468 }
1469 break;
1470
1471 case SO_TXTIME:
1472 if (optlen != sizeof(struct sock_txtime)) {
1473 ret = -EINVAL;
1474 break;
1475 } else if (copy_from_sockptr(&sk_txtime, optval,
1476 sizeof(struct sock_txtime))) {
1477 ret = -EFAULT;
1478 break;
1479 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1480 ret = -EINVAL;
1481 break;
1482 }
1483 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1484 * scheduler has enough safe guards.
1485 */
1486 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1487 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1488 ret = -EPERM;
1489 break;
1490 }
1491 sock_valbool_flag(sk, SOCK_TXTIME, true);
1492 sk->sk_clockid = sk_txtime.clockid;
1493 sk->sk_txtime_deadline_mode =
1494 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1495 sk->sk_txtime_report_errors =
1496 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1497 break;
1498
1499 case SO_BINDTOIFINDEX:
1500 ret = sock_bindtoindex_locked(sk, val);
1501 break;
1502
1503 case SO_BUF_LOCK:
1504 if (val & ~SOCK_BUF_LOCK_MASK) {
1505 ret = -EINVAL;
1506 break;
1507 }
1508 sk->sk_userlocks = val | (sk->sk_userlocks &
1509 ~SOCK_BUF_LOCK_MASK);
1510 break;
1511
1512 case SO_RESERVE_MEM:
1513 {
1514 int delta;
1515
1516 if (val < 0) {
1517 ret = -EINVAL;
1518 break;
1519 }
1520
1521 delta = val - sk->sk_reserved_mem;
1522 if (delta < 0)
1523 sock_release_reserved_memory(sk, -delta);
1524 else
1525 ret = sock_reserve_memory(sk, delta);
1526 break;
1527 }
1528
1529 case SO_TXREHASH:
1530 if (val < -1 || val > 1) {
1531 ret = -EINVAL;
1532 break;
1533 }
1534 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1535 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1536 /* Paired with READ_ONCE() in tcp_rtx_synack() */
1537 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1538 break;
1539
1540 default:
1541 ret = -ENOPROTOOPT;
1542 break;
1543 }
1544 sockopt_release_sock(sk);
1545 return ret;
1546}
1547
1548int sock_setsockopt(struct socket *sock, int level, int optname,
1549 sockptr_t optval, unsigned int optlen)
1550{
1551 return sk_setsockopt(sock->sk, level, optname,
1552 optval, optlen);
1553}
1554EXPORT_SYMBOL(sock_setsockopt);
1555
1556static const struct cred *sk_get_peer_cred(struct sock *sk)
1557{
1558 const struct cred *cred;
1559
1560 spin_lock(&sk->sk_peer_lock);
1561 cred = get_cred(sk->sk_peer_cred);
1562 spin_unlock(&sk->sk_peer_lock);
1563
1564 return cred;
1565}
1566
1567static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1568 struct ucred *ucred)
1569{
1570 ucred->pid = pid_vnr(pid);
1571 ucred->uid = ucred->gid = -1;
1572 if (cred) {
1573 struct user_namespace *current_ns = current_user_ns();
1574
1575 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1576 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1577 }
1578}
1579
1580static int groups_to_user(sockptr_t dst, const struct group_info *src)
1581{
1582 struct user_namespace *user_ns = current_user_ns();
1583 int i;
1584
1585 for (i = 0; i < src->ngroups; i++) {
1586 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1587
1588 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1589 return -EFAULT;
1590 }
1591
1592 return 0;
1593}
1594
1595int sk_getsockopt(struct sock *sk, int level, int optname,
1596 sockptr_t optval, sockptr_t optlen)
1597{
1598 struct socket *sock = sk->sk_socket;
1599
1600 union {
1601 int val;
1602 u64 val64;
1603 unsigned long ulval;
1604 struct linger ling;
1605 struct old_timeval32 tm32;
1606 struct __kernel_old_timeval tm;
1607 struct __kernel_sock_timeval stm;
1608 struct sock_txtime txtime;
1609 struct so_timestamping timestamping;
1610 } v;
1611
1612 int lv = sizeof(int);
1613 int len;
1614
1615 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1616 return -EFAULT;
1617 if (len < 0)
1618 return -EINVAL;
1619
1620 memset(&v, 0, sizeof(v));
1621
1622 switch (optname) {
1623 case SO_DEBUG:
1624 v.val = sock_flag(sk, SOCK_DBG);
1625 break;
1626
1627 case SO_DONTROUTE:
1628 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1629 break;
1630
1631 case SO_BROADCAST:
1632 v.val = sock_flag(sk, SOCK_BROADCAST);
1633 break;
1634
1635 case SO_SNDBUF:
1636 v.val = sk->sk_sndbuf;
1637 break;
1638
1639 case SO_RCVBUF:
1640 v.val = sk->sk_rcvbuf;
1641 break;
1642
1643 case SO_REUSEADDR:
1644 v.val = sk->sk_reuse;
1645 break;
1646
1647 case SO_REUSEPORT:
1648 v.val = sk->sk_reuseport;
1649 break;
1650
1651 case SO_KEEPALIVE:
1652 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1653 break;
1654
1655 case SO_TYPE:
1656 v.val = sk->sk_type;
1657 break;
1658
1659 case SO_PROTOCOL:
1660 v.val = sk->sk_protocol;
1661 break;
1662
1663 case SO_DOMAIN:
1664 v.val = sk->sk_family;
1665 break;
1666
1667 case SO_ERROR:
1668 v.val = -sock_error(sk);
1669 if (v.val == 0)
1670 v.val = xchg(&sk->sk_err_soft, 0);
1671 break;
1672
1673 case SO_OOBINLINE:
1674 v.val = sock_flag(sk, SOCK_URGINLINE);
1675 break;
1676
1677 case SO_NO_CHECK:
1678 v.val = sk->sk_no_check_tx;
1679 break;
1680
1681 case SO_PRIORITY:
1682 v.val = sk->sk_priority;
1683 break;
1684
1685 case SO_LINGER:
1686 lv = sizeof(v.ling);
1687 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1688 v.ling.l_linger = sk->sk_lingertime / HZ;
1689 break;
1690
1691 case SO_BSDCOMPAT:
1692 break;
1693
1694 case SO_TIMESTAMP_OLD:
1695 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1696 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1697 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1698 break;
1699
1700 case SO_TIMESTAMPNS_OLD:
1701 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1702 break;
1703
1704 case SO_TIMESTAMP_NEW:
1705 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1706 break;
1707
1708 case SO_TIMESTAMPNS_NEW:
1709 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1710 break;
1711
1712 case SO_TIMESTAMPING_OLD:
1713 lv = sizeof(v.timestamping);
1714 v.timestamping.flags = sk->sk_tsflags;
1715 v.timestamping.bind_phc = sk->sk_bind_phc;
1716 break;
1717
1718 case SO_RCVTIMEO_OLD:
1719 case SO_RCVTIMEO_NEW:
1720 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1721 break;
1722
1723 case SO_SNDTIMEO_OLD:
1724 case SO_SNDTIMEO_NEW:
1725 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1726 break;
1727
1728 case SO_RCVLOWAT:
1729 v.val = sk->sk_rcvlowat;
1730 break;
1731
1732 case SO_SNDLOWAT:
1733 v.val = 1;
1734 break;
1735
1736 case SO_PASSCRED:
1737 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1738 break;
1739
1740 case SO_PEERCRED:
1741 {
1742 struct ucred peercred;
1743 if (len > sizeof(peercred))
1744 len = sizeof(peercred);
1745
1746 spin_lock(&sk->sk_peer_lock);
1747 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1748 spin_unlock(&sk->sk_peer_lock);
1749
1750 if (copy_to_sockptr(optval, &peercred, len))
1751 return -EFAULT;
1752 goto lenout;
1753 }
1754
1755 case SO_PEERGROUPS:
1756 {
1757 const struct cred *cred;
1758 int ret, n;
1759
1760 cred = sk_get_peer_cred(sk);
1761 if (!cred)
1762 return -ENODATA;
1763
1764 n = cred->group_info->ngroups;
1765 if (len < n * sizeof(gid_t)) {
1766 len = n * sizeof(gid_t);
1767 put_cred(cred);
1768 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1769 }
1770 len = n * sizeof(gid_t);
1771
1772 ret = groups_to_user(optval, cred->group_info);
1773 put_cred(cred);
1774 if (ret)
1775 return ret;
1776 goto lenout;
1777 }
1778
1779 case SO_PEERNAME:
1780 {
1781 char address[128];
1782
1783 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1784 if (lv < 0)
1785 return -ENOTCONN;
1786 if (lv < len)
1787 return -EINVAL;
1788 if (copy_to_sockptr(optval, address, len))
1789 return -EFAULT;
1790 goto lenout;
1791 }
1792
1793 /* Dubious BSD thing... Probably nobody even uses it, but
1794 * the UNIX standard wants it for whatever reason... -DaveM
1795 */
1796 case SO_ACCEPTCONN:
1797 v.val = sk->sk_state == TCP_LISTEN;
1798 break;
1799
1800 case SO_PASSSEC:
1801 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1802 break;
1803
1804 case SO_PEERSEC:
1805 return security_socket_getpeersec_stream(sock,
1806 optval, optlen, len);
1807
1808 case SO_MARK:
1809 v.val = sk->sk_mark;
1810 break;
1811
1812 case SO_RCVMARK:
1813 v.val = sock_flag(sk, SOCK_RCVMARK);
1814 break;
1815
1816 case SO_RXQ_OVFL:
1817 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1818 break;
1819
1820 case SO_WIFI_STATUS:
1821 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1822 break;
1823
1824 case SO_PEEK_OFF:
1825 if (!sock->ops->set_peek_off)
1826 return -EOPNOTSUPP;
1827
1828 v.val = sk->sk_peek_off;
1829 break;
1830 case SO_NOFCS:
1831 v.val = sock_flag(sk, SOCK_NOFCS);
1832 break;
1833
1834 case SO_BINDTODEVICE:
1835 return sock_getbindtodevice(sk, optval, optlen, len);
1836
1837 case SO_GET_FILTER:
1838 len = sk_get_filter(sk, optval, len);
1839 if (len < 0)
1840 return len;
1841
1842 goto lenout;
1843
1844 case SO_LOCK_FILTER:
1845 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1846 break;
1847
1848 case SO_BPF_EXTENSIONS:
1849 v.val = bpf_tell_extensions();
1850 break;
1851
1852 case SO_SELECT_ERR_QUEUE:
1853 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1854 break;
1855
1856#ifdef CONFIG_NET_RX_BUSY_POLL
1857 case SO_BUSY_POLL:
1858 v.val = sk->sk_ll_usec;
1859 break;
1860 case SO_PREFER_BUSY_POLL:
1861 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1862 break;
1863#endif
1864
1865 case SO_MAX_PACING_RATE:
1866 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1867 lv = sizeof(v.ulval);
1868 v.ulval = sk->sk_max_pacing_rate;
1869 } else {
1870 /* 32bit version */
1871 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1872 }
1873 break;
1874
1875 case SO_INCOMING_CPU:
1876 v.val = READ_ONCE(sk->sk_incoming_cpu);
1877 break;
1878
1879 case SO_MEMINFO:
1880 {
1881 u32 meminfo[SK_MEMINFO_VARS];
1882
1883 sk_get_meminfo(sk, meminfo);
1884
1885 len = min_t(unsigned int, len, sizeof(meminfo));
1886 if (copy_to_sockptr(optval, &meminfo, len))
1887 return -EFAULT;
1888
1889 goto lenout;
1890 }
1891
1892#ifdef CONFIG_NET_RX_BUSY_POLL
1893 case SO_INCOMING_NAPI_ID:
1894 v.val = READ_ONCE(sk->sk_napi_id);
1895
1896 /* aggregate non-NAPI IDs down to 0 */
1897 if (v.val < MIN_NAPI_ID)
1898 v.val = 0;
1899
1900 break;
1901#endif
1902
1903 case SO_COOKIE:
1904 lv = sizeof(u64);
1905 if (len < lv)
1906 return -EINVAL;
1907 v.val64 = sock_gen_cookie(sk);
1908 break;
1909
1910 case SO_ZEROCOPY:
1911 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1912 break;
1913
1914 case SO_TXTIME:
1915 lv = sizeof(v.txtime);
1916 v.txtime.clockid = sk->sk_clockid;
1917 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1918 SOF_TXTIME_DEADLINE_MODE : 0;
1919 v.txtime.flags |= sk->sk_txtime_report_errors ?
