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 <linux/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/udp.h>
111#include <linux/init.h>
112#include <linux/highmem.h>
113#include <linux/user_namespace.h>
114#include <linux/static_key.h>
115#include <linux/memcontrol.h>
116#include <linux/prefetch.h>
117#include <linux/compat.h>
118#include <linux/mroute.h>
119#include <linux/mroute6.h>
120#include <linux/icmpv6.h>
121
122#include <linux/uaccess.h>
123
124#include <linux/netdevice.h>
125#include <net/protocol.h>
126#include <linux/skbuff.h>
127#include <linux/skbuff_ref.h>
128#include <net/net_namespace.h>
129#include <net/request_sock.h>
130#include <net/sock.h>
131#include <net/proto_memory.h>
132#include <linux/net_tstamp.h>
133#include <net/xfrm.h>
134#include <linux/ipsec.h>
135#include <net/cls_cgroup.h>
136#include <net/netprio_cgroup.h>
137#include <linux/sock_diag.h>
138
139#include <linux/filter.h>
140#include <net/sock_reuseport.h>
141#include <net/bpf_sk_storage.h>
142
143#include <trace/events/sock.h>
144
145#include <net/tcp.h>
146#include <net/busy_poll.h>
147#include <net/phonet/phonet.h>
148
149#include <linux/ethtool.h>
150
151#include "dev.h"
152
153static DEFINE_MUTEX(proto_list_mutex);
154static LIST_HEAD(proto_list);
155
156static void sock_def_write_space_wfree(struct sock *sk);
157static void sock_def_write_space(struct sock *sk);
158
159/**
160 * sk_ns_capable - General socket capability test
161 * @sk: Socket to use a capability on or through
162 * @user_ns: The user namespace of the capability to use
163 * @cap: The capability to use
164 *
165 * Test to see if the opener of the socket had when the socket was
166 * created and the current process has the capability @cap in the user
167 * namespace @user_ns.
168 */
169bool sk_ns_capable(const struct sock *sk,
170 struct user_namespace *user_ns, int cap)
171{
172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
173 ns_capable(user_ns, cap);
174}
175EXPORT_SYMBOL(sk_ns_capable);
176
177/**
178 * sk_capable - Socket global capability test
179 * @sk: Socket to use a capability on or through
180 * @cap: The global capability to use
181 *
182 * Test to see if the opener of the socket had when the socket was
183 * created and the current process has the capability @cap in all user
184 * namespaces.
185 */
186bool sk_capable(const struct sock *sk, int cap)
187{
188 return sk_ns_capable(sk, &init_user_ns, cap);
189}
190EXPORT_SYMBOL(sk_capable);
191
192/**
193 * sk_net_capable - Network namespace socket capability test
194 * @sk: Socket to use a capability on or through
195 * @cap: The capability to use
196 *
197 * Test to see if the opener of the socket had when the socket was created
198 * and the current process has the capability @cap over the network namespace
199 * the socket is a member of.
200 */
201bool sk_net_capable(const struct sock *sk, int cap)
202{
203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
204}
205EXPORT_SYMBOL(sk_net_capable);
206
207/*
208 * Each address family might have different locking rules, so we have
209 * one slock key per address family and separate keys for internal and
210 * userspace sockets.
211 */
212static struct lock_class_key af_family_keys[AF_MAX];
213static struct lock_class_key af_family_kern_keys[AF_MAX];
214static struct lock_class_key af_family_slock_keys[AF_MAX];
215static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
216
217/*
218 * Make lock validator output more readable. (we pre-construct these
219 * strings build-time, so that runtime initialization of socket
220 * locks is fast):
221 */
222
223#define _sock_locks(x) \
224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
233 x "27" , x "28" , x "AF_CAN" , \
234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
239 x "AF_MCTP" , \
240 x "AF_MAX"
241
242static const char *const af_family_key_strings[AF_MAX+1] = {
243 _sock_locks("sk_lock-")
244};
245static const char *const af_family_slock_key_strings[AF_MAX+1] = {
246 _sock_locks("slock-")
247};
248static const char *const af_family_clock_key_strings[AF_MAX+1] = {
249 _sock_locks("clock-")
250};
251
252static const char *const af_family_kern_key_strings[AF_MAX+1] = {
253 _sock_locks("k-sk_lock-")
254};
255static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
256 _sock_locks("k-slock-")
257};
258static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
259 _sock_locks("k-clock-")
260};
261static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
262 _sock_locks("rlock-")
263};
264static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
265 _sock_locks("wlock-")
266};
267static const char *const af_family_elock_key_strings[AF_MAX+1] = {
268 _sock_locks("elock-")
269};
270
271/*
272 * sk_callback_lock and sk queues locking rules are per-address-family,
273 * so split the lock classes by using a per-AF key:
274 */
275static struct lock_class_key af_callback_keys[AF_MAX];
276static struct lock_class_key af_rlock_keys[AF_MAX];
277static struct lock_class_key af_wlock_keys[AF_MAX];
278static struct lock_class_key af_elock_keys[AF_MAX];
279static struct lock_class_key af_kern_callback_keys[AF_MAX];
280
281/* Run time adjustable parameters. */
282__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
283EXPORT_SYMBOL(sysctl_wmem_max);
284__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
285EXPORT_SYMBOL(sysctl_rmem_max);
286__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
287__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
288
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) >= READ_ONCE(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, READ_ONCE(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 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
824 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
825 if (val) {
826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
827 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
828 }
829}
830
831void sock_enable_timestamps(struct sock *sk)
832{
833 lock_sock(sk);
834 __sock_set_timestamps(sk, true, false, true);
835 release_sock(sk);
836}
837EXPORT_SYMBOL(sock_enable_timestamps);
838
839void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
840{
841 switch (optname) {
842 case SO_TIMESTAMP_OLD:
843 __sock_set_timestamps(sk, valbool, false, false);
844 break;
845 case SO_TIMESTAMP_NEW:
846 __sock_set_timestamps(sk, valbool, true, false);
847 break;
848 case SO_TIMESTAMPNS_OLD:
849 __sock_set_timestamps(sk, valbool, false, true);
850 break;
851 case SO_TIMESTAMPNS_NEW:
852 __sock_set_timestamps(sk, valbool, true, true);
853 break;
854 }
855}
856
857static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
858{
859 struct net *net = sock_net(sk);
860 struct net_device *dev = NULL;
861 bool match = false;
862 int *vclock_index;
863 int i, num;
864
865 if (sk->sk_bound_dev_if)
866 dev = dev_get_by_index(net, sk->sk_bound_dev_if);
867
868 if (!dev) {
869 pr_err("%s: sock not bind to device\n", __func__);
870 return -EOPNOTSUPP;
871 }
872
873 num = ethtool_get_phc_vclocks(dev, &vclock_index);
874 dev_put(dev);
875
876 for (i = 0; i < num; i++) {
877 if (*(vclock_index + i) == phc_index) {
878 match = true;
879 break;
880 }
881 }
882
883 if (num > 0)
884 kfree(vclock_index);
885
886 if (!match)
887 return -EINVAL;
888
889 WRITE_ONCE(sk->sk_bind_phc, phc_index);
890
891 return 0;
892}
893
894int sock_set_timestamping(struct sock *sk, int optname,
895 struct so_timestamping timestamping)
896{
897 int val = timestamping.flags;
898 int ret;
899
900 if (val & ~SOF_TIMESTAMPING_MASK)
901 return -EINVAL;
902
903 if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
904 !(val & SOF_TIMESTAMPING_OPT_ID))
905 return -EINVAL;
906
907 if (val & SOF_TIMESTAMPING_OPT_ID &&
908 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
909 if (sk_is_tcp(sk)) {
910 if ((1 << sk->sk_state) &
911 (TCPF_CLOSE | TCPF_LISTEN))
912 return -EINVAL;
913 if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
915 else
916 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
917 } else {
918 atomic_set(&sk->sk_tskey, 0);
919 }
920 }
921
922 if (val & SOF_TIMESTAMPING_OPT_STATS &&
923 !(val & SOF_TIMESTAMPING_OPT_TSONLY))
924 return -EINVAL;
925
926 if (val & SOF_TIMESTAMPING_BIND_PHC) {
927 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
928 if (ret)
929 return ret;
930 }
931
932 WRITE_ONCE(sk->sk_tsflags, val);
933 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
934
935 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
936 sock_enable_timestamp(sk,
937 SOCK_TIMESTAMPING_RX_SOFTWARE);
938 else
939 sock_disable_timestamp(sk,
940 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
941 return 0;
942}
943
944void sock_set_keepalive(struct sock *sk)
945{
946 lock_sock(sk);
947 if (sk->sk_prot->keepalive)
948 sk->sk_prot->keepalive(sk, true);
949 sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
950 release_sock(sk);
951}
952EXPORT_SYMBOL(sock_set_keepalive);
953
954static void __sock_set_rcvbuf(struct sock *sk, int val)
955{
956 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it
957 * as a negative value.
958 */
959 val = min_t(int, val, INT_MAX / 2);
960 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
961
962 /* We double it on the way in to account for "struct sk_buff" etc.
963 * overhead. Applications assume that the SO_RCVBUF setting they make
964 * will allow that much actual data to be received on that socket.
965 *
966 * Applications are unaware that "struct sk_buff" and other overheads
967 * allocate from the receive buffer during socket buffer allocation.
968 *
969 * And after considering the possible alternatives, returning the value
970 * we actually used in getsockopt is the most desirable behavior.
971 */
972 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
973}
974
975void sock_set_rcvbuf(struct sock *sk, int val)
976{
977 lock_sock(sk);
978 __sock_set_rcvbuf(sk, val);
979 release_sock(sk);
980}
981EXPORT_SYMBOL(sock_set_rcvbuf);
982
983static void __sock_set_mark(struct sock *sk, u32 val)
984{
985 if (val != sk->sk_mark) {
986 WRITE_ONCE(sk->sk_mark, val);
987 sk_dst_reset(sk);
988 }
989}
990
991void sock_set_mark(struct sock *sk, u32 val)
992{
993 lock_sock(sk);
994 __sock_set_mark(sk, val);
995 release_sock(sk);
996}
997EXPORT_SYMBOL(sock_set_mark);
998
999static void sock_release_reserved_memory(struct sock *sk, int bytes)
1000{
1001 /* Round down bytes to multiple of pages */
1002 bytes = round_down(bytes, PAGE_SIZE);
1003
1004 WARN_ON(bytes > sk->sk_reserved_mem);
1005 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1006 sk_mem_reclaim(sk);
1007}
1008
1009static int sock_reserve_memory(struct sock *sk, int bytes)
1010{
1011 long allocated;
1012 bool charged;
1013 int pages;
1014
1015 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1016 return -EOPNOTSUPP;
1017
1018 if (!bytes)
1019 return 0;
1020
1021 pages = sk_mem_pages(bytes);
1022
1023 /* pre-charge to memcg */
1024 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1025 GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1026 if (!charged)
1027 return -ENOMEM;
1028
1029 /* pre-charge to forward_alloc */
1030 sk_memory_allocated_add(sk, pages);
1031 allocated = sk_memory_allocated(sk);
1032 /* If the system goes into memory pressure with this
1033 * precharge, give up and return error.
1034 */
1035 if (allocated > sk_prot_mem_limits(sk, 1)) {
1036 sk_memory_allocated_sub(sk, pages);
1037 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1038 return -ENOMEM;
1039 }
1040 sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1041
1042 WRITE_ONCE(sk->sk_reserved_mem,
1043 sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1044
1045 return 0;
1046}
1047
1048#ifdef CONFIG_PAGE_POOL
1049
1050/* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1051 * in 1 syscall. The limit exists to limit the amount of memory the kernel
1052 * allocates to copy these tokens, and to prevent looping over the frags for
1053 * too long.
1054 */
1055#define MAX_DONTNEED_TOKENS 128
1056#define MAX_DONTNEED_FRAGS 1024
1057
1058static noinline_for_stack int
1059sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1060{
1061 unsigned int num_tokens, i, j, k, netmem_num = 0;
1062 struct dmabuf_token *tokens;
1063 int ret = 0, num_frags = 0;
1064 netmem_ref netmems[16];
1065
1066 if (!sk_is_tcp(sk))
1067 return -EBADF;
1068
1069 if (optlen % sizeof(*tokens) ||
1070 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1071 return -EINVAL;
1072
1073 num_tokens = optlen / sizeof(*tokens);
1074 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL);
1075 if (!tokens)
1076 return -ENOMEM;
1077
1078 if (copy_from_sockptr(tokens, optval, optlen)) {
1079 kvfree(tokens);
1080 return -EFAULT;
1081 }
1082
1083 xa_lock_bh(&sk->sk_user_frags);
1084 for (i = 0; i < num_tokens; i++) {
1085 for (j = 0; j < tokens[i].token_count; j++) {
1086 if (++num_frags > MAX_DONTNEED_FRAGS)
1087 goto frag_limit_reached;
1088
1089 netmem_ref netmem = (__force netmem_ref)__xa_erase(
1090 &sk->sk_user_frags, tokens[i].token_start + j);
1091
1092 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1093 continue;
1094
1095 netmems[netmem_num++] = netmem;
1096 if (netmem_num == ARRAY_SIZE(netmems)) {
1097 xa_unlock_bh(&sk->sk_user_frags);
1098 for (k = 0; k < netmem_num; k++)
1099 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1100 netmem_num = 0;
1101 xa_lock_bh(&sk->sk_user_frags);
1102 }
1103 ret++;
1104 }
1105 }
1106
1107frag_limit_reached:
1108 xa_unlock_bh(&sk->sk_user_frags);
1109 for (k = 0; k < netmem_num; k++)
1110 WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1111
1112 kvfree(tokens);
1113 return ret;
1114}
1115#endif
1116
1117void sockopt_lock_sock(struct sock *sk)
1118{
1119 /* When current->bpf_ctx is set, the setsockopt is called from
1120 * a bpf prog. bpf has ensured the sk lock has been
1121 * acquired before calling setsockopt().
1122 */
1123 if (has_current_bpf_ctx())
1124 return;
1125
1126 lock_sock(sk);
1127}
1128EXPORT_SYMBOL(sockopt_lock_sock);
1129
1130void sockopt_release_sock(struct sock *sk)
1131{
1132 if (has_current_bpf_ctx())
1133 return;
1134
1135 release_sock(sk);
1136}
1137EXPORT_SYMBOL(sockopt_release_sock);
1138
1139bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1140{
1141 return has_current_bpf_ctx() || ns_capable(ns, cap);
1142}
1143EXPORT_SYMBOL(sockopt_ns_capable);
1144
1145bool sockopt_capable(int cap)
1146{
1147 return has_current_bpf_ctx() || capable(cap);
1148}
1149EXPORT_SYMBOL(sockopt_capable);
1150
1151static int sockopt_validate_clockid(__kernel_clockid_t value)
1152{
1153 switch (value) {
1154 case CLOCK_REALTIME:
1155 case CLOCK_MONOTONIC:
1156 case CLOCK_TAI:
1157 return 0;
1158 }
1159 return -EINVAL;
1160}
1161
1162/*
1163 * This is meant for all protocols to use and covers goings on
1164 * at the socket level. Everything here is generic.
1165 */
1166
1167int sk_setsockopt(struct sock *sk, int level, int optname,
1168 sockptr_t optval, unsigned int optlen)
1169{
1170 struct so_timestamping timestamping;
1171 struct socket *sock = sk->sk_socket;
1172 struct sock_txtime sk_txtime;
1173 int val;
1174 int valbool;
1175 struct linger ling;
1176 int ret = 0;
1177
1178 /*
1179 * Options without arguments
1180 */
1181
1182 if (optname == SO_BINDTODEVICE)
1183 return sock_setbindtodevice(sk, optval, optlen);
1184
1185 if (optlen < sizeof(int))
1186 return -EINVAL;
1187
1188 if (copy_from_sockptr(&val, optval, sizeof(val)))
1189 return -EFAULT;
1190
1191 valbool = val ? 1 : 0;
1192
1193 /* handle options which do not require locking the socket. */
1194 switch (optname) {
1195 case SO_PRIORITY:
1196 if ((val >= 0 && val <= 6) ||
1197 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
1198 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1199 sock_set_priority(sk, val);
1200 return 0;
1201 }
1202 return -EPERM;
1203 case SO_PASSSEC:
1204 assign_bit(SOCK_PASSSEC, &sock->flags, valbool);
1205 return 0;
1206 case SO_PASSCRED:
1207 assign_bit(SOCK_PASSCRED, &sock->flags, valbool);
1208 return 0;
1209 case SO_PASSPIDFD:
1210 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool);
1211 return 0;
1212 case SO_TYPE:
1213 case SO_PROTOCOL:
1214 case SO_DOMAIN:
1215 case SO_ERROR:
1216 return -ENOPROTOOPT;
1217#ifdef CONFIG_NET_RX_BUSY_POLL
1218 case SO_BUSY_POLL:
1219 if (val < 0)
1220 return -EINVAL;
1221 WRITE_ONCE(sk->sk_ll_usec, val);
1222 return 0;
1223 case SO_PREFER_BUSY_POLL:
1224 if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1225 return -EPERM;
1226 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1227 return 0;
1228 case SO_BUSY_POLL_BUDGET:
1229 if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1230 !sockopt_capable(CAP_NET_ADMIN))
1231 return -EPERM;
1232 if (val < 0 || val > U16_MAX)
1233 return -EINVAL;
1234 WRITE_ONCE(sk->sk_busy_poll_budget, val);
1235 return 0;
1236#endif
1237 case SO_MAX_PACING_RATE:
1238 {
1239 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1240 unsigned long pacing_rate;
1241
1242 if (sizeof(ulval) != sizeof(val) &&
1243 optlen >= sizeof(ulval) &&
1244 copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1245 return -EFAULT;
1246 }
1247 if (ulval != ~0UL)
1248 cmpxchg(&sk->sk_pacing_status,
1249 SK_PACING_NONE,
1250 SK_PACING_NEEDED);
1251 /* Pairs with READ_ONCE() from sk_getsockopt() */
1252 WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1253 pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1254 if (ulval < pacing_rate)
1255 WRITE_ONCE(sk->sk_pacing_rate, ulval);
1256 return 0;
1257 }
1258 case SO_TXREHASH:
1259 if (val < -1 || val > 1)
1260 return -EINVAL;
1261 if ((u8)val == SOCK_TXREHASH_DEFAULT)
1262 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1263 /* Paired with READ_ONCE() in tcp_rtx_synack()
1264 * and sk_getsockopt().
1265 */
1266 WRITE_ONCE(sk->sk_txrehash, (u8)val);
1267 return 0;
1268 case SO_PEEK_OFF:
1269 {
1270 int (*set_peek_off)(struct sock *sk, int val);
1271
1272 set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1273 if (set_peek_off)
1274 ret = set_peek_off(sk, val);
1275 else
1276 ret = -EOPNOTSUPP;
1277 return ret;
1278 }
1279#ifdef CONFIG_PAGE_POOL
1280 case SO_DEVMEM_DONTNEED:
1281 return sock_devmem_dontneed(sk, optval, optlen);
1282#endif
1283 }
1284
1285 sockopt_lock_sock(sk);
1286
1287 switch (optname) {
1288 case SO_DEBUG:
1289 if (val && !sockopt_capable(CAP_NET_ADMIN))
1290 ret = -EACCES;
1291 else
1292 sock_valbool_flag(sk, SOCK_DBG, valbool);
1293 break;
1294 case SO_REUSEADDR:
1295 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1296 break;
1297 case SO_REUSEPORT:
1298 if (valbool && !sk_is_inet(sk))
1299 ret = -EOPNOTSUPP;
1300 else
1301 sk->sk_reuseport = valbool;
1302 break;
1303 case SO_DONTROUTE:
1304 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1305 sk_dst_reset(sk);
1306 break;
1307 case SO_BROADCAST:
1308 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1309 break;
1310 case SO_SNDBUF:
1311 /* Don't error on this BSD doesn't and if you think
1312 * about it this is right. Otherwise apps have to
1313 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1314 * are treated in BSD as hints
1315 */
1316 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1317set_sndbuf:
1318 /* Ensure val * 2 fits into an int, to prevent max_t()
1319 * from treating it as a negative value.
1320 */
1321 val = min_t(int, val, INT_MAX / 2);
1322 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1323 WRITE_ONCE(sk->sk_sndbuf,
1324 max_t(int, val * 2, SOCK_MIN_SNDBUF));
1325 /* Wake up sending tasks if we upped the value. */
1326 sk->sk_write_space(sk);
1327 break;
1328
1329 case SO_SNDBUFFORCE:
1330 if (!sockopt_capable(CAP_NET_ADMIN)) {
1331 ret = -EPERM;
1332 break;
1333 }
1334
1335 /* No negative values (to prevent underflow, as val will be
1336 * multiplied by 2).
1337 */
1338 if (val < 0)
1339 val = 0;
1340 goto set_sndbuf;
1341
1342 case SO_RCVBUF:
1343 /* Don't error on this BSD doesn't and if you think
1344 * about it this is right. Otherwise apps have to
1345 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1346 * are treated in BSD as hints
1347 */
1348 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1349 break;
1350
1351 case SO_RCVBUFFORCE:
1352 if (!sockopt_capable(CAP_NET_ADMIN)) {
1353 ret = -EPERM;
1354 break;
1355 }
1356
1357 /* No negative values (to prevent underflow, as val will be
1358 * multiplied by 2).
