?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net/compat: Add missing sock updates for SCM_RIGHTS
[linux/fpc-iii.git] / net / core / sock.c
blob919f1a1739e90b95c091b8c6acdae9c983b25702
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88 #include <asm/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
94 #include <linux/in.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/init.h>
111 #include <linux/highmem.h>
112 #include <linux/user_namespace.h>
113 #include <linux/static_key.h>
114 #include <linux/memcontrol.h>
115 #include <linux/prefetch.h>
116
117 #include <linux/uaccess.h>
118
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/net_namespace.h>
123 #include <net/request_sock.h>
124 #include <net/sock.h>
125 #include <linux/net_tstamp.h>
126 #include <net/xfrm.h>
127 #include <linux/ipsec.h>
128 #include <net/cls_cgroup.h>
129 #include <net/netprio_cgroup.h>
130 #include <linux/sock_diag.h>
131
132 #include <linux/filter.h>
133 #include <net/sock_reuseport.h>
134 #include <net/bpf_sk_storage.h>
135
136 #include <trace/events/sock.h>
137
138 #include <net/tcp.h>
139 #include <net/busy_poll.h>
140
141 static DEFINE_MUTEX(proto_list_mutex);
142 static LIST_HEAD(proto_list);
143
144 static void sock_inuse_add(struct net *net, int val);
145
146 /**
147 * sk_ns_capable - General socket capability test
148 * @sk: Socket to use a capability on or through
149 * @user_ns: The user namespace of the capability to use
150 * @cap: The capability to use
151 *
152 * Test to see if the opener of the socket had when the socket was
153 * created and the current process has the capability @cap in the user
154 * namespace @user_ns.
155 */
156 bool sk_ns_capable(const struct sock *sk,
157 struct user_namespace *user_ns, int cap)
158 {
159 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
160 ns_capable(user_ns, cap);
161 }
162 EXPORT_SYMBOL(sk_ns_capable);
163
164 /**
165 * sk_capable - Socket global capability test
166 * @sk: Socket to use a capability on or through
167 * @cap: The global capability to use
168 *
169 * Test to see if the opener of the socket had when the socket was
170 * created and the current process has the capability @cap in all user
171 * namespaces.
172 */
173 bool sk_capable(const struct sock *sk, int cap)
174 {
175 return sk_ns_capable(sk, &init_user_ns, cap);
176 }
177 EXPORT_SYMBOL(sk_capable);
178
179 /**
180 * sk_net_capable - Network namespace socket capability test
181 * @sk: Socket to use a capability on or through
182 * @cap: The capability to use
183 *
184 * Test to see if the opener of the socket had when the socket was created
185 * and the current process has the capability @cap over the network namespace
186 * the socket is a member of.
187 */
188 bool sk_net_capable(const struct sock *sk, int cap)
189 {
190 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
191 }
192 EXPORT_SYMBOL(sk_net_capable);
193
194 /*
195 * Each address family might have different locking rules, so we have
196 * one slock key per address family and separate keys for internal and
197 * userspace sockets.