1920 SOF_TXTIME_REPORT_ERRORS : 0;
1921 break;
1922
1923 case SO_BINDTOIFINDEX:
1924 v.val = READ_ONCE(sk->sk_bound_dev_if);
1925 break;
1926
1927 case SO_NETNS_COOKIE:
1928 lv = sizeof(u64);
1929 if (len != lv)
1930 return -EINVAL;
1931 v.val64 = sock_net(sk)->net_cookie;
1932 break;
1933
1934 case SO_BUF_LOCK:
1935 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1936 break;
1937
1938 case SO_RESERVE_MEM:
1939 v.val = sk->sk_reserved_mem;
1940 break;
1941
1942 case SO_TXREHASH:
1943 v.val = sk->sk_txrehash;
1944 break;
1945
1946 default:
1947 /* We implement the SO_SNDLOWAT etc to not be settable
1948 * (1003.1g 7).
1949 */
1950 return -ENOPROTOOPT;
1951 }
1952
1953 if (len > lv)
1954 len = lv;
1955 if (copy_to_sockptr(optval, &v, len))
1956 return -EFAULT;
1957lenout:
1958 if (copy_to_sockptr(optlen, &len, sizeof(int)))
1959 return -EFAULT;
1960 return 0;
1961}
1962
1963int sock_getsockopt(struct socket *sock, int level, int optname,
1964 char __user *optval, int __user *optlen)
1965{
1966 return sk_getsockopt(sock->sk, level, optname,
1967 USER_SOCKPTR(optval),
1968 USER_SOCKPTR(optlen));
1969}
1970
1971/*
1972 * Initialize an sk_lock.
1973 *
1974 * (We also register the sk_lock with the lock validator.)
1975 */
1976static inline void sock_lock_init(struct sock *sk)
1977{
1978 if (sk->sk_kern_sock)
1979 sock_lock_init_class_and_name(
1980 sk,
1981 af_family_kern_slock_key_strings[sk->sk_family],
1982 af_family_kern_slock_keys + sk->sk_family,
1983 af_family_kern_key_strings[sk->sk_family],
1984 af_family_kern_keys + sk->sk_family);
1985 else
1986 sock_lock_init_class_and_name(
1987 sk,
1988 af_family_slock_key_strings[sk->sk_family],
1989 af_family_slock_keys + sk->sk_family,
1990 af_family_key_strings[sk->sk_family],
1991 af_family_keys + sk->sk_family);
1992}
1993
1994/*
1995 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1996 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1997 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1998 */
1999static void sock_copy(struct sock *nsk, const struct sock *osk)
2000{
2001 const struct proto *prot = READ_ONCE(osk->sk_prot);
2002#ifdef CONFIG_SECURITY_NETWORK
2003 void *sptr = nsk->sk_security;
2004#endif
2005
2006 /* If we move sk_tx_queue_mapping out of the private section,
2007 * we must check if sk_tx_queue_clear() is called after
2008 * sock_copy() in sk_clone_lock().
2009 */
2010 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2011 offsetof(struct sock, sk_dontcopy_begin) ||
2012 offsetof(struct sock, sk_tx_queue_mapping) >=
2013 offsetof(struct sock, sk_dontcopy_end));
2014
2015 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2016
2017 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2018 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2019
2020#ifdef CONFIG_SECURITY_NETWORK
2021 nsk->sk_security = sptr;
2022 security_sk_clone(osk, nsk);
2023#endif
2024}
2025
2026static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2027 int family)
2028{
2029 struct sock *sk;
2030 struct kmem_cache *slab;
2031
2032 slab = prot->slab;
2033 if (slab != NULL) {
2034 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2035 if (!sk)
2036 return sk;
2037 if (want_init_on_alloc(priority))
2038 sk_prot_clear_nulls(sk, prot->obj_size);
2039 } else
2040 sk = kmalloc(prot->obj_size, priority);
2041
2042 if (sk != NULL) {
2043 if (security_sk_alloc(sk, family, priority))
2044 goto out_free;
2045
2046 if (!try_module_get(prot->owner))
2047 goto out_free_sec;
2048 }
2049
2050 return sk;
2051
2052out_free_sec:
2053 security_sk_free(sk);
2054out_free:
2055 if (slab != NULL)
2056 kmem_cache_free(slab, sk);
2057 else
2058 kfree(sk);
2059 return NULL;
2060}
2061
2062static void sk_prot_free(struct proto *prot, struct sock *sk)
2063{
2064 struct kmem_cache *slab;
2065 struct module *owner;
2066
2067 owner = prot->owner;
2068 slab = prot->slab;
2069
2070 cgroup_sk_free(&sk->sk_cgrp_data);
2071 mem_cgroup_sk_free(sk);
2072 security_sk_free(sk);
2073 if (slab != NULL)
2074 kmem_cache_free(slab, sk);
2075 else
2076 kfree(sk);
2077 module_put(owner);
2078}
2079
2080/**
2081 * sk_alloc - All socket objects are allocated here
2082 * @net: the applicable net namespace
2083 * @family: protocol family
2084 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2085 * @prot: struct proto associated with this new sock instance
2086 * @kern: is this to be a kernel socket?
2087 */
2088struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2089 struct proto *prot, int kern)
2090{
2091 struct sock *sk;
2092
2093 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2094 if (sk) {
2095 sk->sk_family = family;
2096 /*
2097 * See comment in struct sock definition to understand
2098 * why we need sk_prot_creator -acme
2099 */
2100 sk->sk_prot = sk->sk_prot_creator = prot;
2101 sk->sk_kern_sock = kern;
2102 sock_lock_init(sk);
2103 sk->sk_net_refcnt = kern ? 0 : 1;
2104 if (likely(sk->sk_net_refcnt)) {
2105 get_net_track(net, &sk->ns_tracker, priority);
2106 sock_inuse_add(net, 1);
2107 } else {
2108 __netns_tracker_alloc(net, &sk->ns_tracker,
2109 false, priority);
2110 }
2111
2112 sock_net_set(sk, net);
2113 refcount_set(&sk->sk_wmem_alloc, 1);
2114
2115 mem_cgroup_sk_alloc(sk);
2116 cgroup_sk_alloc(&sk->sk_cgrp_data);
2117 sock_update_classid(&sk->sk_cgrp_data);
2118 sock_update_netprioidx(&sk->sk_cgrp_data);
2119 sk_tx_queue_clear(sk);
2120 }
2121
2122 return sk;
2123}
2124EXPORT_SYMBOL(sk_alloc);
2125
2126/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2127 * grace period. This is the case for UDP sockets and TCP listeners.