1359 */
1360 __sock_set_rcvbuf(sk, max(val, 0));
1361 break;
1362
1363 case SO_KEEPALIVE:
1364 if (sk->sk_prot->keepalive)
1365 sk->sk_prot->keepalive(sk, valbool);
1366 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1367 break;
1368
1369 case SO_OOBINLINE:
1370 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1371 break;
1372
1373 case SO_NO_CHECK:
1374 sk->sk_no_check_tx = valbool;
1375 break;
1376
1377 case SO_LINGER:
1378 if (optlen < sizeof(ling)) {
1379 ret = -EINVAL; /* 1003.1g */
1380 break;
1381 }
1382 if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1383 ret = -EFAULT;
1384 break;
1385 }
1386 if (!ling.l_onoff) {
1387 sock_reset_flag(sk, SOCK_LINGER);
1388 } else {
1389 unsigned long t_sec = ling.l_linger;
1390
1391 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1392 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1393 else
1394 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1395 sock_set_flag(sk, SOCK_LINGER);
1396 }
1397 break;
1398
1399 case SO_BSDCOMPAT:
1400 break;
1401
1402 case SO_TIMESTAMP_OLD:
1403 case SO_TIMESTAMP_NEW:
1404 case SO_TIMESTAMPNS_OLD:
1405 case SO_TIMESTAMPNS_NEW:
1406 sock_set_timestamp(sk, optname, valbool);
1407 break;
1408
1409 case SO_TIMESTAMPING_NEW:
1410 case SO_TIMESTAMPING_OLD:
1411 if (optlen == sizeof(timestamping)) {
1412 if (copy_from_sockptr(×tamping, optval,
1413 sizeof(timestamping))) {
1414 ret = -EFAULT;
1415 break;
1416 }
1417 } else {
1418 memset(×tamping, 0, sizeof(timestamping));
1419 timestamping.flags = val;
1420 }
1421 ret = sock_set_timestamping(sk, optname, timestamping);
1422 break;
1423
1424 case SO_RCVLOWAT:
1425 {
1426 int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1427
1428 if (val < 0)
1429 val = INT_MAX;
1430 if (sock)
1431 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1432 if (set_rcvlowat)
1433 ret = set_rcvlowat(sk, val);
1434 else
1435 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1436 break;
1437 }
1438 case SO_RCVTIMEO_OLD:
1439 case SO_RCVTIMEO_NEW:
1440 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval,
1441 optlen, optname == SO_RCVTIMEO_OLD);
1442 break;
1443
1444 case SO_SNDTIMEO_OLD:
1445 case SO_SNDTIMEO_NEW:
1446 ret = sock_set_timeout(&sk->sk_sndtimeo, optval,
1447 optlen, optname == SO_SNDTIMEO_OLD);
1448 break;
1449
1450 case SO_ATTACH_FILTER: {
1451 struct sock_fprog fprog;
1452
1453 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1454 if (!ret)
1455 ret = sk_attach_filter(&fprog, sk);
1456 break;
1457 }
1458 case SO_ATTACH_BPF:
1459 ret = -EINVAL;
1460 if (optlen == sizeof(u32)) {
1461 u32 ufd;
1462
1463 ret = -EFAULT;
1464 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1465 break;
1466
1467 ret = sk_attach_bpf(ufd, sk);
1468 }
1469 break;
1470
1471 case SO_ATTACH_REUSEPORT_CBPF: {
1472 struct sock_fprog fprog;
1473
1474 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1475 if (!ret)
1476 ret = sk_reuseport_attach_filter(&fprog, sk);
1477 break;
1478 }
1479 case SO_ATTACH_REUSEPORT_EBPF:
1480 ret = -EINVAL;
1481 if (optlen == sizeof(u32)) {
1482 u32 ufd;
1483
1484 ret = -EFAULT;
1485 if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1486 break;
1487
1488 ret = sk_reuseport_attach_bpf(ufd, sk);
1489 }
1490 break;
1491
1492 case SO_DETACH_REUSEPORT_BPF:
1493 ret = reuseport_detach_prog(sk);
1494 break;
1495
1496 case SO_DETACH_FILTER:
1497 ret = sk_detach_filter(sk);
1498 break;
1499
1500 case SO_LOCK_FILTER:
1501 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1502 ret = -EPERM;
1503 else
1504 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1505 break;
1506
1507 case SO_MARK:
1508 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1509 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1510 ret = -EPERM;
1511 break;
1512 }
1513
1514 __sock_set_mark(sk, val);
1515 break;
1516 case SO_RCVMARK:
1517 sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1518 break;
1519
1520 case SO_RXQ_OVFL:
1521 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1522 break;
1523
1524 case SO_WIFI_STATUS:
1525 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1526 break;
1527
1528 case SO_NOFCS:
1529 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1530 break;
1531
1532 case SO_SELECT_ERR_QUEUE:
1533 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1534 break;
1535
1536
1537 case SO_INCOMING_CPU:
1538 reuseport_update_incoming_cpu(sk, val);
1539 break;
1540
1541 case SO_CNX_ADVICE:
1542 if (val == 1)
1543 dst_negative_advice(sk);
1544 break;
1545
1546 case SO_ZEROCOPY:
1547 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1548 if (!(sk_is_tcp(sk) ||
1549 (sk->sk_type == SOCK_DGRAM &&
1550 sk->sk_protocol == IPPROTO_UDP)))
1551 ret = -EOPNOTSUPP;
1552 } else if (sk->sk_family != PF_RDS) {
1553 ret = -EOPNOTSUPP;
1554 }
1555 if (!ret) {
1556 if (val < 0 || val > 1)
1557 ret = -EINVAL;
1558 else
1559 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1560 }
1561 break;
1562
1563 case SO_TXTIME:
1564 if (optlen != sizeof(struct sock_txtime)) {
1565 ret = -EINVAL;
1566 break;
1567 } else if (copy_from_sockptr(&sk_txtime, optval,
1568 sizeof(struct sock_txtime))) {
1569 ret = -EFAULT;
1570 break;
1571 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1572 ret = -EINVAL;
1573 break;
1574 }
1575 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1576 * scheduler has enough safe guards.
1577 */
1578 if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1579 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1580 ret = -EPERM;
1581 break;
1582 }
1583
1584 ret = sockopt_validate_clockid(sk_txtime.clockid);
1585 if (ret)
1586 break;
1587
1588 sock_valbool_flag(sk, SOCK_TXTIME, true);
1589 sk->sk_clockid = sk_txtime.clockid;
1590 sk->sk_txtime_deadline_mode =
1591 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1592 sk->sk_txtime_report_errors =
1593 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1594 break;
1595
1596 case SO_BINDTOIFINDEX:
1597 ret = sock_bindtoindex_locked(sk, val);
1598 break;
1599
1600 case SO_BUF_LOCK:
1601 if (val & ~SOCK_BUF_LOCK_MASK) {
1602 ret = -EINVAL;
1603 break;
1604 }
1605 sk->sk_userlocks = val | (sk->sk_userlocks &
1606 ~SOCK_BUF_LOCK_MASK);
1607 break;
1608
1609 case SO_RESERVE_MEM:
1610 {
1611 int delta;
1612
1613 if (val < 0) {
1614 ret = -EINVAL;
1615 break;
1616 }
1617
1618 delta = val - sk->sk_reserved_mem;
1619 if (delta < 0)
1620 sock_release_reserved_memory(sk, -delta);
1621 else
1622 ret = sock_reserve_memory(sk, delta);
1623 break;
1624 }
1625
1626 default:
1627 ret = -ENOPROTOOPT;
1628 break;
1629 }
1630 sockopt_release_sock(sk);
1631 return ret;
1632}
1633
1634int sock_setsockopt(struct socket *sock, int level, int optname,
1635 sockptr_t optval, unsigned int optlen)
1636{
1637 return sk_setsockopt(sock->sk, level, optname,
1638 optval, optlen);
1639}
1640EXPORT_SYMBOL(sock_setsockopt);
1641
1642static const struct cred *sk_get_peer_cred(struct sock *sk)
1643{
1644 const struct cred *cred;
1645
1646 spin_lock(&sk->sk_peer_lock);
1647 cred = get_cred(sk->sk_peer_cred);
1648 spin_unlock(&sk->sk_peer_lock);
1649
1650 return cred;
1651}
1652
1653static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1654 struct ucred *ucred)
1655{
1656 ucred->pid = pid_vnr(pid);
1657 ucred->uid = ucred->gid = -1;
1658 if (cred) {
1659 struct user_namespace *current_ns = current_user_ns();
1660
1661 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1662 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1663 }
1664}
1665
1666static int groups_to_user(sockptr_t dst, const struct group_info *src)
1667{
1668 struct user_namespace *user_ns = current_user_ns();
1669 int i;
1670
1671 for (i = 0; i < src->ngroups; i++) {
1672 gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1673
1674 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1675 return -EFAULT;
1676 }
1677
1678 return 0;
1679}
1680
1681int sk_getsockopt(struct sock *sk, int level, int optname,
1682 sockptr_t optval, sockptr_t optlen)
1683{
1684 struct socket *sock = sk->sk_socket;
1685
1686 union {
1687 int val;
1688 u64 val64;
1689 unsigned long ulval;
1690 struct linger ling;
1691 struct old_timeval32 tm32;
1692 struct __kernel_old_timeval tm;
1693 struct __kernel_sock_timeval stm;
1694 struct sock_txtime txtime;
1695 struct so_timestamping timestamping;
1696 } v;
1697
1698 int lv = sizeof(int);
1699 int len;
1700
1701 if (copy_from_sockptr(&len, optlen, sizeof(int)))
1702 return -EFAULT;
1703 if (len < 0)
1704 return -EINVAL;
1705
1706 memset(&v, 0, sizeof(v));
1707
1708 switch (optname) {
1709 case SO_DEBUG:
1710 v.val = sock_flag(sk, SOCK_DBG);
1711 break;
1712
1713 case SO_DONTROUTE:
1714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1715 break;
1716
1717 case SO_BROADCAST:
1718 v.val = sock_flag(sk, SOCK_BROADCAST);
1719 break;
1720
1721 case SO_SNDBUF:
1722 v.val = READ_ONCE(sk->sk_sndbuf);
1723 break;
1724
1725 case SO_RCVBUF:
1726 v.val = READ_ONCE(sk->sk_rcvbuf);
1727 break;
1728
1729 case SO_REUSEADDR:
1730 v.val = sk->sk_reuse;
1731 break;
1732
1733 case SO_REUSEPORT:
1734 v.val = sk->sk_reuseport;
1735 break;
1736
1737 case SO_KEEPALIVE:
1738 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1739 break;
1740
1741 case SO_TYPE:
1742 v.val = sk->sk_type;
1743 break;
1744
1745 case SO_PROTOCOL:
1746 v.val = sk->sk_protocol;
1747 break;
1748
1749 case SO_DOMAIN:
1750 v.val = sk->sk_family;
1751 break;
1752
1753 case SO_ERROR:
1754 v.val = -sock_error(sk);
1755 if (v.val == 0)
1756 v.val = xchg(&sk->sk_err_soft, 0);
1757 break;
1758
1759 case SO_OOBINLINE:
1760 v.val = sock_flag(sk, SOCK_URGINLINE);
1761 break;
1762
1763 case SO_NO_CHECK:
1764 v.val = sk->sk_no_check_tx;
1765 break;
1766
1767 case SO_PRIORITY:
1768 v.val = READ_ONCE(sk->sk_priority);
1769 break;
1770
1771 case SO_LINGER:
1772 lv = sizeof(v.ling);
1773 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1774 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ;
1775 break;
1776
1777 case SO_BSDCOMPAT:
1778 break;
1779
1780 case SO_TIMESTAMP_OLD:
1781 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1782 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1783 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1784 break;
1785
1786 case SO_TIMESTAMPNS_OLD:
1787 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1788 break;
1789
1790 case SO_TIMESTAMP_NEW:
1791 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1792 break;
1793
1794 case SO_TIMESTAMPNS_NEW:
1795 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1796 break;
1797
1798 case SO_TIMESTAMPING_OLD:
1799 case SO_TIMESTAMPING_NEW:
1800 lv = sizeof(v.timestamping);
1801 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1802 * returning the flags when they were set through the same option.
1803 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1804 */
1805 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1806 v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1807 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1808 }
1809 break;
1810
1811 case SO_RCVTIMEO_OLD:
1812 case SO_RCVTIMEO_NEW:
1813 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1814 SO_RCVTIMEO_OLD == optname);
1815 break;
1816
1817 case SO_SNDTIMEO_OLD:
1818 case SO_SNDTIMEO_NEW:
1819 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1820 SO_SNDTIMEO_OLD == optname);
1821 break;
1822
1823 case SO_RCVLOWAT:
1824 v.val = READ_ONCE(sk->sk_rcvlowat);
1825 break;
1826
1827 case SO_SNDLOWAT:
1828 v.val = 1;
1829 break;
1830
1831 case SO_PASSCRED:
1832 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1833 break;
1834
1835 case SO_PASSPIDFD:
1836 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags);
1837 break;
1838
1839 case SO_PEERCRED:
1840 {
1841 struct ucred peercred;
1842 if (len > sizeof(peercred))
1843 len = sizeof(peercred);
1844
1845 spin_lock(&sk->sk_peer_lock);
1846 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1847 spin_unlock(&sk->sk_peer_lock);
1848
1849 if (copy_to_sockptr(optval, &peercred, len))
1850 return -EFAULT;
1851 goto lenout;
1852 }
1853
1854 case SO_PEERPIDFD:
1855 {
1856 struct pid *peer_pid;
1857 struct file *pidfd_file = NULL;
1858 int pidfd;
1859
1860 if (len > sizeof(pidfd))
1861 len = sizeof(pidfd);
1862
1863 spin_lock(&sk->sk_peer_lock);
1864 peer_pid = get_pid(sk->sk_peer_pid);
1865 spin_unlock(&sk->sk_peer_lock);
1866
1867 if (!peer_pid)
1868 return -ENODATA;
1869
1870 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file);
1871 put_pid(peer_pid);
1872 if (pidfd < 0)
1873 return pidfd;
1874
1875 if (copy_to_sockptr(optval, &pidfd, len) ||
1876 copy_to_sockptr(optlen, &len, sizeof(int))) {
1877 put_unused_fd(pidfd);
1878 fput(pidfd_file);
1879
1880 return -EFAULT;
1881 }
1882
1883 fd_install(pidfd, pidfd_file);
1884 return 0;
1885 }
1886
1887 case SO_PEERGROUPS:
1888 {
1889 const struct cred *cred;
1890 int ret, n;
1891
1892 cred = sk_get_peer_cred(sk);
1893 if (!cred)
1894 return -ENODATA;
1895
1896 n = cred->group_info->ngroups;
1897 if (len < n * sizeof(gid_t)) {
1898 len = n * sizeof(gid_t);
1899 put_cred(cred);
1900 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1901 }
1902 len = n * sizeof(gid_t);
1903
1904 ret = groups_to_user(optval, cred->group_info);
1905 put_cred(cred);
1906 if (ret)
1907 return ret;
1908 goto lenout;
1909 }
1910
1911 case SO_PEERNAME:
1912 {
1913 struct sockaddr_storage address;
1914
1915 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1916 if (lv < 0)
1917 return -ENOTCONN;
1918 if (lv < len)
1919 return -EINVAL;
1920 if (copy_to_sockptr(optval, &address, len))
1921 return -EFAULT;
1922 goto lenout;
1923 }
1924
1925 /* Dubious BSD thing... Probably nobody even uses it, but
1926 * the UNIX standard wants it for whatever reason... -DaveM
1927 */
1928 case SO_ACCEPTCONN:
1929 v.val = sk->sk_state == TCP_LISTEN;
1930 break;
1931
1932 case SO_PASSSEC:
1933 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1934 break;
1935
1936 case SO_PEERSEC:
1937 return security_socket_getpeersec_stream(sock,
1938 optval, optlen, len);
1939
1940 case SO_MARK:
1941 v.val = READ_ONCE(sk->sk_mark);
1942 break;
1943
1944 case SO_RCVMARK:
1945 v.val = sock_flag(sk, SOCK_RCVMARK);
1946 break;
1947
1948 case SO_RXQ_OVFL:
1949 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1950 break;
1951
1952 case SO_WIFI_STATUS:
1953 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1954 break;
1955
1956 case SO_PEEK_OFF:
1957 if (!READ_ONCE(sock->ops)->set_peek_off)
1958 return -EOPNOTSUPP;
1959
1960 v.val = READ_ONCE(sk->sk_peek_off);
1961 break;
1962 case SO_NOFCS:
1963 v.val = sock_flag(sk, SOCK_NOFCS);
1964 break;
1965
1966 case SO_BINDTODEVICE:
1967 return sock_getbindtodevice(sk, optval, optlen, len);
1968
1969 case SO_GET_FILTER:
1970 len = sk_get_filter(sk, optval, len);
1971 if (len < 0)
1972 return len;
1973
1974 goto lenout;
1975
1976 case SO_LOCK_FILTER:
1977 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1978 break;
1979
1980 case SO_BPF_EXTENSIONS:
1981 v.val = bpf_tell_extensions();
1982 break;
1983
1984 case SO_SELECT_ERR_QUEUE:
1985 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1986 break;
1987
1988#ifdef CONFIG_NET_RX_BUSY_POLL
1989 case SO_BUSY_POLL:
1990 v.val = READ_ONCE(sk->sk_ll_usec);
1991 break;
1992 case SO_PREFER_BUSY_POLL:
1993 v.val = READ_ONCE(sk->sk_prefer_busy_poll);
1994 break;
1995#endif
1996
1997 case SO_MAX_PACING_RATE:
1998 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
1999 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2000 lv = sizeof(v.ulval);
2001 v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2002 } else {
2003 /* 32bit version */
2004 v.val = min_t(unsigned long, ~0U,
2005 READ_ONCE(sk->sk_max_pacing_rate));
2006 }
2007 break;
2008
2009 case SO_INCOMING_CPU:
2010 v.val = READ_ONCE(sk->sk_incoming_cpu);
2011 break;
2012
2013 case SO_MEMINFO:
2014 {
2015 u32 meminfo[SK_MEMINFO_VARS];
2016
2017 sk_get_meminfo(sk, meminfo);
2018
2019 len = min_t(unsigned int, len, sizeof(meminfo));
2020 if (copy_to_sockptr(optval, &meminfo, len))
2021 return -EFAULT;
2022
2023 goto lenout;
2024 }
2025
2026#ifdef CONFIG_NET_RX_BUSY_POLL
2027 case SO_INCOMING_NAPI_ID:
2028 v.val = READ_ONCE(sk->sk_napi_id);
2029
2030 /* aggregate non-NAPI IDs down to 0 */
2031 if (v.val < MIN_NAPI_ID)
2032 v.val = 0;
2033
2034 break;
2035#endif
2036
2037 case SO_COOKIE:
2038 lv = sizeof(u64);
2039 if (len < lv)
2040 return -EINVAL;
2041 v.val64 = sock_gen_cookie(sk);
2042 break;
2043
2044 case SO_ZEROCOPY:
2045 v.val = sock_flag(sk, SOCK_ZEROCOPY);
2046 break;
2047
2048 case SO_TXTIME:
2049 lv = sizeof(v.txtime);
2050 v.txtime.clockid = sk->sk_clockid;
2051 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2052 SOF_TXTIME_DEADLINE_MODE : 0;
2053 v.txtime.flags |= sk->sk_txtime_report_errors ?
2054 SOF_TXTIME_REPORT_ERRORS : 0;
2055 break;
2056
2057 case SO_BINDTOIFINDEX:
2058 v.val = READ_ONCE(sk->sk_bound_dev_if);
2059 break;
2060
2061 case SO_NETNS_COOKIE:
2062 lv = sizeof(u64);
2063 if (len != lv)
2064 return -EINVAL;
2065 v.val64 = sock_net(sk)->net_cookie;
2066 break;
2067
2068 case SO_BUF_LOCK:
2069 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2070 break;
2071
2072 case SO_RESERVE_MEM:
2073 v.val = READ_ONCE(sk->sk_reserved_mem);
2074 break;
2075
2076 case SO_TXREHASH:
2077 /* Paired with WRITE_ONCE() in sk_setsockopt() */
2078 v.val = READ_ONCE(sk->sk_txrehash);
2079 break;
2080
2081 default:
2082 /* We implement the SO_SNDLOWAT etc to not be settable
2083 * (1003.1g 7).
2084 */
2085 return -ENOPROTOOPT;
2086 }
2087
2088 if (len > lv)
2089 len = lv;
2090 if (copy_to_sockptr(optval, &v, len))
2091 return -EFAULT;
2092lenout:
2093 if (copy_to_sockptr(optlen, &len, sizeof(int)))
2094 return -EFAULT;
2095 return 0;
2096}
2097
2098/*
2099 * Initialize an sk_lock.
2100 *
2101 * (We also register the sk_lock with the lock validator.)
2102 */
2103static inline void sock_lock_init(struct sock *sk)
2104{
2105 if (sk->sk_kern_sock)
2106 sock_lock_init_class_and_name(
2107 sk,
2108 af_family_kern_slock_key_strings[sk->sk_family],
2109 af_family_kern_slock_keys + sk->sk_family,
2110 af_family_kern_key_strings[sk->sk_family],
2111 af_family_kern_keys + sk->sk_family);
2112 else
2113 sock_lock_init_class_and_name(
2114 sk,
2115 af_family_slock_key_strings[sk->sk_family],
2116 af_family_slock_keys + sk->sk_family,
2117 af_family_key_strings[sk->sk_family],
2118 af_family_keys + sk->sk_family);
2119}
2120
2121/*
2122 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2123 * even temporarily, because of RCU lookups. sk_node should also be left as is.
2124 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2125 */
2126static void sock_copy(struct sock *nsk, const struct sock *osk)
2127{
2128 const struct proto *prot = READ_ONCE(osk->sk_prot);
2129#ifdef CONFIG_SECURITY_NETWORK
2130 void *sptr = nsk->sk_security;
2131#endif
2132
2133 /* If we move sk_tx_queue_mapping out of the private section,
2134 * we must check if sk_tx_queue_clear() is called after
2135 * sock_copy() in sk_clone_lock().
2136 */
2137 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2138 offsetof(struct sock, sk_dontcopy_begin) ||
2139 offsetof(struct sock, sk_tx_queue_mapping) >=
2140 offsetof(struct sock, sk_dontcopy_end));
2141
2142 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2143
2144 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2145 prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2146 /* alloc is larger than struct, see sk_prot_alloc() */);
2147
2148#ifdef CONFIG_SECURITY_NETWORK
2149 nsk->sk_security = sptr;
2150 security_sk_clone(osk, nsk);
2151#endif
2152}
2153
2154static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2155 int family)
2156{
2157 struct sock *sk;
2158 struct kmem_cache *slab;
2159
2160 slab = prot->slab;
2161 if (slab != NULL) {
2162 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2163 if (!sk)
2164 return sk;
2165 if (want_init_on_alloc(priority))
2166 sk_prot_clear_nulls(sk, prot->obj_size);
2167 } else
2168 sk = kmalloc(prot->obj_size, priority);
2169
2170 if (sk != NULL) {
2171 if (security_sk_alloc(sk, family, priority))
2172 goto out_free;
2173
2174 if (!try_module_get(prot->owner))
2175 goto out_free_sec;
2176 }
2177
2178 return sk;
2179
2180out_free_sec:
2181 security_sk_free(sk);
2182out_free:
2183 if (slab != NULL)
2184 kmem_cache_free(slab, sk);
2185 else
2186 kfree(sk);
2187 return NULL;
2188}
2189
2190static void sk_prot_free(struct proto *prot, struct sock *sk)
2191{
2192 struct kmem_cache *slab;
2193 struct module *owner;
2194
2195 owner = prot->owner;
2196 slab = prot->slab;
2197
2198 cgroup_sk_free(&sk->sk_cgrp_data);
2199 mem_cgroup_sk_free(sk);
2200 security_sk_free(sk);
2201 if (slab != NULL)
2202 kmem_cache_free(slab, sk);
2203 else
2204 kfree(sk);
2205 module_put(owner);
2206}
2207
2208/**
2209 * sk_alloc - All socket objects are allocated here
2210 * @net: the applicable net namespace
2211 * @family: protocol family
2212 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2213 * @prot: struct proto associated with this new sock instance
2214 * @kern: is this to be a kernel socket?
