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