2128 */
2129static void __sk_destruct(struct rcu_head *head)
2130{
2131 struct sock *sk = container_of(head, struct sock, sk_rcu);
2132 struct sk_filter *filter;
2133
2134 if (sk->sk_destruct)
2135 sk->sk_destruct(sk);
2136
2137 filter = rcu_dereference_check(sk->sk_filter,
2138 refcount_read(&sk->sk_wmem_alloc) == 0);
2139 if (filter) {
2140 sk_filter_uncharge(sk, filter);
2141 RCU_INIT_POINTER(sk->sk_filter, NULL);
2142 }
2143
2144 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2145
2146#ifdef CONFIG_BPF_SYSCALL
2147 bpf_sk_storage_free(sk);
2148#endif
2149
2150 if (atomic_read(&sk->sk_omem_alloc))
2151 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2152 __func__, atomic_read(&sk->sk_omem_alloc));
2153
2154 if (sk->sk_frag.page) {
2155 put_page(sk->sk_frag.page);
2156 sk->sk_frag.page = NULL;
2157 }
2158
2159 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2160 put_cred(sk->sk_peer_cred);
2161 put_pid(sk->sk_peer_pid);
2162
2163 if (likely(sk->sk_net_refcnt))
2164 put_net_track(sock_net(sk), &sk->ns_tracker);
2165 else
2166 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2167
2168 sk_prot_free(sk->sk_prot_creator, sk);
2169}
2170
2171void sk_destruct(struct sock *sk)
2172{
2173 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2174
2175 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2176 reuseport_detach_sock(sk);
2177 use_call_rcu = true;
2178 }
2179
2180 if (use_call_rcu)
2181 call_rcu(&sk->sk_rcu, __sk_destruct);
2182 else
2183 __sk_destruct(&sk->sk_rcu);
2184}
2185
2186static void __sk_free(struct sock *sk)
2187{
2188 if (likely(sk->sk_net_refcnt))
2189 sock_inuse_add(sock_net(sk), -1);
2190
2191 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2192 sock_diag_broadcast_destroy(sk);
2193 else
2194 sk_destruct(sk);
2195}
2196
2197void sk_free(struct sock *sk)
2198{
2199 /*
2200 * We subtract one from sk_wmem_alloc and can know if
2201 * some packets are still in some tx queue.
2202 * If not null, sock_wfree() will call __sk_free(sk) later
2203 */
2204 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2205 __sk_free(sk);
2206}
2207EXPORT_SYMBOL(sk_free);
2208
2209static void sk_init_common(struct sock *sk)
2210{
2211 skb_queue_head_init(&sk->sk_receive_queue);
2212 skb_queue_head_init(&sk->sk_write_queue);
2213 skb_queue_head_init(&sk->sk_error_queue);
2214
2215 rwlock_init(&sk->sk_callback_lock);
2216 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2217 af_rlock_keys + sk->sk_family,
2218 af_family_rlock_key_strings[sk->sk_family]);
2219 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2220 af_wlock_keys + sk->sk_family,
2221 af_family_wlock_key_strings[sk->sk_family]);
2222 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2223 af_elock_keys + sk->sk_family,
2224 af_family_elock_key_strings[sk->sk_family]);
2225 lockdep_set_class_and_name(&sk->sk_callback_lock,
2226 af_callback_keys + sk->sk_family,
2227 af_family_clock_key_strings[sk->sk_family]);
2228}
2229
2230/**
2231 * sk_clone_lock - clone a socket, and lock its clone
2232 * @sk: the socket to clone
2233 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2234 *
2235 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2236 */
2237struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2238{
2239 struct proto *prot = READ_ONCE(sk->sk_prot);
2240 struct sk_filter *filter;
2241 bool is_charged = true;
2242 struct sock *newsk;
2243
2244 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2245 if (!newsk)
2246 goto out;
2247
2248 sock_copy(newsk, sk);
2249
2250 newsk->sk_prot_creator = prot;
2251
2252 /* SANITY */
2253 if (likely(newsk->sk_net_refcnt)) {
2254 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2255 sock_inuse_add(sock_net(newsk), 1);
2256 } else {
2257 /* Kernel sockets are not elevating the struct net refcount.
2258 * Instead, use a tracker to more easily detect if a layer
2259 * is not properly dismantling its kernel sockets at netns
2260 * destroy time.
2261 */
2262 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2263 false, priority);
2264 }
2265 sk_node_init(&newsk->sk_node);
2266 sock_lock_init(newsk);
2267 bh_lock_sock(newsk);
2268 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2269 newsk->sk_backlog.len = 0;
2270
2271 atomic_set(&newsk->sk_rmem_alloc, 0);
2272
2273 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2274 refcount_set(&newsk->sk_wmem_alloc, 1);
2275
2276 atomic_set(&newsk->sk_omem_alloc, 0);
2277 sk_init_common(newsk);
2278
2279 newsk->sk_dst_cache = NULL;
2280 newsk->sk_dst_pending_confirm = 0;
2281 newsk->sk_wmem_queued = 0;
2282 newsk->sk_forward_alloc = 0;
2283 newsk->sk_reserved_mem = 0;
2284 atomic_set(&newsk->sk_drops, 0);
2285 newsk->sk_send_head = NULL;
2286 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2287 atomic_set(&newsk->sk_zckey, 0);
2288
2289 sock_reset_flag(newsk, SOCK_DONE);
2290
2291 /* sk->sk_memcg will be populated at accept() time */
2292 newsk->sk_memcg = NULL;
2293
2294 cgroup_sk_clone(&newsk->sk_cgrp_data);
2295
2296 rcu_read_lock();
2297 filter = rcu_dereference(sk->sk_filter);
2298 if (filter != NULL)
2299 /* though it's an empty new sock, the charging may fail
2300 * if sysctl_optmem_max was changed between creation of
2301 * original socket and cloning
2302 */
2303 is_charged = sk_filter_charge(newsk, filter);
2304 RCU_INIT_POINTER(newsk->sk_filter, filter);
2305 rcu_read_unlock();
2306
2307 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2308 /* We need to make sure that we don't uncharge the new
2309 * socket if we couldn't charge it in the first place
2310 * as otherwise we uncharge the parent's filter.
2311 */
2312 if (!is_charged)
2313 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2314 sk_free_unlock_clone(newsk);
2315 newsk = NULL;
2316 goto out;
2317 }
2318 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2319
2320 if (bpf_sk_storage_clone(sk, newsk)) {
2321 sk_free_unlock_clone(newsk);
2322 newsk = NULL;
2323 goto out;
2324 }
2325
2326 /* Clear sk_user_data if parent had the pointer tagged
2327 * as not suitable for copying when cloning.
2328 */
2329 if (sk_user_data_is_nocopy(newsk))
2330 newsk->sk_user_data = NULL;
2331
2332 newsk->sk_err = 0;
2333 newsk->sk_err_soft = 0;
2334 newsk->sk_priority = 0;
2335 newsk->sk_incoming_cpu = raw_smp_processor_id();
2336
2337 /* Before updating sk_refcnt, we must commit prior changes to memory
2338 * (Documentation/RCU/rculist_nulls.rst for details)
2339 */
2340 smp_wmb();
2341 refcount_set(&newsk->sk_refcnt, 2);
2342
2343 /* Increment the counter in the same struct proto as the master
2344 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
2345 * is the same as sk->sk_prot->socks, as this field was copied
2346 * with memcpy).
2347 *
2348 * This _changes_ the previous behaviour, where
2349 * tcp_create_openreq_child always was incrementing the
2350 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
2351 * to be taken into account in all callers. -acme
2352 */
2353 sk_refcnt_debug_inc(newsk);
2354 sk_set_socket(newsk, NULL);
2355 sk_tx_queue_clear(newsk);
2356 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2357
2358 if (newsk->sk_prot->sockets_allocated)
2359 sk_sockets_allocated_inc(newsk);
2360
2361 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2362 net_enable_timestamp();
2363out:
2364 return newsk;
2365}
2366EXPORT_SYMBOL_GPL(sk_clone_lock);
2367
2368void sk_free_unlock_clone(struct sock *sk)
2369{
2370 /* It is still raw copy of parent, so invalidate
2371 * destructor and make plain sk_free() */
2372 sk->sk_destruct = NULL;
2373 bh_unlock_sock(sk);
2374 sk_free(sk);
2375}
2376EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2377
2378static void sk_trim_gso_size(struct sock *sk)
2379{
2380 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE)
2381 return;
2382#if IS_ENABLED(CONFIG_IPV6)
2383 if (sk->sk_family == AF_INET6 &&
2384 sk_is_tcp(sk) &&
2385 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
2386 return;
2387#endif
2388 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE;
2389}
2390
2391void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2392{
2393 u32 max_segs = 1;
2394
2395 sk_dst_set(sk, dst);
2396 sk->sk_route_caps = dst->dev->features;
2397 if (sk_is_tcp(sk))
2398 sk->sk_route_caps |= NETIF_F_GSO;
2399 if (sk->sk_route_caps & NETIF_F_GSO)
2400 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2401 if (unlikely(sk->sk_gso_disabled))
2402 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2403 if (sk_can_gso(sk)) {
2404 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2405 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2406 } else {
2407 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2408 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */
2409 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size);
2410 sk_trim_gso_size(sk);
2411 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1);
2412 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2413 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2414 }
2415 }
2416 sk->sk_gso_max_segs = max_segs;
2417}
2418EXPORT_SYMBOL_GPL(sk_setup_caps);
2419
2420/*
2421 * Simple resource managers for sockets.
2422 */
2423
2424
2425/*
2426 * Write buffer destructor automatically called from kfree_skb.
2427 */
2428void sock_wfree(struct sk_buff *skb)
2429{
2430 struct sock *sk = skb->sk;
2431 unsigned int len = skb->truesize;
2432 bool free;
2433
2434 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2435 if (sock_flag(sk, SOCK_RCU_FREE) &&
2436 sk->sk_write_space == sock_def_write_space) {
2437 rcu_read_lock();
2438 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2439 sock_def_write_space_wfree(sk);
2440 rcu_read_unlock();
2441 if (unlikely(free))
2442 __sk_free(sk);
2443 return;
2444 }
2445
2446 /*
2447 * Keep a reference on sk_wmem_alloc, this will be released
2448 * after sk_write_space() call
2449 */
2450 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2451 sk->sk_write_space(sk);
2452 len = 1;
2453 }
2454 /*
2455 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2456 * could not do because of in-flight packets
2457 */
2458 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2459 __sk_free(sk);
2460}
2461EXPORT_SYMBOL(sock_wfree);
2462
2463/* This variant of sock_wfree() is used by TCP,
2464 * since it sets SOCK_USE_WRITE_QUEUE.