2215 */
2216struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2217 struct proto *prot, int kern)
2218{
2219 struct sock *sk;
2220
2221 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2222 if (sk) {
2223 sk->sk_family = family;
2224 /*
2225 * See comment in struct sock definition to understand
2226 * why we need sk_prot_creator -acme
2227 */
2228 sk->sk_prot = sk->sk_prot_creator = prot;
2229 sk->sk_kern_sock = kern;
2230 sock_lock_init(sk);
2231 sk->sk_net_refcnt = kern ? 0 : 1;
2232 if (likely(sk->sk_net_refcnt)) {
2233 get_net_track(net, &sk->ns_tracker, priority);
2234 sock_inuse_add(net, 1);
2235 } else {
2236 net_passive_inc(net);
2237 __netns_tracker_alloc(net, &sk->ns_tracker,
2238 false, priority);
2239 }
2240
2241 sock_net_set(sk, net);
2242 refcount_set(&sk->sk_wmem_alloc, 1);
2243
2244 mem_cgroup_sk_alloc(sk);
2245 cgroup_sk_alloc(&sk->sk_cgrp_data);
2246 sock_update_classid(&sk->sk_cgrp_data);
2247 sock_update_netprioidx(&sk->sk_cgrp_data);
2248 sk_tx_queue_clear(sk);
2249 }
2250
2251 return sk;
2252}
2253EXPORT_SYMBOL(sk_alloc);
2254
2255/* Sockets having SOCK_RCU_FREE will call this function after one RCU
2256 * grace period. This is the case for UDP sockets and TCP listeners.
2257 */
2258static void __sk_destruct(struct rcu_head *head)
2259{
2260 struct sock *sk = container_of(head, struct sock, sk_rcu);
2261 struct net *net = sock_net(sk);
2262 struct sk_filter *filter;
2263
2264 if (sk->sk_destruct)
2265 sk->sk_destruct(sk);
2266
2267 filter = rcu_dereference_check(sk->sk_filter,
2268 refcount_read(&sk->sk_wmem_alloc) == 0);
2269 if (filter) {
2270 sk_filter_uncharge(sk, filter);
2271 RCU_INIT_POINTER(sk->sk_filter, NULL);
2272 }
2273
2274 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2275
2276#ifdef CONFIG_BPF_SYSCALL
2277 bpf_sk_storage_free(sk);
2278#endif
2279
2280 if (atomic_read(&sk->sk_omem_alloc))
2281 pr_debug("%s: optmem leakage (%d bytes) detected\n",
2282 __func__, atomic_read(&sk->sk_omem_alloc));
2283
2284 if (sk->sk_frag.page) {
2285 put_page(sk->sk_frag.page);
2286 sk->sk_frag.page = NULL;
2287 }
2288
2289 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2290 put_cred(sk->sk_peer_cred);
2291 put_pid(sk->sk_peer_pid);
2292
2293 if (likely(sk->sk_net_refcnt)) {
2294 put_net_track(net, &sk->ns_tracker);
2295 } else {
2296 __netns_tracker_free(net, &sk->ns_tracker, false);
2297 net_passive_dec(net);
2298 }
2299 sk_prot_free(sk->sk_prot_creator, sk);
2300}
2301
2302void sk_net_refcnt_upgrade(struct sock *sk)
2303{
2304 struct net *net = sock_net(sk);
2305
2306 WARN_ON_ONCE(sk->sk_net_refcnt);
2307 __netns_tracker_free(net, &sk->ns_tracker, false);
2308 net_passive_dec(net);
2309 sk->sk_net_refcnt = 1;
2310 get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
2311 sock_inuse_add(net, 1);
2312}
2313EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade);
2314
2315void sk_destruct(struct sock *sk)
2316{
2317 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2318
2319 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2320 reuseport_detach_sock(sk);
2321 use_call_rcu = true;
2322 }
2323
2324 if (use_call_rcu)
2325 call_rcu(&sk->sk_rcu, __sk_destruct);
2326 else
2327 __sk_destruct(&sk->sk_rcu);
2328}
2329
2330static void __sk_free(struct sock *sk)
2331{
2332 if (likely(sk->sk_net_refcnt))
2333 sock_inuse_add(sock_net(sk), -1);
2334
2335 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2336 sock_diag_broadcast_destroy(sk);
2337 else
2338 sk_destruct(sk);
2339}
2340
2341void sk_free(struct sock *sk)
2342{
2343 /*
2344 * We subtract one from sk_wmem_alloc and can know if
2345 * some packets are still in some tx queue.
2346 * If not null, sock_wfree() will call __sk_free(sk) later
2347 */
2348 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2349 __sk_free(sk);
2350}
2351EXPORT_SYMBOL(sk_free);
2352
2353static void sk_init_common(struct sock *sk)
2354{
2355 skb_queue_head_init(&sk->sk_receive_queue);
2356 skb_queue_head_init(&sk->sk_write_queue);
2357 skb_queue_head_init(&sk->sk_error_queue);
2358
2359 rwlock_init(&sk->sk_callback_lock);
2360 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2361 af_rlock_keys + sk->sk_family,
2362 af_family_rlock_key_strings[sk->sk_family]);
2363 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2364 af_wlock_keys + sk->sk_family,
2365 af_family_wlock_key_strings[sk->sk_family]);
2366 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2367 af_elock_keys + sk->sk_family,
2368 af_family_elock_key_strings[sk->sk_family]);
2369 if (sk->sk_kern_sock)
2370 lockdep_set_class_and_name(&sk->sk_callback_lock,
2371 af_kern_callback_keys + sk->sk_family,
2372 af_family_kern_clock_key_strings[sk->sk_family]);
2373 else
2374 lockdep_set_class_and_name(&sk->sk_callback_lock,
2375 af_callback_keys + sk->sk_family,
2376 af_family_clock_key_strings[sk->sk_family]);
2377}
2378
2379/**
2380 * sk_clone_lock - clone a socket, and lock its clone
2381 * @sk: the socket to clone
2382 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2383 *
2384 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2385 */
2386struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2387{
2388 struct proto *prot = READ_ONCE(sk->sk_prot);
2389 struct sk_filter *filter;
2390 bool is_charged = true;
2391 struct sock *newsk;
2392
2393 newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2394 if (!newsk)
2395 goto out;
2396
2397 sock_copy(newsk, sk);
2398
2399 newsk->sk_prot_creator = prot;
2400
2401 /* SANITY */
2402 if (likely(newsk->sk_net_refcnt)) {
2403 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2404 sock_inuse_add(sock_net(newsk), 1);
2405 } else {
2406 /* Kernel sockets are not elevating the struct net refcount.
2407 * Instead, use a tracker to more easily detect if a layer
2408 * is not properly dismantling its kernel sockets at netns
2409 * destroy time.
2410 */
2411 net_passive_inc(sock_net(newsk));
2412 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2413 false, priority);
2414 }
2415 sk_node_init(&newsk->sk_node);
2416 sock_lock_init(newsk);
2417 bh_lock_sock(newsk);
2418 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
2419 newsk->sk_backlog.len = 0;
2420
2421 atomic_set(&newsk->sk_rmem_alloc, 0);
2422
2423 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2424 refcount_set(&newsk->sk_wmem_alloc, 1);
2425
2426 atomic_set(&newsk->sk_omem_alloc, 0);
2427 sk_init_common(newsk);
2428
2429 newsk->sk_dst_cache = NULL;
2430 newsk->sk_dst_pending_confirm = 0;
2431 newsk->sk_wmem_queued = 0;
2432 newsk->sk_forward_alloc = 0;
2433 newsk->sk_reserved_mem = 0;
2434 atomic_set(&newsk->sk_drops, 0);
2435 newsk->sk_send_head = NULL;
2436 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2437 atomic_set(&newsk->sk_zckey, 0);
2438
2439 sock_reset_flag(newsk, SOCK_DONE);
2440
2441 /* sk->sk_memcg will be populated at accept() time */
2442 newsk->sk_memcg = NULL;
2443
2444 cgroup_sk_clone(&newsk->sk_cgrp_data);
2445
2446 rcu_read_lock();
2447 filter = rcu_dereference(sk->sk_filter);
2448 if (filter != NULL)
2449 /* though it's an empty new sock, the charging may fail
2450 * if sysctl_optmem_max was changed between creation of
2451 * original socket and cloning
2452 */
2453 is_charged = sk_filter_charge(newsk, filter);
2454 RCU_INIT_POINTER(newsk->sk_filter, filter);
2455 rcu_read_unlock();
2456
2457 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2458 /* We need to make sure that we don't uncharge the new
2459 * socket if we couldn't charge it in the first place
2460 * as otherwise we uncharge the parent's filter.
2461 */
2462 if (!is_charged)
2463 RCU_INIT_POINTER(newsk->sk_filter, NULL);
2464 sk_free_unlock_clone(newsk);
2465 newsk = NULL;
2466 goto out;
2467 }
2468 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2469
2470 if (bpf_sk_storage_clone(sk, newsk)) {
2471 sk_free_unlock_clone(newsk);
2472 newsk = NULL;
2473 goto out;
2474 }
2475
2476 /* Clear sk_user_data if parent had the pointer tagged
2477 * as not suitable for copying when cloning.
2478 */
2479 if (sk_user_data_is_nocopy(newsk))
2480 newsk->sk_user_data = NULL;
2481
2482 newsk->sk_err = 0;
2483 newsk->sk_err_soft = 0;
2484 newsk->sk_priority = 0;
2485 newsk->sk_incoming_cpu = raw_smp_processor_id();
2486
2487 /* Before updating sk_refcnt, we must commit prior changes to memory
2488 * (Documentation/RCU/rculist_nulls.rst for details)
2489 */
2490 smp_wmb();
2491 refcount_set(&newsk->sk_refcnt, 2);
2492
2493 sk_set_socket(newsk, NULL);
2494 sk_tx_queue_clear(newsk);
2495 RCU_INIT_POINTER(newsk->sk_wq, NULL);
2496
2497 if (newsk->sk_prot->sockets_allocated)
2498 sk_sockets_allocated_inc(newsk);
2499
2500 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2501 net_enable_timestamp();
2502out:
2503 return newsk;
2504}
2505EXPORT_SYMBOL_GPL(sk_clone_lock);
2506
2507void sk_free_unlock_clone(struct sock *sk)
2508{
2509 /* It is still raw copy of parent, so invalidate
2510 * destructor and make plain sk_free() */
2511 sk->sk_destruct = NULL;
2512 bh_unlock_sock(sk);
2513 sk_free(sk);
2514}
2515EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
2516
2517static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2518{
2519 bool is_ipv6 = false;
2520 u32 max_size;
2521
2522#if IS_ENABLED(CONFIG_IPV6)
2523 is_ipv6 = (sk->sk_family == AF_INET6 &&
2524 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2525#endif
2526 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2527 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) :
2528 READ_ONCE(dst->dev->gso_ipv4_max_size);
2529 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2530 max_size = GSO_LEGACY_MAX_SIZE;
2531
2532 return max_size - (MAX_TCP_HEADER + 1);
2533}
2534
2535void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2536{
2537 u32 max_segs = 1;
2538
2539 sk->sk_route_caps = dst->dev->features;
2540 if (sk_is_tcp(sk))
2541 sk->sk_route_caps |= NETIF_F_GSO;
2542 if (sk->sk_route_caps & NETIF_F_GSO)
2543 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2544 if (unlikely(sk->sk_gso_disabled))
2545 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2546 if (sk_can_gso(sk)) {
2547 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2548 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2549 } else {
2550 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2551 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2552 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2553 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1);
2554 }
2555 }
2556 sk->sk_gso_max_segs = max_segs;
2557 sk_dst_set(sk, dst);
2558}
2559EXPORT_SYMBOL_GPL(sk_setup_caps);
2560
2561/*
2562 * Simple resource managers for sockets.
2563 */
2564
2565
2566/*
2567 * Write buffer destructor automatically called from kfree_skb.
2568 */
2569void sock_wfree(struct sk_buff *skb)
2570{
2571 struct sock *sk = skb->sk;
2572 unsigned int len = skb->truesize;
2573 bool free;
2574
2575 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2576 if (sock_flag(sk, SOCK_RCU_FREE) &&
2577 sk->sk_write_space == sock_def_write_space) {
2578 rcu_read_lock();
2579 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2580 sock_def_write_space_wfree(sk);
2581 rcu_read_unlock();
2582 if (unlikely(free))
2583 __sk_free(sk);
2584 return;
2585 }
2586
2587 /*
2588 * Keep a reference on sk_wmem_alloc, this will be released
2589 * after sk_write_space() call
2590 */
2591 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2592 sk->sk_write_space(sk);
2593 len = 1;
2594 }
2595 /*
2596 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2597 * could not do because of in-flight packets
2598 */
2599 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2600 __sk_free(sk);
2601}
2602EXPORT_SYMBOL(sock_wfree);
2603
2604/* This variant of sock_wfree() is used by TCP,
2605 * since it sets SOCK_USE_WRITE_QUEUE.
2606 */
2607void __sock_wfree(struct sk_buff *skb)
2608{
2609 struct sock *sk = skb->sk;
2610
2611 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2612 __sk_free(sk);
2613}
2614
2615void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2616{
2617 skb_orphan(skb);
2618#ifdef CONFIG_INET
2619 if (unlikely(!sk_fullsock(sk)))
2620 return skb_set_owner_edemux(skb, sk);
2621#endif
2622 skb->sk = sk;
2623 skb->destructor = sock_wfree;
2624 skb_set_hash_from_sk(skb, sk);
2625 /*
2626 * We used to take a refcount on sk, but following operation
2627 * is enough to guarantee sk_free() won't free this sock until
2628 * all in-flight packets are completed
2629 */
2630 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2631}
2632EXPORT_SYMBOL(skb_set_owner_w);
2633
2634static bool can_skb_orphan_partial(const struct sk_buff *skb)
2635{
2636 /* Drivers depend on in-order delivery for crypto offload,
2637 * partial orphan breaks out-of-order-OK logic.
2638 */
2639 if (skb_is_decrypted(skb))
2640 return false;
2641
2642 return (skb->destructor == sock_wfree ||
2643 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2644}
2645
2646/* This helper is used by netem, as it can hold packets in its
2647 * delay queue. We want to allow the owner socket to send more
2648 * packets, as if they were already TX completed by a typical driver.
2649 * But we also want to keep skb->sk set because some packet schedulers
2650 * rely on it (sch_fq for example).
2651 */
2652void skb_orphan_partial(struct sk_buff *skb)
2653{
2654 if (skb_is_tcp_pure_ack(skb))
2655 return;
2656
2657 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2658 return;
2659
2660 skb_orphan(skb);
2661}
2662EXPORT_SYMBOL(skb_orphan_partial);
2663
2664/*
2665 * Read buffer destructor automatically called from kfree_skb.
2666 */
2667void sock_rfree(struct sk_buff *skb)
2668{
2669 struct sock *sk = skb->sk;
2670 unsigned int len = skb->truesize;
2671
2672 atomic_sub(len, &sk->sk_rmem_alloc);
2673 sk_mem_uncharge(sk, len);
2674}
2675EXPORT_SYMBOL(sock_rfree);
2676
2677/*
2678 * Buffer destructor for skbs that are not used directly in read or write
2679 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2680 */
2681void sock_efree(struct sk_buff *skb)
2682{
2683 sock_put(skb->sk);
2684}
2685EXPORT_SYMBOL(sock_efree);
2686
2687/* Buffer destructor for prefetch/receive path where reference count may
2688 * not be held, e.g. for listen sockets.
2689 */
2690#ifdef CONFIG_INET
2691void sock_pfree(struct sk_buff *skb)
2692{
2693 struct sock *sk = skb->sk;
2694
2695 if (!sk_is_refcounted(sk))
2696 return;
2697
2698 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2699 inet_reqsk(sk)->rsk_listener = NULL;
2700 reqsk_free(inet_reqsk(sk));
2701 return;
2702 }
2703
2704 sock_gen_put(sk);
2705}
2706EXPORT_SYMBOL(sock_pfree);
2707#endif /* CONFIG_INET */
2708
2709kuid_t sock_i_uid(struct sock *sk)
2710{
2711 kuid_t uid;
2712
2713 read_lock_bh(&sk->sk_callback_lock);
2714 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2715 read_unlock_bh(&sk->sk_callback_lock);
2716 return uid;
2717}
2718EXPORT_SYMBOL(sock_i_uid);
2719
2720unsigned long __sock_i_ino(struct sock *sk)
2721{
2722 unsigned long ino;
2723
2724 read_lock(&sk->sk_callback_lock);
2725 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2726 read_unlock(&sk->sk_callback_lock);
2727 return ino;
2728}
2729EXPORT_SYMBOL(__sock_i_ino);
2730
2731unsigned long sock_i_ino(struct sock *sk)
2732{
2733 unsigned long ino;
2734
2735 local_bh_disable();
2736 ino = __sock_i_ino(sk);
2737 local_bh_enable();
2738 return ino;
2739}
2740EXPORT_SYMBOL(sock_i_ino);
2741
2742/*
2743 * Allocate a skb from the socket's send buffer.
2744 */
2745struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2746 gfp_t priority)
2747{
2748 if (force ||
2749 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2750 struct sk_buff *skb = alloc_skb(size, priority);
2751
2752 if (skb) {
2753 skb_set_owner_w(skb, sk);
2754 return skb;
2755 }
2756 }
2757 return NULL;
2758}
2759EXPORT_SYMBOL(sock_wmalloc);
2760
2761static void sock_ofree(struct sk_buff *skb)
2762{
2763 struct sock *sk = skb->sk;
2764
2765 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2766}
2767
2768struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2769 gfp_t priority)
2770{
2771 struct sk_buff *skb;
2772
2773 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2774 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2775 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2776 return NULL;
2777
2778 skb = alloc_skb(size, priority);
2779 if (!skb)
2780 return NULL;
2781
2782 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2783 skb->sk = sk;
2784 skb->destructor = sock_ofree;
2785 return skb;
2786}
2787
2788/*
2789 * Allocate a memory block from the socket's option memory buffer.
2790 */
2791void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2792{
2793 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2794
2795 if ((unsigned int)size <= optmem_max &&
2796 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2797 void *mem;
2798 /* First do the add, to avoid the race if kmalloc
2799 * might sleep.
2800 */
2801 atomic_add(size, &sk->sk_omem_alloc);
2802 mem = kmalloc(size, priority);
2803 if (mem)
2804 return mem;
2805 atomic_sub(size, &sk->sk_omem_alloc);
2806 }
2807 return NULL;
2808}
2809EXPORT_SYMBOL(sock_kmalloc);
2810
2811/* Free an option memory block. Note, we actually want the inline
2812 * here as this allows gcc to detect the nullify and fold away the
2813 * condition entirely.
2814 */
2815static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2816 const bool nullify)
2817{
2818 if (WARN_ON_ONCE(!mem))
2819 return;
2820 if (nullify)
2821 kfree_sensitive(mem);
2822 else
2823 kfree(mem);
2824 atomic_sub(size, &sk->sk_omem_alloc);
2825}
2826
2827void sock_kfree_s(struct sock *sk, void *mem, int size)
2828{
2829 __sock_kfree_s(sk, mem, size, false);
2830}
2831EXPORT_SYMBOL(sock_kfree_s);
2832
2833void sock_kzfree_s(struct sock *sk, void *mem, int size)
2834{
2835 __sock_kfree_s(sk, mem, size, true);
2836}
2837EXPORT_SYMBOL(sock_kzfree_s);
2838
2839/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2840 I think, these locks should be removed for datagram sockets.
2841 */
2842static long sock_wait_for_wmem(struct sock *sk, long timeo)
2843{
2844 DEFINE_WAIT(wait);
2845
2846 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2847 for (;;) {
2848 if (!timeo)
2849 break;
2850 if (signal_pending(current))
2851 break;
2852 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2853 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2854 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2855 break;
2856 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2857 break;
2858 if (READ_ONCE(sk->sk_err))
2859 break;
2860 timeo = schedule_timeout(timeo);
2861 }
2862 finish_wait(sk_sleep(sk), &wait);
2863 return timeo;
2864}
2865
2866
2867/*
2868 * Generic send/receive buffer handlers
2869 */
2870
2871struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2872 unsigned long data_len, int noblock,
2873 int *errcode, int max_page_order)
2874{
2875 struct sk_buff *skb;
2876 long timeo;
2877 int err;
2878
2879 timeo = sock_sndtimeo(sk, noblock);
2880 for (;;) {
2881 err = sock_error(sk);
2882 if (err != 0)
2883 goto failure;
2884
2885 err = -EPIPE;
2886 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2887 goto failure;
2888
2889 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2890 break;
2891
2892 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2893 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2894 err = -EAGAIN;
2895 if (!timeo)
2896 goto failure;
2897 if (signal_pending(current))
2898 goto interrupted;
2899 timeo = sock_wait_for_wmem(sk, timeo);
2900 }
2901 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2902 errcode, sk->sk_allocation);
2903 if (skb)
2904 skb_set_owner_w(skb, sk);
2905 return skb;
2906
2907interrupted:
2908 err = sock_intr_errno(timeo);
2909failure:
2910 *errcode = err;
2911 return NULL;
2912}
2913EXPORT_SYMBOL(sock_alloc_send_pskb);
2914
2915int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2916 struct sockcm_cookie *sockc)
2917{
2918 u32 tsflags;
2919
2920 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
2921
2922 switch (cmsg->cmsg_type) {
2923 case SO_MARK:
2924 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2925 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2926 return -EPERM;
2927 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2928 return -EINVAL;
2929 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2930 break;
2931 case SO_TIMESTAMPING_OLD:
2932 case SO_TIMESTAMPING_NEW:
2933 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2934 return -EINVAL;
2935
2936 tsflags = *(u32 *)CMSG_DATA(cmsg);
2937 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2938 return -EINVAL;
2939
2940 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2941 sockc->tsflags |= tsflags;
2942 break;
2943 case SCM_TXTIME:
2944 if (!sock_flag(sk, SOCK_TXTIME))
2945 return -EINVAL;
2946 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2947 return -EINVAL;
2948 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2949 break;
2950 case SCM_TS_OPT_ID:
2951 if (sk_is_tcp(sk))
2952 return -EINVAL;
2953 tsflags = READ_ONCE(sk->sk_tsflags);
2954 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
2955 return -EINVAL;
2956 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2957 return -EINVAL;
2958 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
2959 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
2960 break;
2961 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2962 case SCM_RIGHTS:
2963 case SCM_CREDENTIALS:
2964 break;
2965 default:
2966 return -EINVAL;
2967 }
2968 return 0;
2969}
2970EXPORT_SYMBOL(__sock_cmsg_send);
2971
2972int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2973 struct sockcm_cookie *sockc)
2974{
2975 struct cmsghdr *cmsg;
2976 int ret;
2977
2978 for_each_cmsghdr(cmsg, msg) {
2979 if (!CMSG_OK(msg, cmsg))
2980 return -EINVAL;
2981 if (cmsg->cmsg_level != SOL_SOCKET)
2982 continue;
2983 ret = __sock_cmsg_send(sk, cmsg, sockc);
2984 if (ret)
2985 return ret;
2986 }
2987 return 0;
2988}
2989EXPORT_SYMBOL(sock_cmsg_send);
2990
2991static void sk_enter_memory_pressure(struct sock *sk)
2992{
2993 if (!sk->sk_prot->enter_memory_pressure)
2994 return;
2995
2996 sk->sk_prot->enter_memory_pressure(sk);
2997}
2998
2999static void sk_leave_memory_pressure(struct sock *sk)
3000{
3001 if (sk->sk_prot->leave_memory_pressure) {
3002 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3003 tcp_leave_memory_pressure, sk);
3004 } else {
3005 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3006
3007 if (memory_pressure && READ_ONCE(*memory_pressure))
3008 WRITE_ONCE(*memory_pressure, 0);
3009 }
3010}
3011
3012DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3013
3014/**
3015 * skb_page_frag_refill - check that a page_frag contains enough room
3016 * @sz: minimum size of the fragment we want to get
3017 * @pfrag: pointer to page_frag
3018 * @gfp: priority for memory allocation
3019 *
3020 * Note: While this allocator tries to use high order pages, there is
3021 * no guarantee that allocations succeed. Therefore, @sz MUST be
3022 * less or equal than PAGE_SIZE.