198 */
199 static struct lock_class_key af_family_keys[AF_MAX];
200 static struct lock_class_key af_family_kern_keys[AF_MAX];
201 static struct lock_class_key af_family_slock_keys[AF_MAX];
202 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
203
204 /*
205 * Make lock validator output more readable. (we pre-construct these
206 * strings build-time, so that runtime initialization of socket
207 * locks is fast):
208 */
209
210 #define _sock_locks(x) \
211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
220 x "27" , x "28" , x "AF_CAN" , \
221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
226 x "AF_MAX"
227
228 static const char *const af_family_key_strings[AF_MAX+1] = {
229 _sock_locks("sk_lock-")
230 };
231 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
232 _sock_locks("slock-")
233 };
234 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 _sock_locks("clock-")
236 };
237
238 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
239 _sock_locks("k-sk_lock-")
240 };
241 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
242 _sock_locks("k-slock-")
243 };
244 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-clock-")
246 };
247 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
248 _sock_locks("rlock-")
249 };
250 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
251 _sock_locks("wlock-")
252 };
253 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
254 _sock_locks("elock-")
255 };
256
257 /*
258 * sk_callback_lock and sk queues locking rules are per-address-family,
259 * so split the lock classes by using a per-AF key:
260 */
261 static struct lock_class_key af_callback_keys[AF_MAX];
262 static struct lock_class_key af_rlock_keys[AF_MAX];
263 static struct lock_class_key af_wlock_keys[AF_MAX];
264 static struct lock_class_key af_elock_keys[AF_MAX];
265 static struct lock_class_key af_kern_callback_keys[AF_MAX];
266
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269 EXPORT_SYMBOL(sysctl_wmem_max);
270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271 EXPORT_SYMBOL(sysctl_rmem_max);
272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
274
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277 EXPORT_SYMBOL(sysctl_optmem_max);
278
279 int sysctl_tstamp_allow_data __read_mostly = 1;
280
281 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
282 EXPORT_SYMBOL_GPL(memalloc_socks_key);
283
284 /**
285 * sk_set_memalloc - sets %SOCK_MEMALLOC
286 * @sk: socket to set it on
287 *
288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
289 * It's the responsibility of the admin to adjust min_free_kbytes
290 * to meet the requirements
291 */
292 void sk_set_memalloc(struct sock *sk)
293 {
294 sock_set_flag(sk, SOCK_MEMALLOC);
295 sk->sk_allocation |= __GFP_MEMALLOC;
296 static_branch_inc(&memalloc_socks_key);
297 }
298 EXPORT_SYMBOL_GPL(sk_set_memalloc);
299
300 void sk_clear_memalloc(struct sock *sk)
301 {
302 sock_reset_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation &= ~__GFP_MEMALLOC;
304 static_branch_dec(&memalloc_socks_key);
305
306 /*
307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
308 * progress of swapping. SOCK_MEMALLOC may be cleared while
309 * it has rmem allocations due to the last swapfile being deactivated
310 * but there is a risk that the socket is unusable due to exceeding
311 * the rmem limits. Reclaim the reserves and obey rmem limits again.
312 */
313 sk_mem_reclaim(sk);
314 }
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
316
317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
318 {
319 int ret;
320 unsigned int noreclaim_flag;
321
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
324
325 noreclaim_flag = memalloc_noreclaim_save();
326 ret = sk->sk_backlog_rcv(sk, skb);
327 memalloc_noreclaim_restore(noreclaim_flag);
328
329 return ret;
330 }
331 EXPORT_SYMBOL(__sk_backlog_rcv);
332
333 static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
334 {
335 struct __kernel_sock_timeval tv;
336 int size;
337
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
339 tv.tv_sec = 0;
340 tv.tv_usec = 0;
341 } else {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344 }
345
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
349 return sizeof(tv32);
350 }
351
352 if (old_timeval) {
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 size = sizeof(old_tv);
358 } else {
359 *(struct __kernel_sock_timeval *)optval = tv;
360 size = sizeof(tv);
361 }
362
363 return size;
364 }
365
366 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
367 {
368 struct __kernel_sock_timeval tv;
369
370 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
371 struct old_timeval32 tv32;
372
373 if (optlen < sizeof(tv32))
374 return -EINVAL;
375
376 if (copy_from_user(&tv32, optval, sizeof(tv32)))
377 return -EFAULT;
378 tv.tv_sec = tv32.tv_sec;
379 tv.tv_usec = tv32.tv_usec;
380 } else if (old_timeval) {
381 struct __kernel_old_timeval old_tv;
382
383 if (optlen < sizeof(old_tv))
384 return -EINVAL;
385 if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
386 return -EFAULT;
387 tv.tv_sec = old_tv.tv_sec;
388 tv.tv_usec = old_tv.tv_usec;
389 } else {
390 if (optlen < sizeof(tv))
391 return -EINVAL;
392 if (copy_from_user(&tv, optval, sizeof(tv)))
393 return -EFAULT;