2465 */
2466void __sock_wfree(struct sk_buff *skb)
2467{
2468 struct sock *sk = skb->sk;
2469
2470 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2471 __sk_free(sk);
2472}
2473
2474void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2475{
2476 skb_orphan(skb);
2477 skb->sk = sk;
2478#ifdef CONFIG_INET
2479 if (unlikely(!sk_fullsock(sk))) {
2480 skb->destructor = sock_edemux;
2481 sock_hold(sk);
2482 return;
2483 }
2484#endif
2485 skb->destructor = sock_wfree;
2486 skb_set_hash_from_sk(skb, sk);
2487 /*
2488 * We used to take a refcount on sk, but following operation
2489 * is enough to guarantee sk_free() wont free this sock until
2490 * all in-flight packets are completed
2491 */
2492 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2493}
2494EXPORT_SYMBOL(skb_set_owner_w);
2495
2496static bool can_skb_orphan_partial(const struct sk_buff *skb)
2497{
2498#ifdef CONFIG_TLS_DEVICE
2499 /* Drivers depend on in-order delivery for crypto offload,
2500 * partial orphan breaks out-of-order-OK logic.
2501 */
2502 if (skb->decrypted)
2503 return false;
2504#endif
2505 return (skb->destructor == sock_wfree ||
2506 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2507}
2508
2509/* This helper is used by netem, as it can hold packets in its
2510 * delay queue. We want to allow the owner socket to send more
2511 * packets, as if they were already TX completed by a typical driver.
2512 * But we also want to keep skb->sk set because some packet schedulers
2513 * rely on it (sch_fq for example).
2514 */
2515void skb_orphan_partial(struct sk_buff *skb)
2516{
2517 if (skb_is_tcp_pure_ack(skb))
2518 return;
2519
2520 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2521 return;
2522
2523 skb_orphan(skb);
2524}
2525EXPORT_SYMBOL(skb_orphan_partial);
2526
2527/*
2528 * Read buffer destructor automatically called from kfree_skb.
2529 */
2530void sock_rfree(struct sk_buff *skb)
2531{
2532 struct sock *sk = skb->sk;
2533 unsigned int len = skb->truesize;
2534
2535 atomic_sub(len, &sk->sk_rmem_alloc);
2536 sk_mem_uncharge(sk, len);
2537}
2538EXPORT_SYMBOL(sock_rfree);
2539
2540/*
2541 * Buffer destructor for skbs that are not used directly in read or write
2542 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2543 */
2544void sock_efree(struct sk_buff *skb)
2545{
2546 sock_put(skb->sk);
2547}
2548EXPORT_SYMBOL(sock_efree);
2549
2550/* Buffer destructor for prefetch/receive path where reference count may
2551 * not be held, e.g. for listen sockets.
2552 */
2553#ifdef CONFIG_INET
2554void sock_pfree(struct sk_buff *skb)
2555{
2556 if (sk_is_refcounted(skb->sk))
2557 sock_gen_put(skb->sk);
2558}
2559EXPORT_SYMBOL(sock_pfree);
2560#endif /* CONFIG_INET */
2561
2562kuid_t sock_i_uid(struct sock *sk)
2563{
2564 kuid_t uid;
2565
2566 read_lock_bh(&sk->sk_callback_lock);
2567 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2568 read_unlock_bh(&sk->sk_callback_lock);
2569 return uid;
2570}
2571EXPORT_SYMBOL(sock_i_uid);
2572
2573unsigned long sock_i_ino(struct sock *sk)
2574{
2575 unsigned long ino;
2576
2577 read_lock_bh(&sk->sk_callback_lock);
2578 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2579 read_unlock_bh(&sk->sk_callback_lock);
2580 return ino;
2581}
2582EXPORT_SYMBOL(sock_i_ino);
2583
2584/*
2585 * Allocate a skb from the socket's send buffer.
2586 */
2587struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2588 gfp_t priority)
2589{
2590 if (force ||
2591 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2592 struct sk_buff *skb = alloc_skb(size, priority);
2593
2594 if (skb) {
2595 skb_set_owner_w(skb, sk);
2596 return skb;
2597 }
2598 }
2599 return NULL;
2600}
2601EXPORT_SYMBOL(sock_wmalloc);
2602
2603static void sock_ofree(struct sk_buff *skb)
2604{
2605 struct sock *sk = skb->sk;
2606
2607 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2608}
2609
2610struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2611 gfp_t priority)
2612{
2613 struct sk_buff *skb;
2614
2615 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2616 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2617 READ_ONCE(sysctl_optmem_max))
2618 return NULL;
2619
2620 skb = alloc_skb(size, priority);
2621 if (!skb)
2622 return NULL;
2623
2624 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2625 skb->sk = sk;
2626 skb->destructor = sock_ofree;
2627 return skb;
2628}
2629
2630/*
2631 * Allocate a memory block from the socket's option memory buffer.
2632 */
2633void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2634{
2635 int optmem_max = READ_ONCE(sysctl_optmem_max);
2636
2637 if ((unsigned int)size <= optmem_max &&
2638 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2639 void *mem;
2640 /* First do the add, to avoid the race if kmalloc
2641 * might sleep.
2642 */
2643 atomic_add(size, &sk->sk_omem_alloc);
2644 mem = kmalloc(size, priority);
2645 if (mem)
2646 return mem;
2647 atomic_sub(size, &sk->sk_omem_alloc);
2648 }
2649 return NULL;
2650}
2651EXPORT_SYMBOL(sock_kmalloc);
2652
2653/* Free an option memory block. Note, we actually want the inline
2654 * here as this allows gcc to detect the nullify and fold away the
2655 * condition entirely.
2656 */
2657static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2658 const bool nullify)
2659{
2660 if (WARN_ON_ONCE(!mem))
2661 return;
2662 if (nullify)
2663 kfree_sensitive(mem);
2664 else
2665 kfree(mem);
2666 atomic_sub(size, &sk->sk_omem_alloc);
2667}
2668
2669void sock_kfree_s(struct sock *sk, void *mem, int size)
2670{
2671 __sock_kfree_s(sk, mem, size, false);
2672}
2673EXPORT_SYMBOL(sock_kfree_s);
2674
2675void sock_kzfree_s(struct sock *sk, void *mem, int size)
2676{
2677 __sock_kfree_s(sk, mem, size, true);
2678}
2679EXPORT_SYMBOL(sock_kzfree_s);
2680
2681/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2682 I think, these locks should be removed for datagram sockets.
2683 */
2684static long sock_wait_for_wmem(struct sock *sk, long timeo)
2685{
2686 DEFINE_WAIT(wait);
2687
2688 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2689 for (;;) {
2690 if (!timeo)
2691 break;
2692 if (signal_pending(current))
2693 break;
2694 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2695 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2696 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2697 break;
2698 if (sk->sk_shutdown & SEND_SHUTDOWN)
2699 break;
2700 if (sk->sk_err)
2701 break;
2702 timeo = schedule_timeout(timeo);
2703 }
2704 finish_wait(sk_sleep(sk), &wait);
2705 return timeo;
2706}
2707
2708
2709/*
2710 * Generic send/receive buffer handlers
2711 */
2712
2713struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2714 unsigned long data_len, int noblock,
2715 int *errcode, int max_page_order)
2716{
2717 struct sk_buff *skb;
2718 long timeo;
2719 int err;
2720
2721 timeo = sock_sndtimeo(sk, noblock);
2722 for (;;) {
2723 err = sock_error(sk);
2724 if (err != 0)
2725 goto failure;
2726
2727 err = -EPIPE;
2728 if (sk->sk_shutdown & SEND_SHUTDOWN)
2729 goto failure;
2730
2731 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2732 break;
2733
2734 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2735 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2736 err = -EAGAIN;
2737 if (!timeo)
2738 goto failure;
2739 if (signal_pending(current))
2740 goto interrupted;
2741 timeo = sock_wait_for_wmem(sk, timeo);
2742 }
2743 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2744 errcode, sk->sk_allocation);
2745 if (skb)
2746 skb_set_owner_w(skb, sk);
2747 return skb;
2748
2749interrupted:
2750 err = sock_intr_errno(timeo);
2751failure:
2752 *errcode = err;
2753 return NULL;
2754}
2755EXPORT_SYMBOL(sock_alloc_send_pskb);
2756
2757int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2758 struct sockcm_cookie *sockc)
2759{
2760 u32 tsflags;
2761
2762 switch (cmsg->cmsg_type) {
2763 case SO_MARK:
2764 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2765 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2766 return -EPERM;
2767 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2768 return -EINVAL;
2769 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2770 break;
2771 case SO_TIMESTAMPING_OLD:
2772 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2773 return -EINVAL;
2774
2775 tsflags = *(u32 *)CMSG_DATA(cmsg);
2776 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2777 return -EINVAL;
2778
2779 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2780 sockc->tsflags |= tsflags;
2781 break;
2782 case SCM_TXTIME:
2783 if (!sock_flag(sk, SOCK_TXTIME))
2784 return -EINVAL;
2785 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2786 return -EINVAL;
2787 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2788 break;
2789 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2790 case SCM_RIGHTS:
2791 case SCM_CREDENTIALS:
2792 break;
2793 default:
2794 return -EINVAL;
2795 }
2796 return 0;
2797}
2798EXPORT_SYMBOL(__sock_cmsg_send);
2799
2800int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2801 struct sockcm_cookie *sockc)
2802{
2803 struct cmsghdr *cmsg;
2804 int ret;
2805
2806 for_each_cmsghdr(cmsg, msg) {
2807 if (!CMSG_OK(msg, cmsg))
2808 return -EINVAL;
2809 if (cmsg->cmsg_level != SOL_SOCKET)
2810 continue;
2811 ret = __sock_cmsg_send(sk, cmsg, sockc);
2812 if (ret)
2813 return ret;
2814 }
2815 return 0;
2816}
2817EXPORT_SYMBOL(sock_cmsg_send);
2818
2819static void sk_enter_memory_pressure(struct sock *sk)
2820{
2821 if (!sk->sk_prot->enter_memory_pressure)
2822 return;
2823
2824 sk->sk_prot->enter_memory_pressure(sk);
2825}
2826
2827static void sk_leave_memory_pressure(struct sock *sk)
2828{
2829 if (sk->sk_prot->leave_memory_pressure) {
2830 sk->sk_prot->leave_memory_pressure(sk);
2831 } else {
2832 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2833
2834 if (memory_pressure && READ_ONCE(*memory_pressure))
2835 WRITE_ONCE(*memory_pressure, 0);
2836 }
2837}
2838
2839DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2840
2841/**
2842 * skb_page_frag_refill - check that a page_frag contains enough room
2843 * @sz: minimum size of the fragment we want to get
2844 * @pfrag: pointer to page_frag
2845 * @gfp: priority for memory allocation
2846 *
2847 * Note: While this allocator tries to use high order pages, there is
2848 * no guarantee that allocations succeed. Therefore, @sz MUST be
2849 * less or equal than PAGE_SIZE.