3023 */
3024bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3025{
3026 if (pfrag->page) {
3027 if (page_ref_count(pfrag->page) == 1) {
3028 pfrag->offset = 0;
3029 return true;
3030 }
3031 if (pfrag->offset + sz <= pfrag->size)
3032 return true;
3033 put_page(pfrag->page);
3034 }
3035
3036 pfrag->offset = 0;
3037 if (SKB_FRAG_PAGE_ORDER &&
3038 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3039 /* Avoid direct reclaim but allow kswapd to wake */
3040 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3041 __GFP_COMP | __GFP_NOWARN |
3042 __GFP_NORETRY,
3043 SKB_FRAG_PAGE_ORDER);
3044 if (likely(pfrag->page)) {
3045 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3046 return true;
3047 }
3048 }
3049 pfrag->page = alloc_page(gfp);
3050 if (likely(pfrag->page)) {
3051 pfrag->size = PAGE_SIZE;
3052 return true;
3053 }
3054 return false;
3055}
3056EXPORT_SYMBOL(skb_page_frag_refill);
3057
3058bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3059{
3060 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3061 return true;
3062
3063 sk_enter_memory_pressure(sk);
3064 sk_stream_moderate_sndbuf(sk);
3065 return false;
3066}
3067EXPORT_SYMBOL(sk_page_frag_refill);
3068
3069void __lock_sock(struct sock *sk)
3070 __releases(&sk->sk_lock.slock)
3071 __acquires(&sk->sk_lock.slock)
3072{
3073 DEFINE_WAIT(wait);
3074
3075 for (;;) {
3076 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3077 TASK_UNINTERRUPTIBLE);
3078 spin_unlock_bh(&sk->sk_lock.slock);
3079 schedule();
3080 spin_lock_bh(&sk->sk_lock.slock);
3081 if (!sock_owned_by_user(sk))
3082 break;
3083 }
3084 finish_wait(&sk->sk_lock.wq, &wait);
3085}
3086
3087void __release_sock(struct sock *sk)
3088 __releases(&sk->sk_lock.slock)
3089 __acquires(&sk->sk_lock.slock)
3090{
3091 struct sk_buff *skb, *next;
3092
3093 while ((skb = sk->sk_backlog.head) != NULL) {
3094 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3095
3096 spin_unlock_bh(&sk->sk_lock.slock);
3097
3098 do {
3099 next = skb->next;
3100 prefetch(next);
3101 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3102 skb_mark_not_on_list(skb);
3103 sk_backlog_rcv(sk, skb);
3104
3105 cond_resched();
3106
3107 skb = next;
3108 } while (skb != NULL);
3109
3110 spin_lock_bh(&sk->sk_lock.slock);
3111 }
3112
3113 /*
3114 * Doing the zeroing here guarantee we can not loop forever
3115 * while a wild producer attempts to flood us.
3116 */
3117 sk->sk_backlog.len = 0;
3118}
3119
3120void __sk_flush_backlog(struct sock *sk)
3121{
3122 spin_lock_bh(&sk->sk_lock.slock);
3123 __release_sock(sk);
3124
3125 if (sk->sk_prot->release_cb)
3126 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3127 tcp_release_cb, sk);
3128
3129 spin_unlock_bh(&sk->sk_lock.slock);
3130}
3131EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3132
3133/**
3134 * sk_wait_data - wait for data to arrive at sk_receive_queue
3135 * @sk: sock to wait on
3136 * @timeo: for how long
3137 * @skb: last skb seen on sk_receive_queue
3138 *
3139 * Now socket state including sk->sk_err is changed only under lock,
3140 * hence we may omit checks after joining wait queue.
3141 * We check receive queue before schedule() only as optimization;
3142 * it is very likely that release_sock() added new data.
3143 */
3144int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3145{
3146 DEFINE_WAIT_FUNC(wait, woken_wake_function);
3147 int rc;
3148
3149 add_wait_queue(sk_sleep(sk), &wait);
3150 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3151 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3152 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3153 remove_wait_queue(sk_sleep(sk), &wait);
3154 return rc;
3155}
3156EXPORT_SYMBOL(sk_wait_data);
3157
3158/**
3159 * __sk_mem_raise_allocated - increase memory_allocated
3160 * @sk: socket
3161 * @size: memory size to allocate
3162 * @amt: pages to allocate
3163 * @kind: allocation type
3164 *
3165 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3166 *
3167 * Unlike the globally shared limits among the sockets under same protocol,
3168 * consuming the budget of a memcg won't have direct effect on other ones.
3169 * So be optimistic about memcg's tolerance, and leave the callers to decide
3170 * whether or not to raise allocated through sk_under_memory_pressure() or
3171 * its variants.
3172 */
3173int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3174{
3175 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3176 struct proto *prot = sk->sk_prot;
3177 bool charged = false;
3178 long allocated;
3179
3180 sk_memory_allocated_add(sk, amt);
3181 allocated = sk_memory_allocated(sk);
3182
3183 if (memcg) {
3184 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge()))
3185 goto suppress_allocation;
3186 charged = true;
3187 }
3188
3189 /* Under limit. */
3190 if (allocated <= sk_prot_mem_limits(sk, 0)) {
3191 sk_leave_memory_pressure(sk);
3192 return 1;
3193 }
3194
3195 /* Under pressure. */
3196 if (allocated > sk_prot_mem_limits(sk, 1))
3197 sk_enter_memory_pressure(sk);
3198
3199 /* Over hard limit. */
3200 if (allocated > sk_prot_mem_limits(sk, 2))
3201 goto suppress_allocation;
3202
3203 /* Guarantee minimum buffer size under pressure (either global
3204 * or memcg) to make sure features described in RFC 7323 (TCP
3205 * Extensions for High Performance) work properly.
3206 *
3207 * This rule does NOT stand when exceeds global or memcg's hard
3208 * limit, or else a DoS attack can be taken place by spawning
3209 * lots of sockets whose usage are under minimum buffer size.
3210 */
3211 if (kind == SK_MEM_RECV) {
3212 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3213 return 1;
3214
3215 } else { /* SK_MEM_SEND */
3216 int wmem0 = sk_get_wmem0(sk, prot);
3217
3218 if (sk->sk_type == SOCK_STREAM) {
3219 if (sk->sk_wmem_queued < wmem0)
3220 return 1;
3221 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3222 return 1;
3223 }
3224 }
3225
3226 if (sk_has_memory_pressure(sk)) {
3227 u64 alloc;
3228
3229 /* The following 'average' heuristic is within the
3230 * scope of global accounting, so it only makes
3231 * sense for global memory pressure.
3232 */
3233 if (!sk_under_global_memory_pressure(sk))
3234 return 1;
3235
3236 /* Try to be fair among all the sockets under global
3237 * pressure by allowing the ones that below average
3238 * usage to raise.
3239 */
3240 alloc = sk_sockets_allocated_read_positive(sk);
3241 if (sk_prot_mem_limits(sk, 2) > alloc *
3242 sk_mem_pages(sk->sk_wmem_queued +
3243 atomic_read(&sk->sk_rmem_alloc) +
3244 sk->sk_forward_alloc))
3245 return 1;
3246 }
3247
3248suppress_allocation:
3249
3250 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3251 sk_stream_moderate_sndbuf(sk);
3252
3253 /* Fail only if socket is _under_ its sndbuf.
3254 * In this case we cannot block, so that we have to fail.
3255 */
3256 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3257 /* Force charge with __GFP_NOFAIL */
3258 if (memcg && !charged) {
3259 mem_cgroup_charge_skmem(memcg, amt,
3260 gfp_memcg_charge() | __GFP_NOFAIL);
3261 }
3262 return 1;
3263 }
3264 }
3265
3266 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3267 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3268
3269 sk_memory_allocated_sub(sk, amt);
3270
3271 if (charged)
3272 mem_cgroup_uncharge_skmem(memcg, amt);
3273
3274 return 0;
3275}
3276
3277/**
3278 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3279 * @sk: socket
3280 * @size: memory size to allocate
3281 * @kind: allocation type
3282 *
3283 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3284 * rmem allocation. This function assumes that protocols which have
3285 * memory_pressure use sk_wmem_queued as write buffer accounting.
3286 */
3287int __sk_mem_schedule(struct sock *sk, int size, int kind)
3288{
3289 int ret, amt = sk_mem_pages(size);
3290
3291 sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3292 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3293 if (!ret)
3294 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3295 return ret;
3296}
3297EXPORT_SYMBOL(__sk_mem_schedule);
3298
3299/**
3300 * __sk_mem_reduce_allocated - reclaim memory_allocated
3301 * @sk: socket
3302 * @amount: number of quanta
3303 *
3304 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3305 */
3306void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3307{
3308 sk_memory_allocated_sub(sk, amount);
3309
3310 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3311 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3312
3313 if (sk_under_global_memory_pressure(sk) &&
3314 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3315 sk_leave_memory_pressure(sk);
3316}
3317
3318/**
3319 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3320 * @sk: socket
3321 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3322 */
3323void __sk_mem_reclaim(struct sock *sk, int amount)
3324{
3325 amount >>= PAGE_SHIFT;
3326 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3327 __sk_mem_reduce_allocated(sk, amount);
3328}
3329EXPORT_SYMBOL(__sk_mem_reclaim);
3330
3331int sk_set_peek_off(struct sock *sk, int val)
3332{
3333 WRITE_ONCE(sk->sk_peek_off, val);
3334 return 0;
3335}
3336EXPORT_SYMBOL_GPL(sk_set_peek_off);
3337
3338/*
3339 * Set of default routines for initialising struct proto_ops when
3340 * the protocol does not support a particular function. In certain
3341 * cases where it makes no sense for a protocol to have a "do nothing"
3342 * function, some default processing is provided.
3343 */
3344
3345int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3346{
3347 return -EOPNOTSUPP;
3348}
3349EXPORT_SYMBOL(sock_no_bind);
3350
3351int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3352 int len, int flags)
3353{
3354 return -EOPNOTSUPP;
3355}
3356EXPORT_SYMBOL(sock_no_connect);
3357
3358int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3359{
3360 return -EOPNOTSUPP;
3361}
3362EXPORT_SYMBOL(sock_no_socketpair);
3363
3364int sock_no_accept(struct socket *sock, struct socket *newsock,
3365 struct proto_accept_arg *arg)
3366{
3367 return -EOPNOTSUPP;
3368}
3369EXPORT_SYMBOL(sock_no_accept);
3370
3371int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3372 int peer)
3373{
3374 return -EOPNOTSUPP;
3375}
3376EXPORT_SYMBOL(sock_no_getname);
3377
3378int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3379{
3380 return -EOPNOTSUPP;
3381}
3382EXPORT_SYMBOL(sock_no_ioctl);
3383
3384int sock_no_listen(struct socket *sock, int backlog)
3385{
3386 return -EOPNOTSUPP;
3387}
3388EXPORT_SYMBOL(sock_no_listen);
3389
3390int sock_no_shutdown(struct socket *sock, int how)
3391{
3392 return -EOPNOTSUPP;
3393}
3394EXPORT_SYMBOL(sock_no_shutdown);
3395
3396int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3397{
3398 return -EOPNOTSUPP;
3399}
3400EXPORT_SYMBOL(sock_no_sendmsg);
3401
3402int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3403{
3404 return -EOPNOTSUPP;
3405}
3406EXPORT_SYMBOL(sock_no_sendmsg_locked);
3407
3408int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3409 int flags)
3410{
3411 return -EOPNOTSUPP;
3412}
3413EXPORT_SYMBOL(sock_no_recvmsg);
3414
3415int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3416{
3417 /* Mirror missing mmap method error code */
3418 return -ENODEV;
3419}
3420EXPORT_SYMBOL(sock_no_mmap);
3421
3422/*
3423 * When a file is received (via SCM_RIGHTS, etc), we must bump the
3424 * various sock-based usage counts.
3425 */
3426void __receive_sock(struct file *file)
3427{
3428 struct socket *sock;
3429
3430 sock = sock_from_file(file);
3431 if (sock) {
3432 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3433 sock_update_classid(&sock->sk->sk_cgrp_data);
3434 }
3435}
3436
3437/*
3438 * Default Socket Callbacks
3439 */
3440
3441static void sock_def_wakeup(struct sock *sk)
3442{
3443 struct socket_wq *wq;
3444
3445 rcu_read_lock();
3446 wq = rcu_dereference(sk->sk_wq);
3447 if (skwq_has_sleeper(wq))
3448 wake_up_interruptible_all(&wq->wait);
3449 rcu_read_unlock();
3450}
3451
3452static void sock_def_error_report(struct sock *sk)
3453{
3454 struct socket_wq *wq;
3455
3456 rcu_read_lock();
3457 wq = rcu_dereference(sk->sk_wq);
3458 if (skwq_has_sleeper(wq))
3459 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3460 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3461 rcu_read_unlock();
3462}
3463
3464void sock_def_readable(struct sock *sk)
3465{
3466 struct socket_wq *wq;
3467
3468 trace_sk_data_ready(sk);
3469
3470 rcu_read_lock();
3471 wq = rcu_dereference(sk->sk_wq);
3472 if (skwq_has_sleeper(wq))
3473 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3474 EPOLLRDNORM | EPOLLRDBAND);
3475 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3476 rcu_read_unlock();
3477}
3478
3479static void sock_def_write_space(struct sock *sk)
3480{
3481 struct socket_wq *wq;
3482
3483 rcu_read_lock();
3484
3485 /* Do not wake up a writer until he can make "significant"
3486 * progress. --DaveM
3487 */
3488 if (sock_writeable(sk)) {
3489 wq = rcu_dereference(sk->sk_wq);
3490 if (skwq_has_sleeper(wq))
3491 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3492 EPOLLWRNORM | EPOLLWRBAND);
3493
3494 /* Should agree with poll, otherwise some programs break */
3495 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3496 }
3497
3498 rcu_read_unlock();
3499}
3500
3501/* An optimised version of sock_def_write_space(), should only be called
3502 * for SOCK_RCU_FREE sockets under RCU read section and after putting
3503 * ->sk_wmem_alloc.
3504 */
3505static void sock_def_write_space_wfree(struct sock *sk)
3506{
3507 /* Do not wake up a writer until he can make "significant"
3508 * progress. --DaveM
3509 */
3510 if (sock_writeable(sk)) {
3511 struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3512
3513 /* rely on refcount_sub from sock_wfree() */
3514 smp_mb__after_atomic();
3515 if (wq && waitqueue_active(&wq->wait))
3516 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3517 EPOLLWRNORM | EPOLLWRBAND);
3518
3519 /* Should agree with poll, otherwise some programs break */
3520 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3521 }
3522}
3523
3524static void sock_def_destruct(struct sock *sk)
3525{
3526}
3527
3528void sk_send_sigurg(struct sock *sk)
3529{
3530 if (sk->sk_socket && sk->sk_socket->file)
3531 if (send_sigurg(sk->sk_socket->file))
3532 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3533}
3534EXPORT_SYMBOL(sk_send_sigurg);
3535
3536void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3537 unsigned long expires)
3538{
3539 if (!mod_timer(timer, expires))
3540 sock_hold(sk);
3541}
3542EXPORT_SYMBOL(sk_reset_timer);
3543
3544void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3545{
3546 if (del_timer(timer))
3547 __sock_put(sk);
3548}
3549EXPORT_SYMBOL(sk_stop_timer);
3550
3551void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3552{
3553 if (del_timer_sync(timer))
3554 __sock_put(sk);
3555}
3556EXPORT_SYMBOL(sk_stop_timer_sync);
3557
3558void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3559{
3560 sk_init_common(sk);
3561 sk->sk_send_head = NULL;
3562
3563 timer_setup(&sk->sk_timer, NULL, 0);
3564
3565 sk->sk_allocation = GFP_KERNEL;
3566 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default);
3567 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
3568 sk->sk_state = TCP_CLOSE;
3569 sk->sk_use_task_frag = true;
3570 sk_set_socket(sk, sock);
3571
3572 sock_set_flag(sk, SOCK_ZAPPED);
3573
3574 if (sock) {
3575 sk->sk_type = sock->type;
3576 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3577 sock->sk = sk;
3578 } else {
3579 RCU_INIT_POINTER(sk->sk_wq, NULL);
3580 }
3581 sk->sk_uid = uid;
3582
3583 sk->sk_state_change = sock_def_wakeup;
3584 sk->sk_data_ready = sock_def_readable;
3585 sk->sk_write_space = sock_def_write_space;
3586 sk->sk_error_report = sock_def_error_report;
3587 sk->sk_destruct = sock_def_destruct;
3588
3589 sk->sk_frag.page = NULL;
3590 sk->sk_frag.offset = 0;
3591 sk->sk_peek_off = -1;
3592
3593 sk->sk_peer_pid = NULL;
3594 sk->sk_peer_cred = NULL;
3595 spin_lock_init(&sk->sk_peer_lock);
3596
3597 sk->sk_write_pending = 0;
3598 sk->sk_rcvlowat = 1;
3599 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
3600 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
3601
3602 sk->sk_stamp = SK_DEFAULT_STAMP;
3603#if BITS_PER_LONG==32
3604 seqlock_init(&sk->sk_stamp_seq);
3605#endif
3606 atomic_set(&sk->sk_zckey, 0);
3607
3608#ifdef CONFIG_NET_RX_BUSY_POLL
3609 sk->sk_napi_id = 0;
3610 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read);
3611#endif
3612
3613 sk->sk_max_pacing_rate = ~0UL;
3614 sk->sk_pacing_rate = ~0UL;
3615 WRITE_ONCE(sk->sk_pacing_shift, 10);
3616 sk->sk_incoming_cpu = -1;
3617
3618 sk_rx_queue_clear(sk);
3619 /*
3620 * Before updating sk_refcnt, we must commit prior changes to memory
3621 * (Documentation/RCU/rculist_nulls.rst for details)
3622 */
3623 smp_wmb();
3624 refcount_set(&sk->sk_refcnt, 1);
3625 atomic_set(&sk->sk_drops, 0);
3626}
3627EXPORT_SYMBOL(sock_init_data_uid);
3628
3629void sock_init_data(struct socket *sock, struct sock *sk)
3630{
3631 kuid_t uid = sock ?
3632 SOCK_INODE(sock)->i_uid :
3633 make_kuid(sock_net(sk)->user_ns, 0);
3634
3635 sock_init_data_uid(sock, sk, uid);
3636}
3637EXPORT_SYMBOL(sock_init_data);
3638
3639void lock_sock_nested(struct sock *sk, int subclass)
3640{
3641 /* The sk_lock has mutex_lock() semantics here. */
3642 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3643
3644 might_sleep();
3645 spin_lock_bh(&sk->sk_lock.slock);
3646 if (sock_owned_by_user_nocheck(sk))
3647 __lock_sock(sk);
3648 sk->sk_lock.owned = 1;
3649 spin_unlock_bh(&sk->sk_lock.slock);
3650}
3651EXPORT_SYMBOL(lock_sock_nested);
3652
3653void release_sock(struct sock *sk)
3654{
3655 spin_lock_bh(&sk->sk_lock.slock);
3656 if (sk->sk_backlog.tail)
3657 __release_sock(sk);
3658
3659 if (sk->sk_prot->release_cb)
3660 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3661 tcp_release_cb, sk);
3662
3663 sock_release_ownership(sk);
3664 if (waitqueue_active(&sk->sk_lock.wq))
3665 wake_up(&sk->sk_lock.wq);
3666 spin_unlock_bh(&sk->sk_lock.slock);
3667}
3668EXPORT_SYMBOL(release_sock);
3669
3670bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3671{
3672 might_sleep();
3673 spin_lock_bh(&sk->sk_lock.slock);
3674
3675 if (!sock_owned_by_user_nocheck(sk)) {
3676 /*
3677 * Fast path return with bottom halves disabled and
3678 * sock::sk_lock.slock held.
3679 *
3680 * The 'mutex' is not contended and holding
3681 * sock::sk_lock.slock prevents all other lockers to
3682 * proceed so the corresponding unlock_sock_fast() can
3683 * avoid the slow path of release_sock() completely and
3684 * just release slock.
3685 *
3686 * From a semantical POV this is equivalent to 'acquiring'
3687 * the 'mutex', hence the corresponding lockdep
3688 * mutex_release() has to happen in the fast path of
3689 * unlock_sock_fast().
