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/udp.h>
111#include <linux/init.h>
112#include <linux/highmem.h>
113#include <linux/user_namespace.h>
114#include <linux/static_key.h>
115#include <linux/memcontrol.h>
116#include <linux/prefetch.h>
117#include <linux/compat.h>
118#include <linux/mroute.h>
119#include <linux/mroute6.h>
120#include <linux/icmpv6.h>
121
122#include <linux/uaccess.h>
123
124#include <linux/netdevice.h>
125#include <net/protocol.h>
126#include <linux/skbuff.h>
127#include <net/net_namespace.h>
128#include <net/request_sock.h>
129#include <net/sock.h>
130#include <linux/net_tstamp.h>
131#include <net/xfrm.h>
132#include <linux/ipsec.h>
133#include <net/cls_cgroup.h>
134#include <net/netprio_cgroup.h>
135#include <linux/sock_diag.h>
136
137#include <linux/filter.h>
138#include <net/sock_reuseport.h>
139#include <net/bpf_sk_storage.h>
140
141#include <trace/events/sock.h>
142
143#include <net/tcp.h>
144#include <net/busy_poll.h>
145#include <net/phonet/phonet.h>
146
147#include <linux/ethtool.h>
148
149#include "dev.h"
150
151static DEFINE_MUTEX(proto_list_mutex);
152static LIST_HEAD(proto_list);
153
154static void sock_def_write_space_wfree(struct sock *sk);
155static void sock_def_write_space(struct sock *sk);
156
157/**
158 * sk_ns_capable - General socket capability test
159 * @sk: Socket to use a capability on or through
160 * @user_ns: The user namespace of the capability to use
161 * @cap: The capability to use
162 *
163 * Test to see if the opener of the socket had when the socket was
164 * created and the current process has the capability @cap in the user
165 * namespace @user_ns.
166 */
167bool sk_ns_capable(const struct sock *sk,
168 struct user_namespace *user_ns, int cap)
169{
170 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
171 ns_capable(user_ns, cap);
172}
173EXPORT_SYMBOL(sk_ns_capable);
174
175/**
176 * sk_capable - Socket global capability test
177 * @sk: Socket to use a capability on or through
178 * @cap: The global capability to use
179 *
180 * Test to see if the opener of the socket had when the socket was
181 * created and the current process has the capability @cap in all user
182 * namespaces.
183 */
184bool sk_capable(const struct sock *sk, int cap)
185{
186 return sk_ns_capable(sk, &init_user_ns, cap);
187}
188EXPORT_SYMBOL(sk_capable);
189
190/**
191 * sk_net_capable - Network namespace socket capability test
192 * @sk: Socket to use a capability on or through
193 * @cap: The capability to use
194 *
195 * Test to see if the opener of the socket had when the socket was created
196 * and the current process has the capability @cap over the network namespace
197 * the socket is a member of.
198 */
199bool sk_net_capable(const struct sock *sk, int cap)
200{
201 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
202}
203EXPORT_SYMBOL(sk_net_capable);
204
205/*
206 * Each address family might have different locking rules, so we have
207 * one slock key per address family and separate keys for internal and
208 * userspace sockets.
209 */
210static struct lock_class_key af_family_keys[AF_MAX];
211static struct lock_class_key af_family_kern_keys[AF_MAX];
212static struct lock_class_key af_family_slock_keys[AF_MAX];
213static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
214
215/*
216 * Make lock validator output more readable. (we pre-construct these
217 * strings build-time, so that runtime initialization of socket
218 * locks is fast):
219 */
220
221#define _sock_locks(x) \
222 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
223 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
224 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
225 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
226 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
227 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
228 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
229 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
230 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
231 x "27" , x "28" , x "AF_CAN" , \
232 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
233 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
234 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
235 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
236 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
237 x "AF_MCTP" , \
238 x "AF_MAX"
239
240static const char *const af_family_key_strings[AF_MAX+1] = {
241 _sock_locks("sk_lock-")
242};
243static const char *const af_family_slock_key_strings[AF_MAX+1] = {
244 _sock_locks("slock-")
245};
246static const char *const af_family_clock_key_strings[AF_MAX+1] = {
247 _sock_locks("clock-")
248};
249
250static const char *const af_family_kern_key_strings[AF_MAX+1] = {
251 _sock_locks("k-sk_lock-")
252};
253static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
254 _sock_locks("k-slock-")
255};
256static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
257 _sock_locks("k-clock-")
258};
259static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
260 _sock_locks("rlock-")
261};
262static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
263 _sock_locks("wlock-")
264};
265static const char *const af_family_elock_key_strings[AF_MAX+1] = {
266 _sock_locks("elock-")
267};
268
269/*
270 * sk_callback_lock and sk queues locking rules are per-address-family,
271 * so split the lock classes by using a per-AF key:
272 */
273static struct lock_class_key af_callback_keys[AF_MAX];
274static struct lock_class_key af_rlock_keys[AF_MAX];
275static struct lock_class_key af_wlock_keys[AF_MAX];
276static struct lock_class_key af_elock_keys[AF_MAX];
277static struct lock_class_key af_kern_callback_keys[AF_MAX];
278
279/* Run time adjustable parameters. */
280__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
281EXPORT_SYMBOL(sysctl_wmem_max);
282__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
283EXPORT_SYMBOL(sysctl_rmem_max);
284__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
285__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
286
287int sysctl_tstamp_allow_data __read_mostly = 1;
288
289DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
290EXPORT_SYMBOL_GPL(memalloc_socks_key);
291
292/**
293 * sk_set_memalloc - sets %SOCK_MEMALLOC
294 * @sk: socket to set it on
295 *
296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
297 * It's the responsibility of the admin to adjust min_free_kbytes
298 * to meet the requirements
299 */
300void sk_set_memalloc(struct sock *sk)
301{
302 sock_set_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation |= __GFP_MEMALLOC;
304 static_branch_inc(&memalloc_socks_key);
305}
306EXPORT_SYMBOL_GPL(sk_set_memalloc);
307
308void sk_clear_memalloc(struct sock *sk)
309{
310 sock_reset_flag(sk, SOCK_MEMALLOC);
311 sk->sk_allocation &= ~__GFP_MEMALLOC;
312 static_branch_dec(&memalloc_socks_key);
313
314 /*
315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
316 * progress of swapping. SOCK_MEMALLOC may be cleared while
317 * it has rmem allocations due to the last swapfile being deactivated
318 * but there is a risk that the socket is unusable due to exceeding
319 * the rmem limits. Reclaim the reserves and obey rmem limits again.
320 */
321 sk_mem_reclaim(sk);
322}
323EXPORT_SYMBOL_GPL(sk_clear_memalloc);
324
325int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
326{
327 int ret;
328 unsigned int noreclaim_flag;
329
330 /* these should have been dropped before queueing */
331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
332
333 noreclaim_flag = memalloc_noreclaim_save();
334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
335 tcp_v6_do_rcv,
336 tcp_v4_do_rcv,
337 sk, skb);
338 memalloc_noreclaim_restore(noreclaim_flag);
339
340 return ret;
341}
342EXPORT_SYMBOL(__sk_backlog_rcv);
343
344void sk_error_report(struct sock *sk)
345{
346 sk->sk_error_report(sk);
347
348 switch (sk->sk_family) {
349 case AF_INET:
350 fallthrough;
351 case AF_INET6:
352 trace_inet_sk_error_report(sk);
353 break;
354 default:
355 break;
356 }
357}
358EXPORT_SYMBOL(sk_error_report);
359
360int sock_get_timeout(long timeo, void *optval, bool old_timeval)
361{
362 struct __kernel_sock_timeval tv;
363
364 if (timeo == MAX_SCHEDULE_TIMEOUT) {
365 tv.tv_sec = 0;
366 tv.tv_usec = 0;
367 } else {
368 tv.tv_sec = timeo / HZ;
369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
370 }
371
372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
374 *(struct old_timeval32 *)optval = tv32;
375 return sizeof(tv32);
376 }
377
378 if (old_timeval) {
379 struct __kernel_old_timeval old_tv;
380 old_tv.tv_sec = tv.tv_sec;
381 old_tv.tv_usec = tv.tv_usec;
382 *(struct __kernel_old_timeval *)optval = old_tv;
383 return sizeof(old_tv);
384 }
385
386 *(struct __kernel_sock_timeval *)optval = tv;
387 return sizeof(tv);
388}
389EXPORT_SYMBOL(sock_get_timeout);
390
391int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
392 sockptr_t optval, int optlen, bool old_timeval)
393{
394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
395 struct old_timeval32 tv32;
396
397 if (optlen < sizeof(tv32))
398 return -EINVAL;
399
400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
401 return -EFAULT;
402 tv->tv_sec = tv32.tv_sec;
403 tv->tv_usec = tv32.tv_usec;
404 } else if (old_timeval) {
405 struct __kernel_old_timeval old_tv;
406
407 if (optlen < sizeof(old_tv))
408 return -EINVAL;
409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
410 return -EFAULT;
411 tv->tv_sec = old_tv.tv_sec;
412 tv->tv_usec = old_tv.tv_usec;
413 } else {
414 if (optlen < sizeof(*tv))
415 return -EINVAL;
416 if (copy_from_sockptr(tv, optval, sizeof(*tv)))
417 return -EFAULT;
418 }
419
420 return 0;
421}
422EXPORT_SYMBOL(sock_copy_user_timeval);
423
424static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
425 bool old_timeval)
426{
427 struct __kernel_sock_timeval tv;
428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
429 long val;
430
431 if (err)
432 return err;
433
434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
435 return -EDOM;
436
437 if (tv.tv_sec < 0) {
438 static int warned __read_mostly;
439
440 WRITE_ONCE(*timeo_p, 0);
441 if (warned < 10 && net_ratelimit()) {
442 warned++;
443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
444 __func__, current->comm, task_pid_nr(current));
445 }
446 return 0;
447 }
448 val = MAX_SCHEDULE_TIMEOUT;
449 if ((tv.tv_sec || tv.tv_usec) &&
450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
452 USEC_PER_SEC / HZ);
453 WRITE_ONCE(*timeo_p, val);
454 return 0;
455}
456
457static bool sock_needs_netstamp(const struct sock *sk)
458{
459 switch (sk->sk_family) {
460 case AF_UNSPEC:
461 case AF_UNIX:
462 return false;
463 default:
464 return true;
465 }
466}
467
468static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
469{
470 if (sk->sk_flags & flags) {
471 sk->sk_flags &= ~flags;
472 if (sock_needs_netstamp(sk) &&
473 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
474 net_disable_timestamp();
475 }
476}
477
478
479int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
480{
481 unsigned long flags;
482 struct sk_buff_head *list = &sk->sk_receive_queue;
483
484 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
485 atomic_inc(&sk->sk_drops);
486 trace_sock_rcvqueue_full(sk, skb);
487 return -ENOMEM;
488 }
489
490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
491 atomic_inc(&sk->sk_drops);
492 return -ENOBUFS;
493 }
494
495 skb->dev = NULL;
496 skb_set_owner_r(skb, sk);
497
498 /* we escape from rcu protected region, make sure we dont leak
499 * a norefcounted dst
500 */
501 skb_dst_force(skb);
502
503 spin_lock_irqsave(&list->lock, flags);
504 sock_skb_set_dropcount(sk, skb);
505 __skb_queue_tail(list, skb);
506 spin_unlock_irqrestore(&list->lock, flags);
507
508 if (!sock_flag(sk, SOCK_DEAD))
509 sk->sk_data_ready(sk);
510 return 0;
511}
512EXPORT_SYMBOL(__sock_queue_rcv_skb);
513
514int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
515 enum skb_drop_reason *reason)
516{
517 enum skb_drop_reason drop_reason;
518 int err;
519
520 err = sk_filter(sk, skb);
521 if (err) {
522 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
523 goto out;
524 }
525 err = __sock_queue_rcv_skb(sk, skb);
526 switch (err) {
527 case -ENOMEM:
528 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
529 break;
530 case -ENOBUFS:
531 drop_reason = SKB_DROP_REASON_PROTO_MEM;
532 break;
533 default:
534 drop_reason = SKB_NOT_DROPPED_YET;
535 break;
536 }
537out:
538 if (reason)
539 *reason = drop_reason;
540 return err;
541}
542EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
543
544int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
545 const int nested, unsigned int trim_cap, bool refcounted)
546{
547 int rc = NET_RX_SUCCESS;
548
549 if (sk_filter_trim_cap(sk, skb, trim_cap))
550 goto discard_and_relse;
551
552 skb->dev = NULL;
553
554 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
555 atomic_inc(&sk->sk_drops);
556 goto discard_and_relse;
557 }
558 if (nested)
559 bh_lock_sock_nested(sk);
560 else
561 bh_lock_sock(sk);
562 if (!sock_owned_by_user(sk)) {
563 /*
564 * trylock + unlock semantics:
565 */
566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
567
568 rc = sk_backlog_rcv(sk, skb);
569
570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
572 bh_unlock_sock(sk);
573 atomic_inc(&sk->sk_drops);
574 goto discard_and_relse;
575 }
576
577 bh_unlock_sock(sk);
578out:
579 if (refcounted)
580 sock_put(sk);
581 return rc;
582discard_and_relse:
583 kfree_skb(skb);
584 goto out;
585}
586EXPORT_SYMBOL(__sk_receive_skb);
587
588INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
589 u32));
590INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
591 u32));
592struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
593{
594 struct dst_entry *dst = __sk_dst_get(sk);
595
596 if (dst && dst->obsolete &&
597 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
598 dst, cookie) == NULL) {
599 sk_tx_queue_clear(sk);
600 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
601 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
602 dst_release(dst);
603 return NULL;
604 }
605
606 return dst;
607}
608EXPORT_SYMBOL(__sk_dst_check);
609
610struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
611{
612 struct dst_entry *dst = sk_dst_get(sk);
613
614 if (dst && dst->obsolete &&
615 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
616 dst, cookie) == NULL) {
617 sk_dst_reset(sk);
618 dst_release(dst);
619 return NULL;
620 }
621
622 return dst;
623}
624EXPORT_SYMBOL(sk_dst_check);
625
626static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
627{
628 int ret = -ENOPROTOOPT;
629#ifdef CONFIG_NETDEVICES
630 struct net *net = sock_net(sk);
631
632 /* Sorry... */
633 ret = -EPERM;
634 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
635 goto out;
636
637 ret = -EINVAL;
638 if (ifindex < 0)
639 goto out;
640
641 /* Paired with all READ_ONCE() done locklessly. */
642 WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
643
644 if (sk->sk_prot->rehash)
645 sk->sk_prot->rehash(sk);
646 sk_dst_reset(sk);
647
648 ret = 0;
649
650out:
651#endif
652
653 return ret;
654}
655
656int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
657{
658 int ret;
659
660 if (lock_sk)
661 lock_sock(sk);
662 ret = sock_bindtoindex_locked(sk, ifindex);
663 if (lock_sk)
664 release_sock(sk);
665
666 return ret;
667}
668EXPORT_SYMBOL(sock_bindtoindex);
669
670static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
671{
672 int ret = -ENOPROTOOPT;
673#ifdef CONFIG_NETDEVICES
674 struct net *net = sock_net(sk);
675 char devname[IFNAMSIZ];
676 int index;
677
678 ret = -EINVAL;
679 if (optlen < 0)
680 goto out;
681
682 /* Bind this socket to a particular device like "eth0",
683 * as specified in the passed interface name. If the
684 * name is "" or the option length is zero the socket
685 * is not bound.
686 */
687 if (optlen > IFNAMSIZ - 1)
688 optlen = IFNAMSIZ - 1;
689 memset(devname, 0, sizeof(devname));
690
691 ret = -EFAULT;
692 if (copy_from_sockptr(devname, optval, optlen))
693 goto out;
694
695 index = 0;
696 if (devname[0] != '\0') {
697 struct net_device *dev;
698
699 rcu_read_lock();
700 dev = dev_get_by_name_rcu(net, devname);
701 if (dev)
702 index = dev->ifindex;
703 rcu_read_unlock();
704 ret = -ENODEV;
705 if (!dev)
706 goto out;
707 }
708
709 sockopt_lock_sock(sk);
710 ret = sock_bindtoindex_locked(sk, index);
711 sockopt_release_sock(sk);
712out:
713#endif
714
715 return ret;
716}
717
718static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
719 sockptr_t optlen, int len)
720{
721 int ret = -ENOPROTOOPT;
722#ifdef CONFIG_NETDEVICES
723 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
724 struct net *net = sock_net(sk);
725 char devname[IFNAMSIZ];
726
727 if (bound_dev_if == 0) {
728 len = 0;
729 goto zero;
730 }
731
732 ret = -EINVAL;
733 if (len < IFNAMSIZ)
734 goto out;
735
736 ret = netdev_get_name(net, devname, bound_dev_if);
737 if (ret)
738 goto out;
739
740 len = strlen(devname) + 1;
741
742 ret = -EFAULT;
743 if (copy_to_sockptr(optval, devname, len))
744 goto out;
745
746zero:
747 ret = -EFAULT;
748 if (copy_to_sockptr(optlen, &len, sizeof(int)))
749 goto out;
750
751 ret = 0;
752
753out:
754#endif
755
756 return ret;
757}
758
759bool sk_mc_loop(const struct sock *sk)
760{
761 if (dev_recursion_level())
762 return false;
763 if (!sk)
764 return true;
765 /* IPV6_ADDRFORM can change sk->sk_family under us. */
766 switch (READ_ONCE(sk->sk_family)) {
767 case AF_INET:
768 return inet_test_bit(MC_LOOP, sk);
769#if IS_ENABLED(CONFIG_IPV6)
770 case AF_INET6:
771 return inet6_test_bit(MC6_LOOP, sk);
772#endif
773 }
774 WARN_ON_ONCE(1);
775 return true;
776}
777EXPORT_SYMBOL(sk_mc_loop);
778
779void sock_set_reuseaddr(struct sock *sk)
780{
781 lock_sock(sk);
782 sk->sk_reuse = SK_CAN_REUSE;
783 release_sock(sk);
784}
785EXPORT_SYMBOL(sock_set_reuseaddr);
786
787void sock_set_reuseport(struct sock *sk)
788{
789 lock_sock(sk);
790 sk->sk_reuseport = true;
791 release_sock(sk);
792}
793EXPORT_SYMBOL(sock_set_reuseport);
794
795void sock_no_linger(struct sock *sk)
796{
797 lock_sock(sk);
798 WRITE_ONCE(sk->sk_lingertime, 0);
799 sock_set_flag(sk, SOCK_LINGER);
800 release_sock(sk);
801}
802EXPORT_SYMBOL(sock_no_linger);
803
804void sock_set_priority(struct sock *sk, u32 priority)
805{
806 WRITE_ONCE(sk->sk_priority, priority);
807}
808EXPORT_SYMBOL(sock_set_priority);
809
810void sock_set_sndtimeo(struct sock *sk, s64 secs)
811{
812 lock_sock(sk);
813 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
814 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
815 else
816 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
817 release_sock(sk);
818}
819EXPORT_SYMBOL(sock_set_sndtimeo);
820
821static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
822{
823 if (val) {
824 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
825 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns);
826 sock_set_flag(sk, SOCK_RCVTSTAMP);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
828 } else {
829 sock_reset_flag(sk, SOCK_RCVTSTAMP);
830 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
831 }
832}
833
834void sock_enable_timestamps(struct sock *sk)
835{
836 lock_sock(sk);
837 __sock_set_timestamps(sk, true, false, true);
838 release_sock(sk);
839}
840EXPORT_SYMBOL(sock_enable_timestamps);
841
842void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
843{
844 switch (optname) {
845 case SO_TIMESTAMP_OLD:
846 __sock_set_timestamps(sk, valbool, false, false);
847 break;
848 case SO_TIMESTAMP_NEW:
849 __sock_set_timestamps(sk, valbool, true, false);
850 break;
851 case SO_TIMESTAMPNS_OLD:
852 __sock_set_timestamps(sk, valbool, false, true);
853 break;
854 case SO_TIMESTAMPNS_NEW:
855 __sock_set_timestamps(sk, valbool, true, true);
856 break;
857 }
858}
859
860static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
861{
862 struct net *net = sock_net(sk);
863 struct net_device *dev = NULL;
864 bool match = false;
865 int *vclock_index;
866 int i, num;
867
868 if (sk->sk_bound_dev_if)
869 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
870
871 if (!dev) {
872 pr_err("%s: sock not bind to device\n", __func__);
873 return -EOPNOTSUPP;
874 }
875
876 num = ethtool_get_phc_vclocks(dev, &vclock_index);
877 dev_put(dev);
878
879 for (i = 0; i < num; i++) {
880 if (*(vclock_index + i) == phc_index) {
881 match = true;
882 break;
883 }
884 }
885
886 if (num > 0)
887 kfree(vclock_index);
888
889 if (!match)
890 return -EINVAL;
891
892 WRITE_ONCE(sk->sk_bind_phc, phc_index);
893
894 return 0;
895}
896
897int sock_set_timestamping(struct sock *sk, int optname,
898 struct so_timestamping timestamping)
899{
900 int val = timestamping.flags;
901 int ret;
902
903 if (val & ~SOF_TIMESTAMPING_MASK)
904 return -EINVAL;
905
906 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
907 !(val & SOF_TIMESTAMPING_OPT_ID))
908 return -EINVAL;
909
910 if (val & SOF_TIMESTAMPING_OPT_ID &&
911 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
912 if (sk_is_tcp(sk)) {
913 if ((1 << sk->sk_state) &
914 (TCPF_CLOSE | TCPF_LISTEN))
915 return -EINVAL;
916 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
918 else
919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
920 } else {
921 atomic_set(&sk->sk_tskey, 0);
922 }
923 }
924
925 if (val & SOF_TIMESTAMPING_OPT_STATS &&
926 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
927 return -EINVAL;
928
929 if (val & SOF_TIMESTAMPING_BIND_PHC) {
930 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
931 if (ret)
932 return ret;
933 }
934
935 WRITE_ONCE(sk->sk_tsflags, val);
936 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
937
938 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
939 sock_enable_timestamp(sk,
940 SOCK_TIMESTAMPING_RX_SOFTWARE);
941 else
942 sock_disable_timestamp(sk,
943 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
944 return 0;
945}
946
947void sock_set_keepalive(struct sock *sk)
948{
949 lock_sock(sk);
950 if (sk->sk_prot->keepalive)
951 sk->sk_prot->keepalive(sk, true);
952 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
953 release_sock(sk);
954}
955EXPORT_SYMBOL(sock_set_keepalive);
956
957static void __sock_set_rcvbuf(struct sock *sk, int val)
958{
959 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
960 * as a negative value.
961 */
962 val = min_t(int, val, INT_MAX / 2);
963 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
964
965 /* We double it on the way in to account for "struct sk_buff" etc.
966 * overhead. Applications assume that the SO_RCVBUF setting they make
967 * will allow that much actual data to be received on that socket.
968 *
969 * Applications are unaware that "struct sk_buff" and other overheads
970 * allocate from the receive buffer during socket buffer allocation.
971 *
972 * And after considering the possible alternatives, returning the value
973 * we actually used in getsockopt is the most desirable behavior.