394 }
395 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
396 return -EDOM;
397
398 if (tv.tv_sec < 0) {
399 static int warned __read_mostly;
400
401 *timeo_p = 0;
402 if (warned < 10 && net_ratelimit()) {
403 warned++;
404 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
405 __func__, current->comm, task_pid_nr(current));
406 }
407 return 0;
408 }
409 *timeo_p = MAX_SCHEDULE_TIMEOUT;
410 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 return 0;
412 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
413 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
414 return 0;
415 }
416
417 static void sock_warn_obsolete_bsdism(const char *name)
418 {
419 static int warned;
420 static char warncomm[TASK_COMM_LEN];
421 if (strcmp(warncomm, current->comm) && warned < 5) {
422 strcpy(warncomm, current->comm);
423 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
424 warncomm, name);
425 warned++;
426 }
427 }
428
429 static bool sock_needs_netstamp(const struct sock *sk)
430 {
431 switch (sk->sk_family) {
432 case AF_UNSPEC:
433 case AF_UNIX:
434 return false;
435 default:
436 return true;
437 }
438 }
439
440 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
441 {
442 if (sk->sk_flags & flags) {
443 sk->sk_flags &= ~flags;
444 if (sock_needs_netstamp(sk) &&
445 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
446 net_disable_timestamp();
447 }
448 }
449
450
451 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
452 {
453 unsigned long flags;
454 struct sk_buff_head *list = &sk->sk_receive_queue;
455
456 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
457 atomic_inc(&sk->sk_drops);
458 trace_sock_rcvqueue_full(sk, skb);
459 return -ENOMEM;
460 }
461
462 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
463 atomic_inc(&sk->sk_drops);
464 return -ENOBUFS;
465 }
466
467 skb->dev = NULL;
468 skb_set_owner_r(skb, sk);
469
470 /* we escape from rcu protected region, make sure we dont leak
471 * a norefcounted dst
472 */
473 skb_dst_force(skb);
474
475 spin_lock_irqsave(&list->lock, flags);
476 sock_skb_set_dropcount(sk, skb);
477 __skb_queue_tail(list, skb);
478 spin_unlock_irqrestore(&list->lock, flags);
479
480 if (!sock_flag(sk, SOCK_DEAD))
481 sk->sk_data_ready(sk);
482 return 0;
483 }
484 EXPORT_SYMBOL(__sock_queue_rcv_skb);
485
486 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
487 {
488 int err;
489
490 err = sk_filter(sk, skb);
491 if (err)
492 return err;
493
494 return __sock_queue_rcv_skb(sk, skb);
495 }
496 EXPORT_SYMBOL(sock_queue_rcv_skb);
497
498 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
499 const int nested, unsigned int trim_cap, bool refcounted)
500 {
501 int rc = NET_RX_SUCCESS;
502
503 if (sk_filter_trim_cap(sk, skb, trim_cap))
504 goto discard_and_relse;
505
506 skb->dev = NULL;
507
508 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
509 atomic_inc(&sk->sk_drops);
510 goto discard_and_relse;
511 }
512 if (nested)
513 bh_lock_sock_nested(sk);
514 else
515 bh_lock_sock(sk);
516 if (!sock_owned_by_user(sk)) {
517 /*
518 * trylock + unlock semantics:
519 */
520 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
521
522 rc = sk_backlog_rcv(sk, skb);
523
524 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
525 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
526 bh_unlock_sock(sk);
527 atomic_inc(&sk->sk_drops);
528 goto discard_and_relse;
529 }
530
531 bh_unlock_sock(sk);
532 out:
533 if (refcounted)
534 sock_put(sk);
535 return rc;
536 discard_and_relse:
537 kfree_skb(skb);
538 goto out;
539 }
540 EXPORT_SYMBOL(__sk_receive_skb);
541
542 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
543 {
544 struct dst_entry *dst = __sk_dst_get(sk);
545
546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
547 sk_tx_queue_clear(sk);
548 sk->sk_dst_pending_confirm = 0;
549 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
550 dst_release(dst);
551 return NULL;
552 }
553
554 return dst;
555 }
556 EXPORT_SYMBOL(__sk_dst_check);
557
558 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
559 {
560 struct dst_entry *dst = sk_dst_get(sk);
561
562 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
563 sk_dst_reset(sk);
564 dst_release(dst);
565 return NULL;
566 }
567
568 return dst;
569 }
570 EXPORT_SYMBOL(sk_dst_check);
571
572 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
573 {
574 int ret = -ENOPROTOOPT;
575 #ifdef CONFIG_NETDEVICES
576 struct net *net = sock_net(sk);
577
578 /* Sorry... */
579 ret = -EPERM;
580 if (!ns_capable(net->user_ns, CAP_NET_RAW))
581 goto out;
582
583 ret = -EINVAL;
584 if (ifindex < 0)
585 goto out;
586
587 sk->sk_bound_dev_if = ifindex;
588 if (sk->sk_prot->rehash)
589 sk->sk_prot->rehash(sk);
590 sk_dst_reset(sk);
591
592 ret = 0;
593
594 out:
595 #endif
596
597 return ret;
598 }
599
600 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
601 int optlen)
602 {
603 int ret = -ENOPROTOOPT;
604 #ifdef CONFIG_NETDEVICES
605 struct net *net = sock_net(sk);
606 char devname[IFNAMSIZ];
607 int index;
608
609 ret = -EINVAL;
610 if (optlen < 0)
611 goto out;
612
613 /* Bind this socket to a particular device like "eth0",
614 * as specified in the passed interface name. If the
615 * name is "" or the option length is zero the socket
616 * is not bound.