2850 */
2851bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2852{
2853 if (pfrag->page) {
2854 if (page_ref_count(pfrag->page) == 1) {
2855 pfrag->offset = 0;
2856 return true;
2857 }
2858 if (pfrag->offset + sz <= pfrag->size)
2859 return true;
2860 put_page(pfrag->page);
2861 }
2862
2863 pfrag->offset = 0;
2864 if (SKB_FRAG_PAGE_ORDER &&
2865 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2866 /* Avoid direct reclaim but allow kswapd to wake */
2867 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2868 __GFP_COMP | __GFP_NOWARN |
2869 __GFP_NORETRY,
2870 SKB_FRAG_PAGE_ORDER);
2871 if (likely(pfrag->page)) {
2872 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2873 return true;
2874 }
2875 }
2876 pfrag->page = alloc_page(gfp);
2877 if (likely(pfrag->page)) {
2878 pfrag->size = PAGE_SIZE;
2879 return true;
2880 }
2881 return false;
2882}
2883EXPORT_SYMBOL(skb_page_frag_refill);
2884
2885bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2886{
2887 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2888 return true;
2889
2890 sk_enter_memory_pressure(sk);
2891 sk_stream_moderate_sndbuf(sk);
2892 return false;
2893}
2894EXPORT_SYMBOL(sk_page_frag_refill);
2895
2896void __lock_sock(struct sock *sk)
2897 __releases(&sk->sk_lock.slock)
2898 __acquires(&sk->sk_lock.slock)
2899{
2900 DEFINE_WAIT(wait);
2901
2902 for (;;) {
2903 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2904 TASK_UNINTERRUPTIBLE);
2905 spin_unlock_bh(&sk->sk_lock.slock);
2906 schedule();
2907 spin_lock_bh(&sk->sk_lock.slock);
2908 if (!sock_owned_by_user(sk))
2909 break;
2910 }
2911 finish_wait(&sk->sk_lock.wq, &wait);
2912}
2913
2914void __release_sock(struct sock *sk)
2915 __releases(&sk->sk_lock.slock)
2916 __acquires(&sk->sk_lock.slock)
2917{
2918 struct sk_buff *skb, *next;
2919
2920 while ((skb = sk->sk_backlog.head) != NULL) {
2921 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2922
2923 spin_unlock_bh(&sk->sk_lock.slock);
2924
2925 do {
2926 next = skb->next;
2927 prefetch(next);
2928 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2929 skb_mark_not_on_list(skb);
2930 sk_backlog_rcv(sk, skb);
2931
2932 cond_resched();
2933
2934 skb = next;
2935 } while (skb != NULL);
2936
2937 spin_lock_bh(&sk->sk_lock.slock);
2938 }
2939
2940 /*
2941 * Doing the zeroing here guarantee we can not loop forever
2942 * while a wild producer attempts to flood us.
2943 */
2944 sk->sk_backlog.len = 0;
2945}
2946
2947void __sk_flush_backlog(struct sock *sk)
2948{
2949 spin_lock_bh(&sk->sk_lock.slock);
2950 __release_sock(sk);
2951 spin_unlock_bh(&sk->sk_lock.slock);
2952}
2953EXPORT_SYMBOL_GPL(__sk_flush_backlog);
2954
2955/**
2956 * sk_wait_data - wait for data to arrive at sk_receive_queue
2957 * @sk: sock to wait on
2958 * @timeo: for how long
2959 * @skb: last skb seen on sk_receive_queue
2960 *
2961 * Now socket state including sk->sk_err is changed only under lock,
2962 * hence we may omit checks after joining wait queue.
2963 * We check receive queue before schedule() only as optimization;
2964 * it is very likely that release_sock() added new data.
2965 */
2966int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2967{
2968 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2969 int rc;
2970
2971 add_wait_queue(sk_sleep(sk), &wait);
2972 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2973 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2974 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2975 remove_wait_queue(sk_sleep(sk), &wait);
2976 return rc;
2977}
2978EXPORT_SYMBOL(sk_wait_data);
2979
2980/**
2981 * __sk_mem_raise_allocated - increase memory_allocated
2982 * @sk: socket
2983 * @size: memory size to allocate
2984 * @amt: pages to allocate
2985 * @kind: allocation type
2986 *
2987 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2988 */
2989int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2990{
2991 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg;
2992 struct proto *prot = sk->sk_prot;
2993 bool charged = true;
2994 long allocated;
2995
2996 sk_memory_allocated_add(sk, amt);
2997 allocated = sk_memory_allocated(sk);
2998 if (memcg_charge &&
2999 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3000 gfp_memcg_charge())))
3001 goto suppress_allocation;
3002
3003 /* Under limit. */
3004 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3005 sk_leave_memory_pressure(sk);
3006 return 1;
3007 }
3008
3009 /* Under pressure. */
3010 if (allocated > sk_prot_mem_limits(sk, 1))
3011 sk_enter_memory_pressure(sk);
3012
3013 /* Over hard limit. */
3014 if (allocated > sk_prot_mem_limits(sk, 2))
3015 goto suppress_allocation;
3016
3017 /* guarantee minimum buffer size under pressure */
3018 if (kind == SK_MEM_RECV) {
3019 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3020 return 1;
3021
3022 } else { /* SK_MEM_SEND */
3023 int wmem0 = sk_get_wmem0(sk, prot);
3024
3025 if (sk->sk_type == SOCK_STREAM) {
3026 if (sk->sk_wmem_queued < wmem0)
3027 return 1;
3028 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3029 return 1;
3030 }
3031 }
3032
3033 if (sk_has_memory_pressure(sk)) {
3034 u64 alloc;
3035
3036 if (!sk_under_memory_pressure(sk))
3037 return 1;
3038 alloc = sk_sockets_allocated_read_positive(sk);
3039 if (sk_prot_mem_limits(sk, 2) > alloc *
3040 sk_mem_pages(sk->sk_wmem_queued +
3041 atomic_read(&sk->sk_rmem_alloc) +
3042 sk->sk_forward_alloc))
3043 return 1;
3044 }
3045
3046suppress_allocation:
3047
3048 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3049 sk_stream_moderate_sndbuf(sk);
3050
3051 /* Fail only if socket is _under_ its sndbuf.
3052 * In this case we cannot block, so that we have to fail.
3053 */
3054 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3055 /* Force charge with __GFP_NOFAIL */
3056 if (memcg_charge && !charged) {
3057 mem_cgroup_charge_skmem(sk->sk_memcg, amt,
3058 gfp_memcg_charge() | __GFP_NOFAIL);
3059 }
3060 return 1;
3061 }
3062 }
3063
3064 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3065 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3066
3067 sk_memory_allocated_sub(sk, amt);
3068
3069 if (memcg_charge && charged)
3070 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
3071
3072 return 0;
3073}
3074
3075/**
3076 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3077 * @sk: socket
3078 * @size: memory size to allocate
3079 * @kind: allocation type
3080 *
3081 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3082 * rmem allocation. This function assumes that protocols which have
3083 * memory_pressure use sk_wmem_queued as write buffer accounting.
3084 */
3085int __sk_mem_schedule(struct sock *sk, int size, int kind)
3086{
3087 int ret, amt = sk_mem_pages(size);
3088
3089 sk->sk_forward_alloc += amt << PAGE_SHIFT;
3090 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3091 if (!ret)
3092 sk->sk_forward_alloc -= amt << PAGE_SHIFT;
3093 return ret;
3094}
3095EXPORT_SYMBOL(__sk_mem_schedule);
3096
3097/**
3098 * __sk_mem_reduce_allocated - reclaim memory_allocated
3099 * @sk: socket
3100 * @amount: number of quanta
3101 *
3102 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3103 */
3104void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3105{
3106 sk_memory_allocated_sub(sk, amount);
3107
3108 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3109 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3110
3111 if (sk_under_memory_pressure(sk) &&
3112 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3113 sk_leave_memory_pressure(sk);
3114}
3115
3116/**
3117 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3118 * @sk: socket
3119 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3120 */
3121void __sk_mem_reclaim(struct sock *sk, int amount)
3122{
3123 amount >>= PAGE_SHIFT;
3124 sk->sk_forward_alloc -= amount << PAGE_SHIFT;
3125 __sk_mem_reduce_allocated(sk, amount);
3126}
3127EXPORT_SYMBOL(__sk_mem_reclaim);
3128
3129int sk_set_peek_off(struct sock *sk, int val)
3130{
3131 sk->sk_peek_off = val;
3132 return 0;
3133}
3134EXPORT_SYMBOL_GPL(sk_set_peek_off);
3135
3136/*
3137 * Set of default routines for initialising struct proto_ops when
3138 * the protocol does not support a particular function. In certain
3139 * cases where it makes no sense for a protocol to have a "do nothing"
3140 * function, some default processing is provided.