974 */
975 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
976}
977
978void sock_set_rcvbuf(struct sock *sk, int val)
979{
980 lock_sock(sk);
981 __sock_set_rcvbuf(sk, val);
982 release_sock(sk);
983}
984EXPORT_SYMBOL(sock_set_rcvbuf);
985
986static void __sock_set_mark(struct sock *sk, u32 val)
987{
988 if (val != sk->sk_mark) {
989 WRITE_ONCE(sk->sk_mark, val);
990 sk_dst_reset(sk);
991 }
992}
993
994void sock_set_mark(struct sock *sk, u32 val)
995{
996 lock_sock(sk);
997 __sock_set_mark(sk, val);
998 release_sock(sk);
999}
1000EXPORT_SYMBOL(sock_set_mark);
1001
1002static void sock_release_reserved_memory(struct sock *sk, int bytes)
1003{
1004 /* Round down bytes to multiple of pages */
1005 bytes = round_down(bytes, PAGE_SIZE);
1006
1007 WARN_ON(bytes > sk->sk_reserved_mem);
1008 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1009 sk_mem_reclaim(sk);
1010}
1011
1012static int sock_reserve_memory(struct sock *sk, int bytes)
1013{
1014 long allocated;
1015 bool charged;
1016 int pages;
1017
1018 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1019 return -EOPNOTSUPP;
1020
1021 if (!bytes)
1022 return 0;
1023
1024 pages = sk_mem_pages(bytes);
1025
1026 /* pre-charge to memcg */
1027 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1028 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1029 if (!charged)
1030 return -ENOMEM;
1031
1032 /* pre-charge to forward_alloc */
1033 sk_memory_allocated_add(sk, pages);
1034 allocated = sk_memory_allocated(sk);
1035 /* If the system goes into memory pressure with this
1036 * precharge, give up and return error.
1037 */
1038 if (allocated > sk_prot_mem_limits(sk, 1)) {
1039 sk_memory_allocated_sub(sk, pages);
1040 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1041 return -ENOMEM;
1042 }
1043 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1044
1045 WRITE_ONCE(sk->sk_reserved_mem,
1046 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1047
1048 return 0;
1049}
1050
1051void sockopt_lock_sock(struct sock *sk)
1052{
1053 /* When current->bpf_ctx is set, the setsockopt is called from
1054 * a bpf prog. bpf has ensured the sk lock has been
1055 * acquired before calling setsockopt().
1056 */
1057 if (has_current_bpf_ctx())
1058 return;
1059
1060 lock_sock(sk);
1061}
1062EXPORT_SYMBOL(sockopt_lock_sock);
1063
1064void sockopt_release_sock(struct sock *sk)
1065{
1066 if (has_current_bpf_ctx())
1067 return;
1068
1069 release_sock(sk);
1070}
1071EXPORT_SYMBOL(sockopt_release_sock);
1072
1073bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1074{
1075 return has_current_bpf_ctx() || ns_capable(ns, cap);
1076}
1077EXPORT_SYMBOL(sockopt_ns_capable);
1078
1079bool sockopt_capable(int cap)
1080{
1081 return has_current_bpf_ctx() || capable(cap);
1082}
1083EXPORT_SYMBOL(sockopt_capable);
1084
1085/*
1086 * This is meant for all protocols to use and covers goings on
1087 * at the socket level. Everything here is generic.
1088 */
1089
1090int sk_setsockopt(struct sock *sk, int level, int optname,
1091 sockptr_t optval, unsigned int optlen)
1092{
1093 struct so_timestamping timestamping;
1094 struct socket *sock = sk->sk_socket;
1095 struct sock_txtime sk_txtime;
1096 int val;
1097 int valbool;
1098 struct linger ling;
1099 int ret = 0;
1100
1101 /*
1102 * Options without arguments
1103 */
1104
1105 if (optname == SO_BINDTODEVICE)
1106 return sock_setbindtodevice(sk, optval, optlen);
1107
1108 if (optlen < sizeof(int))
1109 return -EINVAL;
1110
1111 if (copy_from_sockptr(&val, optval, sizeof(val)))
1112 return -EFAULT;
1113
1114 valbool = val ? 1 : 0;
1115
1116 /* handle options which do not require locking the socket. */
1117 switch (optname) {
1118 case SO_PRIORITY:
1119 if ((val >= 0 && val <= 6) ||
1120 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1121 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1122 sock_set_priority(sk, val);
1123 return 0;
1124 }
1125 return -EPERM;
1126 case SO_PASSSEC:
1127 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1128 return 0;
1129 case SO_PASSCRED:
1130 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1131 return 0;
1132 case SO_PASSPIDFD:
1133 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1134 return 0;
1135 case SO_TYPE:
1136 case SO_PROTOCOL:
1137 case SO_DOMAIN:
1138 case SO_ERROR:
1139 return -ENOPROTOOPT;
1140#ifdef CONFIG_NET_RX_BUSY_POLL
1141 case SO_BUSY_POLL:
1142 if (val < 0)
1143 return -EINVAL;
1144 WRITE_ONCE(sk->sk_ll_usec, val);
1145 return 0;
1146 case SO_PREFER_BUSY_POLL:
1147 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1148 return -EPERM;
1149 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1150 return 0;
1151 case SO_BUSY_POLL_BUDGET:
1152 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1153 !sockopt_capable(CAP_NET_ADMIN))
1154 return -EPERM;
1155 if (val < 0 || val > U16_MAX)
1156 return -EINVAL;
1157 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1158 return 0;
1159#endif
1160 case SO_MAX_PACING_RATE:
1161 {
1162 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1163 unsigned long pacing_rate;
1164
1165 if (sizeof(ulval) != sizeof(val) &&
1166 optlen >= sizeof(ulval) &&
1167 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1168 return -EFAULT;
1169 }
1170 if (ulval != ~0UL)
1171 cmpxchg(&sk->sk_pacing_status,
1172 SK_PACING_NONE,
1173 SK_PACING_NEEDED);
1174 /* Pairs with READ_ONCE() from sk_getsockopt() */
1175 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1176 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1177 if (ulval < pacing_rate)
1178 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1179 return 0;
1180 }
1181 case SO_TXREHASH:
1182 if (val < -1 || val > 1)
1183 return -EINVAL;
1184 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1185 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1186 /* Paired with READ_ONCE() in tcp_rtx_synack()
1187 * and sk_getsockopt().
1188 */
1189 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1190 return 0;
1191 case SO_PEEK_OFF:
1192 {
1193 int (*set_peek_off)(struct sock *sk, int val);
1194
1195 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1196 if (set_peek_off)
1197 ret = set_peek_off(sk, val);
1198 else
1199 ret = -EOPNOTSUPP;
1200 return ret;
1201 }
1202 }
1203
1204 sockopt_lock_sock(sk);
1205
1206 switch (optname) {
1207 case SO_DEBUG:
1208 if (val && !sockopt_capable(CAP_NET_ADMIN))
1209 ret = -EACCES;
1210 else
1211 sock_valbool_flag(sk, SOCK_DBG, valbool);
1212 break;
1213 case SO_REUSEADDR:
1214 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1215 break;
1216 case SO_REUSEPORT:
1217 sk->sk_reuseport = valbool;
1218 break;
1219 case SO_DONTROUTE:
1220 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1221 sk_dst_reset(sk);
1222 break;
1223 case SO_BROADCAST:
1224 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1225 break;
1226 case SO_SNDBUF:
1227 /* Don't error on this BSD doesn't and if you think
1228 * about it this is right. Otherwise apps have to
1229 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1230 * are treated in BSD as hints
1231 */
1232 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1233set_sndbuf:
1234 /* Ensure val * 2 fits into an int, to prevent max_t()
1235 * from treating it as a negative value.
1236 */
1237 val = min_t(int, val, INT_MAX / 2);
1238 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1239 WRITE_ONCE(sk->sk_sndbuf,
1240 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1241 /* Wake up sending tasks if we upped the value. */
1242 sk->sk_write_space(sk);
1243 break;
1244
1245 case SO_SNDBUFFORCE:
1246 if (!sockopt_capable(CAP_NET_ADMIN)) {
1247 ret = -EPERM;
1248 break;
1249 }
1250
1251 /* No negative values (to prevent underflow, as val will be
1252 * multiplied by 2).
1253 */
1254 if (val < 0)
1255 val = 0;
1256 goto set_sndbuf;
1257
1258 case SO_RCVBUF:
1259 /* Don't error on this BSD doesn't and if you think
1260 * about it this is right. Otherwise apps have to
1261 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1262 * are treated in BSD as hints
1263 */
1264 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1265 break;
1266
1267 case SO_RCVBUFFORCE:
1268 if (!sockopt_capable(CAP_NET_ADMIN)) {
1269 ret = -EPERM;
1270 break;
1271 }
1272
1273 /* No negative values (to prevent underflow, as val will be
1274 * multiplied by 2).
1275 */
1276 __sock_set_rcvbuf(sk, max(val, 0));
1277 break;
1278
1279 case SO_KEEPALIVE:
1280 if (sk->sk_prot->keepalive)
1281 sk->sk_prot->keepalive(sk, valbool);
1282 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1283 break;
1284
1285 case SO_OOBINLINE:
1286 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1287 break;
1288
1289 case SO_NO_CHECK:
1290 sk->sk_no_check_tx = valbool;
1291 break;
1292
1293 case SO_LINGER:
1294 if (optlen < sizeof(ling)) {
1295 ret = -EINVAL; /* 1003.1g */
1296 break;
1297 }
1298 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1299 ret = -EFAULT;
1300 break;
1301 }
1302 if (!ling.l_onoff) {
1303 sock_reset_flag(sk, SOCK_LINGER);
1304 } else {
1305 unsigned long t_sec = ling.l_linger;
1306
1307 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1308 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1309 else
1310 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1311 sock_set_flag(sk, SOCK_LINGER);
1312 }
1313 break;
1314
1315 case SO_BSDCOMPAT:
1316 break;
1317
1318 case SO_TIMESTAMP_OLD:
1319 case SO_TIMESTAMP_NEW:
1320 case SO_TIMESTAMPNS_OLD:
1321 case SO_TIMESTAMPNS_NEW:
1322 sock_set_timestamp(sk, optname, valbool);
1323 break;
1324
1325 case SO_TIMESTAMPING_NEW:
1326 case SO_TIMESTAMPING_OLD:
1327 if (optlen == sizeof(timestamping)) {
1328 if (copy_from_sockptr(×tamping, optval,
1329 sizeof(timestamping))) {
1330 ret = -EFAULT;
1331 break;
1332 }
1333 } else {
1334 memset(×tamping, 0, sizeof(timestamping));
1335 timestamping.flags = val;
1336 }
1337 ret = sock_set_timestamping(sk, optname, timestamping);
1338 break;
1339
1340 case SO_RCVLOWAT:
1341 {
1342 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1343
1344 if (val < 0)
1345 val = INT_MAX;
1346 if (sock)
1347 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1348 if (set_rcvlowat)
1349 ret = set_rcvlowat(sk, val);
1350 else
1351 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1352 break;
1353 }
1354 case SO_RCVTIMEO_OLD:
1355 case SO_RCVTIMEO_NEW:
1356 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1357 optlen, optname == SO_RCVTIMEO_OLD);
1358 break;
1359
1360 case SO_SNDTIMEO_OLD:
1361 case SO_SNDTIMEO_NEW:
1362 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1363 optlen, optname == SO_SNDTIMEO_OLD);
1364 break;
1365
1366 case SO_ATTACH_FILTER: {
1367 struct sock_fprog fprog;
1368
1369 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1370 if (!ret)
1371 ret = sk_attach_filter(&fprog, sk);
1372 break;
1373 }
1374 case SO_ATTACH_BPF:
1375 ret = -EINVAL;
1376 if (optlen == sizeof(u32)) {
1377 u32 ufd;
1378
1379 ret = -EFAULT;
1380 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1381 break;
1382
1383 ret = sk_attach_bpf(ufd, sk);
1384 }
1385 break;
1386
1387 case SO_ATTACH_REUSEPORT_CBPF: {
1388 struct sock_fprog fprog;
1389
1390 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1391 if (!ret)
1392 ret = sk_reuseport_attach_filter(&fprog, sk);
1393 break;
1394 }
1395 case SO_ATTACH_REUSEPORT_EBPF:
1396 ret = -EINVAL;
1397 if (optlen == sizeof(u32)) {
1398 u32 ufd;
1399
1400 ret = -EFAULT;
1401 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1402 break;
1403
1404 ret = sk_reuseport_attach_bpf(ufd, sk);
1405 }
1406 break;
1407
1408 case SO_DETACH_REUSEPORT_BPF:
1409 ret = reuseport_detach_prog(sk);
1410 break;
1411
1412 case SO_DETACH_FILTER:
1413 ret = sk_detach_filter(sk);
1414 break;
1415
1416 case SO_LOCK_FILTER:
1417 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1418 ret = -EPERM;
1419 else
1420 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1421 break;
1422
1423 case SO_MARK:
1424 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1425 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1426 ret = -EPERM;
1427 break;
1428 }
1429
1430 __sock_set_mark(sk, val);
1431 break;
1432 case SO_RCVMARK:
1433 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1434 break;
1435
1436 case SO_RXQ_OVFL:
1437 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1438 break;
1439
1440 case SO_WIFI_STATUS:
1441 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1442 break;
1443
1444 case SO_NOFCS:
1445 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1446 break;
1447
1448 case SO_SELECT_ERR_QUEUE:
1449 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1450 break;
1451
1452
1453 case SO_INCOMING_CPU:
1454 reuseport_update_incoming_cpu(sk, val);
1455 break;
1456
1457 case SO_CNX_ADVICE:
1458 if (val == 1)
1459 dst_negative_advice(sk);
1460 break;
1461
1462 case SO_ZEROCOPY:
1463 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1464 if (!(sk_is_tcp(sk) ||
1465 (sk->sk_type == SOCK_DGRAM &&
1466 sk->sk_protocol == IPPROTO_UDP)))
1467 ret = -EOPNOTSUPP;
1468 } else if (sk->sk_family != PF_RDS) {
1469 ret = -EOPNOTSUPP;
1470 }
1471 if (!ret) {
1472 if (val < 0 || val > 1)
1473 ret = -EINVAL;
1474 else
1475 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1476 }
1477 break;
1478
1479 case SO_TXTIME:
1480 if (optlen != sizeof(struct sock_txtime)) {
1481 ret = -EINVAL;
1482 break;
1483 } else if (copy_from_sockptr(&sk_txtime, optval,
1484 sizeof(struct sock_txtime))) {
1485 ret = -EFAULT;
1486 break;
1487 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1488 ret = -EINVAL;
1489 break;
1490 }
1491 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1492 * scheduler has enough safe guards.
1493 */
1494 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1495 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1496 ret = -EPERM;
1497 break;
1498 }
1499 sock_valbool_flag(sk, SOCK_TXTIME, true);
1500 sk->sk_clockid = sk_txtime.clockid;
1501 sk->sk_txtime_deadline_mode =
1502 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1503 sk->sk_txtime_report_errors =
1504 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1505 break;
1506
1507 case SO_BINDTOIFINDEX:
1508 ret = sock_bindtoindex_locked(sk, val);
1509 break;
1510
1511 case SO_BUF_LOCK:
1512 if (val & ~SOCK_BUF_LOCK_MASK) {
1513 ret = -EINVAL;
1514 break;
1515 }
1516 sk->sk_userlocks = val | (sk->sk_userlocks &
1517 ~SOCK_BUF_LOCK_MASK);
1518 break;
1519
1520 case SO_RESERVE_MEM:
1521 {
1522 int delta;
1523
1524 if (val < 0) {
1525 ret = -EINVAL;
1526 break;
1527 }
1528
1529 delta = val - sk->sk_reserved_mem;
1530 if (delta < 0)
1531 sock_release_reserved_memory(sk, -delta);
1532 else
1533 ret = sock_reserve_memory(sk, delta);
1534 break;
1535 }
1536
1537 default:
1538 ret = -ENOPROTOOPT;
1539 break;
1540 }
1541 sockopt_release_sock(sk);
1542 return ret;
1543}
1544
1545int sock_setsockopt(struct socket *sock, int level, int optname,
1546 sockptr_t optval, unsigned int optlen)
1547{
1548 return sk_setsockopt(sock->sk, level, optname,
1549 optval, optlen);
1550}
1551EXPORT_SYMBOL(sock_setsockopt);
1552
1553static const struct cred *sk_get_peer_cred(struct sock *sk)
1554{
1555 const struct cred *cred;
1556
1557 spin_lock(&sk->sk_peer_lock);
1558 cred = get_cred(sk->sk_peer_cred);
1559 spin_unlock(&sk->sk_peer_lock);
1560
1561 return cred;
1562}
1563
1564static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1565 struct ucred *ucred)
1566{
1567 ucred->pid = pid_vnr(pid);
1568 ucred->uid = ucred->gid = -1;
1569 if (cred) {
1570 struct user_namespace *current_ns = current_user_ns();
1571
1572 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1573 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1574 }
1575}
1576
1577static int groups_to_user(sockptr_t dst, const struct group_info *src)
1578{
1579 struct user_namespace *user_ns = current_user_ns();
1580 int i;
1581
1582 for (i = 0; i < src->ngroups; i++) {
1583 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1584
1585 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1586 return -EFAULT;
1587 }
1588
1589 return 0;
1590}
1591
1592int sk_getsockopt(struct sock *sk, int level, int optname,
1593 sockptr_t optval, sockptr_t optlen)
1594{
1595 struct socket *sock = sk->sk_socket;
1596
1597 union {
1598 int val;
1599 u64 val64;
1600 unsigned long ulval;
1601 struct linger ling;
1602 struct old_timeval32 tm32;
1603 struct __kernel_old_timeval tm;
1604 struct __kernel_sock_timeval stm;
1605 struct sock_txtime txtime;
1606 struct so_timestamping timestamping;
1607 } v;
1608
1609 int lv = sizeof(int);
1610 int len;
1611
1612 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1613 return -EFAULT;
1614 if (len < 0)
1615 return -EINVAL;
1616
1617 memset(&v, 0, sizeof(v));
1618
1619 switch (optname) {
1620 case SO_DEBUG:
1621 v.val = sock_flag(sk, SOCK_DBG);
1622 break;
1623
1624 case SO_DONTROUTE:
1625 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1626 break;
1627
1628 case SO_BROADCAST:
1629 v.val = sock_flag(sk, SOCK_BROADCAST);
1630 break;
1631
1632 case SO_SNDBUF:
1633 v.val = READ_ONCE(sk->sk_sndbuf);
1634 break;
1635
1636 case SO_RCVBUF:
1637 v.val = READ_ONCE(sk->sk_rcvbuf);
1638 break;
1639
1640 case SO_REUSEADDR:
1641 v.val = sk->sk_reuse;
1642 break;
1643
1644 case SO_REUSEPORT:
1645 v.val = sk->sk_reuseport;
1646 break;
1647
1648 case SO_KEEPALIVE:
1649 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1650 break;
1651
1652 case SO_TYPE:
1653 v.val = sk->sk_type;
1654 break;
1655
1656 case SO_PROTOCOL:
1657 v.val = sk->sk_protocol;
1658 break;
1659
1660 case SO_DOMAIN:
1661 v.val = sk->sk_family;
1662 break;
1663
1664 case SO_ERROR:
1665 v.val = -sock_error(sk);
1666 if (v.val == 0)
1667 v.val = xchg(&sk->sk_err_soft, 0);
1668 break;
1669
1670 case SO_OOBINLINE:
1671 v.val = sock_flag(sk, SOCK_URGINLINE);
1672 break;
1673
1674 case SO_NO_CHECK:
1675 v.val = sk->sk_no_check_tx;
1676 break;
1677
1678 case SO_PRIORITY:
1679 v.val = READ_ONCE(sk->sk_priority);
1680 break;
1681
1682 case SO_LINGER:
1683 lv = sizeof(v.ling);
1684 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1685 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1686 break;
1687
1688 case SO_BSDCOMPAT:
1689 break;
1690
1691 case SO_TIMESTAMP_OLD:
1692 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1693 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1694 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1695 break;
1696
1697 case SO_TIMESTAMPNS_OLD:
1698 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1699 break;
1700
1701 case SO_TIMESTAMP_NEW:
1702 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1703 break;
1704
1705 case SO_TIMESTAMPNS_NEW:
1706 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1707 break;
1708
1709 case SO_TIMESTAMPING_OLD:
1710 case SO_TIMESTAMPING_NEW:
1711 lv = sizeof(v.timestamping);
1712 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1713 * returning the flags when they were set through the same option.
