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