3141 */
3142
3143int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3144{
3145 return -EOPNOTSUPP;
3146}
3147EXPORT_SYMBOL(sock_no_bind);
3148
3149int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3150 int len, int flags)
3151{
3152 return -EOPNOTSUPP;
3153}
3154EXPORT_SYMBOL(sock_no_connect);
3155
3156int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3157{
3158 return -EOPNOTSUPP;
3159}
3160EXPORT_SYMBOL(sock_no_socketpair);
3161
3162int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3163 bool kern)
3164{
3165 return -EOPNOTSUPP;
3166}
3167EXPORT_SYMBOL(sock_no_accept);
3168
3169int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3170 int peer)
3171{
3172 return -EOPNOTSUPP;
3173}
3174EXPORT_SYMBOL(sock_no_getname);
3175
3176int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3177{
3178 return -EOPNOTSUPP;
3179}
3180EXPORT_SYMBOL(sock_no_ioctl);
3181
3182int sock_no_listen(struct socket *sock, int backlog)
3183{
3184 return -EOPNOTSUPP;
3185}
3186EXPORT_SYMBOL(sock_no_listen);
3187
3188int sock_no_shutdown(struct socket *sock, int how)
3189{
3190 return -EOPNOTSUPP;
3191}
3192EXPORT_SYMBOL(sock_no_shutdown);
3193
3194int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3195{
3196 return -EOPNOTSUPP;
3197}
3198EXPORT_SYMBOL(sock_no_sendmsg);
3199
3200int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3201{
3202 return -EOPNOTSUPP;
3203}
3204EXPORT_SYMBOL(sock_no_sendmsg_locked);
3205
3206int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3207 int flags)
3208{
3209 return -EOPNOTSUPP;
3210}
3211EXPORT_SYMBOL(sock_no_recvmsg);
3212
3213int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3214{
3215 /* Mirror missing mmap method error code */
3216 return -ENODEV;
3217}
3218EXPORT_SYMBOL(sock_no_mmap);
3219
3220/*
3221 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3222 * various sock-based usage counts.
3223 */
3224void __receive_sock(struct file *file)
3225{
3226 struct socket *sock;
3227
3228 sock = sock_from_file(file);
3229 if (sock) {
3230 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3231 sock_update_classid(&sock->sk->sk_cgrp_data);
3232 }
3233}
3234
3235ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
3236{
3237 ssize_t res;
3238 struct msghdr msg = {.msg_flags = flags};
3239 struct kvec iov;
3240 char *kaddr = kmap(page);
3241 iov.iov_base = kaddr + offset;
3242 iov.iov_len = size;
3243 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
3244 kunmap(page);
3245 return res;
3246}
3247EXPORT_SYMBOL(sock_no_sendpage);
3248
3249ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
3250 int offset, size_t size, int flags)
3251{
3252 ssize_t res;
3253 struct msghdr msg = {.msg_flags = flags};
3254 struct kvec iov;
3255 char *kaddr = kmap(page);
3256
3257 iov.iov_base = kaddr + offset;
3258 iov.iov_len = size;
3259 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
3260 kunmap(page);
3261 return res;
3262}
3263EXPORT_SYMBOL(sock_no_sendpage_locked);
3264
3265/*
3266 * Default Socket Callbacks
3267 */
3268
3269static void sock_def_wakeup(struct sock *sk)
3270{
3271 struct socket_wq *wq;
3272
3273 rcu_read_lock();
3274 wq = rcu_dereference(sk->sk_wq);
3275 if (skwq_has_sleeper(wq))
3276 wake_up_interruptible_all(&wq->wait);
3277 rcu_read_unlock();
3278}
3279
3280static void sock_def_error_report(struct sock *sk)
3281{
3282 struct socket_wq *wq;
3283
3284 rcu_read_lock();
3285 wq = rcu_dereference(sk->sk_wq);
3286 if (skwq_has_sleeper(wq))
3287 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3288 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3289 rcu_read_unlock();
3290}
3291
3292void sock_def_readable(struct sock *sk)
3293{
3294 struct socket_wq *wq;
3295
3296 rcu_read_lock();
3297 wq = rcu_dereference(sk->sk_wq);
3298 if (skwq_has_sleeper(wq))
3299 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3300 EPOLLRDNORM | EPOLLRDBAND);
3301 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3302 rcu_read_unlock();
3303}
3304
3305static void sock_def_write_space(struct sock *sk)
3306{
3307 struct socket_wq *wq;
3308
3309 rcu_read_lock();
3310
3311 /* Do not wake up a writer until he can make "significant"
3312 * progress. --DaveM
3313 */
3314 if (sock_writeable(sk)) {
3315 wq = rcu_dereference(sk->sk_wq);
3316 if (skwq_has_sleeper(wq))
3317 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3318 EPOLLWRNORM | EPOLLWRBAND);
3319
3320 /* Should agree with poll, otherwise some programs break */
3321 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3322 }
3323
3324 rcu_read_unlock();
3325}
3326
3327/* An optimised version of sock_def_write_space(), should only be called
3328 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3329 * ->sk_wmem_alloc.
3330 */
3331static void sock_def_write_space_wfree(struct sock *sk)
3332{
3333 /* Do not wake up a writer until he can make "significant"
3334 * progress. --DaveM
3335 */
3336 if (sock_writeable(sk)) {
3337 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3338
3339 /* rely on refcount_sub from sock_wfree() */
3340 smp_mb__after_atomic();
3341 if (wq && waitqueue_active(&wq->wait))
3342 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3343 EPOLLWRNORM | EPOLLWRBAND);
3344
3345 /* Should agree with poll, otherwise some programs break */
3346 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3347 }
3348}
3349
3350static void sock_def_destruct(struct sock *sk)
3351{
3352}
3353
3354void sk_send_sigurg(struct sock *sk)
3355{
3356 if (sk->sk_socket && sk->sk_socket->file)
3357 if (send_sigurg(&sk->sk_socket->file->f_owner))
3358 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3359}
3360EXPORT_SYMBOL(sk_send_sigurg);
3361
3362void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3363 unsigned long expires)
3364{
3365 if (!mod_timer(timer, expires))
3366 sock_hold(sk);
3367}
3368EXPORT_SYMBOL(sk_reset_timer);
3369
3370void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3371{
3372 if (del_timer(timer))
3373 __sock_put(sk);
3374}
3375EXPORT_SYMBOL(sk_stop_timer);
3376
3377void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3378{
3379 if (del_timer_sync(timer))
3380 __sock_put(sk);
3381}
3382EXPORT_SYMBOL(sk_stop_timer_sync);
3383
3384void sock_init_data(struct socket *sock, struct sock *sk)
3385{
3386 sk_init_common(sk);
3387 sk->sk_send_head = NULL;
3388
3389 timer_setup(&sk->sk_timer, NULL, 0);
3390
3391 sk->sk_allocation = GFP_KERNEL;
3392 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3393 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3394 sk->sk_state = TCP_CLOSE;
3395 sk->sk_use_task_frag = true;
3396 sk_set_socket(sk, sock);
3397
3398 sock_set_flag(sk, SOCK_ZAPPED);
3399
3400 if (sock) {
3401 sk->sk_type = sock->type;
3402 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3403 sock->sk = sk;
3404 sk->sk_uid = SOCK_INODE(sock)->i_uid;
3405 } else {
3406 RCU_INIT_POINTER(sk->sk_wq, NULL);
3407 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
3408 }
3409
3410 rwlock_init(&sk->sk_callback_lock);
3411 if (sk->sk_kern_sock)
3412 lockdep_set_class_and_name(
3413 &sk->sk_callback_lock,
3414 af_kern_callback_keys + sk->sk_family,
3415 af_family_kern_clock_key_strings[sk->sk_family]);
3416 else
3417 lockdep_set_class_and_name(
3418 &sk->sk_callback_lock,
3419 af_callback_keys + sk->sk_family,
3420 af_family_clock_key_strings[sk->sk_family]);
3421
3422 sk->sk_state_change = sock_def_wakeup;
3423 sk->sk_data_ready = sock_def_readable;
3424 sk->sk_write_space = sock_def_write_space;
3425 sk->sk_error_report = sock_def_error_report;
3426 sk->sk_destruct = sock_def_destruct;
3427
3428 sk->sk_frag.page = NULL;
3429 sk->sk_frag.offset = 0;
3430 sk->sk_peek_off = -1;
3431
3432 sk->sk_peer_pid = NULL;
3433 sk->sk_peer_cred = NULL;
3434 spin_lock_init(&sk->sk_peer_lock);
3435
3436 sk->sk_write_pending = 0;
3437 sk->sk_rcvlowat = 1;
3438 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3439 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3440
3441 sk->sk_stamp = SK_DEFAULT_STAMP;
3442#if BITS_PER_LONG==32
3443 seqlock_init(&sk->sk_stamp_seq);
3444#endif
3445 atomic_set(&sk->sk_zckey, 0);
3446
3447#ifdef CONFIG_NET_RX_BUSY_POLL
3448 sk->sk_napi_id = 0;
3449 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3450#endif
3451
3452 sk->sk_max_pacing_rate = ~0UL;
3453 sk->sk_pacing_rate = ~0UL;
3454 WRITE_ONCE(sk->sk_pacing_shift, 10);
3455 sk->sk_incoming_cpu = -1;
3456
3457 sk_rx_queue_clear(sk);
3458 /*
3459 * Before updating sk_refcnt, we must commit prior changes to memory
3460 * (Documentation/RCU/rculist_nulls.rst for details)
3461 */
3462 smp_wmb();
3463 refcount_set(&sk->sk_refcnt, 1);
3464 atomic_set(&sk->sk_drops, 0);
3465}
3466EXPORT_SYMBOL(sock_init_data);
3467
3468void lock_sock_nested(struct sock *sk, int subclass)
3469{
3470 /* The sk_lock has mutex_lock() semantics here. */
3471 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3472
3473 might_sleep();
3474 spin_lock_bh(&sk->sk_lock.slock);
3475 if (sock_owned_by_user_nocheck(sk))
3476 __lock_sock(sk);
3477 sk->sk_lock.owned = 1;
3478 spin_unlock_bh(&sk->sk_lock.slock);
3479}
3480EXPORT_SYMBOL(lock_sock_nested);
3481
3482void release_sock(struct sock *sk)
3483{
3484 spin_lock_bh(&sk->sk_lock.slock);
3485 if (sk->sk_backlog.tail)
3486 __release_sock(sk);
3487
3488 /* Warning : release_cb() might need to release sk ownership,
3489 * ie call sock_release_ownership(sk) before us.