1714 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1715 */
1716 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1717 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1718 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1719 }
1720 break;
1721
1722 case SO_RCVTIMEO_OLD:
1723 case SO_RCVTIMEO_NEW:
1724 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1725 SO_RCVTIMEO_OLD == optname);
1726 break;
1727
1728 case SO_SNDTIMEO_OLD:
1729 case SO_SNDTIMEO_NEW:
1730 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1731 SO_SNDTIMEO_OLD == optname);
1732 break;
1733
1734 case SO_RCVLOWAT:
1735 v.val = READ_ONCE(sk->sk_rcvlowat);
1736 break;
1737
1738 case SO_SNDLOWAT:
1739 v.val = 1;
1740 break;
1741
1742 case SO_PASSCRED:
1743 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1744 break;
1745
1746 case SO_PASSPIDFD:
1747 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1748 break;
1749
1750 case SO_PEERCRED:
1751 {
1752 struct ucred peercred;
1753 if (len > sizeof(peercred))
1754 len = sizeof(peercred);
1755
1756 spin_lock(&sk->sk_peer_lock);
1757 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1758 spin_unlock(&sk->sk_peer_lock);
1759
1760 if (copy_to_sockptr(optval, &peercred, len))
1761 return -EFAULT;
1762 goto lenout;
1763 }
1764
1765 case SO_PEERPIDFD:
1766 {
1767 struct pid *peer_pid;
1768 struct file *pidfd_file = NULL;
1769 int pidfd;
1770
1771 if (len > sizeof(pidfd))
1772 len = sizeof(pidfd);
1773
1774 spin_lock(&sk->sk_peer_lock);
1775 peer_pid = get_pid(sk->sk_peer_pid);
1776 spin_unlock(&sk->sk_peer_lock);
1777
1778 if (!peer_pid)
1779 return -ENODATA;
1780
1781 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1782 put_pid(peer_pid);
1783 if (pidfd < 0)
1784 return pidfd;
1785
1786 if (copy_to_sockptr(optval, &pidfd, len) ||
1787 copy_to_sockptr(optlen, &len, sizeof(int))) {
1788 put_unused_fd(pidfd);
1789 fput(pidfd_file);
1790
1791 return -EFAULT;
1792 }
1793
1794 fd_install(pidfd, pidfd_file);
1795 return 0;
1796 }
1797
1798 case SO_PEERGROUPS:
1799 {
1800 const struct cred *cred;
1801 int ret, n;
1802
1803 cred = sk_get_peer_cred(sk);
1804 if (!cred)
1805 return -ENODATA;
1806
1807 n = cred->group_info->ngroups;
1808 if (len < n * sizeof(gid_t)) {
1809 len = n * sizeof(gid_t);
1810 put_cred(cred);
1811 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1812 }
1813 len = n * sizeof(gid_t);
1814
1815 ret = groups_to_user(optval, cred->group_info);
1816 put_cred(cred);
1817 if (ret)
1818 return ret;
1819 goto lenout;
1820 }
1821
1822 case SO_PEERNAME:
1823 {
1824 struct sockaddr_storage address;
1825
1826 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1827 if (lv < 0)
1828 return -ENOTCONN;
1829 if (lv < len)
1830 return -EINVAL;
1831 if (copy_to_sockptr(optval, &address, len))
1832 return -EFAULT;
1833 goto lenout;
1834 }
1835
1836 /* Dubious BSD thing... Probably nobody even uses it, but
1837 * the UNIX standard wants it for whatever reason... -DaveM
1838 */
1839 case SO_ACCEPTCONN:
1840 v.val = sk->sk_state == TCP_LISTEN;
1841 break;
1842
1843 case SO_PASSSEC:
1844 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1845 break;
1846
1847 case SO_PEERSEC:
1848 return security_socket_getpeersec_stream(sock,
1849 optval, optlen, len);
1850
1851 case SO_MARK:
1852 v.val = READ_ONCE(sk->sk_mark);
1853 break;
1854
1855 case SO_RCVMARK:
1856 v.val = sock_flag(sk, SOCK_RCVMARK);
1857 break;
1858
1859 case SO_RXQ_OVFL:
1860 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1861 break;
1862
1863 case SO_WIFI_STATUS:
1864 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1865 break;
1866
1867 case SO_PEEK_OFF:
1868 if (!READ_ONCE(sock->ops)->set_peek_off)
1869 return -EOPNOTSUPP;
1870
1871 v.val = READ_ONCE(sk->sk_peek_off);
1872 break;
1873 case SO_NOFCS:
1874 v.val = sock_flag(sk, SOCK_NOFCS);
1875 break;
1876
1877 case SO_BINDTODEVICE:
1878 return sock_getbindtodevice(sk, optval, optlen, len);
1879
1880 case SO_GET_FILTER:
1881 len = sk_get_filter(sk, optval, len);
1882 if (len < 0)
1883 return len;
1884
1885 goto lenout;
1886
1887 case SO_LOCK_FILTER:
1888 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1889 break;
1890
1891 case SO_BPF_EXTENSIONS:
1892 v.val = bpf_tell_extensions();
1893 break;
1894
1895 case SO_SELECT_ERR_QUEUE:
1896 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1897 break;
1898
1899#ifdef CONFIG_NET_RX_BUSY_POLL
1900 case SO_BUSY_POLL:
1901 v.val = READ_ONCE(sk->sk_ll_usec);
1902 break;
1903 case SO_PREFER_BUSY_POLL:
1904 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1905 break;
1906#endif
1907
1908 case SO_MAX_PACING_RATE:
1909 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1910 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1911 lv = sizeof(v.ulval);
1912 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
1913 } else {
1914 /* 32bit version */
1915 v.val = min_t(unsigned long, ~0U,
1916 READ_ONCE(sk->sk_max_pacing_rate));
1917 }
1918 break;
1919
1920 case SO_INCOMING_CPU:
1921 v.val = READ_ONCE(sk->sk_incoming_cpu);
1922 break;
1923
1924 case SO_MEMINFO:
1925 {
1926 u32 meminfo[SK_MEMINFO_VARS];
1927
1928 sk_get_meminfo(sk, meminfo);
1929
1930 len = min_t(unsigned int, len, sizeof(meminfo));
1931 if (copy_to_sockptr(optval, &meminfo, len))
1932 return -EFAULT;
1933
1934 goto lenout;
1935 }
1936
1937#ifdef CONFIG_NET_RX_BUSY_POLL
1938 case SO_INCOMING_NAPI_ID:
1939 v.val = READ_ONCE(sk->sk_napi_id);
1940
1941 /* aggregate non-NAPI IDs down to 0 */
1942 if (v.val < MIN_NAPI_ID)
1943 v.val = 0;
1944
1945 break;
1946#endif
1947
1948 case SO_COOKIE:
1949 lv = sizeof(u64);
1950 if (len < lv)
1951 return -EINVAL;
1952 v.val64 = sock_gen_cookie(sk);
1953 break;
1954
1955 case SO_ZEROCOPY:
1956 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1957 break;
1958
1959 case SO_TXTIME:
1960 lv = sizeof(v.txtime);
1961 v.txtime.clockid = sk->sk_clockid;
1962 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1963 SOF_TXTIME_DEADLINE_MODE : 0;
1964 v.txtime.flags |= sk->sk_txtime_report_errors ?
1965 SOF_TXTIME_REPORT_ERRORS : 0;
1966 break;
1967
1968 case SO_BINDTOIFINDEX:
1969 v.val = READ_ONCE(sk->sk_bound_dev_if);
1970 break;
1971
1972 case SO_NETNS_COOKIE:
1973 lv = sizeof(u64);
1974 if (len != lv)
1975 return -EINVAL;
1976 v.val64 = sock_net(sk)->net_cookie;
1977 break;
1978
1979 case SO_BUF_LOCK:
1980 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
1981 break;
1982
1983 case SO_RESERVE_MEM:
1984 v.val = READ_ONCE(sk->sk_reserved_mem);
1985 break;
1986
1987 case SO_TXREHASH:
1988 /* Paired with WRITE_ONCE() in sk_setsockopt() */
1989 v.val = READ_ONCE(sk->sk_txrehash);
1990 break;
1991
1992 default:
1993 /* We implement the SO_SNDLOWAT etc to not be settable
1994 * (1003.1g 7).
1995 */
1996 return -ENOPROTOOPT;
1997 }
1998
1999 if (len > lv)
2000 len = lv;
2001 if (copy_to_sockptr(optval, &v, len))
2002 return -EFAULT;
2003lenout:
2004 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2005 return -EFAULT;
2006 return 0;
2007}
2008
2009/*
2010 * Initialize an sk_lock.
2011 *
2012 * (We also register the sk_lock with the lock validator.)
2013 */
2014static inline void sock_lock_init(struct sock *sk)
2015{
2016 if (sk->sk_kern_sock)
2017 sock_lock_init_class_and_name(
2018 sk,
2019 af_family_kern_slock_key_strings[sk->sk_family],
2020 af_family_kern_slock_keys + sk->sk_family,
2021 af_family_kern_key_strings[sk->sk_family],
2022 af_family_kern_keys + sk->sk_family);
2023 else
2024 sock_lock_init_class_and_name(
2025 sk,
2026 af_family_slock_key_strings[sk->sk_family],
2027 af_family_slock_keys + sk->sk_family,
2028 af_family_key_strings[sk->sk_family],
2029 af_family_keys + sk->sk_family);
2030}
2031
2032/*
2033 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2034 * even temporarly, because of RCU lookups. sk_node should also be left as is.
2035 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2036 */
2037static void sock_copy(struct sock *nsk, const struct sock *osk)
2038{
2039 const struct proto *prot = READ_ONCE(osk->sk_prot);
2040#ifdef CONFIG_SECURITY_NETWORK
2041 void *sptr = nsk->sk_security;
2042#endif
2043
2044 /* If we move sk_tx_queue_mapping out of the private section,
2045 * we must check if sk_tx_queue_clear() is called after
2046 * sock_copy() in sk_clone_lock().
2047 */
2048 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2049 offsetof(struct sock, sk_dontcopy_begin) ||
2050 offsetof(struct sock, sk_tx_queue_mapping) >=
2051 offsetof(struct sock, sk_dontcopy_end));
2052
2053 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2054
2055 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2056 prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
2057
2058#ifdef CONFIG_SECURITY_NETWORK
2059 nsk->sk_security = sptr;
2060 security_sk_clone(osk, nsk);
2061#endif
2062}
2063
2064static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2065 int family)
2066{
2067 struct sock *sk;
2068 struct kmem_cache *slab;
2069
2070 slab = prot->slab;
2071 if (slab != NULL) {
2072 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2073 if (!sk)
2074 return sk;
2075 if (want_init_on_alloc(priority))
2076 sk_prot_clear_nulls(sk, prot->obj_size);
2077 } else
2078 sk = kmalloc(prot->obj_size, priority);
2079
2080 if (sk != NULL) {
2081 if (security_sk_alloc(sk, family, priority))
2082 goto out_free;
2083
2084 if (!try_module_get(prot->owner))
2085 goto out_free_sec;
2086 }
2087
2088 return sk;
2089
2090out_free_sec:
2091 security_sk_free(sk);
2092out_free:
2093 if (slab != NULL)
2094 kmem_cache_free(slab, sk);
2095 else
2096 kfree(sk);
2097 return NULL;
2098}
2099
2100static void sk_prot_free(struct proto *prot, struct sock *sk)
2101{
2102 struct kmem_cache *slab;
2103 struct module *owner;
2104
2105 owner = prot->owner;
2106 slab = prot->slab;
2107
2108 cgroup_sk_free(&sk->sk_cgrp_data);
2109 mem_cgroup_sk_free(sk);
2110 security_sk_free(sk);
2111 if (slab != NULL)
2112 kmem_cache_free(slab, sk);
2113 else
2114 kfree(sk);
2115 module_put(owner);
2116}
2117
2118/**
2119 * sk_alloc - All socket objects are allocated here
2120 * @net: the applicable net namespace
2121 * @family: protocol family
2122 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2123 * @prot: struct proto associated with this new sock instance
2124 * @kern: is this to be a kernel socket?
2125 */
2126struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2127 struct proto *prot, int kern)
2128{
2129 struct sock *sk;
2130
2131 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2132 if (sk) {
2133 sk->sk_family = family;
2134 /*
2135 * See comment in struct sock definition to understand
2136 * why we need sk_prot_creator -acme
2137 */
2138 sk->sk_prot = sk->sk_prot_creator = prot;
2139 sk->sk_kern_sock = kern;
2140 sock_lock_init(sk);
2141 sk->sk_net_refcnt = kern ? 0 : 1;
2142 if (likely(sk->sk_net_refcnt)) {
2143 get_net_track(net, &sk->ns_tracker, priority);
2144 sock_inuse_add(net, 1);
2145 } else {
2146 __netns_tracker_alloc(net, &sk->ns_tracker,
2147 false, priority);
2148 }
2149
2150 sock_net_set(sk, net);
2151 refcount_set(&sk->sk_wmem_alloc, 1);
2152
2153 mem_cgroup_sk_alloc(sk);
2154 cgroup_sk_alloc(&sk->sk_cgrp_data);
2155 sock_update_classid(&sk->sk_cgrp_data);
2156 sock_update_netprioidx(&sk->sk_cgrp_data);
2157 sk_tx_queue_clear(sk);
2158 }
2159
2160 return sk;
2161}
2162EXPORT_SYMBOL(sk_alloc);
2163
2164/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2165 * grace period. This is the case for UDP sockets and TCP listeners.
2166 */
2167static void __sk_destruct(struct rcu_head *head)
2168{
2169 struct sock *sk = container_of(head, struct sock, sk_rcu);
2170 struct sk_filter *filter;
2171
2172 if (sk->sk_destruct)
2173 sk->sk_destruct(sk);
2174
2175 filter = rcu_dereference_check(sk->sk_filter,
2176 refcount_read(&sk->sk_wmem_alloc) == 0);
2177 if (filter) {
2178 sk_filter_uncharge(sk, filter);
2179 RCU_INIT_POINTER(sk->sk_filter, NULL);
2180 }
2181
2182 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2183
2184#ifdef CONFIG_BPF_SYSCALL
2185 bpf_sk_storage_free(sk);
2186#endif
2187
2188 if (atomic_read(&sk->sk_omem_alloc))
2189 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2190 __func__, atomic_read(&sk->sk_omem_alloc));
2191
2192 if (sk->sk_frag.page) {
2193 put_page(sk->sk_frag.page);
2194 sk->sk_frag.page = NULL;
2195 }
2196
2197 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2198 put_cred(sk->sk_peer_cred);
2199 put_pid(sk->sk_peer_pid);
2200
2201 if (likely(sk->sk_net_refcnt))
2202 put_net_track(sock_net(sk), &sk->ns_tracker);
2203 else
2204 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
2205
2206 sk_prot_free(sk->sk_prot_creator, sk);
2207}
2208
2209void sk_destruct(struct sock *sk)
2210{
2211 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2212
2213 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2214 reuseport_detach_sock(sk);
2215 use_call_rcu = true;
2216 }
2217
2218 if (use_call_rcu)
2219 call_rcu(&sk->sk_rcu, __sk_destruct);
2220 else
2221 __sk_destruct(&sk->sk_rcu);
2222}
2223
2224static void __sk_free(struct sock *sk)
2225{
2226 if (likely(sk->sk_net_refcnt))
2227 sock_inuse_add(sock_net(sk), -1);
2228
2229 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2230 sock_diag_broadcast_destroy(sk);
2231 else
2232 sk_destruct(sk);
2233}
2234
2235void sk_free(struct sock *sk)
2236{
2237 /*
2238 * We subtract one from sk_wmem_alloc and can know if
2239 * some packets are still in some tx queue.
2240 * If not null, sock_wfree() will call __sk_free(sk) later
2241 */
2242 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2243 __sk_free(sk);
2244}
2245EXPORT_SYMBOL(sk_free);
2246
2247static void sk_init_common(struct sock *sk)
2248{
2249 skb_queue_head_init(&sk->sk_receive_queue);
2250 skb_queue_head_init(&sk->sk_write_queue);
2251 skb_queue_head_init(&sk->sk_error_queue);
2252
2253 rwlock_init(&sk->sk_callback_lock);
2254 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2255 af_rlock_keys + sk->sk_family,
2256 af_family_rlock_key_strings[sk->sk_family]);
2257 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2258 af_wlock_keys + sk->sk_family,
2259 af_family_wlock_key_strings[sk->sk_family]);
2260 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2261 af_elock_keys + sk->sk_family,
2262 af_family_elock_key_strings[sk->sk_family]);
2263 lockdep_set_class_and_name(&sk->sk_callback_lock,
2264 af_callback_keys + sk->sk_family,
2265 af_family_clock_key_strings[sk->sk_family]);
2266}
2267
2268/**
2269 * sk_clone_lock - clone a socket, and lock its clone
2270 * @sk: the socket to clone
2271 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2272 *
2273 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2274 */
2275struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2276{
2277 struct proto *prot = READ_ONCE(sk->sk_prot);
2278 struct sk_filter *filter;
2279 bool is_charged = true;
2280 struct sock *newsk;
2281
2282 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2283 if (!newsk)
2284 goto out;
2285
2286 sock_copy(newsk, sk);
2287
2288 newsk->sk_prot_creator = prot;
2289
2290 /* SANITY */
2291 if (likely(newsk->sk_net_refcnt)) {
2292 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2293 sock_inuse_add(sock_net(newsk), 1);
2294 } else {
2295 /* Kernel sockets are not elevating the struct net refcount.
2296 * Instead, use a tracker to more easily detect if a layer
2297 * is not properly dismantling its kernel sockets at netns
2298 * destroy time.
2299 */
2300 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2301 false, priority);
2302 }
2303 sk_node_init(&newsk->sk_node);
2304 sock_lock_init(newsk);
2305 bh_lock_sock(newsk);
2306 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2307 newsk->sk_backlog.len = 0;
2308
2309 atomic_set(&newsk->sk_rmem_alloc, 0);
2310
2311 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2312 refcount_set(&newsk->sk_wmem_alloc, 1);
2313
2314 atomic_set(&newsk->sk_omem_alloc, 0);
2315 sk_init_common(newsk);
2316
2317 newsk->sk_dst_cache = NULL;
2318 newsk->sk_dst_pending_confirm = 0;
2319 newsk->sk_wmem_queued = 0;
2320 newsk->sk_forward_alloc = 0;
2321 newsk->sk_reserved_mem = 0;
2322 atomic_set(&newsk->sk_drops, 0);
2323 newsk->sk_send_head = NULL;
2324 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2325 atomic_set(&newsk->sk_zckey, 0);
2326
2327 sock_reset_flag(newsk, SOCK_DONE);
2328
2329 /* sk->sk_memcg will be populated at accept() time */
2330 newsk->sk_memcg = NULL;
2331
2332 cgroup_sk_clone(&newsk->sk_cgrp_data);
2333
2334 rcu_read_lock();
2335 filter = rcu_dereference(sk->sk_filter);
2336 if (filter != NULL)
2337 /* though it's an empty new sock, the charging may fail
2338 * if sysctl_optmem_max was changed between creation of
2339 * original socket and cloning
2340 */
2341 is_charged = sk_filter_charge(newsk, filter);
2342 RCU_INIT_POINTER(newsk->sk_filter, filter);
2343 rcu_read_unlock();
2344
2345 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2346 /* We need to make sure that we don't uncharge the new
2347 * socket if we couldn't charge it in the first place
2348 * as otherwise we uncharge the parent's filter.
2349 */
2350 if (!is_charged)
2351 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2352 sk_free_unlock_clone(newsk);
2353 newsk = NULL;
2354 goto out;
2355 }
2356 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2357
2358 if (bpf_sk_storage_clone(sk, newsk)) {
2359 sk_free_unlock_clone(newsk);
2360 newsk = NULL;
2361 goto out;
2362 }
2363
2364 /* Clear sk_user_data if parent had the pointer tagged
2365 * as not suitable for copying when cloning.
2366 */
2367 if (sk_user_data_is_nocopy(newsk))
2368 newsk->sk_user_data = NULL;
2369
2370 newsk->sk_err = 0;
2371 newsk->sk_err_soft = 0;
2372 newsk->sk_priority = 0;
2373 newsk->sk_incoming_cpu = raw_smp_processor_id();
2374
2375 /* Before updating sk_refcnt, we must commit prior changes to memory
2376 * (Documentation/RCU/rculist_nulls.rst for details)
2377 */
2378 smp_wmb();
2379 refcount_set(&newsk->sk_refcnt, 2);
2380
2381 sk_set_socket(newsk, NULL);
2382 sk_tx_queue_clear(newsk);
2383 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2384
2385 if (newsk->sk_prot->sockets_allocated)
2386 sk_sockets_allocated_inc(newsk);
2387
2388 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2389 net_enable_timestamp();
2390out:
2391 return newsk;
2392}
2393EXPORT_SYMBOL_GPL(sk_clone_lock);
2394
2395void sk_free_unlock_clone(struct sock *sk)
2396{
2397 /* It is still raw copy of parent, so invalidate
2398 * destructor and make plain sk_free() */
2399 sk->sk_destruct = NULL;
2400 bh_unlock_sock(sk);
2401 sk_free(sk);
2402}
2403EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2404
2405static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2406{
2407 bool is_ipv6 = false;
2408 u32 max_size;
2409
2410#if IS_ENABLED(CONFIG_IPV6)
2411 is_ipv6 = (sk->sk_family == AF_INET6 &&
2412 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2413#endif
2414 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2415 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2416 READ_ONCE(dst->dev->gso_ipv4_max_size);
2417 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2418 max_size = GSO_LEGACY_MAX_SIZE;
2419
2420 return max_size - (MAX_TCP_HEADER + 1);
2421}
2422
2423void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2424{
2425 u32 max_segs = 1;
2426
2427 sk->sk_route_caps = dst->dev->features;
2428 if (sk_is_tcp(sk))
2429 sk->sk_route_caps |= NETIF_F_GSO;
2430 if (sk->sk_route_caps & NETIF_F_GSO)
2431 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2432 if (unlikely(sk->sk_gso_disabled))
2433 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2434 if (sk_can_gso(sk)) {
2435 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2436 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2437 } else {
2438 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2439 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2440 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2441 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2442 }
2443 }
2444 sk->sk_gso_max_segs = max_segs;
2445 sk_dst_set(sk, dst);
2446}
2447EXPORT_SYMBOL_GPL(sk_setup_caps);
2448
2449/*
2450 * Simple resource managers for sockets.
2451 */
2452
2453
2454/*
2455 * Write buffer destructor automatically called from kfree_skb.
2456 */
2457void sock_wfree(struct sk_buff *skb)
2458{
2459 struct sock *sk = skb->sk;
2460 unsigned int len = skb->truesize;
2461 bool free;
2462
2463 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2464 if (sock_flag(sk, SOCK_RCU_FREE) &&
2465 sk->sk_write_space == sock_def_write_space) {
2466 rcu_read_lock();
2467 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2468 sock_def_write_space_wfree(sk);
2469 rcu_read_unlock();
2470 if (unlikely(free))
2471 __sk_free(sk);
2472 return;
2473 }
2474
2475 /*
2476 * Keep a reference on sk_wmem_alloc, this will be released
2477 * after sk_write_space() call
2478 */
2479 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2480 sk->sk_write_space(sk);
2481 len = 1;
2482 }
2483 /*
2484 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2485 * could not do because of in-flight packets
2486 */
2487 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2488 __sk_free(sk);
2489}
2490EXPORT_SYMBOL(sock_wfree);
2491
2492/* This variant of sock_wfree() is used by TCP,
2493 * since it sets SOCK_USE_WRITE_QUEUE.
2494 */
2495void __sock_wfree(struct sk_buff *skb)
2496{
2497 struct sock *sk = skb->sk;
2498
2499 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2500 __sk_free(sk);
2501}
2502
2503void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2504{
2505 skb_orphan(skb);
2506 skb->sk = sk;
2507#ifdef CONFIG_INET
2508 if (unlikely(!sk_fullsock(sk))) {
2509 skb->destructor = sock_edemux;
2510 sock_hold(sk);
2511 return;
2512 }
2513#endif
2514 skb->destructor = sock_wfree;
2515 skb_set_hash_from_sk(skb, sk);
2516 /*
2517 * We used to take a refcount on sk, but following operation
2518 * is enough to guarantee sk_free() wont free this sock until
2519 * all in-flight packets are completed
2520 */
2521 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2522}
2523EXPORT_SYMBOL(skb_set_owner_w);
2524
2525static bool can_skb_orphan_partial(const struct sk_buff *skb)
2526{
2527#ifdef CONFIG_TLS_DEVICE
2528 /* Drivers depend on in-order delivery for crypto offload,
2529 * partial orphan breaks out-of-order-OK logic.
2530 */
2531 if (skb->decrypted)
2532 return false;
2533#endif
2534 return (skb->destructor == sock_wfree ||
2535 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2536}
2537
2538/* This helper is used by netem, as it can hold packets in its
2539 * delay queue. We want to allow the owner socket to send more
2540 * packets, as if they were already TX completed by a typical driver.
2541 * But we also want to keep skb->sk set because some packet schedulers
2542 * rely on it (sch_fq for example).
2543 */
2544void skb_orphan_partial(struct sk_buff *skb)
2545{
2546 if (skb_is_tcp_pure_ack(skb))
2547 return;
2548
2549 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2550 return;
2551
2552 skb_orphan(skb);
2553}
2554EXPORT_SYMBOL(skb_orphan_partial);
2555
2556/*
2557 * Read buffer destructor automatically called from kfree_skb.
2558 */
2559void sock_rfree(struct sk_buff *skb)
2560{
2561 struct sock *sk = skb->sk;
2562 unsigned int len = skb->truesize;
2563
2564 atomic_sub(len, &sk->sk_rmem_alloc);
2565 sk_mem_uncharge(sk, len);
2566}
2567EXPORT_SYMBOL(sock_rfree);
2568
2569/*
2570 * Buffer destructor for skbs that are not used directly in read or write
2571 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2572 */
2573void sock_efree(struct sk_buff *skb)
2574{
2575 sock_put(skb->sk);
2576}
2577EXPORT_SYMBOL(sock_efree);
2578
2579/* Buffer destructor for prefetch/receive path where reference count may
2580 * not be held, e.g. for listen sockets.