617 */
618 if (optlen > IFNAMSIZ - 1)
619 optlen = IFNAMSIZ - 1;
620 memset(devname, 0, sizeof(devname));
621
622 ret = -EFAULT;
623 if (copy_from_user(devname, optval, optlen))
624 goto out;
625
626 index = 0;
627 if (devname[0] != '\0') {
628 struct net_device *dev;
629
630 rcu_read_lock();
631 dev = dev_get_by_name_rcu(net, devname);
632 if (dev)
633 index = dev->ifindex;
634 rcu_read_unlock();
635 ret = -ENODEV;
636 if (!dev)
637 goto out;
638 }
639
640 lock_sock(sk);
641 ret = sock_setbindtodevice_locked(sk, index);
642 release_sock(sk);
643
644 out:
645 #endif
646
647 return ret;
648 }
649
650 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
651 int __user *optlen, int len)
652 {
653 int ret = -ENOPROTOOPT;
654 #ifdef CONFIG_NETDEVICES
655 struct net *net = sock_net(sk);
656 char devname[IFNAMSIZ];
657
658 if (sk->sk_bound_dev_if == 0) {
659 len = 0;
660 goto zero;
661 }
662
663 ret = -EINVAL;
664 if (len < IFNAMSIZ)
665 goto out;
666
667 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
668 if (ret)
669 goto out;
670
671 len = strlen(devname) + 1;
672
673 ret = -EFAULT;
674 if (copy_to_user(optval, devname, len))
675 goto out;
676
677 zero:
678 ret = -EFAULT;
679 if (put_user(len, optlen))
680 goto out;
681
682 ret = 0;
683
684 out:
685 #endif
686
687 return ret;
688 }
689
690 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
691 {
692 if (valbool)
693 sock_set_flag(sk, bit);
694 else
695 sock_reset_flag(sk, bit);
696 }
697
698 bool sk_mc_loop(struct sock *sk)
699 {
700 if (dev_recursion_level())
701 return false;
702 if (!sk)
703 return true;
704 switch (sk->sk_family) {
705 case AF_INET:
706 return inet_sk(sk)->mc_loop;
707 #if IS_ENABLED(CONFIG_IPV6)
708 case AF_INET6:
709 return inet6_sk(sk)->mc_loop;
710 #endif
711 }
712 WARN_ON_ONCE(1);
713 return true;
714 }
715 EXPORT_SYMBOL(sk_mc_loop);
716
717 /*
718 * This is meant for all protocols to use and covers goings on
719 * at the socket level. Everything here is generic.
720 */
721
722 int sock_setsockopt(struct socket *sock, int level, int optname,
723 char __user *optval, unsigned int optlen)
724 {
725 struct sock_txtime sk_txtime;
726 struct sock *sk = sock->sk;
727 int val;
728 int valbool;
729 struct linger ling;
730 int ret = 0;
731
732 /*
733 * Options without arguments
734 */
735
736 if (optname == SO_BINDTODEVICE)
737 return sock_setbindtodevice(sk, optval, optlen);
738
739 if (optlen < sizeof(int))
740 return -EINVAL;
741
742 if (get_user(val, (int __user *)optval))
743 return -EFAULT;
744
745 valbool = val ? 1 : 0;
746
747 lock_sock(sk);
748
749 switch (optname) {
750 case SO_DEBUG:
751 if (val && !capable(CAP_NET_ADMIN))
752 ret = -EACCES;
753 else
754 sock_valbool_flag(sk, SOCK_DBG, valbool);
755 break;
756 case SO_REUSEADDR:
757 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
758 break;
759 case SO_REUSEPORT:
760 sk->sk_reuseport = valbool;
761 break;
762 case SO_TYPE:
763 case SO_PROTOCOL:
764 case SO_DOMAIN:
765 case SO_ERROR:
766 ret = -ENOPROTOOPT;
767 break;
768 case SO_DONTROUTE:
769 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
770 sk_dst_reset(sk);
771 break;
772 case SO_BROADCAST:
773 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
774 break;
775 case SO_SNDBUF:
776 /* Don't error on this BSD doesn't and if you think
777 * about it this is right. Otherwise apps have to
778 * play 'guess the biggest size' games. RCVBUF/SNDBUF
779 * are treated in BSD as hints
780 */
781 val = min_t(u32, val, sysctl_wmem_max);
782 set_sndbuf:
783 /* Ensure val * 2 fits into an int, to prevent max_t()
784 * from treating it as a negative value.