3490 */
3491 if (sk->sk_prot->release_cb)
3492 sk->sk_prot->release_cb(sk);
3493
3494 sock_release_ownership(sk);
3495 if (waitqueue_active(&sk->sk_lock.wq))
3496 wake_up(&sk->sk_lock.wq);
3497 spin_unlock_bh(&sk->sk_lock.slock);
3498}
3499EXPORT_SYMBOL(release_sock);
3500
3501bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3502{
3503 might_sleep();
3504 spin_lock_bh(&sk->sk_lock.slock);
3505
3506 if (!sock_owned_by_user_nocheck(sk)) {
3507 /*
3508 * Fast path return with bottom halves disabled and
3509 * sock::sk_lock.slock held.
3510 *
3511 * The 'mutex' is not contended and holding
3512 * sock::sk_lock.slock prevents all other lockers to
3513 * proceed so the corresponding unlock_sock_fast() can
3514 * avoid the slow path of release_sock() completely and
3515 * just release slock.
3516 *
3517 * From a semantical POV this is equivalent to 'acquiring'
3518 * the 'mutex', hence the corresponding lockdep
3519 * mutex_release() has to happen in the fast path of
3520 * unlock_sock_fast().
3521 */
3522 return false;
3523 }
3524
3525 __lock_sock(sk);
3526 sk->sk_lock.owned = 1;
3527 __acquire(&sk->sk_lock.slock);
3528 spin_unlock_bh(&sk->sk_lock.slock);
3529 return true;
3530}
3531EXPORT_SYMBOL(__lock_sock_fast);
3532
3533int sock_gettstamp(struct socket *sock, void __user *userstamp,
3534 bool timeval, bool time32)
3535{
3536 struct sock *sk = sock->sk;
3537 struct timespec64 ts;
3538
3539 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3540 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3541 if (ts.tv_sec == -1)
3542 return -ENOENT;
3543 if (ts.tv_sec == 0) {
3544 ktime_t kt = ktime_get_real();
3545 sock_write_timestamp(sk, kt);
3546 ts = ktime_to_timespec64(kt);
3547 }
3548
3549 if (timeval)
3550 ts.tv_nsec /= 1000;
3551
3552#ifdef CONFIG_COMPAT_32BIT_TIME
3553 if (time32)
3554 return put_old_timespec32(&ts, userstamp);
3555#endif
3556#ifdef CONFIG_SPARC64
3557 /* beware of padding in sparc64 timeval */
3558 if (timeval && !in_compat_syscall()) {
3559 struct __kernel_old_timeval __user tv = {
3560 .tv_sec = ts.tv_sec,
3561 .tv_usec = ts.tv_nsec,
3562 };
3563 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3564 return -EFAULT;
3565 return 0;
3566 }
3567#endif
3568 return put_timespec64(&ts, userstamp);
3569}
3570EXPORT_SYMBOL(sock_gettstamp);
3571
3572void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3573{
3574 if (!sock_flag(sk, flag)) {
3575 unsigned long previous_flags = sk->sk_flags;
3576
3577 sock_set_flag(sk, flag);
3578 /*
3579 * we just set one of the two flags which require net
3580 * time stamping, but time stamping might have been on
3581 * already because of the other one
3582 */
3583 if (sock_needs_netstamp(sk) &&
3584 !(previous_flags & SK_FLAGS_TIMESTAMP))
3585 net_enable_timestamp();
3586 }
3587}
3588
3589int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3590 int level, int type)
3591{
3592 struct sock_exterr_skb *serr;
3593 struct sk_buff *skb;
3594 int copied, err;
3595
3596 err = -EAGAIN;
3597 skb = sock_dequeue_err_skb(sk);
3598 if (skb == NULL)
3599 goto out;
3600
3601 copied = skb->len;
3602 if (copied > len) {
3603 msg->msg_flags |= MSG_TRUNC;
3604 copied = len;
3605 }
3606 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3607 if (err)
3608 goto out_free_skb;
3609
3610 sock_recv_timestamp(msg, sk, skb);
3611
3612 serr = SKB_EXT_ERR(skb);
3613 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3614
3615 msg->msg_flags |= MSG_ERRQUEUE;
3616 err = copied;
3617
3618out_free_skb:
3619 kfree_skb(skb);
3620out:
3621 return err;
3622}
3623EXPORT_SYMBOL(sock_recv_errqueue);
3624
3625/*
3626 * Get a socket option on an socket.
3627 *
3628 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3629 * asynchronous errors should be reported by getsockopt. We assume
3630 * this means if you specify SO_ERROR (otherwise whats the point of it).
3631 */
3632int sock_common_getsockopt(struct socket *sock, int level, int optname,
3633 char __user *optval, int __user *optlen)
3634{
3635 struct sock *sk = sock->sk;
3636
3637 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3638 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3639}
3640EXPORT_SYMBOL(sock_common_getsockopt);
3641
3642int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3643 int flags)
3644{
3645 struct sock *sk = sock->sk;
3646 int addr_len = 0;
3647 int err;
3648
3649 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3650 if (err >= 0)
3651 msg->msg_namelen = addr_len;
3652 return err;
3653}
3654EXPORT_SYMBOL(sock_common_recvmsg);
3655
3656/*
3657 * Set socket options on an inet socket.
3658 */
3659int sock_common_setsockopt(struct socket *sock, int level, int optname,
3660 sockptr_t optval, unsigned int optlen)
3661{
3662 struct sock *sk = sock->sk;
3663
3664 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3665 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3666}
3667EXPORT_SYMBOL(sock_common_setsockopt);
3668
3669void sk_common_release(struct sock *sk)
3670{
3671 if (sk->sk_prot->destroy)
3672 sk->sk_prot->destroy(sk);
3673
3674 /*
3675 * Observation: when sk_common_release is called, processes have
3676 * no access to socket. But net still has.
3677 * Step one, detach it from networking:
3678 *
3679 * A. Remove from hash tables.
3680 */
3681
3682 sk->sk_prot->unhash(sk);
3683
3684 /*
3685 * In this point socket cannot receive new packets, but it is possible
3686 * that some packets are in flight because some CPU runs receiver and
3687 * did hash table lookup before we unhashed socket. They will achieve
3688 * receive queue and will be purged by socket destructor.
3689 *
3690 * Also we still have packets pending on receive queue and probably,
3691 * our own packets waiting in device queues. sock_destroy will drain
3692 * receive queue, but transmitted packets will delay socket destruction
3693 * until the last reference will be released.