2581 */
2582#ifdef CONFIG_INET
2583void sock_pfree(struct sk_buff *skb)
2584{
2585 if (sk_is_refcounted(skb->sk))
2586 sock_gen_put(skb->sk);
2587}
2588EXPORT_SYMBOL(sock_pfree);
2589#endif /* CONFIG_INET */
2590
2591kuid_t sock_i_uid(struct sock *sk)
2592{
2593 kuid_t uid;
2594
2595 read_lock_bh(&sk->sk_callback_lock);
2596 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2597 read_unlock_bh(&sk->sk_callback_lock);
2598 return uid;
2599}
2600EXPORT_SYMBOL(sock_i_uid);
2601
2602unsigned long __sock_i_ino(struct sock *sk)
2603{
2604 unsigned long ino;
2605
2606 read_lock(&sk->sk_callback_lock);
2607 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2608 read_unlock(&sk->sk_callback_lock);
2609 return ino;
2610}
2611EXPORT_SYMBOL(__sock_i_ino);
2612
2613unsigned long sock_i_ino(struct sock *sk)
2614{
2615 unsigned long ino;
2616
2617 local_bh_disable();
2618 ino = __sock_i_ino(sk);
2619 local_bh_enable();
2620 return ino;
2621}
2622EXPORT_SYMBOL(sock_i_ino);
2623
2624/*
2625 * Allocate a skb from the socket's send buffer.
2626 */
2627struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2628 gfp_t priority)
2629{
2630 if (force ||
2631 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2632 struct sk_buff *skb = alloc_skb(size, priority);
2633
2634 if (skb) {
2635 skb_set_owner_w(skb, sk);
2636 return skb;
2637 }
2638 }
2639 return NULL;
2640}
2641EXPORT_SYMBOL(sock_wmalloc);
2642
2643static void sock_ofree(struct sk_buff *skb)
2644{
2645 struct sock *sk = skb->sk;
2646
2647 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2648}
2649
2650struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2651 gfp_t priority)
2652{
2653 struct sk_buff *skb;
2654
2655 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2656 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2657 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2658 return NULL;
2659
2660 skb = alloc_skb(size, priority);
2661 if (!skb)
2662 return NULL;
2663
2664 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2665 skb->sk = sk;
2666 skb->destructor = sock_ofree;
2667 return skb;
2668}
2669
2670/*
2671 * Allocate a memory block from the socket's option memory buffer.
2672 */
2673void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2674{
2675 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2676
2677 if ((unsigned int)size <= optmem_max &&
2678 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2679 void *mem;
2680 /* First do the add, to avoid the race if kmalloc
2681 * might sleep.
2682 */
2683 atomic_add(size, &sk->sk_omem_alloc);
2684 mem = kmalloc(size, priority);
2685 if (mem)
2686 return mem;
2687 atomic_sub(size, &sk->sk_omem_alloc);
2688 }
2689 return NULL;
2690}
2691EXPORT_SYMBOL(sock_kmalloc);
2692
2693/* Free an option memory block. Note, we actually want the inline
2694 * here as this allows gcc to detect the nullify and fold away the
2695 * condition entirely.
2696 */
2697static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2698 const bool nullify)
2699{
2700 if (WARN_ON_ONCE(!mem))
2701 return;
2702 if (nullify)
2703 kfree_sensitive(mem);
2704 else
2705 kfree(mem);
2706 atomic_sub(size, &sk->sk_omem_alloc);
2707}
2708
2709void sock_kfree_s(struct sock *sk, void *mem, int size)
2710{
2711 __sock_kfree_s(sk, mem, size, false);
2712}
2713EXPORT_SYMBOL(sock_kfree_s);
2714
2715void sock_kzfree_s(struct sock *sk, void *mem, int size)
2716{
2717 __sock_kfree_s(sk, mem, size, true);
2718}
2719EXPORT_SYMBOL(sock_kzfree_s);
2720
2721/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2722 I think, these locks should be removed for datagram sockets.
2723 */
2724static long sock_wait_for_wmem(struct sock *sk, long timeo)
2725{
2726 DEFINE_WAIT(wait);
2727
2728 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2729 for (;;) {
2730 if (!timeo)
2731 break;
2732 if (signal_pending(current))
2733 break;
2734 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2735 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2736 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2737 break;
2738 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2739 break;
2740 if (READ_ONCE(sk->sk_err))
2741 break;
2742 timeo = schedule_timeout(timeo);
2743 }
2744 finish_wait(sk_sleep(sk), &wait);
2745 return timeo;
2746}
2747
2748
2749/*
2750 * Generic send/receive buffer handlers
2751 */
2752
2753struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2754 unsigned long data_len, int noblock,
2755 int *errcode, int max_page_order)
2756{
2757 struct sk_buff *skb;
2758 long timeo;
2759 int err;
2760
2761 timeo = sock_sndtimeo(sk, noblock);
2762 for (;;) {
2763 err = sock_error(sk);
2764 if (err != 0)
2765 goto failure;
2766
2767 err = -EPIPE;
2768 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2769 goto failure;
2770
2771 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2772 break;
2773
2774 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2775 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2776 err = -EAGAIN;
2777 if (!timeo)
2778 goto failure;
2779 if (signal_pending(current))
2780 goto interrupted;
2781 timeo = sock_wait_for_wmem(sk, timeo);
2782 }
2783 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2784 errcode, sk->sk_allocation);
2785 if (skb)
2786 skb_set_owner_w(skb, sk);
2787 return skb;
2788
2789interrupted:
2790 err = sock_intr_errno(timeo);
2791failure:
2792 *errcode = err;
2793 return NULL;
2794}
2795EXPORT_SYMBOL(sock_alloc_send_pskb);
2796
2797int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2798 struct sockcm_cookie *sockc)
2799{
2800 u32 tsflags;
2801
2802 switch (cmsg->cmsg_type) {
2803 case SO_MARK:
2804 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2805 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2806 return -EPERM;
2807 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2808 return -EINVAL;
2809 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2810 break;
2811 case SO_TIMESTAMPING_OLD:
2812 case SO_TIMESTAMPING_NEW:
2813 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2814 return -EINVAL;
2815
2816 tsflags = *(u32 *)CMSG_DATA(cmsg);
2817 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2818 return -EINVAL;
2819
2820 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2821 sockc->tsflags |= tsflags;
2822 break;
2823 case SCM_TXTIME:
2824 if (!sock_flag(sk, SOCK_TXTIME))
2825 return -EINVAL;
2826 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2827 return -EINVAL;
2828 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2829 break;
2830 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2831 case SCM_RIGHTS:
2832 case SCM_CREDENTIALS:
2833 break;
2834 default:
2835 return -EINVAL;
2836 }
2837 return 0;
2838}
2839EXPORT_SYMBOL(__sock_cmsg_send);
2840
2841int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2842 struct sockcm_cookie *sockc)
2843{
2844 struct cmsghdr *cmsg;
2845 int ret;
2846
2847 for_each_cmsghdr(cmsg, msg) {
2848 if (!CMSG_OK(msg, cmsg))
2849 return -EINVAL;
2850 if (cmsg->cmsg_level != SOL_SOCKET)
2851 continue;
2852 ret = __sock_cmsg_send(sk, cmsg, sockc);
2853 if (ret)
2854 return ret;
2855 }
2856 return 0;
2857}
2858EXPORT_SYMBOL(sock_cmsg_send);
2859
2860static void sk_enter_memory_pressure(struct sock *sk)
2861{
2862 if (!sk->sk_prot->enter_memory_pressure)
2863 return;
2864
2865 sk->sk_prot->enter_memory_pressure(sk);
2866}
2867
2868static void sk_leave_memory_pressure(struct sock *sk)
2869{
2870 if (sk->sk_prot->leave_memory_pressure) {
2871 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
2872 tcp_leave_memory_pressure, sk);
2873 } else {
2874 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2875
2876 if (memory_pressure && READ_ONCE(*memory_pressure))
2877 WRITE_ONCE(*memory_pressure, 0);
2878 }
2879}
2880
2881DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2882
2883/**
2884 * skb_page_frag_refill - check that a page_frag contains enough room
2885 * @sz: minimum size of the fragment we want to get
2886 * @pfrag: pointer to page_frag
2887 * @gfp: priority for memory allocation
2888 *
2889 * Note: While this allocator tries to use high order pages, there is
2890 * no guarantee that allocations succeed. Therefore, @sz MUST be
2891 * less or equal than PAGE_SIZE.
2892 */
2893bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2894{
2895 if (pfrag->page) {
2896 if (page_ref_count(pfrag->page) == 1) {
2897 pfrag->offset = 0;
2898 return true;
2899 }
2900 if (pfrag->offset + sz <= pfrag->size)
2901 return true;
2902 put_page(pfrag->page);
2903 }
2904
2905 pfrag->offset = 0;
2906 if (SKB_FRAG_PAGE_ORDER &&
2907 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2908 /* Avoid direct reclaim but allow kswapd to wake */
2909 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2910 __GFP_COMP | __GFP_NOWARN |
2911 __GFP_NORETRY,
2912 SKB_FRAG_PAGE_ORDER);
2913 if (likely(pfrag->page)) {
2914 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2915 return true;
2916 }
2917 }
2918 pfrag->page = alloc_page(gfp);
2919 if (likely(pfrag->page)) {
2920 pfrag->size = PAGE_SIZE;
2921 return true;
2922 }
2923 return false;
2924}
2925EXPORT_SYMBOL(skb_page_frag_refill);
2926
2927bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2928{
2929 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2930 return true;
2931
2932 sk_enter_memory_pressure(sk);
2933 sk_stream_moderate_sndbuf(sk);
2934 return false;
2935}
2936EXPORT_SYMBOL(sk_page_frag_refill);
2937
2938void __lock_sock(struct sock *sk)
2939 __releases(&sk->sk_lock.slock)
2940 __acquires(&sk->sk_lock.slock)
2941{
2942 DEFINE_WAIT(wait);
2943
2944 for (;;) {
2945 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2946 TASK_UNINTERRUPTIBLE);
2947 spin_unlock_bh(&sk->sk_lock.slock);
2948 schedule();
2949 spin_lock_bh(&sk->sk_lock.slock);
2950 if (!sock_owned_by_user(sk))
2951 break;
2952 }
2953 finish_wait(&sk->sk_lock.wq, &wait);
2954}
2955
2956void __release_sock(struct sock *sk)
2957 __releases(&sk->sk_lock.slock)
2958 __acquires(&sk->sk_lock.slock)
2959{
2960 struct sk_buff *skb, *next;
2961
2962 while ((skb = sk->sk_backlog.head) != NULL) {
2963 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2964
2965 spin_unlock_bh(&sk->sk_lock.slock);
2966
2967 do {
2968 next = skb->next;
2969 prefetch(next);
2970 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
2971 skb_mark_not_on_list(skb);
2972 sk_backlog_rcv(sk, skb);
2973
2974 cond_resched();
2975
2976 skb = next;
2977 } while (skb != NULL);
2978
2979 spin_lock_bh(&sk->sk_lock.slock);
2980 }
2981
2982 /*
2983 * Doing the zeroing here guarantee we can not loop forever
2984 * while a wild producer attempts to flood us.
2985 */
2986 sk->sk_backlog.len = 0;
2987}
2988
2989void __sk_flush_backlog(struct sock *sk)
2990{
2991 spin_lock_bh(&sk->sk_lock.slock);
2992 __release_sock(sk);
2993
2994 if (sk->sk_prot->release_cb)
2995 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
2996 tcp_release_cb, sk);
2997
2998 spin_unlock_bh(&sk->sk_lock.slock);
2999}
3000EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3001
3002/**
3003 * sk_wait_data - wait for data to arrive at sk_receive_queue
3004 * @sk: sock to wait on
3005 * @timeo: for how long
3006 * @skb: last skb seen on sk_receive_queue
3007 *
3008 * Now socket state including sk->sk_err is changed only under lock,
3009 * hence we may omit checks after joining wait queue.
3010 * We check receive queue before schedule() only as optimization;
3011 * it is very likely that release_sock() added new data.
3012 */
3013int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3014{
3015 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3016 int rc;
3017
3018 add_wait_queue(sk_sleep(sk), &wait);
3019 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3020 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3021 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3022 remove_wait_queue(sk_sleep(sk), &wait);
3023 return rc;
3024}
3025EXPORT_SYMBOL(sk_wait_data);
3026
3027/**
3028 * __sk_mem_raise_allocated - increase memory_allocated
3029 * @sk: socket
3030 * @size: memory size to allocate
3031 * @amt: pages to allocate
3032 * @kind: allocation type
3033 *
3034 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3035 *
3036 * Unlike the globally shared limits among the sockets under same protocol,
3037 * consuming the budget of a memcg won't have direct effect on other ones.
3038 * So be optimistic about memcg's tolerance, and leave the callers to decide
3039 * whether or not to raise allocated through sk_under_memory_pressure() or
3040 * its variants.
3041 */
3042int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3043{
3044 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3045 struct proto *prot = sk->sk_prot;
3046 bool charged = false;
3047 long allocated;
3048
3049 sk_memory_allocated_add(sk, amt);
3050 allocated = sk_memory_allocated(sk);
3051
3052 if (memcg) {
3053 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3054 goto suppress_allocation;
3055 charged = true;
3056 }
3057
3058 /* Under limit. */
3059 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3060 sk_leave_memory_pressure(sk);
3061 return 1;
3062 }
3063
3064 /* Under pressure. */
3065 if (allocated > sk_prot_mem_limits(sk, 1))
3066 sk_enter_memory_pressure(sk);
3067
3068 /* Over hard limit. */
3069 if (allocated > sk_prot_mem_limits(sk, 2))
3070 goto suppress_allocation;
3071
3072 /* Guarantee minimum buffer size under pressure (either global
3073 * or memcg) to make sure features described in RFC 7323 (TCP
3074 * Extensions for High Performance) work properly.
3075 *
3076 * This rule does NOT stand when exceeds global or memcg's hard
3077 * limit, or else a DoS attack can be taken place by spawning
3078 * lots of sockets whose usage are under minimum buffer size.
3079 */
3080 if (kind == SK_MEM_RECV) {
3081 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3082 return 1;
3083
3084 } else { /* SK_MEM_SEND */
3085 int wmem0 = sk_get_wmem0(sk, prot);
3086
3087 if (sk->sk_type == SOCK_STREAM) {
3088 if (sk->sk_wmem_queued < wmem0)
3089 return 1;
3090 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3091 return 1;
3092 }
3093 }
3094
3095 if (sk_has_memory_pressure(sk)) {
3096 u64 alloc;
3097
3098 /* The following 'average' heuristic is within the
3099 * scope of global accounting, so it only makes
3100 * sense for global memory pressure.
3101 */
3102 if (!sk_under_global_memory_pressure(sk))
3103 return 1;
3104
3105 /* Try to be fair among all the sockets under global
3106 * pressure by allowing the ones that below average
3107 * usage to raise.
3108 */
3109 alloc = sk_sockets_allocated_read_positive(sk);
3110 if (sk_prot_mem_limits(sk, 2) > alloc *
3111 sk_mem_pages(sk->sk_wmem_queued +
3112 atomic_read(&sk->sk_rmem_alloc) +
3113 sk->sk_forward_alloc))
3114 return 1;
3115 }
3116
3117suppress_allocation:
3118
3119 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3120 sk_stream_moderate_sndbuf(sk);
3121
3122 /* Fail only if socket is _under_ its sndbuf.
3123 * In this case we cannot block, so that we have to fail.
3124 */
3125 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3126 /* Force charge with __GFP_NOFAIL */
3127 if (memcg && !charged) {
3128 mem_cgroup_charge_skmem(memcg, amt,
3129 gfp_memcg_charge() | __GFP_NOFAIL);
3130 }
3131 return 1;
3132 }
3133 }
3134
3135 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3136 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3137
3138 sk_memory_allocated_sub(sk, amt);
3139
3140 if (charged)
3141 mem_cgroup_uncharge_skmem(memcg, amt);
3142
3143 return 0;
3144}
3145
3146/**
3147 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3148 * @sk: socket
3149 * @size: memory size to allocate
3150 * @kind: allocation type
3151 *
3152 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3153 * rmem allocation. This function assumes that protocols which have
3154 * memory_pressure use sk_wmem_queued as write buffer accounting.
3155 */
3156int __sk_mem_schedule(struct sock *sk, int size, int kind)
3157{
3158 int ret, amt = sk_mem_pages(size);
3159
3160 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3161 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3162 if (!ret)
3163 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3164 return ret;
3165}
3166EXPORT_SYMBOL(__sk_mem_schedule);
3167
3168/**
3169 * __sk_mem_reduce_allocated - reclaim memory_allocated
3170 * @sk: socket
3171 * @amount: number of quanta
3172 *
3173 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3174 */
3175void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3176{
3177 sk_memory_allocated_sub(sk, amount);
3178
3179 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3180 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3181
3182 if (sk_under_global_memory_pressure(sk) &&
3183 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3184 sk_leave_memory_pressure(sk);
3185}
3186
3187/**
3188 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3189 * @sk: socket
3190 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3191 */
3192void __sk_mem_reclaim(struct sock *sk, int amount)
3193{
3194 amount >>= PAGE_SHIFT;
3195 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3196 __sk_mem_reduce_allocated(sk, amount);
3197}
3198EXPORT_SYMBOL(__sk_mem_reclaim);
3199
3200int sk_set_peek_off(struct sock *sk, int val)
3201{
3202 WRITE_ONCE(sk->sk_peek_off, val);
3203 return 0;
3204}
3205EXPORT_SYMBOL_GPL(sk_set_peek_off);
3206
3207/*
3208 * Set of default routines for initialising struct proto_ops when
3209 * the protocol does not support a particular function. In certain
3210 * cases where it makes no sense for a protocol to have a "do nothing"
3211 * function, some default processing is provided.
3212 */
3213
3214int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3215{
3216 return -EOPNOTSUPP;
3217}
3218EXPORT_SYMBOL(sock_no_bind);
3219
3220int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3221 int len, int flags)
3222{
3223 return -EOPNOTSUPP;
3224}
3225EXPORT_SYMBOL(sock_no_connect);
3226
3227int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3228{
3229 return -EOPNOTSUPP;
3230}
3231EXPORT_SYMBOL(sock_no_socketpair);
3232
3233int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
3234 bool kern)
3235{
3236 return -EOPNOTSUPP;
3237}
3238EXPORT_SYMBOL(sock_no_accept);
3239
3240int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3241 int peer)
3242{
3243 return -EOPNOTSUPP;
3244}
3245EXPORT_SYMBOL(sock_no_getname);
3246
3247int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3248{
3249 return -EOPNOTSUPP;
3250}
3251EXPORT_SYMBOL(sock_no_ioctl);
3252
3253int sock_no_listen(struct socket *sock, int backlog)
3254{
3255 return -EOPNOTSUPP;
3256}
3257EXPORT_SYMBOL(sock_no_listen);
3258
3259int sock_no_shutdown(struct socket *sock, int how)
3260{
3261 return -EOPNOTSUPP;
3262}
3263EXPORT_SYMBOL(sock_no_shutdown);
3264
3265int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3266{
3267 return -EOPNOTSUPP;
3268}
3269EXPORT_SYMBOL(sock_no_sendmsg);
3270
3271int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3272{
3273 return -EOPNOTSUPP;
3274}
3275EXPORT_SYMBOL(sock_no_sendmsg_locked);
3276
3277int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3278 int flags)
3279{
3280 return -EOPNOTSUPP;
3281}
3282EXPORT_SYMBOL(sock_no_recvmsg);
3283
3284int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3285{
3286 /* Mirror missing mmap method error code */
3287 return -ENODEV;
3288}
3289EXPORT_SYMBOL(sock_no_mmap);
3290
3291/*
3292 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3293 * various sock-based usage counts.
3294 */
3295void __receive_sock(struct file *file)
3296{
3297 struct socket *sock;
3298
3299 sock = sock_from_file(file);
3300 if (sock) {
3301 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3302 sock_update_classid(&sock->sk->sk_cgrp_data);
3303 }
3304}
3305
3306/*
3307 * Default Socket Callbacks
3308 */
3309
3310static void sock_def_wakeup(struct sock *sk)
3311{
3312 struct socket_wq *wq;
3313
3314 rcu_read_lock();
3315 wq = rcu_dereference(sk->sk_wq);
3316 if (skwq_has_sleeper(wq))
3317 wake_up_interruptible_all(&wq->wait);
3318 rcu_read_unlock();
3319}
3320
3321static void sock_def_error_report(struct sock *sk)
3322{
3323 struct socket_wq *wq;
3324
3325 rcu_read_lock();
3326 wq = rcu_dereference(sk->sk_wq);
3327 if (skwq_has_sleeper(wq))
3328 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3329 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
3330 rcu_read_unlock();
3331}
3332
3333void sock_def_readable(struct sock *sk)
3334{
3335 struct socket_wq *wq;
3336
3337 trace_sk_data_ready(sk);
3338
3339 rcu_read_lock();
3340 wq = rcu_dereference(sk->sk_wq);
3341 if (skwq_has_sleeper(wq))
3342 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3343 EPOLLRDNORM | EPOLLRDBAND);
3344 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
3345 rcu_read_unlock();
3346}
3347
3348static void sock_def_write_space(struct sock *sk)
3349{
3350 struct socket_wq *wq;
3351
3352 rcu_read_lock();
3353
3354 /* Do not wake up a writer until he can make "significant"
3355 * progress. --DaveM
3356 */
3357 if (sock_writeable(sk)) {
3358 wq = rcu_dereference(sk->sk_wq);
3359 if (skwq_has_sleeper(wq))
3360 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3361 EPOLLWRNORM | EPOLLWRBAND);
3362
3363 /* Should agree with poll, otherwise some programs break */
3364 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3365 }
3366
3367 rcu_read_unlock();
3368}
3369
3370/* An optimised version of sock_def_write_space(), should only be called
3371 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3372 * ->sk_wmem_alloc.