3690 */
3691 return false;
3692 }
3693
3694 __lock_sock(sk);
3695 sk->sk_lock.owned = 1;
3696 __acquire(&sk->sk_lock.slock);
3697 spin_unlock_bh(&sk->sk_lock.slock);
3698 return true;
3699}
3700EXPORT_SYMBOL(__lock_sock_fast);
3701
3702int sock_gettstamp(struct socket *sock, void __user *userstamp,
3703 bool timeval, bool time32)
3704{
3705 struct sock *sk = sock->sk;
3706 struct timespec64 ts;
3707
3708 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3709 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3710 if (ts.tv_sec == -1)
3711 return -ENOENT;
3712 if (ts.tv_sec == 0) {
3713 ktime_t kt = ktime_get_real();
3714 sock_write_timestamp(sk, kt);
3715 ts = ktime_to_timespec64(kt);
3716 }
3717
3718 if (timeval)
3719 ts.tv_nsec /= 1000;
3720
3721#ifdef CONFIG_COMPAT_32BIT_TIME
3722 if (time32)
3723 return put_old_timespec32(&ts, userstamp);
3724#endif
3725#ifdef CONFIG_SPARC64
3726 /* beware of padding in sparc64 timeval */
3727 if (timeval && !in_compat_syscall()) {
3728 struct __kernel_old_timeval __user tv = {
3729 .tv_sec = ts.tv_sec,
3730 .tv_usec = ts.tv_nsec,
3731 };
3732 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3733 return -EFAULT;
3734 return 0;
3735 }
3736#endif
3737 return put_timespec64(&ts, userstamp);
3738}
3739EXPORT_SYMBOL(sock_gettstamp);
3740
3741void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3742{
3743 if (!sock_flag(sk, flag)) {
3744 unsigned long previous_flags = sk->sk_flags;
3745
3746 sock_set_flag(sk, flag);
3747 /*
3748 * we just set one of the two flags which require net
3749 * time stamping, but time stamping might have been on
3750 * already because of the other one
3751 */
3752 if (sock_needs_netstamp(sk) &&
3753 !(previous_flags & SK_FLAGS_TIMESTAMP))
3754 net_enable_timestamp();
3755 }
3756}
3757
3758int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3759 int level, int type)
3760{
3761 struct sock_exterr_skb *serr;
3762 struct sk_buff *skb;
3763 int copied, err;
3764
3765 err = -EAGAIN;
3766 skb = sock_dequeue_err_skb(sk);
3767 if (skb == NULL)
3768 goto out;
3769
3770 copied = skb->len;
3771 if (copied > len) {
3772 msg->msg_flags |= MSG_TRUNC;
3773 copied = len;
3774 }
3775 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3776 if (err)
3777 goto out_free_skb;
3778
3779 sock_recv_timestamp(msg, sk, skb);
3780
3781 serr = SKB_EXT_ERR(skb);
3782 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3783
3784 msg->msg_flags |= MSG_ERRQUEUE;
3785 err = copied;
3786
3787out_free_skb:
3788 kfree_skb(skb);
3789out:
3790 return err;
3791}
3792EXPORT_SYMBOL(sock_recv_errqueue);
3793
3794/*
3795 * Get a socket option on an socket.
3796 *
3797 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3798 * asynchronous errors should be reported by getsockopt. We assume
3799 * this means if you specify SO_ERROR (otherwise what is the point of it).
3800 */
3801int sock_common_getsockopt(struct socket *sock, int level, int optname,
3802 char __user *optval, int __user *optlen)
3803{
3804 struct sock *sk = sock->sk;
3805
3806 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3807 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3808}
3809EXPORT_SYMBOL(sock_common_getsockopt);
3810
3811int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3812 int flags)
3813{
3814 struct sock *sk = sock->sk;
3815 int addr_len = 0;
3816 int err;
3817
3818 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3819 if (err >= 0)
3820 msg->msg_namelen = addr_len;
3821 return err;
3822}
3823EXPORT_SYMBOL(sock_common_recvmsg);
3824
3825/*
3826 * Set socket options on an inet socket.
3827 */
3828int sock_common_setsockopt(struct socket *sock, int level, int optname,
3829 sockptr_t optval, unsigned int optlen)
3830{
3831 struct sock *sk = sock->sk;
3832
3833 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
3834 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3835}
3836EXPORT_SYMBOL(sock_common_setsockopt);
3837
3838void sk_common_release(struct sock *sk)
3839{
3840 if (sk->sk_prot->destroy)
3841 sk->sk_prot->destroy(sk);
3842
3843 /*
3844 * Observation: when sk_common_release is called, processes have
3845 * no access to socket. But net still has.
3846 * Step one, detach it from networking:
3847 *
3848 * A. Remove from hash tables.
3849 */
3850
3851 sk->sk_prot->unhash(sk);
3852
3853 /*
3854 * In this point socket cannot receive new packets, but it is possible
3855 * that some packets are in flight because some CPU runs receiver and
3856 * did hash table lookup before we unhashed socket. They will achieve
3857 * receive queue and will be purged by socket destructor.
3858 *
3859 * Also we still have packets pending on receive queue and probably,
3860 * our own packets waiting in device queues. sock_destroy will drain
3861 * receive queue, but transmitted packets will delay socket destruction
3862 * until the last reference will be released.
3863 */
3864
3865 sock_orphan(sk);
3866
3867 xfrm_sk_free_policy(sk);
3868
3869 sock_put(sk);
3870}
3871EXPORT_SYMBOL(sk_common_release);
3872
3873void sk_get_meminfo(const struct sock *sk, u32 *mem)
3874{
3875 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3876
3877 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3878 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3879 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3880 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3881 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk);
3882 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3883 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3884 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3885 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3886}
3887
3888#ifdef CONFIG_PROC_FS
3889static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3890
3891int sock_prot_inuse_get(struct net *net, struct proto *prot)
3892{
3893 int cpu, idx = prot->inuse_idx;
3894 int res = 0;
3895
3896 for_each_possible_cpu(cpu)
3897 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3898
3899 return res >= 0 ? res : 0;
3900}
3901EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3902
3903int sock_inuse_get(struct net *net)
3904{
3905 int cpu, res = 0;
3906
3907 for_each_possible_cpu(cpu)
3908 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3909
3910 return res;
3911}
3912
3913EXPORT_SYMBOL_GPL(sock_inuse_get);
3914
3915static int __net_init sock_inuse_init_net(struct net *net)
3916{
3917 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3918 if (net->core.prot_inuse == NULL)
3919 return -ENOMEM;
3920 return 0;
3921}
3922
3923static void __net_exit sock_inuse_exit_net(struct net *net)
3924{
3925 free_percpu(net->core.prot_inuse);
3926}
3927
3928static struct pernet_operations net_inuse_ops = {
3929 .init = sock_inuse_init_net,
3930 .exit = sock_inuse_exit_net,
3931};
3932
3933static __init int net_inuse_init(void)
3934{
3935 if (register_pernet_subsys(&net_inuse_ops))
3936 panic("Cannot initialize net inuse counters");
3937
3938 return 0;
3939}
3940
3941core_initcall(net_inuse_init);
3942
3943static int assign_proto_idx(struct proto *prot)
3944{
3945 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3946
3947 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3948 pr_err("PROTO_INUSE_NR exhausted\n");
3949 return -ENOSPC;
3950 }
3951
3952 set_bit(prot->inuse_idx, proto_inuse_idx);
3953 return 0;
3954}
3955
3956static void release_proto_idx(struct proto *prot)
3957{
3958 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3959 clear_bit(prot->inuse_idx, proto_inuse_idx);
3960}
3961#else
3962static inline int assign_proto_idx(struct proto *prot)
3963{
3964 return 0;
3965}
3966
3967static inline void release_proto_idx(struct proto *prot)
3968{
3969}
3970
3971#endif
3972
3973static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3974{
3975 if (!twsk_prot)
3976 return;
3977 kfree(twsk_prot->twsk_slab_name);
3978 twsk_prot->twsk_slab_name = NULL;
3979 kmem_cache_destroy(twsk_prot->twsk_slab);
3980 twsk_prot->twsk_slab = NULL;
3981}
3982
3983static int tw_prot_init(const struct proto *prot)
3984{
3985 struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
3986
3987 if (!twsk_prot)
3988 return 0;
3989
3990 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
3991 prot->name);
3992 if (!twsk_prot->twsk_slab_name)
3993 return -ENOMEM;
3994
3995 twsk_prot->twsk_slab =
3996 kmem_cache_create(twsk_prot->twsk_slab_name,
3997 twsk_prot->twsk_obj_size, 0,
3998 SLAB_ACCOUNT | prot->slab_flags,
3999 NULL);
4000 if (!twsk_prot->twsk_slab) {
4001 pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4002 prot->name);
4003 return -ENOMEM;
4004 }
4005
4006 return 0;
4007}
4008
4009static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4010{
4011 if (!rsk_prot)
4012 return;
4013 kfree(rsk_prot->slab_name);
4014 rsk_prot->slab_name = NULL;
4015 kmem_cache_destroy(rsk_prot->slab);
4016 rsk_prot->slab = NULL;
4017}
4018
4019static int req_prot_init(const struct proto *prot)
4020{
4021 struct request_sock_ops *rsk_prot = prot->rsk_prot;
4022
4023 if (!rsk_prot)
4024 return 0;
4025
4026 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4027 prot->name);
4028 if (!rsk_prot->slab_name)
4029 return -ENOMEM;
4030
4031 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4032 rsk_prot->obj_size, 0,
4033 SLAB_ACCOUNT | prot->slab_flags,
4034 NULL);
4035
4036 if (!rsk_prot->slab) {
4037 pr_crit("%s: Can't create request sock SLAB cache!\n",
4038 prot->name);
4039 return -ENOMEM;
4040 }
4041 return 0;
4042}
4043
4044int proto_register(struct proto *prot, int alloc_slab)
4045{
4046 int ret = -ENOBUFS;
4047
4048 if (prot->memory_allocated && !prot->sysctl_mem) {
4049 pr_err("%s: missing sysctl_mem\n", prot->name);
4050 return -EINVAL;
4051 }
4052 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4053 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4054 return -EINVAL;
4055 }
4056 if (alloc_slab) {
4057 prot->slab = kmem_cache_create_usercopy(prot->name,
4058 prot->obj_size, 0,
4059 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4060 prot->slab_flags,
4061 prot->useroffset, prot->usersize,
4062 NULL);
4063
4064 if (prot->slab == NULL) {
4065 pr_crit("%s: Can't create sock SLAB cache!\n",
4066 prot->name);
4067 goto out;
4068 }
4069
4070 if (req_prot_init(prot))
4071 goto out_free_request_sock_slab;
4072
4073 if (tw_prot_init(prot))
4074 goto out_free_timewait_sock_slab;
4075 }
4076
4077 mutex_lock(&proto_list_mutex);
4078 ret = assign_proto_idx(prot);
4079 if (ret) {
4080 mutex_unlock(&proto_list_mutex);
4081 goto out_free_timewait_sock_slab;
4082 }
4083 list_add(&prot->node, &proto_list);
4084 mutex_unlock(&proto_list_mutex);
4085 return ret;
4086
4087out_free_timewait_sock_slab:
4088 if (alloc_slab)
4089 tw_prot_cleanup(prot->twsk_prot);
4090out_free_request_sock_slab:
4091 if (alloc_slab) {
4092 req_prot_cleanup(prot->rsk_prot);
4093
4094 kmem_cache_destroy(prot->slab);
4095 prot->slab = NULL;
4096 }
4097out:
4098 return ret;
4099}
4100EXPORT_SYMBOL(proto_register);
4101
4102void proto_unregister(struct proto *prot)
4103{
4104 mutex_lock(&proto_list_mutex);
4105 release_proto_idx(prot);
4106 list_del(&prot->node);
4107 mutex_unlock(&proto_list_mutex);
4108
4109 kmem_cache_destroy(prot->slab);
4110 prot->slab = NULL;
4111
4112 req_prot_cleanup(prot->rsk_prot);
4113 tw_prot_cleanup(prot->twsk_prot);
4114}
4115EXPORT_SYMBOL(proto_unregister);
4116
4117int sock_load_diag_module(int family, int protocol)
4118{
4119 if (!protocol) {
4120 if (!sock_is_registered(family))
4121 return -ENOENT;
4122
4123 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4124 NETLINK_SOCK_DIAG, family);
4125 }
4126
4127#ifdef CONFIG_INET
4128 if (family == AF_INET &&
4129 protocol != IPPROTO_RAW &&
4130 protocol < MAX_INET_PROTOS &&
4131 !rcu_access_pointer(inet_protos[protocol]))
4132 return -ENOENT;
4133#endif
4134
4135 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4136 NETLINK_SOCK_DIAG, family, protocol);
4137}
4138EXPORT_SYMBOL(sock_load_diag_module);
4139
4140#ifdef CONFIG_PROC_FS
4141static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4142 __acquires(proto_list_mutex)
4143{
4144 mutex_lock(&proto_list_mutex);
4145 return seq_list_start_head(&proto_list, *pos);
4146}
4147
4148static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4149{
4150 return seq_list_next(v, &proto_list, pos);
4151}
4152
4153static void proto_seq_stop(struct seq_file *seq, void *v)
4154 __releases(proto_list_mutex)
4155{
4156 mutex_unlock(&proto_list_mutex);
4157}
4158
4159static char proto_method_implemented(const void *method)
4160{
4161 return method == NULL ? 'n' : 'y';
4162}
4163static long sock_prot_memory_allocated(struct proto *proto)
4164{
4165 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4166}
4167
4168static const char *sock_prot_memory_pressure(struct proto *proto)
4169{
4170 return proto->memory_pressure != NULL ?
4171 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4172}
4173
4174static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4175{
4176
4177 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
4178 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4179 proto->name,
4180 proto->obj_size,
4181 sock_prot_inuse_get(seq_file_net(seq), proto),
4182 sock_prot_memory_allocated(proto),
4183 sock_prot_memory_pressure(proto),
4184 proto->max_header,
4185 proto->slab == NULL ? "no" : "yes",
4186 module_name(proto->owner),
4187 proto_method_implemented(proto->close),
4188 proto_method_implemented(proto->connect),
4189 proto_method_implemented(proto->disconnect),
4190 proto_method_implemented(proto->accept),
4191 proto_method_implemented(proto->ioctl),
4192 proto_method_implemented(proto->init),
4193 proto_method_implemented(proto->destroy),
4194 proto_method_implemented(proto->shutdown),
4195 proto_method_implemented(proto->setsockopt),
4196 proto_method_implemented(proto->getsockopt),
4197 proto_method_implemented(proto->sendmsg),
4198 proto_method_implemented(proto->recvmsg),
4199 proto_method_implemented(proto->bind),
4200 proto_method_implemented(proto->backlog_rcv),
4201 proto_method_implemented(proto->hash),
4202 proto_method_implemented(proto->unhash),
4203 proto_method_implemented(proto->get_port),
4204 proto_method_implemented(proto->enter_memory_pressure));
4205}
4206
4207static int proto_seq_show(struct seq_file *seq, void *v)
4208{
4209 if (v == &proto_list)
4210 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4211 "protocol",
4212 "size",
4213 "sockets",
4214 "memory",
4215 "press",
4216 "maxhdr",
4217 "slab",
4218 "module",
4219 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4220 else
4221 proto_seq_printf(seq, list_entry(v, struct proto, node));
4222 return 0;
4223}
4224
4225static const struct seq_operations proto_seq_ops = {
4226 .start = proto_seq_start,
4227 .next = proto_seq_next,
4228 .stop = proto_seq_stop,
4229 .show = proto_seq_show,
4230};
4231
4232static __net_init int proto_init_net(struct net *net)
4233{
4234 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4235 sizeof(struct seq_net_private)))
4236 return -ENOMEM;
4237
4238 return 0;
4239}
4240
4241static __net_exit void proto_exit_net(struct net *net)
4242{
4243 remove_proc_entry("protocols", net->proc_net);
4244}
4245
4246
4247static __net_initdata struct pernet_operations proto_net_ops = {
4248 .init = proto_init_net,
4249 .exit = proto_exit_net,
4250};
4251
4252static int __init proto_init(void)
4253{
4254 return register_pernet_subsys(&proto_net_ops);
4255}
4256
4257subsys_initcall(proto_init);
4258
4259#endif /* PROC_FS */
4260
4261#ifdef CONFIG_NET_RX_BUSY_POLL
4262bool sk_busy_loop_end(void *p, unsigned long start_time)
4263{
4264 struct sock *sk = p;
4265
4266 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4267 return true;
4268
4269 if (sk_is_udp(sk) &&
4270 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4271 return true;
4272
4273 return sk_busy_loop_timeout(sk, start_time);
4274}
4275EXPORT_SYMBOL(sk_busy_loop_end);
4276#endif /* CONFIG_NET_RX_BUSY_POLL */
4277
4278int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4279{
4280 if (!sk->sk_prot->bind_add)
4281 return -EOPNOTSUPP;
4282 return sk->sk_prot->bind_add(sk, addr, addr_len);
4283}
4284EXPORT_SYMBOL(sock_bind_add);
4285
4286/* Copy 'size' bytes from userspace and return `size` back to userspace */
4287int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4288 void __user *arg, void *karg, size_t size)
4289{
4290 int ret;
4291
4292 if (copy_from_user(karg, arg, size))
4293 return -EFAULT;
4294
4295 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4296 if (ret)
4297 return ret;
4298
4299 if (copy_to_user(arg, karg, size))
4300 return -EFAULT;
4301
4302 return 0;
4303}
4304EXPORT_SYMBOL(sock_ioctl_inout);
4305
4306/* This is the most common ioctl prep function, where the result (4 bytes) is
4307 * copied back to userspace if the ioctl() returns successfully. No input is
4308 * copied from userspace as input argument.
4309 */
4310static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4311{
4312 int ret, karg = 0;
4313
4314 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4315 if (ret)
4316 return ret;
4317
4318 return put_user(karg, (int __user *)arg);
4319}
4320
4321/* A wrapper around sock ioctls, which copies the data from userspace
4322 * (depending on the protocol/ioctl), and copies back the result to userspace.
4323 * The main motivation for this function is to pass kernel memory to the
4324 * protocol ioctl callbacks, instead of userspace memory.
4325 */
4326int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4327{
4328 int rc = 1;
4329
4330 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4331 rc = ipmr_sk_ioctl(sk, cmd, arg);
4332 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4333 rc = ip6mr_sk_ioctl(sk, cmd, arg);
4334 else if (sk_is_phonet(sk))
4335 rc = phonet_sk_ioctl(sk, cmd, arg);
4336
4337 /* If ioctl was processed, returns its value */
4338 if (rc <= 0)
4339 return rc;
4340
4341 /* Otherwise call the default handler */
4342 return sock_ioctl_out(sk, cmd, arg);
4343}
4344EXPORT_SYMBOL(sk_ioctl);
4345
4346static int __init sock_struct_check(void)
4347{
4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4353
4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4363
4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4367
4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4372
4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4385 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4386 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4387 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4388 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4389
4390 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4391 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4392 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4393 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4394 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4395 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4396 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4397 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4398 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4399 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4400 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4401 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4402 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4403 return 0;
4404}
4405
4406core_initcall(sock_struct_check);
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94#include <linux/capability.h>
95#include <linux/errno.h>
96#include <linux/types.h>
97#include <linux/socket.h>
98#include <linux/in.h>
99#include <linux/kernel.h>
100#include <linux/module.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
103#include <linux/sched.h>
104#include <linux/timer.h>
105#include <linux/string.h>
106#include <linux/sockios.h>
107#include <linux/net.h>
108#include <linux/mm.h>
109#include <linux/slab.h>
110#include <linux/interrupt.h>
111#include <linux/poll.h>
112#include <linux/tcp.h>
113#include <linux/init.h>
114#include <linux/highmem.h>
115#include <linux/user_namespace.h>
116#include <linux/static_key.h>
117#include <linux/memcontrol.h>
118#include <linux/prefetch.h>
119
120#include <asm/uaccess.h>
121
122#include <linux/netdevice.h>
123#include <net/protocol.h>
124#include <linux/skbuff.h>
125#include <net/net_namespace.h>
126#include <net/request_sock.h>
127#include <net/sock.h>
128#include <linux/net_tstamp.h>
129#include <net/xfrm.h>
130#include <linux/ipsec.h>
131#include <net/cls_cgroup.h>
132#include <net/netprio_cgroup.h>
133
134#include <linux/filter.h>
135
136#include <trace/events/sock.h>
137
138#ifdef CONFIG_INET
139#include <net/tcp.h>
140#endif
141
142static DEFINE_MUTEX(proto_list_mutex);
143static LIST_HEAD(proto_list);
144
145#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
146int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
147{
148 struct proto *proto;
149 int ret = 0;
150
151 mutex_lock(&proto_list_mutex);
152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
154 ret = proto->init_cgroup(memcg, ss);
155 if (ret)
156 goto out;
157 }
158 }
159
160 mutex_unlock(&proto_list_mutex);
161 return ret;
162out:
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
165 proto->destroy_cgroup(memcg);
166 mutex_unlock(&proto_list_mutex);
167 return ret;
168}
169
170void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
171{
172 struct proto *proto;
173
174 mutex_lock(&proto_list_mutex);
175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
177 proto->destroy_cgroup(memcg);
178 mutex_unlock(&proto_list_mutex);
179}
180#endif
181
182/*
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
185 */
186static struct lock_class_key af_family_keys[AF_MAX];
187static struct lock_class_key af_family_slock_keys[AF_MAX];
188
189struct static_key memcg_socket_limit_enabled;
190EXPORT_SYMBOL(memcg_socket_limit_enabled);
191
192/*
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
195 * locks is fast):
196 */
197static const char *const af_family_key_strings[AF_MAX+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
212};
213static const char *const af_family_slock_key_strings[AF_MAX+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
228};
229static const char *const af_family_clock_key_strings[AF_MAX+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
244};
245
246/*
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
249 */
250static struct lock_class_key af_callback_keys[AF_MAX];
251
252/* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
256 */
257#define _SK_MEM_PACKETS 256
258#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
261
262/* Run time adjustable parameters. */
263__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
264EXPORT_SYMBOL(sysctl_wmem_max);
265__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
266EXPORT_SYMBOL(sysctl_rmem_max);
267__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
269
270/* Maximal space eaten by iovec or ancillary data plus some space */
271int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
272EXPORT_SYMBOL(sysctl_optmem_max);
273
274#if defined(CONFIG_CGROUPS)
275#if !defined(CONFIG_NET_CLS_CGROUP)
276int net_cls_subsys_id = -1;
277EXPORT_SYMBOL_GPL(net_cls_subsys_id);
278#endif
279#if !defined(CONFIG_NETPRIO_CGROUP)
280int net_prio_subsys_id = -1;
281EXPORT_SYMBOL_GPL(net_prio_subsys_id);
282#endif
283#endif
284
285static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
286{
287 struct timeval tv;
288
289 if (optlen < sizeof(tv))
290 return -EINVAL;
291 if (copy_from_user(&tv, optval, sizeof(tv)))
292 return -EFAULT;
293 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
294 return -EDOM;
295
296 if (tv.tv_sec < 0) {
297 static int warned __read_mostly;
298
299 *timeo_p = 0;
300 if (warned < 10 && net_ratelimit()) {
301 warned++;
302 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
303 __func__, current->comm, task_pid_nr(current));
304 }
305 return 0;
306 }
307 *timeo_p = MAX_SCHEDULE_TIMEOUT;
308 if (tv.tv_sec == 0 && tv.tv_usec == 0)
309 return 0;
310 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
311 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
312 return 0;
313}
314
315static void sock_warn_obsolete_bsdism(const char *name)
316{
317 static int warned;
318 static char warncomm[TASK_COMM_LEN];
319 if (strcmp(warncomm, current->comm) && warned < 5) {
320 strcpy(warncomm, current->comm);
321 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
322 warncomm, name);
323 warned++;
324 }
325}
326
327#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
328
329static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
330{
331 if (sk->sk_flags & flags) {
332 sk->sk_flags &= ~flags;
333 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
334 net_disable_timestamp();
335 }
336}
337
338
339int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
340{
341 int err;
342 int skb_len;
343 unsigned long flags;
344 struct sk_buff_head *list = &sk->sk_receive_queue;
345
346 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
347 atomic_inc(&sk->sk_drops);
348 trace_sock_rcvqueue_full(sk, skb);
349 return -ENOMEM;
350 }
351
352 err = sk_filter(sk, skb);
353 if (err)
354 return err;
355
356 if (!sk_rmem_schedule(sk, skb->truesize)) {
357 atomic_inc(&sk->sk_drops);
358 return -ENOBUFS;
359 }
360
361 skb->dev = NULL;
362 skb_set_owner_r(skb, sk);
363
364 /* Cache the SKB length before we tack it onto the receive
365 * queue. Once it is added it no longer belongs to us and
366 * may be freed by other threads of control pulling packets
367 * from the queue.