3694 */
3695
3696 sock_orphan(sk);
3697
3698 xfrm_sk_free_policy(sk);
3699
3700 sk_refcnt_debug_release(sk);
3701
3702 sock_put(sk);
3703}
3704EXPORT_SYMBOL(sk_common_release);
3705
3706void sk_get_meminfo(const struct sock *sk, u32 *mem)
3707{
3708 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3709
3710 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3711 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3712 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3713 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3714 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3715 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3716 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3717 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3718 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3719}
3720
3721#ifdef CONFIG_PROC_FS
3722static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3723
3724int sock_prot_inuse_get(struct net *net, struct proto *prot)
3725{
3726 int cpu, idx = prot->inuse_idx;
3727 int res = 0;
3728
3729 for_each_possible_cpu(cpu)
3730 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3731
3732 return res >= 0 ? res : 0;
3733}
3734EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3735
3736int sock_inuse_get(struct net *net)
3737{
3738 int cpu, res = 0;
3739
3740 for_each_possible_cpu(cpu)
3741 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3742
3743 return res;
3744}
3745
3746EXPORT_SYMBOL_GPL(sock_inuse_get);
3747
3748static int __net_init sock_inuse_init_net(struct net *net)
3749{
3750 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3751 if (net->core.prot_inuse == NULL)
3752 return -ENOMEM;
3753 return 0;
3754}
3755
3756static void __net_exit sock_inuse_exit_net(struct net *net)
3757{
3758 free_percpu(net->core.prot_inuse);
3759}
3760
3761static struct pernet_operations net_inuse_ops = {
3762 .init = sock_inuse_init_net,
3763 .exit = sock_inuse_exit_net,
3764};
3765
3766static __init int net_inuse_init(void)
3767{
3768 if (register_pernet_subsys(&net_inuse_ops))
3769 panic("Cannot initialize net inuse counters");
3770
3771 return 0;
3772}
3773
3774core_initcall(net_inuse_init);
3775
3776static int assign_proto_idx(struct proto *prot)
3777{
3778 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3779
3780 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3781 pr_err("PROTO_INUSE_NR exhausted\n");
3782 return -ENOSPC;
3783 }
3784
3785 set_bit(prot->inuse_idx, proto_inuse_idx);
3786 return 0;
3787}
3788
3789static void release_proto_idx(struct proto *prot)
3790{
3791 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3792 clear_bit(prot->inuse_idx, proto_inuse_idx);
3793}
3794#else
3795static inline int assign_proto_idx(struct proto *prot)
3796{
3797 return 0;
3798}
3799
3800static inline void release_proto_idx(struct proto *prot)
3801{
3802}
3803
3804#endif
3805
3806static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3807{
3808 if (!twsk_prot)
3809 return;
3810 kfree(twsk_prot->twsk_slab_name);
3811 twsk_prot->twsk_slab_name = NULL;
3812 kmem_cache_destroy(twsk_prot->twsk_slab);
3813 twsk_prot->twsk_slab = NULL;
3814}
3815
3816static int tw_prot_init(const struct proto *prot)
3817{
3818 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3819
3820 if (!twsk_prot)
3821 return 0;
3822
3823 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3824 prot->name);
3825 if (!twsk_prot->twsk_slab_name)
3826 return -ENOMEM;
3827
3828 twsk_prot->twsk_slab =
3829 kmem_cache_create(twsk_prot->twsk_slab_name,
3830 twsk_prot->twsk_obj_size, 0,
3831 SLAB_ACCOUNT | prot->slab_flags,
3832 NULL);
3833 if (!twsk_prot->twsk_slab) {
3834 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3835 prot->name);
3836 return -ENOMEM;
3837 }
3838
3839 return 0;
3840}
3841
3842static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3843{
3844 if (!rsk_prot)
3845 return;
3846 kfree(rsk_prot->slab_name);
3847 rsk_prot->slab_name = NULL;
3848 kmem_cache_destroy(rsk_prot->slab);
3849 rsk_prot->slab = NULL;
3850}
3851
3852static int req_prot_init(const struct proto *prot)
3853{
3854 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3855
3856 if (!rsk_prot)
3857 return 0;
3858
3859 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3860 prot->name);
3861 if (!rsk_prot->slab_name)
3862 return -ENOMEM;
3863
3864 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3865 rsk_prot->obj_size, 0,
3866 SLAB_ACCOUNT | prot->slab_flags,
3867 NULL);
3868
3869 if (!rsk_prot->slab) {
3870 pr_crit("%s: Can't create request sock SLAB cache!\n",
3871 prot->name);
3872 return -ENOMEM;
3873 }
3874 return 0;
3875}
3876
3877int proto_register(struct proto *prot, int alloc_slab)
3878{
3879 int ret = -ENOBUFS;
3880
3881 if (prot->memory_allocated && !prot->sysctl_mem) {
3882 pr_err("%s: missing sysctl_mem\n", prot->name);
3883 return -EINVAL;
3884 }
3885 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3886 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3887 return -EINVAL;
3888 }
3889 if (alloc_slab) {
3890 prot->slab = kmem_cache_create_usercopy(prot->name,
3891 prot->obj_size, 0,
3892 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3893 prot->slab_flags,
3894 prot->useroffset, prot->usersize,
3895 NULL);
3896
3897 if (prot->slab == NULL) {
3898 pr_crit("%s: Can't create sock SLAB cache!\n",
3899 prot->name);
3900 goto out;
3901 }
3902
3903 if (req_prot_init(prot))
3904 goto out_free_request_sock_slab;
3905
3906 if (tw_prot_init(prot))
3907 goto out_free_timewait_sock_slab;
3908 }
3909
3910 mutex_lock(&proto_list_mutex);
3911 ret = assign_proto_idx(prot);
3912 if (ret) {
3913 mutex_unlock(&proto_list_mutex);
3914 goto out_free_timewait_sock_slab;
3915 }
3916 list_add(&prot->node, &proto_list);
3917 mutex_unlock(&proto_list_mutex);
3918 return ret;
3919
3920out_free_timewait_sock_slab:
3921 if (alloc_slab)
3922 tw_prot_cleanup(prot->twsk_prot);
3923out_free_request_sock_slab:
3924 if (alloc_slab) {
3925 req_prot_cleanup(prot->rsk_prot);
3926
3927 kmem_cache_destroy(prot->slab);
3928 prot->slab = NULL;
3929 }
3930out:
3931 return ret;
3932}
3933EXPORT_SYMBOL(proto_register);
3934
3935void proto_unregister(struct proto *prot)
3936{
3937 mutex_lock(&proto_list_mutex);
3938 release_proto_idx(prot);
3939 list_del(&prot->node);
3940 mutex_unlock(&proto_list_mutex);
3941
3942 kmem_cache_destroy(prot->slab);
3943 prot->slab = NULL;
3944
3945 req_prot_cleanup(prot->rsk_prot);
3946 tw_prot_cleanup(prot->twsk_prot);
3947}
3948EXPORT_SYMBOL(proto_unregister);
3949
3950int sock_load_diag_module(int family, int protocol)
3951{
3952 if (!protocol) {
3953 if (!sock_is_registered(family))
3954 return -ENOENT;
3955
3956 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3957 NETLINK_SOCK_DIAG, family);
3958 }
3959
3960#ifdef CONFIG_INET
3961 if (family == AF_INET &&
3962 protocol != IPPROTO_RAW &&
3963 protocol < MAX_INET_PROTOS &&
3964 !rcu_access_pointer(inet_protos[protocol]))
3965 return -ENOENT;
3966#endif
3967
3968 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3969 NETLINK_SOCK_DIAG, family, protocol);
3970}
3971EXPORT_SYMBOL(sock_load_diag_module);
3972
3973#ifdef CONFIG_PROC_FS
3974static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3975 __acquires(proto_list_mutex)
3976{
3977 mutex_lock(&proto_list_mutex);
3978 return seq_list_start_head(&proto_list, *pos);
3979}
3980
3981static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3982{
3983 return seq_list_next(v, &proto_list, pos);
3984}
3985
3986static void proto_seq_stop(struct seq_file *seq, void *v)
3987 __releases(proto_list_mutex)
3988{
3989 mutex_unlock(&proto_list_mutex);
3990}
3991
3992static char proto_method_implemented(const void *method)
3993{
3994 return method == NULL ? 'n' : 'y';
3995}
3996static long sock_prot_memory_allocated(struct proto *proto)
3997{
3998 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3999}
4000
4001static const char *sock_prot_memory_pressure(struct proto *proto)
4002{
4003 return proto->memory_pressure != NULL ?
4004 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4005}
4006
4007static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4008{
4009
4010 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4011 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4012 proto->name,
4013 proto->obj_size,
4014 sock_prot_inuse_get(seq_file_net(seq), proto),
4015 sock_prot_memory_allocated(proto),
4016 sock_prot_memory_pressure(proto),
4017 proto->max_header,
4018 proto->slab == NULL ? "no" : "yes",
4019 module_name(proto->owner),
4020 proto_method_implemented(proto->close),
4021 proto_method_implemented(proto->connect),
4022 proto_method_implemented(proto->disconnect),
4023 proto_method_implemented(proto->accept),
4024 proto_method_implemented(proto->ioctl),
4025 proto_method_implemented(proto->init),
4026 proto_method_implemented(proto->destroy),
4027 proto_method_implemented(proto->shutdown),
4028 proto_method_implemented(proto->setsockopt),
4029 proto_method_implemented(proto->getsockopt),
4030 proto_method_implemented(proto->sendmsg),
4031 proto_method_implemented(proto->recvmsg),
4032 proto_method_implemented(proto->sendpage),
4033 proto_method_implemented(proto->bind),
4034 proto_method_implemented(proto->backlog_rcv),
4035 proto_method_implemented(proto->hash),
4036 proto_method_implemented(proto->unhash),
4037 proto_method_implemented(proto->get_port),
4038 proto_method_implemented(proto->enter_memory_pressure));
4039}
4040
4041static int proto_seq_show(struct seq_file *seq, void *v)
4042{
4043 if (v == &proto_list)
4044 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4045 "protocol",
4046 "size",
4047 "sockets",
4048 "memory",
4049 "press",
4050 "maxhdr",
4051 "slab",
4052 "module",
4053 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
4054 else
4055 proto_seq_printf(seq, list_entry(v, struct proto, node));
4056 return 0;
4057}
4058
4059static const struct seq_operations proto_seq_ops = {
4060 .start = proto_seq_start,
4061 .next = proto_seq_next,
4062 .stop = proto_seq_stop,
4063 .show = proto_seq_show,
4064};
4065
4066static __net_init int proto_init_net(struct net *net)
4067{
4068 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4069 sizeof(struct seq_net_private)))
4070 return -ENOMEM;
4071
4072 return 0;
4073}
4074
4075static __net_exit void proto_exit_net(struct net *net)
4076{
4077 remove_proc_entry("protocols", net->proc_net);
4078}
4079
4080
4081static __net_initdata struct pernet_operations proto_net_ops = {
4082 .init = proto_init_net,
4083 .exit = proto_exit_net,
4084};
4085
4086static int __init proto_init(void)
4087{
4088 return register_pernet_subsys(&proto_net_ops);
4089}
4090
4091subsys_initcall(proto_init);
4092
4093#endif /* PROC_FS */
4094
4095#ifdef CONFIG_NET_RX_BUSY_POLL
4096bool sk_busy_loop_end(void *p, unsigned long start_time)
4097{
4098 struct sock *sk = p;
4099
4100 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
4101 sk_busy_loop_timeout(sk, start_time);
4102}
4103EXPORT_SYMBOL(sk_busy_loop_end);
4104#endif /* CONFIG_NET_RX_BUSY_POLL */
4105
4106int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4107{
4108 if (!sk->sk_prot->bind_add)
4109 return -EOPNOTSUPP;
4110 return sk->sk_prot->bind_add(sk, addr, addr_len);
4111}
4112EXPORT_SYMBOL(sock_bind_add);