3373 */
3374static void sock_def_write_space_wfree(struct sock *sk)
3375{
3376 /* Do not wake up a writer until he can make "significant"
3377 * progress. --DaveM
3378 */
3379 if (sock_writeable(sk)) {
3380 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3381
3382 /* rely on refcount_sub from sock_wfree() */
3383 smp_mb__after_atomic();
3384 if (wq && waitqueue_active(&wq->wait))
3385 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3386 EPOLLWRNORM | EPOLLWRBAND);
3387
3388 /* Should agree with poll, otherwise some programs break */
3389 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
3390 }
3391}
3392
3393static void sock_def_destruct(struct sock *sk)
3394{
3395}
3396
3397void sk_send_sigurg(struct sock *sk)
3398{
3399 if (sk->sk_socket && sk->sk_socket->file)
3400 if (send_sigurg(&sk->sk_socket->file->f_owner))
3401 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3402}
3403EXPORT_SYMBOL(sk_send_sigurg);
3404
3405void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3406 unsigned long expires)
3407{
3408 if (!mod_timer(timer, expires))
3409 sock_hold(sk);
3410}
3411EXPORT_SYMBOL(sk_reset_timer);
3412
3413void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3414{
3415 if (del_timer(timer))
3416 __sock_put(sk);
3417}
3418EXPORT_SYMBOL(sk_stop_timer);
3419
3420void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3421{
3422 if (del_timer_sync(timer))
3423 __sock_put(sk);
3424}
3425EXPORT_SYMBOL(sk_stop_timer_sync);
3426
3427void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3428{
3429 sk_init_common(sk);
3430 sk->sk_send_head = NULL;
3431
3432 timer_setup(&sk->sk_timer, NULL, 0);
3433
3434 sk->sk_allocation = GFP_KERNEL;
3435 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3436 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3437 sk->sk_state = TCP_CLOSE;
3438 sk->sk_use_task_frag = true;
3439 sk_set_socket(sk, sock);
3440
3441 sock_set_flag(sk, SOCK_ZAPPED);
3442
3443 if (sock) {
3444 sk->sk_type = sock->type;
3445 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3446 sock->sk = sk;
3447 } else {
3448 RCU_INIT_POINTER(sk->sk_wq, NULL);
3449 }
3450 sk->sk_uid = uid;
3451
3452 rwlock_init(&sk->sk_callback_lock);
3453 if (sk->sk_kern_sock)
3454 lockdep_set_class_and_name(
3455 &sk->sk_callback_lock,
3456 af_kern_callback_keys + sk->sk_family,
3457 af_family_kern_clock_key_strings[sk->sk_family]);
3458 else
3459 lockdep_set_class_and_name(
3460 &sk->sk_callback_lock,
3461 af_callback_keys + sk->sk_family,
3462 af_family_clock_key_strings[sk->sk_family]);
3463
3464 sk->sk_state_change = sock_def_wakeup;
3465 sk->sk_data_ready = sock_def_readable;
3466 sk->sk_write_space = sock_def_write_space;
3467 sk->sk_error_report = sock_def_error_report;
3468 sk->sk_destruct = sock_def_destruct;
3469
3470 sk->sk_frag.page = NULL;
3471 sk->sk_frag.offset = 0;
3472 sk->sk_peek_off = -1;
3473
3474 sk->sk_peer_pid = NULL;
3475 sk->sk_peer_cred = NULL;
3476 spin_lock_init(&sk->sk_peer_lock);
3477
3478 sk->sk_write_pending = 0;
3479 sk->sk_rcvlowat = 1;
3480 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3481 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3482
3483 sk->sk_stamp = SK_DEFAULT_STAMP;
3484#if BITS_PER_LONG==32
3485 seqlock_init(&sk->sk_stamp_seq);
3486#endif
3487 atomic_set(&sk->sk_zckey, 0);
3488
3489#ifdef CONFIG_NET_RX_BUSY_POLL
3490 sk->sk_napi_id = 0;
3491 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3492#endif
3493
3494 sk->sk_max_pacing_rate = ~0UL;
3495 sk->sk_pacing_rate = ~0UL;
3496 WRITE_ONCE(sk->sk_pacing_shift, 10);
3497 sk->sk_incoming_cpu = -1;
3498
3499 sk_rx_queue_clear(sk);
3500 /*
3501 * Before updating sk_refcnt, we must commit prior changes to memory
3502 * (Documentation/RCU/rculist_nulls.rst for details)
3503 */
3504 smp_wmb();
3505 refcount_set(&sk->sk_refcnt, 1);
3506 atomic_set(&sk->sk_drops, 0);
3507}
3508EXPORT_SYMBOL(sock_init_data_uid);
3509
3510void sock_init_data(struct socket *sock, struct sock *sk)
3511{
3512 kuid_t uid = sock ?
3513 SOCK_INODE(sock)->i_uid :
3514 make_kuid(sock_net(sk)->user_ns, 0);
3515
3516 sock_init_data_uid(sock, sk, uid);
3517}
3518EXPORT_SYMBOL(sock_init_data);
3519
3520void lock_sock_nested(struct sock *sk, int subclass)
3521{
3522 /* The sk_lock has mutex_lock() semantics here. */
3523 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3524
3525 might_sleep();
3526 spin_lock_bh(&sk->sk_lock.slock);
3527 if (sock_owned_by_user_nocheck(sk))
3528 __lock_sock(sk);
3529 sk->sk_lock.owned = 1;
3530 spin_unlock_bh(&sk->sk_lock.slock);
3531}
3532EXPORT_SYMBOL(lock_sock_nested);
3533
3534void release_sock(struct sock *sk)
3535{
3536 spin_lock_bh(&sk->sk_lock.slock);
3537 if (sk->sk_backlog.tail)
3538 __release_sock(sk);
3539
3540 if (sk->sk_prot->release_cb)
3541 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3542 tcp_release_cb, sk);
3543
3544 sock_release_ownership(sk);
3545 if (waitqueue_active(&sk->sk_lock.wq))
3546 wake_up(&sk->sk_lock.wq);
3547 spin_unlock_bh(&sk->sk_lock.slock);
3548}
3549EXPORT_SYMBOL(release_sock);
3550
3551bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3552{
3553 might_sleep();
3554 spin_lock_bh(&sk->sk_lock.slock);
3555
3556 if (!sock_owned_by_user_nocheck(sk)) {
3557 /*
3558 * Fast path return with bottom halves disabled and
3559 * sock::sk_lock.slock held.
3560 *
3561 * The 'mutex' is not contended and holding
3562 * sock::sk_lock.slock prevents all other lockers to
3563 * proceed so the corresponding unlock_sock_fast() can
3564 * avoid the slow path of release_sock() completely and
3565 * just release slock.
3566 *
3567 * From a semantical POV this is equivalent to 'acquiring'
3568 * the 'mutex', hence the corresponding lockdep
3569 * mutex_release() has to happen in the fast path of
3570 * unlock_sock_fast().
3571 */
3572 return false;
3573 }
3574
3575 __lock_sock(sk);
3576 sk->sk_lock.owned = 1;
3577 __acquire(&sk->sk_lock.slock);
3578 spin_unlock_bh(&sk->sk_lock.slock);
3579 return true;
3580}
3581EXPORT_SYMBOL(__lock_sock_fast);
3582
3583int sock_gettstamp(struct socket *sock, void __user *userstamp,
3584 bool timeval, bool time32)
3585{
3586 struct sock *sk = sock->sk;
3587 struct timespec64 ts;
3588
3589 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3590 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3591 if (ts.tv_sec == -1)
3592 return -ENOENT;
3593 if (ts.tv_sec == 0) {
3594 ktime_t kt = ktime_get_real();
3595 sock_write_timestamp(sk, kt);
3596 ts = ktime_to_timespec64(kt);
3597 }
3598
3599 if (timeval)
3600 ts.tv_nsec /= 1000;
3601
3602#ifdef CONFIG_COMPAT_32BIT_TIME
3603 if (time32)
3604 return put_old_timespec32(&ts, userstamp);
3605#endif
3606#ifdef CONFIG_SPARC64
3607 /* beware of padding in sparc64 timeval */
3608 if (timeval && !in_compat_syscall()) {
3609 struct __kernel_old_timeval __user tv = {
3610 .tv_sec = ts.tv_sec,
3611 .tv_usec = ts.tv_nsec,
3612 };
3613 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3614 return -EFAULT;
3615 return 0;
3616 }
3617#endif
3618 return put_timespec64(&ts, userstamp);
3619}
3620EXPORT_SYMBOL(sock_gettstamp);
3621
3622void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3623{
3624 if (!sock_flag(sk, flag)) {
3625 unsigned long previous_flags = sk->sk_flags;
3626
3627 sock_set_flag(sk, flag);
3628 /*
3629 * we just set one of the two flags which require net
3630 * time stamping, but time stamping might have been on
3631 * already because of the other one
3632 */
3633 if (sock_needs_netstamp(sk) &&
3634 !(previous_flags & SK_FLAGS_TIMESTAMP))
3635 net_enable_timestamp();
3636 }
3637}
3638
3639int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3640 int level, int type)
3641{
3642 struct sock_exterr_skb *serr;
3643 struct sk_buff *skb;
3644 int copied, err;
3645
3646 err = -EAGAIN;
3647 skb = sock_dequeue_err_skb(sk);
3648 if (skb == NULL)
3649 goto out;
3650
3651 copied = skb->len;
3652 if (copied > len) {
3653 msg->msg_flags |= MSG_TRUNC;
3654 copied = len;
3655 }
3656 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3657 if (err)
3658 goto out_free_skb;
3659
3660 sock_recv_timestamp(msg, sk, skb);
3661
3662 serr = SKB_EXT_ERR(skb);
3663 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3664
3665 msg->msg_flags |= MSG_ERRQUEUE;
3666 err = copied;
3667
3668out_free_skb:
3669 kfree_skb(skb);
3670out:
3671 return err;
3672}
3673EXPORT_SYMBOL(sock_recv_errqueue);
3674
3675/*
3676 * Get a socket option on an socket.
3677 *
3678 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3679 * asynchronous errors should be reported by getsockopt. We assume
3680 * this means if you specify SO_ERROR (otherwise whats the point of it).
3681 */
3682int sock_common_getsockopt(struct socket *sock, int level, int optname,
3683 char __user *optval, int __user *optlen)
3684{
3685 struct sock *sk = sock->sk;
3686
3687 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3688 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3689}
3690EXPORT_SYMBOL(sock_common_getsockopt);
3691
3692int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3693 int flags)
3694{
3695 struct sock *sk = sock->sk;
3696 int addr_len = 0;
3697 int err;
3698
3699 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3700 if (err >= 0)
3701 msg->msg_namelen = addr_len;
3702 return err;
3703}
3704EXPORT_SYMBOL(sock_common_recvmsg);
3705
3706/*
3707 * Set socket options on an inet socket.
3708 */
3709int sock_common_setsockopt(struct socket *sock, int level, int optname,
3710 sockptr_t optval, unsigned int optlen)
3711{
3712 struct sock *sk = sock->sk;
3713
3714 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3715 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3716}
3717EXPORT_SYMBOL(sock_common_setsockopt);
3718
3719void sk_common_release(struct sock *sk)
3720{
3721 if (sk->sk_prot->destroy)
3722 sk->sk_prot->destroy(sk);
3723
3724 /*
3725 * Observation: when sk_common_release is called, processes have
3726 * no access to socket. But net still has.
3727 * Step one, detach it from networking:
3728 *
3729 * A. Remove from hash tables.
3730 */
3731
3732 sk->sk_prot->unhash(sk);
3733
3734 /*
3735 * In this point socket cannot receive new packets, but it is possible
3736 * that some packets are in flight because some CPU runs receiver and
3737 * did hash table lookup before we unhashed socket. They will achieve
3738 * receive queue and will be purged by socket destructor.
3739 *
3740 * Also we still have packets pending on receive queue and probably,
3741 * our own packets waiting in device queues. sock_destroy will drain
3742 * receive queue, but transmitted packets will delay socket destruction
3743 * until the last reference will be released.
3744 */
3745
3746 sock_orphan(sk);
3747
3748 xfrm_sk_free_policy(sk);
3749
3750 sock_put(sk);
3751}
3752EXPORT_SYMBOL(sk_common_release);
3753
3754void sk_get_meminfo(const struct sock *sk, u32 *mem)
3755{
3756 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3757
3758 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3759 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3760 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3761 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3762 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3763 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3764 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3765 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3766 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3767}
3768
3769#ifdef CONFIG_PROC_FS
3770static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3771
3772int sock_prot_inuse_get(struct net *net, struct proto *prot)
3773{
3774 int cpu, idx = prot->inuse_idx;
3775 int res = 0;
3776
3777 for_each_possible_cpu(cpu)
3778 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3779
3780 return res >= 0 ? res : 0;
3781}
3782EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3783
3784int sock_inuse_get(struct net *net)
3785{
3786 int cpu, res = 0;
3787
3788 for_each_possible_cpu(cpu)
3789 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3790
3791 return res;
3792}
3793
3794EXPORT_SYMBOL_GPL(sock_inuse_get);
3795
3796static int __net_init sock_inuse_init_net(struct net *net)
3797{
3798 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3799 if (net->core.prot_inuse == NULL)
3800 return -ENOMEM;
3801 return 0;
3802}
3803
3804static void __net_exit sock_inuse_exit_net(struct net *net)
3805{
3806 free_percpu(net->core.prot_inuse);
3807}
3808
3809static struct pernet_operations net_inuse_ops = {
3810 .init = sock_inuse_init_net,
3811 .exit = sock_inuse_exit_net,
3812};
3813
3814static __init int net_inuse_init(void)
3815{
3816 if (register_pernet_subsys(&net_inuse_ops))
3817 panic("Cannot initialize net inuse counters");
3818
3819 return 0;
3820}
3821
3822core_initcall(net_inuse_init);
3823
3824static int assign_proto_idx(struct proto *prot)
3825{
3826 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3827
3828 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3829 pr_err("PROTO_INUSE_NR exhausted\n");
3830 return -ENOSPC;
3831 }
3832
3833 set_bit(prot->inuse_idx, proto_inuse_idx);
3834 return 0;
3835}
3836
3837static void release_proto_idx(struct proto *prot)
3838{
3839 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3840 clear_bit(prot->inuse_idx, proto_inuse_idx);
3841}
3842#else
3843static inline int assign_proto_idx(struct proto *prot)
3844{
3845 return 0;
3846}
3847
3848static inline void release_proto_idx(struct proto *prot)
3849{
3850}
3851
3852#endif
3853
3854static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3855{
3856 if (!twsk_prot)
3857 return;
3858 kfree(twsk_prot->twsk_slab_name);
3859 twsk_prot->twsk_slab_name = NULL;
3860 kmem_cache_destroy(twsk_prot->twsk_slab);
3861 twsk_prot->twsk_slab = NULL;
3862}
3863
3864static int tw_prot_init(const struct proto *prot)
3865{
3866 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3867
3868 if (!twsk_prot)
3869 return 0;
3870
3871 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3872 prot->name);
3873 if (!twsk_prot->twsk_slab_name)
3874 return -ENOMEM;
3875
3876 twsk_prot->twsk_slab =
3877 kmem_cache_create(twsk_prot->twsk_slab_name,
3878 twsk_prot->twsk_obj_size, 0,
3879 SLAB_ACCOUNT | prot->slab_flags,
3880 NULL);
3881 if (!twsk_prot->twsk_slab) {
3882 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
3883 prot->name);
3884 return -ENOMEM;
3885 }
3886
3887 return 0;
3888}
3889
3890static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3891{
3892 if (!rsk_prot)
3893 return;
3894 kfree(rsk_prot->slab_name);
3895 rsk_prot->slab_name = NULL;
3896 kmem_cache_destroy(rsk_prot->slab);
3897 rsk_prot->slab = NULL;
3898}
3899
3900static int req_prot_init(const struct proto *prot)
3901{
3902 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3903
3904 if (!rsk_prot)
3905 return 0;
3906
3907 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3908 prot->name);
3909 if (!rsk_prot->slab_name)
3910 return -ENOMEM;
3911
3912 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3913 rsk_prot->obj_size, 0,
3914 SLAB_ACCOUNT | prot->slab_flags,
3915 NULL);
3916
3917 if (!rsk_prot->slab) {
3918 pr_crit("%s: Can't create request sock SLAB cache!\n",
3919 prot->name);
3920 return -ENOMEM;
3921 }
3922 return 0;
3923}
3924
3925int proto_register(struct proto *prot, int alloc_slab)
3926{
3927 int ret = -ENOBUFS;
3928
3929 if (prot->memory_allocated && !prot->sysctl_mem) {
3930 pr_err("%s: missing sysctl_mem\n", prot->name);
3931 return -EINVAL;
3932 }
3933 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
3934 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
3935 return -EINVAL;
3936 }
3937 if (alloc_slab) {
3938 prot->slab = kmem_cache_create_usercopy(prot->name,
3939 prot->obj_size, 0,
3940 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3941 prot->slab_flags,
3942 prot->useroffset, prot->usersize,
3943 NULL);
3944
3945 if (prot->slab == NULL) {
3946 pr_crit("%s: Can't create sock SLAB cache!\n",
3947 prot->name);
3948 goto out;
3949 }
3950
3951 if (req_prot_init(prot))
3952 goto out_free_request_sock_slab;
3953
3954 if (tw_prot_init(prot))
3955 goto out_free_timewait_sock_slab;
3956 }
3957
3958 mutex_lock(&proto_list_mutex);
3959 ret = assign_proto_idx(prot);
3960 if (ret) {
3961 mutex_unlock(&proto_list_mutex);
3962 goto out_free_timewait_sock_slab;
3963 }
3964 list_add(&prot->node, &proto_list);
3965 mutex_unlock(&proto_list_mutex);
3966 return ret;
3967
3968out_free_timewait_sock_slab:
3969 if (alloc_slab)
3970 tw_prot_cleanup(prot->twsk_prot);
3971out_free_request_sock_slab:
3972 if (alloc_slab) {
3973 req_prot_cleanup(prot->rsk_prot);
3974
3975 kmem_cache_destroy(prot->slab);
3976 prot->slab = NULL;
3977 }
3978out:
3979 return ret;
3980}
3981EXPORT_SYMBOL(proto_register);
3982
3983void proto_unregister(struct proto *prot)
3984{
3985 mutex_lock(&proto_list_mutex);
3986 release_proto_idx(prot);
3987 list_del(&prot->node);
3988 mutex_unlock(&proto_list_mutex);
3989
3990 kmem_cache_destroy(prot->slab);
3991 prot->slab = NULL;
3992
3993 req_prot_cleanup(prot->rsk_prot);
3994 tw_prot_cleanup(prot->twsk_prot);
3995}
3996EXPORT_SYMBOL(proto_unregister);
3997
3998int sock_load_diag_module(int family, int protocol)
3999{
4000 if (!protocol) {
4001 if (!sock_is_registered(family))
4002 return -ENOENT;
4003
4004 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4005 NETLINK_SOCK_DIAG, family);
4006 }
4007
4008#ifdef CONFIG_INET
4009 if (family == AF_INET &&
4010 protocol != IPPROTO_RAW &&
4011 protocol < MAX_INET_PROTOS &&
4012 !rcu_access_pointer(inet_protos[protocol]))
4013 return -ENOENT;
4014#endif
4015
4016 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4017 NETLINK_SOCK_DIAG, family, protocol);
4018}
4019EXPORT_SYMBOL(sock_load_diag_module);
4020
4021#ifdef CONFIG_PROC_FS
4022static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4023 __acquires(proto_list_mutex)
4024{
4025 mutex_lock(&proto_list_mutex);
4026 return seq_list_start_head(&proto_list, *pos);
4027}
4028
4029static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4030{
4031 return seq_list_next(v, &proto_list, pos);
4032}
4033
4034static void proto_seq_stop(struct seq_file *seq, void *v)
4035 __releases(proto_list_mutex)
4036{
4037 mutex_unlock(&proto_list_mutex);
4038}
4039
4040static char proto_method_implemented(const void *method)
4041{
4042 return method == NULL ? 'n' : 'y';
4043}
4044static long sock_prot_memory_allocated(struct proto *proto)
4045{
4046 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4047}
4048
4049static const char *sock_prot_memory_pressure(struct proto *proto)
4050{
4051 return proto->memory_pressure != NULL ?
4052 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4053}
4054
4055static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4056{
4057
4058 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4059 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4060 proto->name,
4061 proto->obj_size,
4062 sock_prot_inuse_get(seq_file_net(seq), proto),
4063 sock_prot_memory_allocated(proto),
4064 sock_prot_memory_pressure(proto),
4065 proto->max_header,
4066 proto->slab == NULL ? "no" : "yes",
4067 module_name(proto->owner),
4068 proto_method_implemented(proto->close),
4069 proto_method_implemented(proto->connect),
4070 proto_method_implemented(proto->disconnect),
4071 proto_method_implemented(proto->accept),
4072 proto_method_implemented(proto->ioctl),
4073 proto_method_implemented(proto->init),
4074 proto_method_implemented(proto->destroy),
4075 proto_method_implemented(proto->shutdown),
4076 proto_method_implemented(proto->setsockopt),
4077 proto_method_implemented(proto->getsockopt),
4078 proto_method_implemented(proto->sendmsg),
4079 proto_method_implemented(proto->recvmsg),
4080 proto_method_implemented(proto->bind),
4081 proto_method_implemented(proto->backlog_rcv),
4082 proto_method_implemented(proto->hash),
4083 proto_method_implemented(proto->unhash),
4084 proto_method_implemented(proto->get_port),
4085 proto_method_implemented(proto->enter_memory_pressure));
4086}
4087
4088static int proto_seq_show(struct seq_file *seq, void *v)
4089{
4090 if (v == &proto_list)
4091 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4092 "protocol",
4093 "size",
4094 "sockets",
4095 "memory",
4096 "press",
4097 "maxhdr",
4098 "slab",
4099 "module",
4100 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4101 else
4102 proto_seq_printf(seq, list_entry(v, struct proto, node));
4103 return 0;
4104}
4105
4106static const struct seq_operations proto_seq_ops = {
4107 .start = proto_seq_start,
4108 .next = proto_seq_next,
4109 .stop = proto_seq_stop,
4110 .show = proto_seq_show,
4111};
4112
4113static __net_init int proto_init_net(struct net *net)
4114{
4115 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4116 sizeof(struct seq_net_private)))
4117 return -ENOMEM;
4118
4119 return 0;
4120}
4121
4122static __net_exit void proto_exit_net(struct net *net)
4123{
4124 remove_proc_entry("protocols", net->proc_net);
4125}
4126
4127
4128static __net_initdata struct pernet_operations proto_net_ops = {
4129 .init = proto_init_net,
4130 .exit = proto_exit_net,
4131};
4132
4133static int __init proto_init(void)
4134{
4135 return register_pernet_subsys(&proto_net_ops);
4136}
4137
4138subsys_initcall(proto_init);
4139
4140#endif /* PROC_FS */
4141
4142#ifdef CONFIG_NET_RX_BUSY_POLL
4143bool sk_busy_loop_end(void *p, unsigned long start_time)
4144{
4145 struct sock *sk = p;
4146
4147 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4148 return true;
4149
4150 if (sk_is_udp(sk) &&
4151 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4152 return true;
4153
4154 return sk_busy_loop_timeout(sk, start_time);
4155}
4156EXPORT_SYMBOL(sk_busy_loop_end);
4157#endif /* CONFIG_NET_RX_BUSY_POLL */
4158
4159int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4160{
4161 if (!sk->sk_prot->bind_add)
4162 return -EOPNOTSUPP;
4163 return sk->sk_prot->bind_add(sk, addr, addr_len);
4164}
4165EXPORT_SYMBOL(sock_bind_add);
4166
4167/* Copy 'size' bytes from userspace and return `size` back to userspace */
4168int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4169 void __user *arg, void *karg, size_t size)
4170{
4171 int ret;
4172
4173 if (copy_from_user(karg, arg, size))
4174 return -EFAULT;
4175
4176 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4177 if (ret)
4178 return ret;
4179
4180 if (copy_to_user(arg, karg, size))
4181 return -EFAULT;
4182
4183 return 0;
4184}
4185EXPORT_SYMBOL(sock_ioctl_inout);
4186
4187/* This is the most common ioctl prep function, where the result (4 bytes) is
4188 * copied back to userspace if the ioctl() returns successfully. No input is
4189 * copied from userspace as input argument.
4190 */
4191static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4192{
4193 int ret, karg = 0;
4194
4195 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4196 if (ret)
4197 return ret;
4198
4199 return put_user(karg, (int __user *)arg);
4200}
4201
4202/* A wrapper around sock ioctls, which copies the data from userspace
4203 * (depending on the protocol/ioctl), and copies back the result to userspace.
4204 * The main motivation for this function is to pass kernel memory to the
4205 * protocol ioctl callbacks, instead of userspace memory.