785 */
786 val = min_t(int, val, INT_MAX / 2);
787 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
788 WRITE_ONCE(sk->sk_sndbuf,
789 max_t(int, val * 2, SOCK_MIN_SNDBUF));
790 /* Wake up sending tasks if we upped the value. */
791 sk->sk_write_space(sk);
792 break;
793
794 case SO_SNDBUFFORCE:
795 if (!capable(CAP_NET_ADMIN)) {
796 ret = -EPERM;
797 break;
798 }
799
800 /* No negative values (to prevent underflow, as val will be
801 * multiplied by 2).
802 */
803 if (val < 0)
804 val = 0;
805 goto set_sndbuf;
806
807 case SO_RCVBUF:
808 /* Don't error on this BSD doesn't and if you think
809 * about it this is right. Otherwise apps have to
810 * play 'guess the biggest size' games. RCVBUF/SNDBUF
811 * are treated in BSD as hints
812 */
813 val = min_t(u32, val, sysctl_rmem_max);
814 set_rcvbuf:
815 /* Ensure val * 2 fits into an int, to prevent max_t()
816 * from treating it as a negative value.
817 */
818 val = min_t(int, val, INT_MAX / 2);
819 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
820 /*
821 * We double it on the way in to account for
822 * "struct sk_buff" etc. overhead. Applications
823 * assume that the SO_RCVBUF setting they make will
824 * allow that much actual data to be received on that
825 * socket.
826 *
827 * Applications are unaware that "struct sk_buff" and
828 * other overheads allocate from the receive buffer
829 * during socket buffer allocation.
830 *
831 * And after considering the possible alternatives,
832 * returning the value we actually used in getsockopt
833 * is the most desirable behavior.
834 */
835 WRITE_ONCE(sk->sk_rcvbuf,
836 max_t(int, val * 2, SOCK_MIN_RCVBUF));
837 break;
838
839 case SO_RCVBUFFORCE:
840 if (!capable(CAP_NET_ADMIN)) {
841 ret = -EPERM;
842 break;
843 }
844
845 /* No negative values (to prevent underflow, as val will be
846 * multiplied by 2).
847 */
848 if (val < 0)
849 val = 0;
850 goto set_rcvbuf;
851
852 case SO_KEEPALIVE:
853 if (sk->sk_prot->keepalive)
854 sk->sk_prot->keepalive(sk, valbool);
855 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
856 break;
857
858 case SO_OOBINLINE:
859 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
860 break;
861
862 case SO_NO_CHECK:
863 sk->sk_no_check_tx = valbool;
864 break;
865
866 case SO_PRIORITY:
867 if ((val >= 0 && val <= 6) ||
868 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
869 sk->sk_priority = val;
870 else
871 ret = -EPERM;
872 break;
873
874 case SO_LINGER:
875 if (optlen < sizeof(ling)) {
876 ret = -EINVAL; /* 1003.1g */
877 break;
878 }
879 if (copy_from_user(&ling, optval, sizeof(ling))) {
880 ret = -EFAULT;
881 break;
882 }
883 if (!ling.l_onoff)
884 sock_reset_flag(sk, SOCK_LINGER);
885 else {
886 #if (BITS_PER_LONG == 32)
887 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
888 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
889 else
890 #endif
891 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
892 sock_set_flag(sk, SOCK_LINGER);
893 }
894 break;
895
896 case SO_BSDCOMPAT:
897 sock_warn_obsolete_bsdism("setsockopt");
898 break;
899
900 case SO_PASSCRED:
901 if (valbool)
902 set_bit(SOCK_PASSCRED, &sock->flags);
903 else
904 clear_bit(SOCK_PASSCRED, &sock->flags);
905 break;
906
907 case SO_TIMESTAMP_OLD:
908 case SO_TIMESTAMP_NEW:
909 case SO_TIMESTAMPNS_OLD:
910 case SO_TIMESTAMPNS_NEW:
911 if (valbool) {