368 */
369 skb_len = skb->len;
370
371 /* we escape from rcu protected region, make sure we dont leak
372 * a norefcounted dst
373 */
374 skb_dst_force(skb);
375
376 spin_lock_irqsave(&list->lock, flags);
377 skb->dropcount = atomic_read(&sk->sk_drops);
378 __skb_queue_tail(list, skb);
379 spin_unlock_irqrestore(&list->lock, flags);
380
381 if (!sock_flag(sk, SOCK_DEAD))
382 sk->sk_data_ready(sk, skb_len);
383 return 0;
384}
385EXPORT_SYMBOL(sock_queue_rcv_skb);
386
387int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
388{
389 int rc = NET_RX_SUCCESS;
390
391 if (sk_filter(sk, skb))
392 goto discard_and_relse;
393
394 skb->dev = NULL;
395
396 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
397 atomic_inc(&sk->sk_drops);
398 goto discard_and_relse;
399 }
400 if (nested)
401 bh_lock_sock_nested(sk);
402 else
403 bh_lock_sock(sk);
404 if (!sock_owned_by_user(sk)) {
405 /*
406 * trylock + unlock semantics:
407 */
408 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
409
410 rc = sk_backlog_rcv(sk, skb);
411
412 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
413 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
414 bh_unlock_sock(sk);
415 atomic_inc(&sk->sk_drops);
416 goto discard_and_relse;
417 }
418
419 bh_unlock_sock(sk);
420out:
421 sock_put(sk);
422 return rc;
423discard_and_relse:
424 kfree_skb(skb);
425 goto out;
426}
427EXPORT_SYMBOL(sk_receive_skb);
428
429void sk_reset_txq(struct sock *sk)
430{
431 sk_tx_queue_clear(sk);
432}
433EXPORT_SYMBOL(sk_reset_txq);
434
435struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
436{
437 struct dst_entry *dst = __sk_dst_get(sk);
438
439 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
440 sk_tx_queue_clear(sk);
441 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
442 dst_release(dst);
443 return NULL;
444 }
445
446 return dst;
447}
448EXPORT_SYMBOL(__sk_dst_check);
449
450struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
451{
452 struct dst_entry *dst = sk_dst_get(sk);
453
454 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
455 sk_dst_reset(sk);
456 dst_release(dst);
457 return NULL;
458 }
459
460 return dst;
461}
462EXPORT_SYMBOL(sk_dst_check);
463
464static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
465{
466 int ret = -ENOPROTOOPT;
467#ifdef CONFIG_NETDEVICES
468 struct net *net = sock_net(sk);
469 char devname[IFNAMSIZ];
470 int index;
471
472 /* Sorry... */
473 ret = -EPERM;
474 if (!capable(CAP_NET_RAW))
475 goto out;
476
477 ret = -EINVAL;
478 if (optlen < 0)
479 goto out;
480
481 /* Bind this socket to a particular device like "eth0",
482 * as specified in the passed interface name. If the
483 * name is "" or the option length is zero the socket
484 * is not bound.
485 */
486 if (optlen > IFNAMSIZ - 1)
487 optlen = IFNAMSIZ - 1;
488 memset(devname, 0, sizeof(devname));
489
490 ret = -EFAULT;
491 if (copy_from_user(devname, optval, optlen))
492 goto out;
493
494 index = 0;
495 if (devname[0] != '\0') {
496 struct net_device *dev;
497
498 rcu_read_lock();
499 dev = dev_get_by_name_rcu(net, devname);
500 if (dev)
501 index = dev->ifindex;
502 rcu_read_unlock();
503 ret = -ENODEV;
504 if (!dev)
505 goto out;
506 }
507
508 lock_sock(sk);
509 sk->sk_bound_dev_if = index;
510 sk_dst_reset(sk);
511 release_sock(sk);
512
513 ret = 0;
514
515out:
516#endif
517
518 return ret;
519}
520
521static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
522{
523 if (valbool)
524 sock_set_flag(sk, bit);
525 else
526 sock_reset_flag(sk, bit);
527}
528
529/*
530 * This is meant for all protocols to use and covers goings on
531 * at the socket level. Everything here is generic.
532 */
533
534int sock_setsockopt(struct socket *sock, int level, int optname,
535 char __user *optval, unsigned int optlen)
536{
537 struct sock *sk = sock->sk;
538 int val;
539 int valbool;
540 struct linger ling;
541 int ret = 0;
542
543 /*
544 * Options without arguments
545 */
546
547 if (optname == SO_BINDTODEVICE)
548 return sock_bindtodevice(sk, optval, optlen);
549
550 if (optlen < sizeof(int))
551 return -EINVAL;
552
553 if (get_user(val, (int __user *)optval))
554 return -EFAULT;
555
556 valbool = val ? 1 : 0;
557
558 lock_sock(sk);
559
560 switch (optname) {
561 case SO_DEBUG:
562 if (val && !capable(CAP_NET_ADMIN))
563 ret = -EACCES;
564 else
565 sock_valbool_flag(sk, SOCK_DBG, valbool);
566 break;
567 case SO_REUSEADDR:
568 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
569 break;
570 case SO_TYPE:
571 case SO_PROTOCOL:
572 case SO_DOMAIN:
573 case SO_ERROR:
574 ret = -ENOPROTOOPT;
575 break;
576 case SO_DONTROUTE:
577 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
578 break;
579 case SO_BROADCAST:
580 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
581 break;
582 case SO_SNDBUF:
583 /* Don't error on this BSD doesn't and if you think
584 * about it this is right. Otherwise apps have to
585 * play 'guess the biggest size' games. RCVBUF/SNDBUF
586 * are treated in BSD as hints
587 */
588 val = min_t(u32, val, sysctl_wmem_max);
589set_sndbuf:
590 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
591 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
592 /* Wake up sending tasks if we upped the value. */
593 sk->sk_write_space(sk);
594 break;
595
596 case SO_SNDBUFFORCE:
597 if (!capable(CAP_NET_ADMIN)) {
598 ret = -EPERM;
599 break;
600 }
601 goto set_sndbuf;
602
603 case SO_RCVBUF:
604 /* Don't error on this BSD doesn't and if you think
605 * about it this is right. Otherwise apps have to
606 * play 'guess the biggest size' games. RCVBUF/SNDBUF
607 * are treated in BSD as hints
608 */
609 val = min_t(u32, val, sysctl_rmem_max);
610set_rcvbuf:
611 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
612 /*
613 * We double it on the way in to account for
614 * "struct sk_buff" etc. overhead. Applications
615 * assume that the SO_RCVBUF setting they make will
616 * allow that much actual data to be received on that
617 * socket.
618 *
619 * Applications are unaware that "struct sk_buff" and
620 * other overheads allocate from the receive buffer
621 * during socket buffer allocation.
622 *
623 * And after considering the possible alternatives,
624 * returning the value we actually used in getsockopt
625 * is the most desirable behavior.
626 */
627 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
628 break;
629
630 case SO_RCVBUFFORCE:
631 if (!capable(CAP_NET_ADMIN)) {
632 ret = -EPERM;
633 break;
634 }
635 goto set_rcvbuf;
636
637 case SO_KEEPALIVE:
638#ifdef CONFIG_INET
639 if (sk->sk_protocol == IPPROTO_TCP)
640 tcp_set_keepalive(sk, valbool);
641#endif
642 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
643 break;
644
645 case SO_OOBINLINE:
646 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
647 break;
648
649 case SO_NO_CHECK:
650 sk->sk_no_check = valbool;
651 break;
652
653 case SO_PRIORITY:
654 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
655 sk->sk_priority = val;
656 else
657 ret = -EPERM;
658 break;
659
660 case SO_LINGER:
661 if (optlen < sizeof(ling)) {
662 ret = -EINVAL; /* 1003.1g */
663 break;
664 }
665 if (copy_from_user(&ling, optval, sizeof(ling))) {
666 ret = -EFAULT;
667 break;
668 }
669 if (!ling.l_onoff)
670 sock_reset_flag(sk, SOCK_LINGER);
671 else {
672#if (BITS_PER_LONG == 32)
673 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
674 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
675 else
676#endif
677 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
678 sock_set_flag(sk, SOCK_LINGER);
679 }
680 break;
681
682 case SO_BSDCOMPAT:
683 sock_warn_obsolete_bsdism("setsockopt");
684 break;
685
686 case SO_PASSCRED:
687 if (valbool)
688 set_bit(SOCK_PASSCRED, &sock->flags);
689 else
690 clear_bit(SOCK_PASSCRED, &sock->flags);
691 break;
692
693 case SO_TIMESTAMP:
694 case SO_TIMESTAMPNS:
695 if (valbool) {
696 if (optname == SO_TIMESTAMP)
697 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
698 else
699 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
700 sock_set_flag(sk, SOCK_RCVTSTAMP);
701 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
702 } else {
703 sock_reset_flag(sk, SOCK_RCVTSTAMP);
704 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
705 }
706 break;
707
708 case SO_TIMESTAMPING:
709 if (val & ~SOF_TIMESTAMPING_MASK) {
710 ret = -EINVAL;
711 break;
712 }
713 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
714 val & SOF_TIMESTAMPING_TX_HARDWARE);
715 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
716 val & SOF_TIMESTAMPING_TX_SOFTWARE);
717 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
718 val & SOF_TIMESTAMPING_RX_HARDWARE);
719 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
720 sock_enable_timestamp(sk,
721 SOCK_TIMESTAMPING_RX_SOFTWARE);
722 else
723 sock_disable_timestamp(sk,
724 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
725 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
726 val & SOF_TIMESTAMPING_SOFTWARE);
727 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
728 val & SOF_TIMESTAMPING_SYS_HARDWARE);
729 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
730 val & SOF_TIMESTAMPING_RAW_HARDWARE);
731 break;
732
733 case SO_RCVLOWAT:
734 if (val < 0)
735 val = INT_MAX;
736 sk->sk_rcvlowat = val ? : 1;
737 break;
738
739 case SO_RCVTIMEO:
740 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
741 break;
742
743 case SO_SNDTIMEO:
744 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
745 break;
746
747 case SO_ATTACH_FILTER:
748 ret = -EINVAL;
749 if (optlen == sizeof(struct sock_fprog)) {
750 struct sock_fprog fprog;
751
752 ret = -EFAULT;
753 if (copy_from_user(&fprog, optval, sizeof(fprog)))
754 break;
755
756 ret = sk_attach_filter(&fprog, sk);
757 }
758 break;
759
760 case SO_DETACH_FILTER:
761 ret = sk_detach_filter(sk);
762 break;
763
764 case SO_PASSSEC:
765 if (valbool)
766 set_bit(SOCK_PASSSEC, &sock->flags);
767 else
768 clear_bit(SOCK_PASSSEC, &sock->flags);
769 break;
770 case SO_MARK:
771 if (!capable(CAP_NET_ADMIN))
772 ret = -EPERM;
773 else
774 sk->sk_mark = val;
775 break;
776
777 /* We implement the SO_SNDLOWAT etc to
778 not be settable (1003.1g 5.3) */
779 case SO_RXQ_OVFL:
780 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
781 break;
782
783 case SO_WIFI_STATUS:
784 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
785 break;
786
787 case SO_PEEK_OFF:
788 if (sock->ops->set_peek_off)
789 sock->ops->set_peek_off(sk, val);
790 else
791 ret = -EOPNOTSUPP;
792 break;
793
794 case SO_NOFCS:
795 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
796 break;
797
798 default:
799 ret = -ENOPROTOOPT;
800 break;
801 }
802 release_sock(sk);
803 return ret;
804}
805EXPORT_SYMBOL(sock_setsockopt);
806
807
808void cred_to_ucred(struct pid *pid, const struct cred *cred,
809 struct ucred *ucred)
810{
811 ucred->pid = pid_vnr(pid);
812 ucred->uid = ucred->gid = -1;
813 if (cred) {
814 struct user_namespace *current_ns = current_user_ns();
815
816 ucred->uid = from_kuid(current_ns, cred->euid);
817 ucred->gid = from_kgid(current_ns, cred->egid);
818 }
819}
820EXPORT_SYMBOL_GPL(cred_to_ucred);
821
822int sock_getsockopt(struct socket *sock, int level, int optname,
823 char __user *optval, int __user *optlen)
824{
825 struct sock *sk = sock->sk;
826
827 union {
828 int val;
829 struct linger ling;
830 struct timeval tm;
831 } v;
832
833 int lv = sizeof(int);
834 int len;
835
836 if (get_user(len, optlen))
837 return -EFAULT;
838 if (len < 0)
839 return -EINVAL;
840
841 memset(&v, 0, sizeof(v));
842
843 switch (optname) {
844 case SO_DEBUG:
845 v.val = sock_flag(sk, SOCK_DBG);
846 break;
847
848 case SO_DONTROUTE:
849 v.val = sock_flag(sk, SOCK_LOCALROUTE);
850 break;
851
852 case SO_BROADCAST:
853 v.val = sock_flag(sk, SOCK_BROADCAST);
854 break;
855
856 case SO_SNDBUF:
857 v.val = sk->sk_sndbuf;
858 break;
859
860 case SO_RCVBUF:
861 v.val = sk->sk_rcvbuf;
862 break;
863
864 case SO_REUSEADDR:
865 v.val = sk->sk_reuse;
866 break;
867
868 case SO_KEEPALIVE:
869 v.val = sock_flag(sk, SOCK_KEEPOPEN);
870 break;
871
872 case SO_TYPE:
873 v.val = sk->sk_type;
874 break;
875
876 case SO_PROTOCOL:
877 v.val = sk->sk_protocol;
878 break;
879
880 case SO_DOMAIN:
881 v.val = sk->sk_family;
882 break;
883
884 case SO_ERROR:
885 v.val = -sock_error(sk);
886 if (v.val == 0)
887 v.val = xchg(&sk->sk_err_soft, 0);
888 break;
889
890 case SO_OOBINLINE:
891 v.val = sock_flag(sk, SOCK_URGINLINE);
892 break;
893
894 case SO_NO_CHECK:
895 v.val = sk->sk_no_check;
896 break;
897
898 case SO_PRIORITY:
899 v.val = sk->sk_priority;
900 break;
901
902 case SO_LINGER:
903 lv = sizeof(v.ling);
904 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
905 v.ling.l_linger = sk->sk_lingertime / HZ;
906 break;
907
908 case SO_BSDCOMPAT:
909 sock_warn_obsolete_bsdism("getsockopt");
910 break;
911
912 case SO_TIMESTAMP:
913 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
914 !sock_flag(sk, SOCK_RCVTSTAMPNS);
915 break;
916
917 case SO_TIMESTAMPNS:
918 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
919 break;
920
921 case SO_TIMESTAMPING:
922 v.val = 0;
923 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
924 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
925 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
926 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
927 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
928 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
929 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
930 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
931 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
932 v.val |= SOF_TIMESTAMPING_SOFTWARE;
933 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
934 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
935 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
936 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
937 break;
938
939 case SO_RCVTIMEO:
940 lv = sizeof(struct timeval);
941 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
942 v.tm.tv_sec = 0;
943 v.tm.tv_usec = 0;
944 } else {
945 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
946 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
947 }
948 break;
949
950 case SO_SNDTIMEO:
951 lv = sizeof(struct timeval);
952 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
953 v.tm.tv_sec = 0;
954 v.tm.tv_usec = 0;
955 } else {
956 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
957 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
958 }
959 break;
960
961 case SO_RCVLOWAT:
962 v.val = sk->sk_rcvlowat;
963 break;
964
965 case SO_SNDLOWAT:
966 v.val = 1;
967 break;
968
969 case SO_PASSCRED:
970 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
971 break;
972
973 case SO_PEERCRED:
974 {
975 struct ucred peercred;
976 if (len > sizeof(peercred))
977 len = sizeof(peercred);
978 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
979 if (copy_to_user(optval, &peercred, len))
980 return -EFAULT;
981 goto lenout;
982 }
983
984 case SO_PEERNAME:
985 {
986 char address[128];
987
988 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
989 return -ENOTCONN;
990 if (lv < len)
991 return -EINVAL;
992 if (copy_to_user(optval, address, len))
993 return -EFAULT;
994 goto lenout;
995 }
996
997 /* Dubious BSD thing... Probably nobody even uses it, but
998 * the UNIX standard wants it for whatever reason... -DaveM
999 */
1000 case SO_ACCEPTCONN:
1001 v.val = sk->sk_state == TCP_LISTEN;
1002 break;
1003
1004 case SO_PASSSEC:
1005 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1006 break;
1007
1008 case SO_PEERSEC:
1009 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1010
1011 case SO_MARK:
1012 v.val = sk->sk_mark;
1013 break;
1014
1015 case SO_RXQ_OVFL:
1016 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1017 break;
1018
1019 case SO_WIFI_STATUS:
1020 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1021 break;
1022
1023 case SO_PEEK_OFF:
1024 if (!sock->ops->set_peek_off)
1025 return -EOPNOTSUPP;
1026
1027 v.val = sk->sk_peek_off;
1028 break;
1029 case SO_NOFCS:
1030 v.val = sock_flag(sk, SOCK_NOFCS);
1031 break;
1032 default:
1033 return -ENOPROTOOPT;
1034 }
1035
1036 if (len > lv)
1037 len = lv;
1038 if (copy_to_user(optval, &v, len))
1039 return -EFAULT;
1040lenout:
1041 if (put_user(len, optlen))
1042 return -EFAULT;
1043 return 0;
1044}
1045
1046/*
1047 * Initialize an sk_lock.
1048 *
1049 * (We also register the sk_lock with the lock validator.)
1050 */
1051static inline void sock_lock_init(struct sock *sk)
1052{
1053 sock_lock_init_class_and_name(sk,
1054 af_family_slock_key_strings[sk->sk_family],
1055 af_family_slock_keys + sk->sk_family,
1056 af_family_key_strings[sk->sk_family],
1057 af_family_keys + sk->sk_family);
1058}
1059
1060/*
1061 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1062 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1063 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1064 */
1065static void sock_copy(struct sock *nsk, const struct sock *osk)
1066{
1067#ifdef CONFIG_SECURITY_NETWORK
1068 void *sptr = nsk->sk_security;
1069#endif
1070 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1071
1072 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1073 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1074
1075#ifdef CONFIG_SECURITY_NETWORK
1076 nsk->sk_security = sptr;
1077 security_sk_clone(osk, nsk);
1078#endif
1079}
1080
1081/*
1082 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1083 * un-modified. Special care is taken when initializing object to zero.