4206 */
4207int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4208{
4209 int rc = 1;
4210
4211 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4212 rc = ipmr_sk_ioctl(sk, cmd, arg);
4213 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4214 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4215 else if (sk_is_phonet(sk))
4216 rc = phonet_sk_ioctl(sk, cmd, arg);
4217
4218 /* If ioctl was processed, returns its value */
4219 if (rc <= 0)
4220 return rc;
4221
4222 /* Otherwise call the default handler */
4223 return sock_ioctl_out(sk, cmd, arg);
4224}
4225EXPORT_SYMBOL(sk_ioctl);
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
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 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92#include <linux/capability.h>
93#include <linux/errno.h>
94#include <linux/types.h>
95#include <linux/socket.h>
96#include <linux/in.h>
97#include <linux/kernel.h>
98#include <linux/module.h>
99#include <linux/proc_fs.h>
100#include <linux/seq_file.h>
101#include <linux/sched.h>
102#include <linux/timer.h>
103#include <linux/string.h>
104#include <linux/sockios.h>
105#include <linux/net.h>
106#include <linux/mm.h>
107#include <linux/slab.h>
108#include <linux/interrupt.h>
109#include <linux/poll.h>
110#include <linux/tcp.h>
111#include <linux/init.h>
112#include <linux/highmem.h>
113#include <linux/user_namespace.h>
114
115#include <asm/uaccess.h>
116#include <asm/system.h>
117
118#include <linux/netdevice.h>
119#include <net/protocol.h>
120#include <linux/skbuff.h>
121#include <net/net_namespace.h>
122#include <net/request_sock.h>
123#include <net/sock.h>
124#include <linux/net_tstamp.h>
125#include <net/xfrm.h>
126#include <linux/ipsec.h>
127#include <net/cls_cgroup.h>
128
129#include <linux/filter.h>
130
131#include <trace/events/sock.h>
132
133#ifdef CONFIG_INET
134#include <net/tcp.h>
135#endif
136
137/*
138 * Each address family might have different locking rules, so we have
139 * one slock key per address family:
140 */
141static struct lock_class_key af_family_keys[AF_MAX];
142static struct lock_class_key af_family_slock_keys[AF_MAX];
143
144/*
145 * Make lock validator output more readable. (we pre-construct these
146 * strings build-time, so that runtime initialization of socket
147 * locks is fast):
148 */
149static const char *const af_family_key_strings[AF_MAX+1] = {
150 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
151 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
152 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
153 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
154 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
155 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
156 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
157 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
158 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
159 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
160 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
161 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
162 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
163 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
164};
165static const char *const af_family_slock_key_strings[AF_MAX+1] = {
166 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
167 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
168 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
169 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
170 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
171 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
172 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
173 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
174 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
175 "slock-27" , "slock-28" , "slock-AF_CAN" ,
176 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
177 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
178 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
179 "slock-AF_NFC" , "slock-AF_MAX"
180};
181static const char *const af_family_clock_key_strings[AF_MAX+1] = {
182 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
183 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
184 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
185 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
186 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
187 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
188 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
189 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
190 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
191 "clock-27" , "clock-28" , "clock-AF_CAN" ,
192 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
193 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
194 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
195 "clock-AF_NFC" , "clock-AF_MAX"
196};
197
198/*
199 * sk_callback_lock locking rules are per-address-family,
200 * so split the lock classes by using a per-AF key:
201 */
202static struct lock_class_key af_callback_keys[AF_MAX];
203
204/* Take into consideration the size of the struct sk_buff overhead in the
205 * determination of these values, since that is non-constant across
206 * platforms. This makes socket queueing behavior and performance
207 * not depend upon such differences.
208 */
209#define _SK_MEM_PACKETS 256
210#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
211#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
213
214/* Run time adjustable parameters. */
215__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
216__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
217__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
218__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
219
220/* Maximal space eaten by iovec or ancillary data plus some space */
221int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
222EXPORT_SYMBOL(sysctl_optmem_max);
223
224#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
225int net_cls_subsys_id = -1;
226EXPORT_SYMBOL_GPL(net_cls_subsys_id);
227#endif
228
229static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
230{
231 struct timeval tv;
232
233 if (optlen < sizeof(tv))
234 return -EINVAL;
235 if (copy_from_user(&tv, optval, sizeof(tv)))
236 return -EFAULT;
237 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
238 return -EDOM;
239
240 if (tv.tv_sec < 0) {
241 static int warned __read_mostly;
242
243 *timeo_p = 0;
244 if (warned < 10 && net_ratelimit()) {
245 warned++;
246 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
247 "tries to set negative timeout\n",
248 current->comm, task_pid_nr(current));
249 }
250 return 0;
251 }
252 *timeo_p = MAX_SCHEDULE_TIMEOUT;
253 if (tv.tv_sec == 0 && tv.tv_usec == 0)
254 return 0;
255 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
256 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
257 return 0;
258}
259
260static void sock_warn_obsolete_bsdism(const char *name)
261{
262 static int warned;
263 static char warncomm[TASK_COMM_LEN];
264 if (strcmp(warncomm, current->comm) && warned < 5) {
265 strcpy(warncomm, current->comm);
266 printk(KERN_WARNING "process `%s' is using obsolete "
267 "%s SO_BSDCOMPAT\n", warncomm, name);
268 warned++;
269 }
270}
271
272static void sock_disable_timestamp(struct sock *sk, int flag)
273{
274 if (sock_flag(sk, flag)) {
275 sock_reset_flag(sk, flag);
276 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
277 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
278 net_disable_timestamp();
279 }
280 }
281}
282
283
284int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
285{
286 int err;
287 int skb_len;
288 unsigned long flags;
289 struct sk_buff_head *list = &sk->sk_receive_queue;
290
291 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
292 number of warnings when compiling with -W --ANK
293 */
294 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
295 (unsigned)sk->sk_rcvbuf) {
296 atomic_inc(&sk->sk_drops);
297 trace_sock_rcvqueue_full(sk, skb);
298 return -ENOMEM;
299 }
300
301 err = sk_filter(sk, skb);
302 if (err)
303 return err;
304
305 if (!sk_rmem_schedule(sk, skb->truesize)) {
306 atomic_inc(&sk->sk_drops);
307 return -ENOBUFS;
308 }
309
310 skb->dev = NULL;
311 skb_set_owner_r(skb, sk);
312
313 /* Cache the SKB length before we tack it onto the receive
314 * queue. Once it is added it no longer belongs to us and
315 * may be freed by other threads of control pulling packets
316 * from the queue.
317 */
318 skb_len = skb->len;
319
320 /* we escape from rcu protected region, make sure we dont leak
321 * a norefcounted dst
322 */
323 skb_dst_force(skb);
324
325 spin_lock_irqsave(&list->lock, flags);
326 skb->dropcount = atomic_read(&sk->sk_drops);
327 __skb_queue_tail(list, skb);
328 spin_unlock_irqrestore(&list->lock, flags);
329
330 if (!sock_flag(sk, SOCK_DEAD))
331 sk->sk_data_ready(sk, skb_len);
332 return 0;
333}
334EXPORT_SYMBOL(sock_queue_rcv_skb);
335
336int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
337{
338 int rc = NET_RX_SUCCESS;
339
340 if (sk_filter(sk, skb))
341 goto discard_and_relse;
342
343 skb->dev = NULL;
344
345 if (sk_rcvqueues_full(sk, skb)) {
346 atomic_inc(&sk->sk_drops);
347 goto discard_and_relse;
348 }
349 if (nested)
350 bh_lock_sock_nested(sk);
351 else
352 bh_lock_sock(sk);
353 if (!sock_owned_by_user(sk)) {
354 /*
355 * trylock + unlock semantics:
356 */
357 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
358
359 rc = sk_backlog_rcv(sk, skb);
360
361 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
362 } else if (sk_add_backlog(sk, skb)) {
363 bh_unlock_sock(sk);
364 atomic_inc(&sk->sk_drops);
365 goto discard_and_relse;
366 }
367
368 bh_unlock_sock(sk);
369out:
370 sock_put(sk);
371 return rc;
372discard_and_relse:
373 kfree_skb(skb);
374 goto out;
375}
376EXPORT_SYMBOL(sk_receive_skb);
377
378void sk_reset_txq(struct sock *sk)
379{
380 sk_tx_queue_clear(sk);
381}
382EXPORT_SYMBOL(sk_reset_txq);
383
384struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
385{
386 struct dst_entry *dst = __sk_dst_get(sk);
387
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
389 sk_tx_queue_clear(sk);
390 rcu_assign_pointer(sk->sk_dst_cache, NULL);
391 dst_release(dst);
392 return NULL;
393 }
394
395 return dst;
396}
397EXPORT_SYMBOL(__sk_dst_check);
398
399struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
400{
401 struct dst_entry *dst = sk_dst_get(sk);
402
403 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
404 sk_dst_reset(sk);
405 dst_release(dst);
406 return NULL;
407 }
408
409 return dst;
410}
411EXPORT_SYMBOL(sk_dst_check);
412
413static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
414{
415 int ret = -ENOPROTOOPT;
416#ifdef CONFIG_NETDEVICES
417 struct net *net = sock_net(sk);
418 char devname[IFNAMSIZ];
419 int index;
420
421 /* Sorry... */
422 ret = -EPERM;
423 if (!capable(CAP_NET_RAW))
424 goto out;
425
426 ret = -EINVAL;
427 if (optlen < 0)
428 goto out;
429
430 /* Bind this socket to a particular device like "eth0",
431 * as specified in the passed interface name. If the
432 * name is "" or the option length is zero the socket
433 * is not bound.
434 */
435 if (optlen > IFNAMSIZ - 1)
436 optlen = IFNAMSIZ - 1;
437 memset(devname, 0, sizeof(devname));
438
439 ret = -EFAULT;
440 if (copy_from_user(devname, optval, optlen))
441 goto out;
442
443 index = 0;
444 if (devname[0] != '\0') {
445 struct net_device *dev;
446
447 rcu_read_lock();
448 dev = dev_get_by_name_rcu(net, devname);
449 if (dev)
450 index = dev->ifindex;
451 rcu_read_unlock();
452 ret = -ENODEV;
453 if (!dev)
454 goto out;
455 }
456
457 lock_sock(sk);
458 sk->sk_bound_dev_if = index;
459 sk_dst_reset(sk);
460 release_sock(sk);
461
462 ret = 0;
463
464out:
465#endif
466
467 return ret;
468}
469
470static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
471{
472 if (valbool)
473 sock_set_flag(sk, bit);
474 else
475 sock_reset_flag(sk, bit);
476}
477
478/*
479 * This is meant for all protocols to use and covers goings on
480 * at the socket level. Everything here is generic.
481 */
482
483int sock_setsockopt(struct socket *sock, int level, int optname,
484 char __user *optval, unsigned int optlen)
485{
486 struct sock *sk = sock->sk;
487 int val;
488 int valbool;
489 struct linger ling;
490 int ret = 0;
491
492 /*
493 * Options without arguments
494 */
495
496 if (optname == SO_BINDTODEVICE)
497 return sock_bindtodevice(sk, optval, optlen);
498
499 if (optlen < sizeof(int))
500 return -EINVAL;
501
502 if (get_user(val, (int __user *)optval))
503 return -EFAULT;
504
505 valbool = val ? 1 : 0;
506
507 lock_sock(sk);
508
509 switch (optname) {
510 case SO_DEBUG:
511 if (val && !capable(CAP_NET_ADMIN))
512 ret = -EACCES;
513 else
514 sock_valbool_flag(sk, SOCK_DBG, valbool);
515 break;
516 case SO_REUSEADDR:
517 sk->sk_reuse = valbool;
518 break;
519 case SO_TYPE:
520 case SO_PROTOCOL:
521 case SO_DOMAIN:
522 case SO_ERROR:
523 ret = -ENOPROTOOPT;
524 break;
525 case SO_DONTROUTE:
526 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
527 break;
528 case SO_BROADCAST:
529 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
530 break;
531 case SO_SNDBUF:
532 /* Don't error on this BSD doesn't and if you think
533 about it this is right. Otherwise apps have to
534 play 'guess the biggest size' games. RCVBUF/SNDBUF
535 are treated in BSD as hints */
536
537 if (val > sysctl_wmem_max)
538 val = sysctl_wmem_max;
539set_sndbuf:
540 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
541 if ((val * 2) < SOCK_MIN_SNDBUF)
542 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
543 else
544 sk->sk_sndbuf = val * 2;
545
546 /*
547 * Wake up sending tasks if we
548 * upped the value.
549 */
550 sk->sk_write_space(sk);
551 break;
552
553 case SO_SNDBUFFORCE:
554 if (!capable(CAP_NET_ADMIN)) {
555 ret = -EPERM;
556 break;
557 }
558 goto set_sndbuf;
559
560 case SO_RCVBUF:
561 /* Don't error on this BSD doesn't and if you think
562 about it this is right. Otherwise apps have to
563 play 'guess the biggest size' games. RCVBUF/SNDBUF
564 are treated in BSD as hints */
565
566 if (val > sysctl_rmem_max)
567 val = sysctl_rmem_max;
568set_rcvbuf:
569 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
570 /*
571 * We double it on the way in to account for
572 * "struct sk_buff" etc. overhead. Applications
573 * assume that the SO_RCVBUF setting they make will
574 * allow that much actual data to be received on that
575 * socket.
576 *
577 * Applications are unaware that "struct sk_buff" and
578 * other overheads allocate from the receive buffer
579 * during socket buffer allocation.
580 *
581 * And after considering the possible alternatives,
582 * returning the value we actually used in getsockopt
583 * is the most desirable behavior.
584 */
585 if ((val * 2) < SOCK_MIN_RCVBUF)
586 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
587 else
588 sk->sk_rcvbuf = val * 2;
589 break;
590
591 case SO_RCVBUFFORCE:
592 if (!capable(CAP_NET_ADMIN)) {
593 ret = -EPERM;
594 break;
595 }
596 goto set_rcvbuf;
597
598 case SO_KEEPALIVE:
599#ifdef CONFIG_INET
600 if (sk->sk_protocol == IPPROTO_TCP)
601 tcp_set_keepalive(sk, valbool);
602#endif
603 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
604 break;
605
606 case SO_OOBINLINE:
607 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
608 break;
609
610 case SO_NO_CHECK:
611 sk->sk_no_check = valbool;
612 break;
613
614 case SO_PRIORITY:
615 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
616 sk->sk_priority = val;
617 else
618 ret = -EPERM;
619 break;
620
621 case SO_LINGER:
622 if (optlen < sizeof(ling)) {
623 ret = -EINVAL; /* 1003.1g */
624 break;
625 }
626 if (copy_from_user(&ling, optval, sizeof(ling))) {
627 ret = -EFAULT;
628 break;
629 }
630 if (!ling.l_onoff)
631 sock_reset_flag(sk, SOCK_LINGER);
632 else {
633#if (BITS_PER_LONG == 32)
634 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
635 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
636 else
637#endif
638 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
639 sock_set_flag(sk, SOCK_LINGER);
640 }
641 break;
642
643 case SO_BSDCOMPAT:
644 sock_warn_obsolete_bsdism("setsockopt");
645 break;
646
647 case SO_PASSCRED:
648 if (valbool)
649 set_bit(SOCK_PASSCRED, &sock->flags);
650 else
651 clear_bit(SOCK_PASSCRED, &sock->flags);
652 break;
653
654 case SO_TIMESTAMP:
655 case SO_TIMESTAMPNS:
656 if (valbool) {
657 if (optname == SO_TIMESTAMP)
658 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
659 else
660 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
661 sock_set_flag(sk, SOCK_RCVTSTAMP);
662 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
663 } else {
664 sock_reset_flag(sk, SOCK_RCVTSTAMP);
665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
666 }
667 break;
668
669 case SO_TIMESTAMPING:
670 if (val & ~SOF_TIMESTAMPING_MASK) {
671 ret = -EINVAL;
672 break;
673 }
674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
675 val & SOF_TIMESTAMPING_TX_HARDWARE);
676 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
677 val & SOF_TIMESTAMPING_TX_SOFTWARE);
678 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
679 val & SOF_TIMESTAMPING_RX_HARDWARE);
680 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
681 sock_enable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 else
684 sock_disable_timestamp(sk,
685 SOCK_TIMESTAMPING_RX_SOFTWARE);
686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
687 val & SOF_TIMESTAMPING_SOFTWARE);
688 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
689 val & SOF_TIMESTAMPING_SYS_HARDWARE);
690 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
691 val & SOF_TIMESTAMPING_RAW_HARDWARE);
692 break;
693
694 case SO_RCVLOWAT:
695 if (val < 0)
696 val = INT_MAX;
697 sk->sk_rcvlowat = val ? : 1;
698 break;
699
700 case SO_RCVTIMEO:
701 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
702 break;
703
704 case SO_SNDTIMEO:
705 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
706 break;
707
708 case SO_ATTACH_FILTER:
709 ret = -EINVAL;
710 if (optlen == sizeof(struct sock_fprog)) {
711 struct sock_fprog fprog;
712
713 ret = -EFAULT;
714 if (copy_from_user(&fprog, optval, sizeof(fprog)))
715 break;
716
717 ret = sk_attach_filter(&fprog, sk);
718 }
719 break;
720
721 case SO_DETACH_FILTER:
722 ret = sk_detach_filter(sk);
723 break;
724
725 case SO_PASSSEC:
726 if (valbool)
727 set_bit(SOCK_PASSSEC, &sock->flags);
728 else
729 clear_bit(SOCK_PASSSEC, &sock->flags);
730 break;
731 case SO_MARK:
732 if (!capable(CAP_NET_ADMIN))
733 ret = -EPERM;
734 else
735 sk->sk_mark = val;
736 break;
737
738 /* We implement the SO_SNDLOWAT etc to
739 not be settable (1003.1g 5.3) */
740 case SO_RXQ_OVFL:
741 if (valbool)
742 sock_set_flag(sk, SOCK_RXQ_OVFL);
743 else
744 sock_reset_flag(sk, SOCK_RXQ_OVFL);
745 break;
746 default:
747 ret = -ENOPROTOOPT;
748 break;
749 }
750 release_sock(sk);
751 return ret;
752}
753EXPORT_SYMBOL(sock_setsockopt);
754
755
756void cred_to_ucred(struct pid *pid, const struct cred *cred,
757 struct ucred *ucred)
758{
759 ucred->pid = pid_vnr(pid);
760 ucred->uid = ucred->gid = -1;
761 if (cred) {
762 struct user_namespace *current_ns = current_user_ns();
763
764 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
765 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
766 }
767}
768EXPORT_SYMBOL_GPL(cred_to_ucred);
769
770int sock_getsockopt(struct socket *sock, int level, int optname,
771 char __user *optval, int __user *optlen)
772{
773 struct sock *sk = sock->sk;
774
775 union {
776 int val;
777 struct linger ling;
778 struct timeval tm;
779 } v;
780
781 int lv = sizeof(int);
782 int len;
783
784 if (get_user(len, optlen))
785 return -EFAULT;
786 if (len < 0)
787 return -EINVAL;
788
789 memset(&v, 0, sizeof(v));
790
791 switch (optname) {
792 case SO_DEBUG:
793 v.val = sock_flag(sk, SOCK_DBG);
794 break;
795
796 case SO_DONTROUTE:
797 v.val = sock_flag(sk, SOCK_LOCALROUTE);
798 break;
799
800 case SO_BROADCAST:
801 v.val = !!sock_flag(sk, SOCK_BROADCAST);
802 break;
803
804 case SO_SNDBUF:
805 v.val = sk->sk_sndbuf;
806 break;
807
808 case SO_RCVBUF:
809 v.val = sk->sk_rcvbuf;
810 break;
811
812 case SO_REUSEADDR:
813 v.val = sk->sk_reuse;
814 break;
815
816 case SO_KEEPALIVE:
817 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
818 break;
819
820 case SO_TYPE:
821 v.val = sk->sk_type;
822 break;
823
824 case SO_PROTOCOL:
825 v.val = sk->sk_protocol;
826 break;
827
828 case SO_DOMAIN:
829 v.val = sk->sk_family;
830 break;
831
832 case SO_ERROR:
833 v.val = -sock_error(sk);
834 if (v.val == 0)
835 v.val = xchg(&sk->sk_err_soft, 0);
836 break;
837
838 case SO_OOBINLINE:
839 v.val = !!sock_flag(sk, SOCK_URGINLINE);
840 break;
841
842 case SO_NO_CHECK:
843 v.val = sk->sk_no_check;
844 break;
845
846 case SO_PRIORITY:
847 v.val = sk->sk_priority;
848 break;
849
850 case SO_LINGER:
851 lv = sizeof(v.ling);
852 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
853 v.ling.l_linger = sk->sk_lingertime / HZ;
854 break;
855
856 case SO_BSDCOMPAT:
857 sock_warn_obsolete_bsdism("getsockopt");
858 break;
859
860 case SO_TIMESTAMP:
861 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
862 !sock_flag(sk, SOCK_RCVTSTAMPNS);
863 break;
864
865 case SO_TIMESTAMPNS:
866 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
867 break;
868
869 case SO_TIMESTAMPING:
870 v.val = 0;
871 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
872 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
873 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
874 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
875 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
876 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
877 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
878 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
879 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
880 v.val |= SOF_TIMESTAMPING_SOFTWARE;
881 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
882 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
883 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
884 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
885 break;
886
887 case SO_RCVTIMEO:
888 lv = sizeof(struct timeval);
889 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
890 v.tm.tv_sec = 0;
891 v.tm.tv_usec = 0;
892 } else {
893 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
894 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
895 }
896 break;
897
898 case SO_SNDTIMEO:
899 lv = sizeof(struct timeval);
900 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
901 v.tm.tv_sec = 0;
902 v.tm.tv_usec = 0;
903 } else {
904 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
905 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
906 }
907 break;
908
909 case SO_RCVLOWAT:
910 v.val = sk->sk_rcvlowat;
911 break;
912
913 case SO_SNDLOWAT:
914 v.val = 1;
915 break;
916
917 case SO_PASSCRED:
918 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
919 break;
920
921 case SO_PEERCRED:
922 {
923 struct ucred peercred;
924 if (len > sizeof(peercred))
925 len = sizeof(peercred);
926 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
927 if (copy_to_user(optval, &peercred, len))
928 return -EFAULT;
929 goto lenout;
930 }
931
932 case SO_PEERNAME:
933 {
934 char address[128];
935
936 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
937 return -ENOTCONN;
938 if (lv < len)
939 return -EINVAL;
940 if (copy_to_user(optval, address, len))
941 return -EFAULT;
942 goto lenout;
943 }
944
945 /* Dubious BSD thing... Probably nobody even uses it, but
946 * the UNIX standard wants it for whatever reason... -DaveM
947 */
948 case SO_ACCEPTCONN:
949 v.val = sk->sk_state == TCP_LISTEN;
950 break;
951
952 case SO_PASSSEC:
953 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
954 break;
955
956 case SO_PEERSEC:
957 return security_socket_getpeersec_stream(sock, optval, optlen, len);
958
959 case SO_MARK:
960 v.val = sk->sk_mark;
961 break;
962
963 case SO_RXQ_OVFL:
964 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
965 break;
966
967 default:
968 return -ENOPROTOOPT;
969 }
970
971 if (len > lv)
972 len = lv;
973 if (copy_to_user(optval, &v, len))
974 return -EFAULT;
975lenout:
976 if (put_user(len, optlen))
977 return -EFAULT;
978 return 0;
979}
980
981/*
982 * Initialize an sk_lock.
983 *
984 * (We also register the sk_lock with the lock validator.)
985 */
986static inline void sock_lock_init(struct sock *sk)
987{
988 sock_lock_init_class_and_name(sk,
989 af_family_slock_key_strings[sk->sk_family],
990 af_family_slock_keys + sk->sk_family,
991 af_family_key_strings[sk->sk_family],
992 af_family_keys + sk->sk_family);
993}
994
995/*
996 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
997 * even temporarly, because of RCU lookups. sk_node should also be left as is.
998 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
999 */
1000static void sock_copy(struct sock *nsk, const struct sock *osk)
1001{
1002#ifdef CONFIG_SECURITY_NETWORK
1003 void *sptr = nsk->sk_security;
1004#endif
1005 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1006
1007 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1008 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1009
1010#ifdef CONFIG_SECURITY_NETWORK
1011 nsk->sk_security = sptr;
1012 security_sk_clone(osk, nsk);
1013#endif
1014}
1015
1016/*
1017 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1018 * un-modified. Special care is taken when initializing object to zero.
1019 */
1020static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1021{
1022 if (offsetof(struct sock, sk_node.next) != 0)
1023 memset(sk, 0, offsetof(struct sock, sk_node.next));
1024 memset(&sk->sk_node.pprev, 0,
1025 size - offsetof(struct sock, sk_node.pprev));
1026}
1027
1028void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1029{
1030 unsigned long nulls1, nulls2;
1031
1032 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1033 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1034 if (nulls1 > nulls2)
1035 swap(nulls1, nulls2);
1036
1037 if (nulls1 != 0)
1038 memset((char *)sk, 0, nulls1);
1039 memset((char *)sk + nulls1 + sizeof(void *), 0,
1040 nulls2 - nulls1 - sizeof(void *));
1041 memset((char *)sk + nulls2 + sizeof(void *), 0,
1042 size - nulls2 - sizeof(void *));
1043}
1044EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1045
1046static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1047 int family)
1048{
1049 struct sock *sk;
1050 struct kmem_cache *slab;
1051
1052 slab = prot->slab;
1053 if (slab != NULL) {
1054 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1055 if (!sk)
1056 return sk;
1057 if (priority & __GFP_ZERO) {
1058 if (prot->clear_sk)
1059 prot->clear_sk(sk, prot->obj_size);
1060 else
1061 sk_prot_clear_nulls(sk, prot->obj_size);
1062 }
1063 } else
1064 sk = kmalloc(prot->obj_size, priority);
1065
1066 if (sk != NULL) {
1067 kmemcheck_annotate_bitfield(sk, flags);
1068
1069 if (security_sk_alloc(sk, family, priority))
1070 goto out_free;
1071
1072 if (!try_module_get(prot->owner))
1073 goto out_free_sec;
1074 sk_tx_queue_clear(sk);
1075 }
1076
1077 return sk;
1078
1079out_free_sec:
1080 security_sk_free(sk);
1081out_free:
1082 if (slab != NULL)
1083 kmem_cache_free(slab, sk);
1084 else
1085 kfree(sk);
1086 return NULL;
1087}
1088
1089static void sk_prot_free(struct proto *prot, struct sock *sk)
1090{
1091 struct kmem_cache *slab;
1092 struct module *owner;
1093
1094 owner = prot->owner;
1095 slab = prot->slab;
1096
1097 security_sk_free(sk);
1098 if (slab != NULL)
1099 kmem_cache_free(slab, sk);
1100 else
1101 kfree(sk);
1102 module_put(owner);
1103}
1104
1105#ifdef CONFIG_CGROUPS
1106void sock_update_classid(struct sock *sk)
1107{
1108 u32 classid;
1109
1110 rcu_read_lock(); /* doing current task, which cannot vanish. */
1111 classid = task_cls_classid(current);
1112 rcu_read_unlock();
1113 if (classid && classid != sk->sk_classid)
1114 sk->sk_classid = classid;
1115}
1116EXPORT_SYMBOL(sock_update_classid);
1117#endif
1118
1119/**
1120 * sk_alloc - All socket objects are allocated here
1121 * @net: the applicable net namespace
1122 * @family: protocol family
1123 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1124 * @prot: struct proto associated with this new sock instance
1125 */
1126struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1127 struct proto *prot)
1128{
1129 struct sock *sk;
1130
1131 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1132 if (sk) {
1133 sk->sk_family = family;
1134 /*
1135 * See comment in struct sock definition to understand
1136 * why we need sk_prot_creator -acme
1137 */
1138 sk->sk_prot = sk->sk_prot_creator = prot;
1139 sock_lock_init(sk);
1140 sock_net_set(sk, get_net(net));
1141 atomic_set(&sk->sk_wmem_alloc, 1);
1142
1143 sock_update_classid(sk);
1144 }
1145
1146 return sk;
1147}
1148EXPORT_SYMBOL(sk_alloc);
1149
1150static void __sk_free(struct sock *sk)
1151{
1152 struct sk_filter *filter;
1153
1154 if (sk->sk_destruct)
1155 sk->sk_destruct(sk);
1156
1157 filter = rcu_dereference_check(sk->sk_filter,
1158 atomic_read(&sk->sk_wmem_alloc) == 0);
1159 if (filter) {
1160 sk_filter_uncharge(sk, filter);
1161 rcu_assign_pointer(sk->sk_filter, NULL);
1162 }
1163
1164 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1165 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1166
1167 if (atomic_read(&sk->sk_omem_alloc))
1168 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1169 __func__, atomic_read(&sk->sk_omem_alloc));
1170
1171 if (sk->sk_peer_cred)
1172 put_cred(sk->sk_peer_cred);
1173 put_pid(sk->sk_peer_pid);
1174 put_net(sock_net(sk));
1175 sk_prot_free(sk->sk_prot_creator, sk);
1176}
1177
1178void sk_free(struct sock *sk)
1179{
1180 /*
1181 * We subtract one from sk_wmem_alloc and can know if
1182 * some packets are still in some tx queue.
1183 * If not null, sock_wfree() will call __sk_free(sk) later
1184 */
1185 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1186 __sk_free(sk);
1187}
1188EXPORT_SYMBOL(sk_free);
1189
1190/*
1191 * Last sock_put should drop reference to sk->sk_net. It has already
1192 * been dropped in sk_change_net. Taking reference to stopping namespace
1193 * is not an option.
1194 * Take reference to a socket to remove it from hash _alive_ and after that
1195 * destroy it in the context of init_net.
1196 */
1197void sk_release_kernel(struct sock *sk)
1198{
1199 if (sk == NULL || sk->sk_socket == NULL)
1200 return;
1201
1202 sock_hold(sk);
1203 sock_release(sk->sk_socket);
1204 release_net(sock_net(sk));
1205 sock_net_set(sk, get_net(&init_net));
1206 sock_put(sk);
1207}
1208EXPORT_SYMBOL(sk_release_kernel);
1209
1210struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1211{
1212 struct sock *newsk;
1213
1214 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1215 if (newsk != NULL) {
1216 struct sk_filter *filter;
1217
1218 sock_copy(newsk, sk);
1219
1220 /* SANITY */
1221 get_net(sock_net(newsk));
1222 sk_node_init(&newsk->sk_node);
1223 sock_lock_init(newsk);
1224 bh_lock_sock(newsk);
1225 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1226 newsk->sk_backlog.len = 0;
1227
1228 atomic_set(&newsk->sk_rmem_alloc, 0);
1229 /*
1230 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1231 */
1232 atomic_set(&newsk->sk_wmem_alloc, 1);
1233 atomic_set(&newsk->sk_omem_alloc, 0);
1234 skb_queue_head_init(&newsk->sk_receive_queue);
1235 skb_queue_head_init(&newsk->sk_write_queue);
1236#ifdef CONFIG_NET_DMA
1237 skb_queue_head_init(&newsk->sk_async_wait_queue);
1238#endif
1239
1240 spin_lock_init(&newsk->sk_dst_lock);
1241 rwlock_init(&newsk->sk_callback_lock);
1242 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1243 af_callback_keys + newsk->sk_family,
1244 af_family_clock_key_strings[newsk->sk_family]);
1245
1246 newsk->sk_dst_cache = NULL;
1247 newsk->sk_wmem_queued = 0;
1248 newsk->sk_forward_alloc = 0;
1249 newsk->sk_send_head = NULL;
1250 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1251
1252 sock_reset_flag(newsk, SOCK_DONE);
1253 skb_queue_head_init(&newsk->sk_error_queue);
1254
1255 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1256 if (filter != NULL)
1257 sk_filter_charge(newsk, filter);
1258
1259 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1260 /* It is still raw copy of parent, so invalidate
1261 * destructor and make plain sk_free() */
1262 newsk->sk_destruct = NULL;
1263 sk_free(newsk);
1264 newsk = NULL;
1265 goto out;
1266 }
1267
1268 newsk->sk_err = 0;
1269 newsk->sk_priority = 0;
1270 /*
1271 * Before updating sk_refcnt, we must commit prior changes to memory
1272 * (Documentation/RCU/rculist_nulls.txt for details)
1273 */
1274 smp_wmb();
1275 atomic_set(&newsk->sk_refcnt, 2);
1276
1277 /*
1278 * Increment the counter in the same struct proto as the master
1279 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1280 * is the same as sk->sk_prot->socks, as this field was copied
1281 * with memcpy).
1282 *
1283 * This _changes_ the previous behaviour, where
1284 * tcp_create_openreq_child always was incrementing the
1285 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1286 * to be taken into account in all callers. -acme
1287 */
1288 sk_refcnt_debug_inc(newsk);
1289 sk_set_socket(newsk, NULL);
1290 newsk->sk_wq = NULL;
1291
1292 if (newsk->sk_prot->sockets_allocated)
1293 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1294
1295 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1296 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1297 net_enable_timestamp();
1298 }
1299out:
1300 return newsk;
1301}
1302EXPORT_SYMBOL_GPL(sk_clone);
1303
1304void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1305{
1306 __sk_dst_set(sk, dst);
1307 sk->sk_route_caps = dst->dev->features;
1308 if (sk->sk_route_caps & NETIF_F_GSO)
1309 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1310 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1311 if (sk_can_gso(sk)) {
1312 if (dst->header_len) {
1313 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1314 } else {
1315 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1316 sk->sk_gso_max_size = dst->dev->gso_max_size;
1317 }
1318 }
1319}
1320EXPORT_SYMBOL_GPL(sk_setup_caps);
1321
1322void __init sk_init(void)
1323{
1324 if (totalram_pages <= 4096) {
1325 sysctl_wmem_max = 32767;
1326 sysctl_rmem_max = 32767;
1327 sysctl_wmem_default = 32767;
1328 sysctl_rmem_default = 32767;
1329 } else if (totalram_pages >= 131072) {
1330 sysctl_wmem_max = 131071;
1331 sysctl_rmem_max = 131071;
1332 }
1333}
1334
1335/*
1336 * Simple resource managers for sockets.
1337 */
1338
1339
1340/*
1341 * Write buffer destructor automatically called from kfree_skb.
1342 */
1343void sock_wfree(struct sk_buff *skb)
1344{
1345 struct sock *sk = skb->sk;
1346 unsigned int len = skb->truesize;
1347
1348 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1349 /*
1350 * Keep a reference on sk_wmem_alloc, this will be released
1351 * after sk_write_space() call
1352 */
1353 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1354 sk->sk_write_space(sk);
1355 len = 1;
1356 }
1357 /*
1358 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1359 * could not do because of in-flight packets
1360 */
1361 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1362 __sk_free(sk);
1363}
1364EXPORT_SYMBOL(sock_wfree);
1365
1366/*
1367 * Read buffer destructor automatically called from kfree_skb.
1368 */
1369void sock_rfree(struct sk_buff *skb)
1370{
1371 struct sock *sk = skb->sk;
1372 unsigned int len = skb->truesize;
1373
1374 atomic_sub(len, &sk->sk_rmem_alloc);
1375 sk_mem_uncharge(sk, len);
1376}
1377EXPORT_SYMBOL(sock_rfree);
1378
1379
1380int sock_i_uid(struct sock *sk)
1381{
1382 int uid;
1383
1384 read_lock_bh(&sk->sk_callback_lock);
1385 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1386 read_unlock_bh(&sk->sk_callback_lock);
1387 return uid;
1388}
1389EXPORT_SYMBOL(sock_i_uid);
1390
1391unsigned long sock_i_ino(struct sock *sk)
1392{
1393 unsigned long ino;
1394
1395 read_lock_bh(&sk->sk_callback_lock);
1396 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1397 read_unlock_bh(&sk->sk_callback_lock);
1398 return ino;
1399}
1400EXPORT_SYMBOL(sock_i_ino);
1401
1402/*
1403 * Allocate a skb from the socket's send buffer.
1404 */
1405struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1406 gfp_t priority)
1407{
1408 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1409 struct sk_buff *skb = alloc_skb(size, priority);
1410 if (skb) {
1411 skb_set_owner_w(skb, sk);
1412 return skb;
1413 }
1414 }
1415 return NULL;
1416}
1417EXPORT_SYMBOL(sock_wmalloc);
1418
1419/*
1420 * Allocate a skb from the socket's receive buffer.
1421 */
1422struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1423 gfp_t priority)
1424{
1425 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1426 struct sk_buff *skb = alloc_skb(size, priority);
1427 if (skb) {
1428 skb_set_owner_r(skb, sk);
1429 return skb;
1430 }
1431 }
1432 return NULL;
1433}
1434
1435/*
1436 * Allocate a memory block from the socket's option memory buffer.
1437 */
1438void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1439{
1440 if ((unsigned)size <= sysctl_optmem_max &&
1441 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1442 void *mem;
1443 /* First do the add, to avoid the race if kmalloc
1444 * might sleep.
1445 */
1446 atomic_add(size, &sk->sk_omem_alloc);
1447 mem = kmalloc(size, priority);
1448 if (mem)
1449 return mem;
1450 atomic_sub(size, &sk->sk_omem_alloc);
1451 }
1452 return NULL;
1453}
1454EXPORT_SYMBOL(sock_kmalloc);
1455
1456/*
1457 * Free an option memory block.
1458 */
1459void sock_kfree_s(struct sock *sk, void *mem, int size)
1460{
1461 kfree(mem);
1462 atomic_sub(size, &sk->sk_omem_alloc);
1463}
1464EXPORT_SYMBOL(sock_kfree_s);
1465
1466/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1467 I think, these locks should be removed for datagram sockets.
1468 */
1469static long sock_wait_for_wmem(struct sock *sk, long timeo)
1470{
1471 DEFINE_WAIT(wait);
1472
1473 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1474 for (;;) {
1475 if (!timeo)
1476 break;
1477 if (signal_pending(current))
1478 break;
1479 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1480 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1481 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1482 break;
1483 if (sk->sk_shutdown & SEND_SHUTDOWN)
1484 break;
1485 if (sk->sk_err)
1486 break;
1487 timeo = schedule_timeout(timeo);
1488 }
1489 finish_wait(sk_sleep(sk), &wait);
1490 return timeo;
1491}
1492
1493
1494/*
1495 * Generic send/receive buffer handlers
1496 */
1497
1498struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1499 unsigned long data_len, int noblock,
1500 int *errcode)
1501{
1502 struct sk_buff *skb;
1503 gfp_t gfp_mask;
1504 long timeo;
1505 int err;
1506
1507 gfp_mask = sk->sk_allocation;
1508 if (gfp_mask & __GFP_WAIT)
1509 gfp_mask |= __GFP_REPEAT;
1510
1511 timeo = sock_sndtimeo(sk, noblock);
1512 while (1) {
1513 err = sock_error(sk);
1514 if (err != 0)
1515 goto failure;
1516
1517 err = -EPIPE;
1518 if (sk->sk_shutdown & SEND_SHUTDOWN)
1519 goto failure;
1520
1521 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1522 skb = alloc_skb(header_len, gfp_mask);
1523 if (skb) {
1524 int npages;
1525 int i;
1526
1527 /* No pages, we're done... */
1528 if (!data_len)
1529 break;
1530
1531 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1532 skb->truesize += data_len;
1533 skb_shinfo(skb)->nr_frags = npages;
1534 for (i = 0; i < npages; i++) {
1535 struct page *page;
1536 skb_frag_t *frag;
1537
1538 page = alloc_pages(sk->sk_allocation, 0);
1539 if (!page) {
1540 err = -ENOBUFS;
1541 skb_shinfo(skb)->nr_frags = i;
1542 kfree_skb(skb);
1543 goto failure;
1544 }
1545
1546 frag = &skb_shinfo(skb)->frags[i];
1547 frag->page = page;
1548 frag->page_offset = 0;
1549 frag->size = (data_len >= PAGE_SIZE ?
1550 PAGE_SIZE :
1551 data_len);
1552 data_len -= PAGE_SIZE;
1553 }
1554
1555 /* Full success... */
1556 break;
1557 }
1558 err = -ENOBUFS;
1559 goto failure;
1560 }
1561 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1562 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1563 err = -EAGAIN;
1564 if (!timeo)
1565 goto failure;
1566 if (signal_pending(current))
1567 goto interrupted;
1568 timeo = sock_wait_for_wmem(sk, timeo);
1569 }
1570
1571 skb_set_owner_w(skb, sk);
1572 return skb;
1573
1574interrupted:
1575 err = sock_intr_errno(timeo);
1576failure:
1577 *errcode = err;
1578 return NULL;
1579}
1580EXPORT_SYMBOL(sock_alloc_send_pskb);
1581
1582struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1583 int noblock, int *errcode)
1584{
1585 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1586}
1587EXPORT_SYMBOL(sock_alloc_send_skb);
1588
1589static void __lock_sock(struct sock *sk)
1590 __releases(&sk->sk_lock.slock)
1591 __acquires(&sk->sk_lock.slock)
1592{
1593 DEFINE_WAIT(wait);
1594
1595 for (;;) {
1596 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1597 TASK_UNINTERRUPTIBLE);
1598 spin_unlock_bh(&sk->sk_lock.slock);
1599 schedule();
1600 spin_lock_bh(&sk->sk_lock.slock);
1601 if (!sock_owned_by_user(sk))
1602 break;
1603 }
1604 finish_wait(&sk->sk_lock.wq, &wait);
1605}
1606
1607static void __release_sock(struct sock *sk)
1608 __releases(&sk->sk_lock.slock)
1609 __acquires(&sk->sk_lock.slock)
1610{
1611 struct sk_buff *skb = sk->sk_backlog.head;
1612
1613 do {
1614 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1615 bh_unlock_sock(sk);
1616
1617 do {
1618 struct sk_buff *next = skb->next;
1619
1620 WARN_ON_ONCE(skb_dst_is_noref(skb));
1621 skb->next = NULL;
1622 sk_backlog_rcv(sk, skb);
1623
1624 /*
1625 * We are in process context here with softirqs
1626 * disabled, use cond_resched_softirq() to preempt.
1627 * This is safe to do because we've taken the backlog
1628 * queue private:
1629 */
1630 cond_resched_softirq();
1631
1632 skb = next;
1633 } while (skb != NULL);
1634
1635 bh_lock_sock(sk);
1636 } while ((skb = sk->sk_backlog.head) != NULL);
1637
1638 /*
1639 * Doing the zeroing here guarantee we can not loop forever
1640 * while a wild producer attempts to flood us.
1641 */
1642 sk->sk_backlog.len = 0;
1643}
1644
1645/**
1646 * sk_wait_data - wait for data to arrive at sk_receive_queue
1647 * @sk: sock to wait on
1648 * @timeo: for how long
1649 *
1650 * Now socket state including sk->sk_err is changed only under lock,
1651 * hence we may omit checks after joining wait queue.
1652 * We check receive queue before schedule() only as optimization;
1653 * it is very likely that release_sock() added new data.
1654 */
1655int sk_wait_data(struct sock *sk, long *timeo)
1656{
1657 int rc;
1658 DEFINE_WAIT(wait);
1659
1660 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1661 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1662 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1663 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1664 finish_wait(sk_sleep(sk), &wait);
1665 return rc;
1666}
1667EXPORT_SYMBOL(sk_wait_data);
1668
1669/**
1670 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1671 * @sk: socket
1672 * @size: memory size to allocate
1673 * @kind: allocation type
1674 *
1675 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1676 * rmem allocation. This function assumes that protocols which have
1677 * memory_pressure use sk_wmem_queued as write buffer accounting.
1678 */
1679int __sk_mem_schedule(struct sock *sk, int size, int kind)
1680{
1681 struct proto *prot = sk->sk_prot;
1682 int amt = sk_mem_pages(size);
1683 long allocated;
1684
1685 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1686 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1687
1688 /* Under limit. */
1689 if (allocated <= prot->sysctl_mem[0]) {
1690 if (prot->memory_pressure && *prot->memory_pressure)
1691 *prot->memory_pressure = 0;
1692 return 1;
1693 }
1694
1695 /* Under pressure. */
1696 if (allocated > prot->sysctl_mem[1])
1697 if (prot->enter_memory_pressure)
1698 prot->enter_memory_pressure(sk);
1699
1700 /* Over hard limit. */
1701 if (allocated > prot->sysctl_mem[2])
1702 goto suppress_allocation;
1703
1704 /* guarantee minimum buffer size under pressure */
1705 if (kind == SK_MEM_RECV) {
1706 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1707 return 1;
1708 } else { /* SK_MEM_SEND */
1709 if (sk->sk_type == SOCK_STREAM) {
1710 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1711 return 1;
1712 } else if (atomic_read(&sk->sk_wmem_alloc) <
1713 prot->sysctl_wmem[0])
1714 return 1;
1715 }
1716
1717 if (prot->memory_pressure) {
1718 int alloc;
1719
1720 if (!*prot->memory_pressure)
1721 return 1;
1722 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1723 if (prot->sysctl_mem[2] > alloc *
1724 sk_mem_pages(sk->sk_wmem_queued +
1725 atomic_read(&sk->sk_rmem_alloc) +
1726 sk->sk_forward_alloc))
1727 return 1;
1728 }
1729
1730suppress_allocation:
1731
1732 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1733 sk_stream_moderate_sndbuf(sk);
1734
1735 /* Fail only if socket is _under_ its sndbuf.
1736 * In this case we cannot block, so that we have to fail.
1737 */
1738 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1739 return 1;
1740 }
1741
1742 trace_sock_exceed_buf_limit(sk, prot, allocated);
1743
1744 /* Alas. Undo changes. */
1745 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1746 atomic_long_sub(amt, prot->memory_allocated);
1747 return 0;
1748}
1749EXPORT_SYMBOL(__sk_mem_schedule);
1750
1751/**
1752 * __sk_reclaim - reclaim memory_allocated
1753 * @sk: socket
1754 */
1755void __sk_mem_reclaim(struct sock *sk)
1756{
1757 struct proto *prot = sk->sk_prot;
1758
1759 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1760 prot->memory_allocated);
1761 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1762
1763 if (prot->memory_pressure && *prot->memory_pressure &&
1764 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1765 *prot->memory_pressure = 0;
1766}
1767EXPORT_SYMBOL(__sk_mem_reclaim);
1768
1769
1770/*
1771 * Set of default routines for initialising struct proto_ops when
1772 * the protocol does not support a particular function. In certain
1773 * cases where it makes no sense for a protocol to have a "do nothing"
1774 * function, some default processing is provided.