1084 */
1085static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1086{
1087 if (offsetof(struct sock, sk_node.next) != 0)
1088 memset(sk, 0, offsetof(struct sock, sk_node.next));
1089 memset(&sk->sk_node.pprev, 0,
1090 size - offsetof(struct sock, sk_node.pprev));
1091}
1092
1093void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1094{
1095 unsigned long nulls1, nulls2;
1096
1097 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1098 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1099 if (nulls1 > nulls2)
1100 swap(nulls1, nulls2);
1101
1102 if (nulls1 != 0)
1103 memset((char *)sk, 0, nulls1);
1104 memset((char *)sk + nulls1 + sizeof(void *), 0,
1105 nulls2 - nulls1 - sizeof(void *));
1106 memset((char *)sk + nulls2 + sizeof(void *), 0,
1107 size - nulls2 - sizeof(void *));
1108}
1109EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1110
1111static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1112 int family)
1113{
1114 struct sock *sk;
1115 struct kmem_cache *slab;
1116
1117 slab = prot->slab;
1118 if (slab != NULL) {
1119 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1120 if (!sk)
1121 return sk;
1122 if (priority & __GFP_ZERO) {
1123 if (prot->clear_sk)
1124 prot->clear_sk(sk, prot->obj_size);
1125 else
1126 sk_prot_clear_nulls(sk, prot->obj_size);
1127 }
1128 } else
1129 sk = kmalloc(prot->obj_size, priority);
1130
1131 if (sk != NULL) {
1132 kmemcheck_annotate_bitfield(sk, flags);
1133
1134 if (security_sk_alloc(sk, family, priority))
1135 goto out_free;
1136
1137 if (!try_module_get(prot->owner))
1138 goto out_free_sec;
1139 sk_tx_queue_clear(sk);
1140 }
1141
1142 return sk;
1143
1144out_free_sec:
1145 security_sk_free(sk);
1146out_free:
1147 if (slab != NULL)
1148 kmem_cache_free(slab, sk);
1149 else
1150 kfree(sk);
1151 return NULL;
1152}
1153
1154static void sk_prot_free(struct proto *prot, struct sock *sk)
1155{
1156 struct kmem_cache *slab;
1157 struct module *owner;
1158
1159 owner = prot->owner;
1160 slab = prot->slab;
1161
1162 security_sk_free(sk);
1163 if (slab != NULL)
1164 kmem_cache_free(slab, sk);
1165 else
1166 kfree(sk);
1167 module_put(owner);
1168}
1169
1170#ifdef CONFIG_CGROUPS
1171void sock_update_classid(struct sock *sk)
1172{
1173 u32 classid;
1174
1175 rcu_read_lock(); /* doing current task, which cannot vanish. */
1176 classid = task_cls_classid(current);
1177 rcu_read_unlock();
1178 if (classid && classid != sk->sk_classid)
1179 sk->sk_classid = classid;
1180}
1181EXPORT_SYMBOL(sock_update_classid);
1182
1183void sock_update_netprioidx(struct sock *sk)
1184{
1185 if (in_interrupt())
1186 return;
1187
1188 sk->sk_cgrp_prioidx = task_netprioidx(current);
1189}
1190EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1191#endif
1192
1193/**
1194 * sk_alloc - All socket objects are allocated here
1195 * @net: the applicable net namespace
1196 * @family: protocol family
1197 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1198 * @prot: struct proto associated with this new sock instance
1199 */
1200struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1201 struct proto *prot)
1202{
1203 struct sock *sk;
1204
1205 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1206 if (sk) {
1207 sk->sk_family = family;
1208 /*
1209 * See comment in struct sock definition to understand
1210 * why we need sk_prot_creator -acme
1211 */
1212 sk->sk_prot = sk->sk_prot_creator = prot;
1213 sock_lock_init(sk);
1214 sock_net_set(sk, get_net(net));
1215 atomic_set(&sk->sk_wmem_alloc, 1);
1216
1217 sock_update_classid(sk);
1218 sock_update_netprioidx(sk);
1219 }
1220
1221 return sk;
1222}
1223EXPORT_SYMBOL(sk_alloc);
1224
1225static void __sk_free(struct sock *sk)
1226{
1227 struct sk_filter *filter;
1228
1229 if (sk->sk_destruct)
1230 sk->sk_destruct(sk);
1231
1232 filter = rcu_dereference_check(sk->sk_filter,
1233 atomic_read(&sk->sk_wmem_alloc) == 0);
1234 if (filter) {
1235 sk_filter_uncharge(sk, filter);
1236 RCU_INIT_POINTER(sk->sk_filter, NULL);
1237 }
1238
1239 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1240
1241 if (atomic_read(&sk->sk_omem_alloc))
1242 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1243 __func__, atomic_read(&sk->sk_omem_alloc));
1244
1245 if (sk->sk_peer_cred)
1246 put_cred(sk->sk_peer_cred);
1247 put_pid(sk->sk_peer_pid);
1248 put_net(sock_net(sk));
1249 sk_prot_free(sk->sk_prot_creator, sk);
1250}
1251
1252void sk_free(struct sock *sk)
1253{
1254 /*
1255 * We subtract one from sk_wmem_alloc and can know if
1256 * some packets are still in some tx queue.
1257 * If not null, sock_wfree() will call __sk_free(sk) later
1258 */
1259 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1260 __sk_free(sk);
1261}
1262EXPORT_SYMBOL(sk_free);
1263
1264/*
1265 * Last sock_put should drop reference to sk->sk_net. It has already
1266 * been dropped in sk_change_net. Taking reference to stopping namespace
1267 * is not an option.
1268 * Take reference to a socket to remove it from hash _alive_ and after that
1269 * destroy it in the context of init_net.
1270 */
1271void sk_release_kernel(struct sock *sk)
1272{
1273 if (sk == NULL || sk->sk_socket == NULL)
1274 return;
1275
1276 sock_hold(sk);
1277 sock_release(sk->sk_socket);
1278 release_net(sock_net(sk));
1279 sock_net_set(sk, get_net(&init_net));
1280 sock_put(sk);
1281}
1282EXPORT_SYMBOL(sk_release_kernel);
1283
1284static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1285{
1286 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1287 sock_update_memcg(newsk);
1288}
1289
1290/**
1291 * sk_clone_lock - clone a socket, and lock its clone
1292 * @sk: the socket to clone
1293 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1294 *
1295 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1296 */
1297struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1298{
1299 struct sock *newsk;
1300
1301 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1302 if (newsk != NULL) {
1303 struct sk_filter *filter;
1304
1305 sock_copy(newsk, sk);
1306
1307 /* SANITY */
1308 get_net(sock_net(newsk));
1309 sk_node_init(&newsk->sk_node);
1310 sock_lock_init(newsk);
1311 bh_lock_sock(newsk);
1312 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1313 newsk->sk_backlog.len = 0;
1314
1315 atomic_set(&newsk->sk_rmem_alloc, 0);
1316 /*
1317 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1318 */
1319 atomic_set(&newsk->sk_wmem_alloc, 1);
1320 atomic_set(&newsk->sk_omem_alloc, 0);
1321 skb_queue_head_init(&newsk->sk_receive_queue);
1322 skb_queue_head_init(&newsk->sk_write_queue);
1323#ifdef CONFIG_NET_DMA
1324 skb_queue_head_init(&newsk->sk_async_wait_queue);
1325#endif
1326
1327 spin_lock_init(&newsk->sk_dst_lock);
1328 rwlock_init(&newsk->sk_callback_lock);
1329 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1330 af_callback_keys + newsk->sk_family,
1331 af_family_clock_key_strings[newsk->sk_family]);
1332
1333 newsk->sk_dst_cache = NULL;
1334 newsk->sk_wmem_queued = 0;
1335 newsk->sk_forward_alloc = 0;
1336 newsk->sk_send_head = NULL;
1337 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1338
1339 sock_reset_flag(newsk, SOCK_DONE);
1340 skb_queue_head_init(&newsk->sk_error_queue);
1341
1342 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1343 if (filter != NULL)
1344 sk_filter_charge(newsk, filter);
1345
1346 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1347 /* It is still raw copy of parent, so invalidate
1348 * destructor and make plain sk_free() */
1349 newsk->sk_destruct = NULL;
1350 bh_unlock_sock(newsk);
1351 sk_free(newsk);
1352 newsk = NULL;
1353 goto out;
1354 }
1355
1356 newsk->sk_err = 0;
1357 newsk->sk_priority = 0;
1358 /*
1359 * Before updating sk_refcnt, we must commit prior changes to memory
1360 * (Documentation/RCU/rculist_nulls.txt for details)
1361 */
1362 smp_wmb();
1363 atomic_set(&newsk->sk_refcnt, 2);
1364
1365 /*
1366 * Increment the counter in the same struct proto as the master
1367 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1368 * is the same as sk->sk_prot->socks, as this field was copied
1369 * with memcpy).
1370 *
1371 * This _changes_ the previous behaviour, where
1372 * tcp_create_openreq_child always was incrementing the
1373 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1374 * to be taken into account in all callers. -acme
1375 */
1376 sk_refcnt_debug_inc(newsk);
1377 sk_set_socket(newsk, NULL);
1378 newsk->sk_wq = NULL;
1379
1380 sk_update_clone(sk, newsk);
1381
1382 if (newsk->sk_prot->sockets_allocated)
1383 sk_sockets_allocated_inc(newsk);
1384
1385 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1386 net_enable_timestamp();
1387 }
1388out:
1389 return newsk;
1390}
1391EXPORT_SYMBOL_GPL(sk_clone_lock);
1392
1393void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1394{
1395 __sk_dst_set(sk, dst);
1396 sk->sk_route_caps = dst->dev->features;
1397 if (sk->sk_route_caps & NETIF_F_GSO)
1398 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1399 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1400 if (sk_can_gso(sk)) {
1401 if (dst->header_len) {
1402 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1403 } else {
1404 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1405 sk->sk_gso_max_size = dst->dev->gso_max_size;
1406 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1407 }
1408 }
1409}
1410EXPORT_SYMBOL_GPL(sk_setup_caps);
1411
1412void __init sk_init(void)
1413{
1414 if (totalram_pages <= 4096) {
1415 sysctl_wmem_max = 32767;
1416 sysctl_rmem_max = 32767;
1417 sysctl_wmem_default = 32767;
1418 sysctl_rmem_default = 32767;
1419 } else if (totalram_pages >= 131072) {
1420 sysctl_wmem_max = 131071;
1421 sysctl_rmem_max = 131071;
1422 }
1423}
1424
1425/*
1426 * Simple resource managers for sockets.
1427 */
1428
1429
1430/*
1431 * Write buffer destructor automatically called from kfree_skb.
1432 */
1433void sock_wfree(struct sk_buff *skb)
1434{
1435 struct sock *sk = skb->sk;
1436 unsigned int len = skb->truesize;
1437
1438 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1439 /*
1440 * Keep a reference on sk_wmem_alloc, this will be released
1441 * after sk_write_space() call
1442 */
1443 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1444 sk->sk_write_space(sk);
1445 len = 1;
1446 }
1447 /*
1448 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1449 * could not do because of in-flight packets
1450 */
1451 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1452 __sk_free(sk);
1453}
1454EXPORT_SYMBOL(sock_wfree);
1455
1456/*
1457 * Read buffer destructor automatically called from kfree_skb.
1458 */
1459void sock_rfree(struct sk_buff *skb)
1460{
1461 struct sock *sk = skb->sk;
1462 unsigned int len = skb->truesize;
1463
1464 atomic_sub(len, &sk->sk_rmem_alloc);
1465 sk_mem_uncharge(sk, len);
1466}
1467EXPORT_SYMBOL(sock_rfree);
1468
1469
1470int sock_i_uid(struct sock *sk)
1471{
1472 int uid;
1473
1474 read_lock_bh(&sk->sk_callback_lock);
1475 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1476 read_unlock_bh(&sk->sk_callback_lock);
1477 return uid;
1478}
1479EXPORT_SYMBOL(sock_i_uid);
1480
1481unsigned long sock_i_ino(struct sock *sk)
1482{
1483 unsigned long ino;
1484
1485 read_lock_bh(&sk->sk_callback_lock);
1486 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1487 read_unlock_bh(&sk->sk_callback_lock);
1488 return ino;
1489}
1490EXPORT_SYMBOL(sock_i_ino);
1491
1492/*
1493 * Allocate a skb from the socket's send buffer.
1494 */
1495struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1496 gfp_t priority)
1497{
1498 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1499 struct sk_buff *skb = alloc_skb(size, priority);
1500 if (skb) {
1501 skb_set_owner_w(skb, sk);
1502 return skb;
1503 }
1504 }
1505 return NULL;
1506}
1507EXPORT_SYMBOL(sock_wmalloc);
1508
1509/*
1510 * Allocate a skb from the socket's receive buffer.
1511 */
1512struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1513 gfp_t priority)
1514{
1515 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1516 struct sk_buff *skb = alloc_skb(size, priority);
1517 if (skb) {
1518 skb_set_owner_r(skb, sk);
1519 return skb;
1520 }
1521 }
1522 return NULL;
1523}
1524
1525/*
1526 * Allocate a memory block from the socket's option memory buffer.
1527 */
1528void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1529{
1530 if ((unsigned int)size <= sysctl_optmem_max &&
1531 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1532 void *mem;
1533 /* First do the add, to avoid the race if kmalloc
1534 * might sleep.
1535 */
1536 atomic_add(size, &sk->sk_omem_alloc);
1537 mem = kmalloc(size, priority);
1538 if (mem)
1539 return mem;
1540 atomic_sub(size, &sk->sk_omem_alloc);
1541 }
1542 return NULL;
1543}
1544EXPORT_SYMBOL(sock_kmalloc);
1545
1546/*
1547 * Free an option memory block.
1548 */
1549void sock_kfree_s(struct sock *sk, void *mem, int size)
1550{
1551 kfree(mem);
1552 atomic_sub(size, &sk->sk_omem_alloc);
1553}
1554EXPORT_SYMBOL(sock_kfree_s);
1555
1556/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1557 I think, these locks should be removed for datagram sockets.
1558 */
1559static long sock_wait_for_wmem(struct sock *sk, long timeo)
1560{
1561 DEFINE_WAIT(wait);
1562
1563 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1564 for (;;) {
1565 if (!timeo)
1566 break;
1567 if (signal_pending(current))
1568 break;
1569 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1570 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1571 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1572 break;
1573 if (sk->sk_shutdown & SEND_SHUTDOWN)
1574 break;
1575 if (sk->sk_err)
1576 break;
1577 timeo = schedule_timeout(timeo);
1578 }
1579 finish_wait(sk_sleep(sk), &wait);
1580 return timeo;
1581}
1582
1583
1584/*
1585 * Generic send/receive buffer handlers
1586 */
1587
1588struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1589 unsigned long data_len, int noblock,
1590 int *errcode)
1591{
1592 struct sk_buff *skb;
1593 gfp_t gfp_mask;
1594 long timeo;
1595 int err;
1596 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1597
1598 err = -EMSGSIZE;
1599 if (npages > MAX_SKB_FRAGS)
1600 goto failure;
1601
1602 gfp_mask = sk->sk_allocation;
1603 if (gfp_mask & __GFP_WAIT)
1604 gfp_mask |= __GFP_REPEAT;
1605
1606 timeo = sock_sndtimeo(sk, noblock);
1607 while (1) {
1608 err = sock_error(sk);
1609 if (err != 0)
1610 goto failure;
1611
1612 err = -EPIPE;
1613 if (sk->sk_shutdown & SEND_SHUTDOWN)
1614 goto failure;
1615
1616 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1617 skb = alloc_skb(header_len, gfp_mask);
1618 if (skb) {
1619 int i;
1620
1621 /* No pages, we're done... */
1622 if (!data_len)
1623 break;
1624
1625 skb->truesize += data_len;
1626 skb_shinfo(skb)->nr_frags = npages;
1627 for (i = 0; i < npages; i++) {
1628 struct page *page;
1629
1630 page = alloc_pages(sk->sk_allocation, 0);
1631 if (!page) {
1632 err = -ENOBUFS;
1633 skb_shinfo(skb)->nr_frags = i;
1634 kfree_skb(skb);
1635 goto failure;
1636 }
1637
1638 __skb_fill_page_desc(skb, i,
1639 page, 0,
1640 (data_len >= PAGE_SIZE ?
1641 PAGE_SIZE :
1642 data_len));
1643 data_len -= PAGE_SIZE;
1644 }
1645
1646 /* Full success... */
1647 break;
1648 }
1649 err = -ENOBUFS;
1650 goto failure;
1651 }
1652 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1653 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1654 err = -EAGAIN;
1655 if (!timeo)
1656 goto failure;
1657 if (signal_pending(current))
1658 goto interrupted;
1659 timeo = sock_wait_for_wmem(sk, timeo);
1660 }
1661
1662 skb_set_owner_w(skb, sk);
1663 return skb;
1664
1665interrupted:
1666 err = sock_intr_errno(timeo);
1667failure:
1668 *errcode = err;
1669 return NULL;
1670}
1671EXPORT_SYMBOL(sock_alloc_send_pskb);
1672
1673struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1674 int noblock, int *errcode)
1675{
1676 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1677}
1678EXPORT_SYMBOL(sock_alloc_send_skb);
1679
1680static void __lock_sock(struct sock *sk)
1681 __releases(&sk->sk_lock.slock)
1682 __acquires(&sk->sk_lock.slock)
1683{
1684 DEFINE_WAIT(wait);
1685
1686 for (;;) {
1687 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1688 TASK_UNINTERRUPTIBLE);
1689 spin_unlock_bh(&sk->sk_lock.slock);
1690 schedule();
1691 spin_lock_bh(&sk->sk_lock.slock);
1692 if (!sock_owned_by_user(sk))
1693 break;
1694 }
1695 finish_wait(&sk->sk_lock.wq, &wait);
1696}
1697
1698static void __release_sock(struct sock *sk)
1699 __releases(&sk->sk_lock.slock)
1700 __acquires(&sk->sk_lock.slock)
1701{
1702 struct sk_buff *skb = sk->sk_backlog.head;
1703
1704 do {
1705 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1706 bh_unlock_sock(sk);
1707
1708 do {
1709 struct sk_buff *next = skb->next;
1710
1711 prefetch(next);
1712 WARN_ON_ONCE(skb_dst_is_noref(skb));
1713 skb->next = NULL;
1714 sk_backlog_rcv(sk, skb);
1715
1716 /*
1717 * We are in process context here with softirqs
1718 * disabled, use cond_resched_softirq() to preempt.
1719 * This is safe to do because we've taken the backlog
1720 * queue private:
1721 */
1722 cond_resched_softirq();
1723
1724 skb = next;
1725 } while (skb != NULL);
1726
1727 bh_lock_sock(sk);
1728 } while ((skb = sk->sk_backlog.head) != NULL);
1729
1730 /*
1731 * Doing the zeroing here guarantee we can not loop forever
1732 * while a wild producer attempts to flood us.
1733 */
1734 sk->sk_backlog.len = 0;
1735}
1736
1737/**
1738 * sk_wait_data - wait for data to arrive at sk_receive_queue
1739 * @sk: sock to wait on
1740 * @timeo: for how long
1741 *
1742 * Now socket state including sk->sk_err is changed only under lock,
1743 * hence we may omit checks after joining wait queue.
1744 * We check receive queue before schedule() only as optimization;
1745 * it is very likely that release_sock() added new data.
1746 */
1747int sk_wait_data(struct sock *sk, long *timeo)
1748{
1749 int rc;
1750 DEFINE_WAIT(wait);
1751
1752 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1753 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1754 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1755 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1756 finish_wait(sk_sleep(sk), &wait);
1757 return rc;
1758}
1759EXPORT_SYMBOL(sk_wait_data);
1760
1761/**
1762 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1763 * @sk: socket
1764 * @size: memory size to allocate
1765 * @kind: allocation type
1766 *
1767 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1768 * rmem allocation. This function assumes that protocols which have
1769 * memory_pressure use sk_wmem_queued as write buffer accounting.
1770 */
1771int __sk_mem_schedule(struct sock *sk, int size, int kind)
1772{
1773 struct proto *prot = sk->sk_prot;
1774 int amt = sk_mem_pages(size);
1775 long allocated;
1776 int parent_status = UNDER_LIMIT;
1777
1778 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1779
1780 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1781
1782 /* Under limit. */
1783 if (parent_status == UNDER_LIMIT &&
1784 allocated <= sk_prot_mem_limits(sk, 0)) {
1785 sk_leave_memory_pressure(sk);
1786 return 1;
1787 }
1788
1789 /* Under pressure. (we or our parents) */
1790 if ((parent_status > SOFT_LIMIT) ||
1791 allocated > sk_prot_mem_limits(sk, 1))
1792 sk_enter_memory_pressure(sk);
1793
1794 /* Over hard limit (we or our parents) */
1795 if ((parent_status == OVER_LIMIT) ||
1796 (allocated > sk_prot_mem_limits(sk, 2)))
1797 goto suppress_allocation;
1798
1799 /* guarantee minimum buffer size under pressure */
1800 if (kind == SK_MEM_RECV) {
1801 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1802 return 1;
1803
1804 } else { /* SK_MEM_SEND */
1805 if (sk->sk_type == SOCK_STREAM) {
1806 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1807 return 1;
1808 } else if (atomic_read(&sk->sk_wmem_alloc) <
1809 prot->sysctl_wmem[0])
1810 return 1;
1811 }
1812
1813 if (sk_has_memory_pressure(sk)) {
1814 int alloc;
1815
1816 if (!sk_under_memory_pressure(sk))
1817 return 1;
1818 alloc = sk_sockets_allocated_read_positive(sk);
1819 if (sk_prot_mem_limits(sk, 2) > alloc *
1820 sk_mem_pages(sk->sk_wmem_queued +
1821 atomic_read(&sk->sk_rmem_alloc) +
1822 sk->sk_forward_alloc))
1823 return 1;
1824 }
1825
1826suppress_allocation:
1827
1828 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1829 sk_stream_moderate_sndbuf(sk);
1830
1831 /* Fail only if socket is _under_ its sndbuf.
1832 * In this case we cannot block, so that we have to fail.
1833 */
1834 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1835 return 1;
1836 }
1837
1838 trace_sock_exceed_buf_limit(sk, prot, allocated);
1839
1840 /* Alas. Undo changes. */
1841 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1842
1843 sk_memory_allocated_sub(sk, amt);
1844
1845 return 0;
1846}
1847EXPORT_SYMBOL(__sk_mem_schedule);
1848
1849/**
1850 * __sk_reclaim - reclaim memory_allocated
1851 * @sk: socket
1852 */
1853void __sk_mem_reclaim(struct sock *sk)
1854{
1855 sk_memory_allocated_sub(sk,
1856 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1857 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1858
1859 if (sk_under_memory_pressure(sk) &&
1860 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1861 sk_leave_memory_pressure(sk);
1862}
1863EXPORT_SYMBOL(__sk_mem_reclaim);
1864
1865
1866/*
1867 * Set of default routines for initialising struct proto_ops when
1868 * the protocol does not support a particular function. In certain
1869 * cases where it makes no sense for a protocol to have a "do nothing"
1870 * function, some default processing is provided.