1775 */
1776
1777int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1778{
1779 return -EOPNOTSUPP;
1780}
1781EXPORT_SYMBOL(sock_no_bind);
1782
1783int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1784 int len, int flags)
1785{
1786 return -EOPNOTSUPP;
1787}
1788EXPORT_SYMBOL(sock_no_connect);
1789
1790int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1791{
1792 return -EOPNOTSUPP;
1793}
1794EXPORT_SYMBOL(sock_no_socketpair);
1795
1796int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1797{
1798 return -EOPNOTSUPP;
1799}
1800EXPORT_SYMBOL(sock_no_accept);
1801
1802int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1803 int *len, int peer)
1804{
1805 return -EOPNOTSUPP;
1806}
1807EXPORT_SYMBOL(sock_no_getname);
1808
1809unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1810{
1811 return 0;
1812}
1813EXPORT_SYMBOL(sock_no_poll);
1814
1815int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1816{
1817 return -EOPNOTSUPP;
1818}
1819EXPORT_SYMBOL(sock_no_ioctl);
1820
1821int sock_no_listen(struct socket *sock, int backlog)
1822{
1823 return -EOPNOTSUPP;
1824}
1825EXPORT_SYMBOL(sock_no_listen);
1826
1827int sock_no_shutdown(struct socket *sock, int how)
1828{
1829 return -EOPNOTSUPP;
1830}
1831EXPORT_SYMBOL(sock_no_shutdown);
1832
1833int sock_no_setsockopt(struct socket *sock, int level, int optname,
1834 char __user *optval, unsigned int optlen)
1835{
1836 return -EOPNOTSUPP;
1837}
1838EXPORT_SYMBOL(sock_no_setsockopt);
1839
1840int sock_no_getsockopt(struct socket *sock, int level, int optname,
1841 char __user *optval, int __user *optlen)
1842{
1843 return -EOPNOTSUPP;
1844}
1845EXPORT_SYMBOL(sock_no_getsockopt);
1846
1847int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1848 size_t len)
1849{
1850 return -EOPNOTSUPP;
1851}
1852EXPORT_SYMBOL(sock_no_sendmsg);
1853
1854int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1855 size_t len, int flags)
1856{
1857 return -EOPNOTSUPP;
1858}
1859EXPORT_SYMBOL(sock_no_recvmsg);
1860
1861int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1862{
1863 /* Mirror missing mmap method error code */
1864 return -ENODEV;
1865}
1866EXPORT_SYMBOL(sock_no_mmap);
1867
1868ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1869{
1870 ssize_t res;
1871 struct msghdr msg = {.msg_flags = flags};
1872 struct kvec iov;
1873 char *kaddr = kmap(page);
1874 iov.iov_base = kaddr + offset;
1875 iov.iov_len = size;
1876 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1877 kunmap(page);
1878 return res;
1879}
1880EXPORT_SYMBOL(sock_no_sendpage);
1881
1882/*
1883 * Default Socket Callbacks
1884 */
1885
1886static void sock_def_wakeup(struct sock *sk)
1887{
1888 struct socket_wq *wq;
1889
1890 rcu_read_lock();
1891 wq = rcu_dereference(sk->sk_wq);
1892 if (wq_has_sleeper(wq))
1893 wake_up_interruptible_all(&wq->wait);
1894 rcu_read_unlock();
1895}
1896
1897static void sock_def_error_report(struct sock *sk)
1898{
1899 struct socket_wq *wq;
1900
1901 rcu_read_lock();
1902 wq = rcu_dereference(sk->sk_wq);
1903 if (wq_has_sleeper(wq))
1904 wake_up_interruptible_poll(&wq->wait, POLLERR);
1905 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1906 rcu_read_unlock();
1907}
1908
1909static void sock_def_readable(struct sock *sk, int len)
1910{
1911 struct socket_wq *wq;
1912
1913 rcu_read_lock();
1914 wq = rcu_dereference(sk->sk_wq);
1915 if (wq_has_sleeper(wq))
1916 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1917 POLLRDNORM | POLLRDBAND);
1918 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1919 rcu_read_unlock();
1920}
1921
1922static void sock_def_write_space(struct sock *sk)
1923{
1924 struct socket_wq *wq;
1925
1926 rcu_read_lock();
1927
1928 /* Do not wake up a writer until he can make "significant"
1929 * progress. --DaveM
1930 */
1931 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1932 wq = rcu_dereference(sk->sk_wq);
1933 if (wq_has_sleeper(wq))
1934 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1935 POLLWRNORM | POLLWRBAND);
1936
1937 /* Should agree with poll, otherwise some programs break */
1938 if (sock_writeable(sk))
1939 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1940 }
1941
1942 rcu_read_unlock();
1943}
1944
1945static void sock_def_destruct(struct sock *sk)
1946{
1947 kfree(sk->sk_protinfo);
1948}
1949
1950void sk_send_sigurg(struct sock *sk)
1951{
1952 if (sk->sk_socket && sk->sk_socket->file)
1953 if (send_sigurg(&sk->sk_socket->file->f_owner))
1954 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1955}
1956EXPORT_SYMBOL(sk_send_sigurg);
1957
1958void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1959 unsigned long expires)
1960{
1961 if (!mod_timer(timer, expires))
1962 sock_hold(sk);
1963}
1964EXPORT_SYMBOL(sk_reset_timer);
1965
1966void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1967{
1968 if (timer_pending(timer) && del_timer(timer))
1969 __sock_put(sk);
1970}
1971EXPORT_SYMBOL(sk_stop_timer);
1972
1973void sock_init_data(struct socket *sock, struct sock *sk)
1974{
1975 skb_queue_head_init(&sk->sk_receive_queue);
1976 skb_queue_head_init(&sk->sk_write_queue);
1977 skb_queue_head_init(&sk->sk_error_queue);
1978#ifdef CONFIG_NET_DMA
1979 skb_queue_head_init(&sk->sk_async_wait_queue);
1980#endif
1981
1982 sk->sk_send_head = NULL;
1983
1984 init_timer(&sk->sk_timer);
1985
1986 sk->sk_allocation = GFP_KERNEL;
1987 sk->sk_rcvbuf = sysctl_rmem_default;
1988 sk->sk_sndbuf = sysctl_wmem_default;
1989 sk->sk_state = TCP_CLOSE;
1990 sk_set_socket(sk, sock);
1991
1992 sock_set_flag(sk, SOCK_ZAPPED);
1993
1994 if (sock) {
1995 sk->sk_type = sock->type;
1996 sk->sk_wq = sock->wq;
1997 sock->sk = sk;
1998 } else
1999 sk->sk_wq = NULL;
2000
2001 spin_lock_init(&sk->sk_dst_lock);
2002 rwlock_init(&sk->sk_callback_lock);
2003 lockdep_set_class_and_name(&sk->sk_callback_lock,
2004 af_callback_keys + sk->sk_family,
2005 af_family_clock_key_strings[sk->sk_family]);
2006
2007 sk->sk_state_change = sock_def_wakeup;
2008 sk->sk_data_ready = sock_def_readable;
2009 sk->sk_write_space = sock_def_write_space;
2010 sk->sk_error_report = sock_def_error_report;
2011 sk->sk_destruct = sock_def_destruct;
2012
2013 sk->sk_sndmsg_page = NULL;
2014 sk->sk_sndmsg_off = 0;
2015
2016 sk->sk_peer_pid = NULL;
2017 sk->sk_peer_cred = NULL;
2018 sk->sk_write_pending = 0;
2019 sk->sk_rcvlowat = 1;
2020 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2021 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2022
2023 sk->sk_stamp = ktime_set(-1L, 0);
2024
2025 /*
2026 * Before updating sk_refcnt, we must commit prior changes to memory
2027 * (Documentation/RCU/rculist_nulls.txt for details)
2028 */
2029 smp_wmb();
2030 atomic_set(&sk->sk_refcnt, 1);
2031 atomic_set(&sk->sk_drops, 0);
2032}
2033EXPORT_SYMBOL(sock_init_data);
2034
2035void lock_sock_nested(struct sock *sk, int subclass)
2036{
2037 might_sleep();
2038 spin_lock_bh(&sk->sk_lock.slock);
2039 if (sk->sk_lock.owned)
2040 __lock_sock(sk);
2041 sk->sk_lock.owned = 1;
2042 spin_unlock(&sk->sk_lock.slock);
2043 /*
2044 * The sk_lock has mutex_lock() semantics here:
2045 */
2046 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2047 local_bh_enable();
2048}
2049EXPORT_SYMBOL(lock_sock_nested);
2050
2051void release_sock(struct sock *sk)
2052{
2053 /*
2054 * The sk_lock has mutex_unlock() semantics:
2055 */
2056 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2057
2058 spin_lock_bh(&sk->sk_lock.slock);
2059 if (sk->sk_backlog.tail)
2060 __release_sock(sk);
2061 sk->sk_lock.owned = 0;
2062 if (waitqueue_active(&sk->sk_lock.wq))
2063 wake_up(&sk->sk_lock.wq);
2064 spin_unlock_bh(&sk->sk_lock.slock);
2065}
2066EXPORT_SYMBOL(release_sock);
2067
2068/**
2069 * lock_sock_fast - fast version of lock_sock
2070 * @sk: socket
2071 *
2072 * This version should be used for very small section, where process wont block
2073 * return false if fast path is taken
2074 * sk_lock.slock locked, owned = 0, BH disabled
2075 * return true if slow path is taken
2076 * sk_lock.slock unlocked, owned = 1, BH enabled
2077 */
2078bool lock_sock_fast(struct sock *sk)
2079{
2080 might_sleep();
2081 spin_lock_bh(&sk->sk_lock.slock);
2082
2083 if (!sk->sk_lock.owned)
2084 /*
2085 * Note : We must disable BH
2086 */
2087 return false;
2088
2089 __lock_sock(sk);
2090 sk->sk_lock.owned = 1;
2091 spin_unlock(&sk->sk_lock.slock);
2092 /*
2093 * The sk_lock has mutex_lock() semantics here:
2094 */
2095 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2096 local_bh_enable();
2097 return true;
2098}
2099EXPORT_SYMBOL(lock_sock_fast);
2100
2101int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2102{
2103 struct timeval tv;
2104 if (!sock_flag(sk, SOCK_TIMESTAMP))
2105 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2106 tv = ktime_to_timeval(sk->sk_stamp);
2107 if (tv.tv_sec == -1)
2108 return -ENOENT;
2109 if (tv.tv_sec == 0) {
2110 sk->sk_stamp = ktime_get_real();
2111 tv = ktime_to_timeval(sk->sk_stamp);
2112 }
2113 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2114}
2115EXPORT_SYMBOL(sock_get_timestamp);
2116
2117int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2118{
2119 struct timespec ts;
2120 if (!sock_flag(sk, SOCK_TIMESTAMP))
2121 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2122 ts = ktime_to_timespec(sk->sk_stamp);
2123 if (ts.tv_sec == -1)
2124 return -ENOENT;
2125 if (ts.tv_sec == 0) {
2126 sk->sk_stamp = ktime_get_real();
2127 ts = ktime_to_timespec(sk->sk_stamp);
2128 }
2129 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2130}
2131EXPORT_SYMBOL(sock_get_timestampns);
2132
2133void sock_enable_timestamp(struct sock *sk, int flag)
2134{
2135 if (!sock_flag(sk, flag)) {
2136 sock_set_flag(sk, flag);
2137 /*
2138 * we just set one of the two flags which require net
2139 * time stamping, but time stamping might have been on
2140 * already because of the other one
2141 */
2142 if (!sock_flag(sk,
2143 flag == SOCK_TIMESTAMP ?
2144 SOCK_TIMESTAMPING_RX_SOFTWARE :
2145 SOCK_TIMESTAMP))
2146 net_enable_timestamp();
2147 }
2148}
2149
2150/*
2151 * Get a socket option on an socket.
2152 *
2153 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2154 * asynchronous errors should be reported by getsockopt. We assume
2155 * this means if you specify SO_ERROR (otherwise whats the point of it).
2156 */
2157int sock_common_getsockopt(struct socket *sock, int level, int optname,
2158 char __user *optval, int __user *optlen)
2159{
2160 struct sock *sk = sock->sk;
2161
2162 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2163}
2164EXPORT_SYMBOL(sock_common_getsockopt);
2165
2166#ifdef CONFIG_COMPAT
2167int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2168 char __user *optval, int __user *optlen)
2169{
2170 struct sock *sk = sock->sk;
2171
2172 if (sk->sk_prot->compat_getsockopt != NULL)
2173 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2174 optval, optlen);
2175 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2176}
2177EXPORT_SYMBOL(compat_sock_common_getsockopt);
2178#endif
2179
2180int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2181 struct msghdr *msg, size_t size, int flags)
2182{
2183 struct sock *sk = sock->sk;
2184 int addr_len = 0;
2185 int err;
2186
2187 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2188 flags & ~MSG_DONTWAIT, &addr_len);
2189 if (err >= 0)
2190 msg->msg_namelen = addr_len;
2191 return err;
2192}
2193EXPORT_SYMBOL(sock_common_recvmsg);
2194
2195/*
2196 * Set socket options on an inet socket.
2197 */
2198int sock_common_setsockopt(struct socket *sock, int level, int optname,
2199 char __user *optval, unsigned int optlen)
2200{
2201 struct sock *sk = sock->sk;
2202
2203 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2204}
2205EXPORT_SYMBOL(sock_common_setsockopt);
2206
2207#ifdef CONFIG_COMPAT
2208int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2209 char __user *optval, unsigned int optlen)
2210{
2211 struct sock *sk = sock->sk;
2212
2213 if (sk->sk_prot->compat_setsockopt != NULL)
2214 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2215 optval, optlen);
2216 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2217}
2218EXPORT_SYMBOL(compat_sock_common_setsockopt);
2219#endif
2220
2221void sk_common_release(struct sock *sk)
2222{
2223 if (sk->sk_prot->destroy)
2224 sk->sk_prot->destroy(sk);
2225
2226 /*
2227 * Observation: when sock_common_release is called, processes have
2228 * no access to socket. But net still has.
2229 * Step one, detach it from networking:
2230 *
2231 * A. Remove from hash tables.
2232 */
2233
2234 sk->sk_prot->unhash(sk);
2235
2236 /*
2237 * In this point socket cannot receive new packets, but it is possible
2238 * that some packets are in flight because some CPU runs receiver and
2239 * did hash table lookup before we unhashed socket. They will achieve
2240 * receive queue and will be purged by socket destructor.
2241 *
2242 * Also we still have packets pending on receive queue and probably,
2243 * our own packets waiting in device queues. sock_destroy will drain
2244 * receive queue, but transmitted packets will delay socket destruction
2245 * until the last reference will be released.
2246 */
2247
2248 sock_orphan(sk);
2249
2250 xfrm_sk_free_policy(sk);
2251
2252 sk_refcnt_debug_release(sk);
2253 sock_put(sk);
2254}
2255EXPORT_SYMBOL(sk_common_release);
2256
2257static DEFINE_RWLOCK(proto_list_lock);
2258static LIST_HEAD(proto_list);
2259
2260#ifdef CONFIG_PROC_FS
2261#define PROTO_INUSE_NR 64 /* should be enough for the first time */
2262struct prot_inuse {
2263 int val[PROTO_INUSE_NR];
2264};
2265
2266static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2267
2268#ifdef CONFIG_NET_NS
2269void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2270{
2271 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2272}
2273EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2274
2275int sock_prot_inuse_get(struct net *net, struct proto *prot)
2276{
2277 int cpu, idx = prot->inuse_idx;
2278 int res = 0;
2279
2280 for_each_possible_cpu(cpu)
2281 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2282
2283 return res >= 0 ? res : 0;
2284}
2285EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2286
2287static int __net_init sock_inuse_init_net(struct net *net)
2288{
2289 net->core.inuse = alloc_percpu(struct prot_inuse);
2290 return net->core.inuse ? 0 : -ENOMEM;
2291}
2292
2293static void __net_exit sock_inuse_exit_net(struct net *net)
2294{
2295 free_percpu(net->core.inuse);
2296}
2297
2298static struct pernet_operations net_inuse_ops = {
2299 .init = sock_inuse_init_net,
2300 .exit = sock_inuse_exit_net,
2301};
2302
2303static __init int net_inuse_init(void)
2304{
2305 if (register_pernet_subsys(&net_inuse_ops))
2306 panic("Cannot initialize net inuse counters");
2307
2308 return 0;
2309}
2310
2311core_initcall(net_inuse_init);
2312#else
2313static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2314
2315void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2316{
2317 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2318}
2319EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2320
2321int sock_prot_inuse_get(struct net *net, struct proto *prot)
2322{
2323 int cpu, idx = prot->inuse_idx;
2324 int res = 0;
2325
2326 for_each_possible_cpu(cpu)
2327 res += per_cpu(prot_inuse, cpu).val[idx];
2328
2329 return res >= 0 ? res : 0;
2330}
2331EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2332#endif
2333
2334static void assign_proto_idx(struct proto *prot)
2335{
2336 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2337
2338 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2339 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2340 return;
2341 }
2342
2343 set_bit(prot->inuse_idx, proto_inuse_idx);
2344}
2345
2346static void release_proto_idx(struct proto *prot)
2347{
2348 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2349 clear_bit(prot->inuse_idx, proto_inuse_idx);
2350}
2351#else
2352static inline void assign_proto_idx(struct proto *prot)
2353{
2354}
2355
2356static inline void release_proto_idx(struct proto *prot)
2357{
2358}
2359#endif
2360
2361int proto_register(struct proto *prot, int alloc_slab)
2362{
2363 if (alloc_slab) {
2364 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2365 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2366 NULL);
2367
2368 if (prot->slab == NULL) {
2369 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2370 prot->name);
2371 goto out;
2372 }
2373
2374 if (prot->rsk_prot != NULL) {
2375 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2376 if (prot->rsk_prot->slab_name == NULL)
2377 goto out_free_sock_slab;
2378
2379 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2380 prot->rsk_prot->obj_size, 0,
2381 SLAB_HWCACHE_ALIGN, NULL);
2382
2383 if (prot->rsk_prot->slab == NULL) {
2384 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2385 prot->name);
2386 goto out_free_request_sock_slab_name;
2387 }
2388 }
2389
2390 if (prot->twsk_prot != NULL) {
2391 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2392
2393 if (prot->twsk_prot->twsk_slab_name == NULL)
2394 goto out_free_request_sock_slab;
2395
2396 prot->twsk_prot->twsk_slab =
2397 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2398 prot->twsk_prot->twsk_obj_size,
2399 0,
2400 SLAB_HWCACHE_ALIGN |
2401 prot->slab_flags,
2402 NULL);
2403 if (prot->twsk_prot->twsk_slab == NULL)
2404 goto out_free_timewait_sock_slab_name;
2405 }
2406 }
2407
2408 write_lock(&proto_list_lock);
2409 list_add(&prot->node, &proto_list);
2410 assign_proto_idx(prot);
2411 write_unlock(&proto_list_lock);
2412 return 0;
2413
2414out_free_timewait_sock_slab_name:
2415 kfree(prot->twsk_prot->twsk_slab_name);
2416out_free_request_sock_slab:
2417 if (prot->rsk_prot && prot->rsk_prot->slab) {
2418 kmem_cache_destroy(prot->rsk_prot->slab);
2419 prot->rsk_prot->slab = NULL;
2420 }
2421out_free_request_sock_slab_name:
2422 if (prot->rsk_prot)
2423 kfree(prot->rsk_prot->slab_name);
2424out_free_sock_slab:
2425 kmem_cache_destroy(prot->slab);
2426 prot->slab = NULL;
2427out:
2428 return -ENOBUFS;
2429}
2430EXPORT_SYMBOL(proto_register);
2431
2432void proto_unregister(struct proto *prot)
2433{
2434 write_lock(&proto_list_lock);
2435 release_proto_idx(prot);
2436 list_del(&prot->node);
2437 write_unlock(&proto_list_lock);
2438
2439 if (prot->slab != NULL) {
2440 kmem_cache_destroy(prot->slab);
2441 prot->slab = NULL;
2442 }
2443
2444 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2445 kmem_cache_destroy(prot->rsk_prot->slab);
2446 kfree(prot->rsk_prot->slab_name);
2447 prot->rsk_prot->slab = NULL;
2448 }
2449
2450 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2451 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2452 kfree(prot->twsk_prot->twsk_slab_name);
2453 prot->twsk_prot->twsk_slab = NULL;
2454 }
2455}
2456EXPORT_SYMBOL(proto_unregister);
2457
2458#ifdef CONFIG_PROC_FS
2459static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2460 __acquires(proto_list_lock)
2461{
2462 read_lock(&proto_list_lock);
2463 return seq_list_start_head(&proto_list, *pos);
2464}
2465
2466static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2467{
2468 return seq_list_next(v, &proto_list, pos);
2469}
2470
2471static void proto_seq_stop(struct seq_file *seq, void *v)
2472 __releases(proto_list_lock)
2473{
2474 read_unlock(&proto_list_lock);
2475}
2476
2477static char proto_method_implemented(const void *method)
2478{
2479 return method == NULL ? 'n' : 'y';
2480}
2481
2482static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2483{
2484 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2485 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2486 proto->name,
2487 proto->obj_size,
2488 sock_prot_inuse_get(seq_file_net(seq), proto),
2489 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2490 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2491 proto->max_header,
2492 proto->slab == NULL ? "no" : "yes",
2493 module_name(proto->owner),
2494 proto_method_implemented(proto->close),
2495 proto_method_implemented(proto->connect),
2496 proto_method_implemented(proto->disconnect),
2497 proto_method_implemented(proto->accept),
2498 proto_method_implemented(proto->ioctl),
2499 proto_method_implemented(proto->init),
2500 proto_method_implemented(proto->destroy),
2501 proto_method_implemented(proto->shutdown),
2502 proto_method_implemented(proto->setsockopt),
2503 proto_method_implemented(proto->getsockopt),
2504 proto_method_implemented(proto->sendmsg),
2505 proto_method_implemented(proto->recvmsg),
2506 proto_method_implemented(proto->sendpage),
2507 proto_method_implemented(proto->bind),
2508 proto_method_implemented(proto->backlog_rcv),
2509 proto_method_implemented(proto->hash),
2510 proto_method_implemented(proto->unhash),
2511 proto_method_implemented(proto->get_port),
2512 proto_method_implemented(proto->enter_memory_pressure));
2513}
2514
2515static int proto_seq_show(struct seq_file *seq, void *v)
2516{
2517 if (v == &proto_list)
2518 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2519 "protocol",
2520 "size",
2521 "sockets",
2522 "memory",
2523 "press",
2524 "maxhdr",
2525 "slab",
2526 "module",
2527 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2528 else
2529 proto_seq_printf(seq, list_entry(v, struct proto, node));
2530 return 0;
2531}
2532
2533static const struct seq_operations proto_seq_ops = {
2534 .start = proto_seq_start,
2535 .next = proto_seq_next,
2536 .stop = proto_seq_stop,
2537 .show = proto_seq_show,
2538};
2539
2540static int proto_seq_open(struct inode *inode, struct file *file)
2541{
2542 return seq_open_net(inode, file, &proto_seq_ops,
2543 sizeof(struct seq_net_private));
2544}
2545
2546static const struct file_operations proto_seq_fops = {
2547 .owner = THIS_MODULE,
2548 .open = proto_seq_open,
2549 .read = seq_read,
2550 .llseek = seq_lseek,
2551 .release = seq_release_net,
2552};
2553
2554static __net_init int proto_init_net(struct net *net)
2555{
2556 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2557 return -ENOMEM;
2558
2559 return 0;
2560}
2561
2562static __net_exit void proto_exit_net(struct net *net)
2563{
2564 proc_net_remove(net, "protocols");
2565}
2566
2567
2568static __net_initdata struct pernet_operations proto_net_ops = {
2569 .init = proto_init_net,
2570 .exit = proto_exit_net,
2571};
2572
2573static int __init proto_init(void)
2574{
2575 return register_pernet_subsys(&proto_net_ops);
2576}
2577
2578subsys_initcall(proto_init);
2579
2580#endif /* PROC_FS */