1871 */
1872
1873int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1874{
1875 return -EOPNOTSUPP;
1876}
1877EXPORT_SYMBOL(sock_no_bind);
1878
1879int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1880 int len, int flags)
1881{
1882 return -EOPNOTSUPP;
1883}
1884EXPORT_SYMBOL(sock_no_connect);
1885
1886int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1887{
1888 return -EOPNOTSUPP;
1889}
1890EXPORT_SYMBOL(sock_no_socketpair);
1891
1892int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1893{
1894 return -EOPNOTSUPP;
1895}
1896EXPORT_SYMBOL(sock_no_accept);
1897
1898int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1899 int *len, int peer)
1900{
1901 return -EOPNOTSUPP;
1902}
1903EXPORT_SYMBOL(sock_no_getname);
1904
1905unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1906{
1907 return 0;
1908}
1909EXPORT_SYMBOL(sock_no_poll);
1910
1911int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1912{
1913 return -EOPNOTSUPP;
1914}
1915EXPORT_SYMBOL(sock_no_ioctl);
1916
1917int sock_no_listen(struct socket *sock, int backlog)
1918{
1919 return -EOPNOTSUPP;
1920}
1921EXPORT_SYMBOL(sock_no_listen);
1922
1923int sock_no_shutdown(struct socket *sock, int how)
1924{
1925 return -EOPNOTSUPP;
1926}
1927EXPORT_SYMBOL(sock_no_shutdown);
1928
1929int sock_no_setsockopt(struct socket *sock, int level, int optname,
1930 char __user *optval, unsigned int optlen)
1931{
1932 return -EOPNOTSUPP;
1933}
1934EXPORT_SYMBOL(sock_no_setsockopt);
1935
1936int sock_no_getsockopt(struct socket *sock, int level, int optname,
1937 char __user *optval, int __user *optlen)
1938{
1939 return -EOPNOTSUPP;
1940}
1941EXPORT_SYMBOL(sock_no_getsockopt);
1942
1943int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1944 size_t len)
1945{
1946 return -EOPNOTSUPP;
1947}
1948EXPORT_SYMBOL(sock_no_sendmsg);
1949
1950int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1951 size_t len, int flags)
1952{
1953 return -EOPNOTSUPP;
1954}
1955EXPORT_SYMBOL(sock_no_recvmsg);
1956
1957int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1958{
1959 /* Mirror missing mmap method error code */
1960 return -ENODEV;
1961}
1962EXPORT_SYMBOL(sock_no_mmap);
1963
1964ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1965{
1966 ssize_t res;
1967 struct msghdr msg = {.msg_flags = flags};
1968 struct kvec iov;
1969 char *kaddr = kmap(page);
1970 iov.iov_base = kaddr + offset;
1971 iov.iov_len = size;
1972 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1973 kunmap(page);
1974 return res;
1975}
1976EXPORT_SYMBOL(sock_no_sendpage);
1977
1978/*
1979 * Default Socket Callbacks
1980 */
1981
1982static void sock_def_wakeup(struct sock *sk)
1983{
1984 struct socket_wq *wq;
1985
1986 rcu_read_lock();
1987 wq = rcu_dereference(sk->sk_wq);
1988 if (wq_has_sleeper(wq))
1989 wake_up_interruptible_all(&wq->wait);
1990 rcu_read_unlock();
1991}
1992
1993static void sock_def_error_report(struct sock *sk)
1994{
1995 struct socket_wq *wq;
1996
1997 rcu_read_lock();
1998 wq = rcu_dereference(sk->sk_wq);
1999 if (wq_has_sleeper(wq))
2000 wake_up_interruptible_poll(&wq->wait, POLLERR);
2001 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2002 rcu_read_unlock();
2003}
2004
2005static void sock_def_readable(struct sock *sk, int len)
2006{
2007 struct socket_wq *wq;
2008
2009 rcu_read_lock();
2010 wq = rcu_dereference(sk->sk_wq);
2011 if (wq_has_sleeper(wq))
2012 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2013 POLLRDNORM | POLLRDBAND);
2014 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2015 rcu_read_unlock();
2016}
2017
2018static void sock_def_write_space(struct sock *sk)
2019{
2020 struct socket_wq *wq;
2021
2022 rcu_read_lock();
2023
2024 /* Do not wake up a writer until he can make "significant"
2025 * progress. --DaveM
2026 */
2027 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2028 wq = rcu_dereference(sk->sk_wq);
2029 if (wq_has_sleeper(wq))
2030 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2031 POLLWRNORM | POLLWRBAND);
2032
2033 /* Should agree with poll, otherwise some programs break */
2034 if (sock_writeable(sk))
2035 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2036 }
2037
2038 rcu_read_unlock();
2039}
2040
2041static void sock_def_destruct(struct sock *sk)
2042{
2043 kfree(sk->sk_protinfo);
2044}
2045
2046void sk_send_sigurg(struct sock *sk)
2047{
2048 if (sk->sk_socket && sk->sk_socket->file)
2049 if (send_sigurg(&sk->sk_socket->file->f_owner))
2050 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2051}
2052EXPORT_SYMBOL(sk_send_sigurg);
2053
2054void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2055 unsigned long expires)
2056{
2057 if (!mod_timer(timer, expires))
2058 sock_hold(sk);
2059}
2060EXPORT_SYMBOL(sk_reset_timer);
2061
2062void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2063{
2064 if (timer_pending(timer) && del_timer(timer))
2065 __sock_put(sk);
2066}
2067EXPORT_SYMBOL(sk_stop_timer);
2068
2069void sock_init_data(struct socket *sock, struct sock *sk)
2070{
2071 skb_queue_head_init(&sk->sk_receive_queue);
2072 skb_queue_head_init(&sk->sk_write_queue);
2073 skb_queue_head_init(&sk->sk_error_queue);
2074#ifdef CONFIG_NET_DMA
2075 skb_queue_head_init(&sk->sk_async_wait_queue);
2076#endif
2077
2078 sk->sk_send_head = NULL;
2079
2080 init_timer(&sk->sk_timer);
2081
2082 sk->sk_allocation = GFP_KERNEL;
2083 sk->sk_rcvbuf = sysctl_rmem_default;
2084 sk->sk_sndbuf = sysctl_wmem_default;
2085 sk->sk_state = TCP_CLOSE;
2086 sk_set_socket(sk, sock);
2087
2088 sock_set_flag(sk, SOCK_ZAPPED);
2089
2090 if (sock) {
2091 sk->sk_type = sock->type;
2092 sk->sk_wq = sock->wq;
2093 sock->sk = sk;
2094 } else
2095 sk->sk_wq = NULL;
2096
2097 spin_lock_init(&sk->sk_dst_lock);
2098 rwlock_init(&sk->sk_callback_lock);
2099 lockdep_set_class_and_name(&sk->sk_callback_lock,
2100 af_callback_keys + sk->sk_family,
2101 af_family_clock_key_strings[sk->sk_family]);
2102
2103 sk->sk_state_change = sock_def_wakeup;
2104 sk->sk_data_ready = sock_def_readable;
2105 sk->sk_write_space = sock_def_write_space;
2106 sk->sk_error_report = sock_def_error_report;
2107 sk->sk_destruct = sock_def_destruct;
2108
2109 sk->sk_sndmsg_page = NULL;
2110 sk->sk_sndmsg_off = 0;
2111 sk->sk_peek_off = -1;
2112
2113 sk->sk_peer_pid = NULL;
2114 sk->sk_peer_cred = NULL;
2115 sk->sk_write_pending = 0;
2116 sk->sk_rcvlowat = 1;
2117 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2118 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2119
2120 sk->sk_stamp = ktime_set(-1L, 0);
2121
2122 /*
2123 * Before updating sk_refcnt, we must commit prior changes to memory
2124 * (Documentation/RCU/rculist_nulls.txt for details)
2125 */
2126 smp_wmb();
2127 atomic_set(&sk->sk_refcnt, 1);
2128 atomic_set(&sk->sk_drops, 0);
2129}
2130EXPORT_SYMBOL(sock_init_data);
2131
2132void lock_sock_nested(struct sock *sk, int subclass)
2133{
2134 might_sleep();
2135 spin_lock_bh(&sk->sk_lock.slock);
2136 if (sk->sk_lock.owned)
2137 __lock_sock(sk);
2138 sk->sk_lock.owned = 1;
2139 spin_unlock(&sk->sk_lock.slock);
2140 /*
2141 * The sk_lock has mutex_lock() semantics here:
2142 */
2143 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2144 local_bh_enable();
2145}
2146EXPORT_SYMBOL(lock_sock_nested);
2147
2148void release_sock(struct sock *sk)
2149{
2150 /*
2151 * The sk_lock has mutex_unlock() semantics:
2152 */
2153 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2154
2155 spin_lock_bh(&sk->sk_lock.slock);
2156 if (sk->sk_backlog.tail)
2157 __release_sock(sk);
2158 sk->sk_lock.owned = 0;
2159 if (waitqueue_active(&sk->sk_lock.wq))
2160 wake_up(&sk->sk_lock.wq);
2161 spin_unlock_bh(&sk->sk_lock.slock);
2162}
2163EXPORT_SYMBOL(release_sock);
2164
2165/**
2166 * lock_sock_fast - fast version of lock_sock
2167 * @sk: socket
2168 *
2169 * This version should be used for very small section, where process wont block
2170 * return false if fast path is taken
2171 * sk_lock.slock locked, owned = 0, BH disabled
2172 * return true if slow path is taken
2173 * sk_lock.slock unlocked, owned = 1, BH enabled
2174 */
2175bool lock_sock_fast(struct sock *sk)
2176{
2177 might_sleep();
2178 spin_lock_bh(&sk->sk_lock.slock);
2179
2180 if (!sk->sk_lock.owned)
2181 /*
2182 * Note : We must disable BH
2183 */
2184 return false;
2185
2186 __lock_sock(sk);
2187 sk->sk_lock.owned = 1;
2188 spin_unlock(&sk->sk_lock.slock);
2189 /*
2190 * The sk_lock has mutex_lock() semantics here:
2191 */
2192 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2193 local_bh_enable();
2194 return true;
2195}
2196EXPORT_SYMBOL(lock_sock_fast);
2197
2198int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2199{
2200 struct timeval tv;
2201 if (!sock_flag(sk, SOCK_TIMESTAMP))
2202 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2203 tv = ktime_to_timeval(sk->sk_stamp);
2204 if (tv.tv_sec == -1)
2205 return -ENOENT;
2206 if (tv.tv_sec == 0) {
2207 sk->sk_stamp = ktime_get_real();
2208 tv = ktime_to_timeval(sk->sk_stamp);
2209 }
2210 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2211}
2212EXPORT_SYMBOL(sock_get_timestamp);
2213
2214int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2215{
2216 struct timespec ts;
2217 if (!sock_flag(sk, SOCK_TIMESTAMP))
2218 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2219 ts = ktime_to_timespec(sk->sk_stamp);
2220 if (ts.tv_sec == -1)
2221 return -ENOENT;
2222 if (ts.tv_sec == 0) {
2223 sk->sk_stamp = ktime_get_real();
2224 ts = ktime_to_timespec(sk->sk_stamp);
2225 }
2226 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2227}
2228EXPORT_SYMBOL(sock_get_timestampns);
2229
2230void sock_enable_timestamp(struct sock *sk, int flag)
2231{
2232 if (!sock_flag(sk, flag)) {
2233 unsigned long previous_flags = sk->sk_flags;
2234
2235 sock_set_flag(sk, flag);
2236 /*
2237 * we just set one of the two flags which require net
2238 * time stamping, but time stamping might have been on
2239 * already because of the other one
2240 */
2241 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2242 net_enable_timestamp();
2243 }
2244}
2245
2246/*
2247 * Get a socket option on an socket.
2248 *
2249 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2250 * asynchronous errors should be reported by getsockopt. We assume
2251 * this means if you specify SO_ERROR (otherwise whats the point of it).
2252 */
2253int sock_common_getsockopt(struct socket *sock, int level, int optname,
2254 char __user *optval, int __user *optlen)
2255{
2256 struct sock *sk = sock->sk;
2257
2258 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2259}
2260EXPORT_SYMBOL(sock_common_getsockopt);
2261
2262#ifdef CONFIG_COMPAT
2263int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2264 char __user *optval, int __user *optlen)
2265{
2266 struct sock *sk = sock->sk;
2267
2268 if (sk->sk_prot->compat_getsockopt != NULL)
2269 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2270 optval, optlen);
2271 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2272}
2273EXPORT_SYMBOL(compat_sock_common_getsockopt);
2274#endif
2275
2276int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2277 struct msghdr *msg, size_t size, int flags)
2278{
2279 struct sock *sk = sock->sk;
2280 int addr_len = 0;
2281 int err;
2282
2283 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2284 flags & ~MSG_DONTWAIT, &addr_len);
2285 if (err >= 0)
2286 msg->msg_namelen = addr_len;
2287 return err;
2288}
2289EXPORT_SYMBOL(sock_common_recvmsg);
2290
2291/*
2292 * Set socket options on an inet socket.
2293 */
2294int sock_common_setsockopt(struct socket *sock, int level, int optname,
2295 char __user *optval, unsigned int optlen)
2296{
2297 struct sock *sk = sock->sk;
2298
2299 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2300}
2301EXPORT_SYMBOL(sock_common_setsockopt);
2302
2303#ifdef CONFIG_COMPAT
2304int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2305 char __user *optval, unsigned int optlen)
2306{
2307 struct sock *sk = sock->sk;
2308
2309 if (sk->sk_prot->compat_setsockopt != NULL)
2310 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2311 optval, optlen);
2312 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2313}
2314EXPORT_SYMBOL(compat_sock_common_setsockopt);
2315#endif
2316
2317void sk_common_release(struct sock *sk)
2318{
2319 if (sk->sk_prot->destroy)
2320 sk->sk_prot->destroy(sk);
2321
2322 /*
2323 * Observation: when sock_common_release is called, processes have
2324 * no access to socket. But net still has.
2325 * Step one, detach it from networking:
2326 *
2327 * A. Remove from hash tables.
2328 */
2329
2330 sk->sk_prot->unhash(sk);
2331
2332 /*
2333 * In this point socket cannot receive new packets, but it is possible
2334 * that some packets are in flight because some CPU runs receiver and
2335 * did hash table lookup before we unhashed socket. They will achieve
2336 * receive queue and will be purged by socket destructor.
2337 *
2338 * Also we still have packets pending on receive queue and probably,
2339 * our own packets waiting in device queues. sock_destroy will drain
2340 * receive queue, but transmitted packets will delay socket destruction
2341 * until the last reference will be released.
2342 */
2343
2344 sock_orphan(sk);
2345
2346 xfrm_sk_free_policy(sk);
2347
2348 sk_refcnt_debug_release(sk);
2349 sock_put(sk);
2350}
2351EXPORT_SYMBOL(sk_common_release);
2352
2353#ifdef CONFIG_PROC_FS
2354#define PROTO_INUSE_NR 64 /* should be enough for the first time */
2355struct prot_inuse {
2356 int val[PROTO_INUSE_NR];
2357};
2358
2359static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2360
2361#ifdef CONFIG_NET_NS
2362void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2363{
2364 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2365}
2366EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2367
2368int sock_prot_inuse_get(struct net *net, struct proto *prot)
2369{
2370 int cpu, idx = prot->inuse_idx;
2371 int res = 0;
2372
2373 for_each_possible_cpu(cpu)
2374 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2375
2376 return res >= 0 ? res : 0;
2377}
2378EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2379
2380static int __net_init sock_inuse_init_net(struct net *net)
2381{
2382 net->core.inuse = alloc_percpu(struct prot_inuse);
2383 return net->core.inuse ? 0 : -ENOMEM;
2384}
2385
2386static void __net_exit sock_inuse_exit_net(struct net *net)
2387{
2388 free_percpu(net->core.inuse);
2389}
2390
2391static struct pernet_operations net_inuse_ops = {
2392 .init = sock_inuse_init_net,
2393 .exit = sock_inuse_exit_net,
2394};
2395
2396static __init int net_inuse_init(void)
2397{
2398 if (register_pernet_subsys(&net_inuse_ops))
2399 panic("Cannot initialize net inuse counters");
2400
2401 return 0;
2402}
2403
2404core_initcall(net_inuse_init);
2405#else
2406static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2407
2408void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2409{
2410 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2411}
2412EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2413
2414int sock_prot_inuse_get(struct net *net, struct proto *prot)
2415{
2416 int cpu, idx = prot->inuse_idx;
2417 int res = 0;
2418
2419 for_each_possible_cpu(cpu)
2420 res += per_cpu(prot_inuse, cpu).val[idx];
2421
2422 return res >= 0 ? res : 0;
2423}
2424EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2425#endif
2426
2427static void assign_proto_idx(struct proto *prot)
2428{
2429 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2430
2431 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2432 pr_err("PROTO_INUSE_NR exhausted\n");
2433 return;
2434 }
2435
2436 set_bit(prot->inuse_idx, proto_inuse_idx);
2437}
2438
2439static void release_proto_idx(struct proto *prot)
2440{
2441 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2442 clear_bit(prot->inuse_idx, proto_inuse_idx);
2443}
2444#else
2445static inline void assign_proto_idx(struct proto *prot)
2446{
2447}
2448
2449static inline void release_proto_idx(struct proto *prot)
2450{
2451}
2452#endif
2453
2454int proto_register(struct proto *prot, int alloc_slab)
2455{
2456 if (alloc_slab) {
2457 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2458 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2459 NULL);
2460
2461 if (prot->slab == NULL) {
2462 pr_crit("%s: Can't create sock SLAB cache!\n",
2463 prot->name);
2464 goto out;
2465 }
2466
2467 if (prot->rsk_prot != NULL) {
2468 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2469 if (prot->rsk_prot->slab_name == NULL)
2470 goto out_free_sock_slab;
2471
2472 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2473 prot->rsk_prot->obj_size, 0,
2474 SLAB_HWCACHE_ALIGN, NULL);
2475
2476 if (prot->rsk_prot->slab == NULL) {
2477 pr_crit("%s: Can't create request sock SLAB cache!\n",
2478 prot->name);
2479 goto out_free_request_sock_slab_name;
2480 }
2481 }
2482
2483 if (prot->twsk_prot != NULL) {
2484 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2485
2486 if (prot->twsk_prot->twsk_slab_name == NULL)
2487 goto out_free_request_sock_slab;
2488
2489 prot->twsk_prot->twsk_slab =
2490 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2491 prot->twsk_prot->twsk_obj_size,
2492 0,
2493 SLAB_HWCACHE_ALIGN |
2494 prot->slab_flags,
2495 NULL);
2496 if (prot->twsk_prot->twsk_slab == NULL)
2497 goto out_free_timewait_sock_slab_name;
2498 }
2499 }
2500
2501 mutex_lock(&proto_list_mutex);
2502 list_add(&prot->node, &proto_list);
2503 assign_proto_idx(prot);
2504 mutex_unlock(&proto_list_mutex);
2505 return 0;
2506
2507out_free_timewait_sock_slab_name:
2508 kfree(prot->twsk_prot->twsk_slab_name);
2509out_free_request_sock_slab:
2510 if (prot->rsk_prot && prot->rsk_prot->slab) {
2511 kmem_cache_destroy(prot->rsk_prot->slab);
2512 prot->rsk_prot->slab = NULL;
2513 }
2514out_free_request_sock_slab_name:
2515 if (prot->rsk_prot)
2516 kfree(prot->rsk_prot->slab_name);
2517out_free_sock_slab:
2518 kmem_cache_destroy(prot->slab);
2519 prot->slab = NULL;
2520out:
2521 return -ENOBUFS;
2522}
2523EXPORT_SYMBOL(proto_register);
2524
2525void proto_unregister(struct proto *prot)
2526{
2527 mutex_lock(&proto_list_mutex);
2528 release_proto_idx(prot);
2529 list_del(&prot->node);
2530 mutex_unlock(&proto_list_mutex);
2531
2532 if (prot->slab != NULL) {
2533 kmem_cache_destroy(prot->slab);
2534 prot->slab = NULL;
2535 }
2536
2537 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2538 kmem_cache_destroy(prot->rsk_prot->slab);
2539 kfree(prot->rsk_prot->slab_name);
2540 prot->rsk_prot->slab = NULL;
2541 }
2542
2543 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2544 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2545 kfree(prot->twsk_prot->twsk_slab_name);
2546 prot->twsk_prot->twsk_slab = NULL;
2547 }
2548}
2549EXPORT_SYMBOL(proto_unregister);
2550
2551#ifdef CONFIG_PROC_FS
2552static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2553 __acquires(proto_list_mutex)
2554{
2555 mutex_lock(&proto_list_mutex);
2556 return seq_list_start_head(&proto_list, *pos);
2557}
2558
2559static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2560{
2561 return seq_list_next(v, &proto_list, pos);
2562}
2563
2564static void proto_seq_stop(struct seq_file *seq, void *v)
2565 __releases(proto_list_mutex)
2566{
2567 mutex_unlock(&proto_list_mutex);
2568}
2569
2570static char proto_method_implemented(const void *method)
2571{
2572 return method == NULL ? 'n' : 'y';
2573}
2574static long sock_prot_memory_allocated(struct proto *proto)
2575{
2576 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2577}
2578
2579static char *sock_prot_memory_pressure(struct proto *proto)
2580{
2581 return proto->memory_pressure != NULL ?
2582 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2583}
2584
2585static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2586{
2587
2588 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2589 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2590 proto->name,
2591 proto->obj_size,
2592 sock_prot_inuse_get(seq_file_net(seq), proto),
2593 sock_prot_memory_allocated(proto),
2594 sock_prot_memory_pressure(proto),
2595 proto->max_header,
2596 proto->slab == NULL ? "no" : "yes",
2597 module_name(proto->owner),
2598 proto_method_implemented(proto->close),
2599 proto_method_implemented(proto->connect),
2600 proto_method_implemented(proto->disconnect),
2601 proto_method_implemented(proto->accept),
2602 proto_method_implemented(proto->ioctl),
2603 proto_method_implemented(proto->init),
2604 proto_method_implemented(proto->destroy),
2605 proto_method_implemented(proto->shutdown),
2606 proto_method_implemented(proto->setsockopt),
2607 proto_method_implemented(proto->getsockopt),
2608 proto_method_implemented(proto->sendmsg),
2609 proto_method_implemented(proto->recvmsg),
2610 proto_method_implemented(proto->sendpage),
2611 proto_method_implemented(proto->bind),
2612 proto_method_implemented(proto->backlog_rcv),
2613 proto_method_implemented(proto->hash),
2614 proto_method_implemented(proto->unhash),
2615 proto_method_implemented(proto->get_port),
2616 proto_method_implemented(proto->enter_memory_pressure));
2617}
2618
2619static int proto_seq_show(struct seq_file *seq, void *v)
2620{
2621 if (v == &proto_list)
2622 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2623 "protocol",
2624 "size",
2625 "sockets",
2626 "memory",
2627 "press",
2628 "maxhdr",
2629 "slab",
2630 "module",
2631 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2632 else
2633 proto_seq_printf(seq, list_entry(v, struct proto, node));
2634 return 0;
2635}
2636
2637static const struct seq_operations proto_seq_ops = {
2638 .start = proto_seq_start,
2639 .next = proto_seq_next,
2640 .stop = proto_seq_stop,
2641 .show = proto_seq_show,
2642};
2643
2644static int proto_seq_open(struct inode *inode, struct file *file)
2645{
2646 return seq_open_net(inode, file, &proto_seq_ops,
2647 sizeof(struct seq_net_private));
2648}
2649
2650static const struct file_operations proto_seq_fops = {
2651 .owner = THIS_MODULE,
2652 .open = proto_seq_open,
2653 .read = seq_read,
2654 .llseek = seq_lseek,
2655 .release = seq_release_net,
2656};
2657
2658static __net_init int proto_init_net(struct net *net)
2659{
2660 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2661 return -ENOMEM;
2662
2663 return 0;
2664}
2665
2666static __net_exit void proto_exit_net(struct net *net)
2667{
2668 proc_net_remove(net, "protocols");
2669}
2670
2671
2672static __net_initdata struct pernet_operations proto_net_ops = {
2673 .init = proto_init_net,
2674 .exit = proto_exit_net,
2675};
2676
2677static int __init proto_init(void)
2678{
2679 return register_pernet_subsys(&proto_net_ops);
2680}
2681
2682subsys_initcall(proto_init);
2683
2684#endif /* PROC_FS */