912 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
913 sock_set_flag(sk, SOCK_TSTAMP_NEW);
914 else
915 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
916
917 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
918 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
919 else
920 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
921 sock_set_flag(sk, SOCK_RCVTSTAMP);
922 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
923 } else {
924 sock_reset_flag(sk, SOCK_RCVTSTAMP);
925 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
926 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
927 }
928 break;
929
930 case SO_TIMESTAMPING_NEW:
931 sock_set_flag(sk, SOCK_TSTAMP_NEW);
932 /* fall through */
933 case SO_TIMESTAMPING_OLD:
934 if (val & ~SOF_TIMESTAMPING_MASK) {
935 ret = -EINVAL;
936 break;
937 }
938
939 if (val & SOF_TIMESTAMPING_OPT_ID &&
940 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
941 if (sk->sk_protocol == IPPROTO_TCP &&
942 sk->sk_type == SOCK_STREAM) {
943 if ((1 << sk->sk_state) &
944 (TCPF_CLOSE | TCPF_LISTEN)) {
945 ret = -EINVAL;
946 break;
947 }
948 sk->sk_tskey = tcp_sk(sk)->snd_una;
949 } else {
950 sk->sk_tskey = 0;
951 }
952 }
953
954 if (val & SOF_TIMESTAMPING_OPT_STATS &&
955 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
956 ret = -EINVAL;
957 break;
958 }
959
960 sk->sk_tsflags = val;
961 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
962 sock_enable_timestamp(sk,
963 SOCK_TIMESTAMPING_RX_SOFTWARE);
964 else {
965 if (optname == SO_TIMESTAMPING_NEW)
966 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
967
968 sock_disable_timestamp(sk,
969 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
970 }
971 break;
972
973 case SO_RCVLOWAT:
974 if (val < 0)
975 val = INT_MAX;
976 if (sock->ops->set_rcvlowat)
977 ret = sock->ops->set_rcvlowat(sk, val);
978 else
979 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
980 break;
981
982 case SO_RCVTIMEO_OLD:
983 case SO_RCVTIMEO_NEW:
984 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
985 break;
986
987 case SO_SNDTIMEO_OLD:
988 case SO_SNDTIMEO_NEW:
989 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
990 break;
991
992 case SO_ATTACH_FILTER:
993 ret = -EINVAL;
994 if (optlen == sizeof(struct sock_fprog)) {
995 struct sock_fprog fprog;
996
997 ret = -EFAULT;
998 if (copy_from_user(&fprog, optval, sizeof(fprog)))
999 break;
1000
1001 ret = sk_attach_filter(&fprog, sk);
1002 }
1003 break;
1004
1005 case SO_ATTACH_BPF:
1006 ret = -EINVAL;
1007 if (optlen == sizeof(u32)) {
1008 u32 ufd;
1009
1010 ret = -EFAULT;
1011 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1012 break;
1013
1014 ret = sk_attach_bpf(ufd, sk);
1015 }
1016 break;
1017
1018 case SO_ATTACH_REUSEPORT_CBPF:
1019 ret = -EINVAL;
1020 if (optlen == sizeof(struct sock_fprog)) {
1021 struct sock_fprog fprog;
1022
1023 ret = -EFAULT;
1024 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1025 break;
1026
1027 ret = sk_reuseport_attach_filter(&fprog, sk);
1028 }
1029 break;
1030
1031 case SO_ATTACH_REUSEPORT_EBPF:
1032 ret = -EINVAL;
1033 if (optlen == sizeof(u32)) {
1034 u32 ufd;
1035
1036 ret = -EFAULT;
1037 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1038 break;
1039
1040 ret = sk_reuseport_attach_bpf(ufd, sk);
1041 }
1042 break;
1043
1044 case SO_DETACH_REUSEPORT_BPF:
1045 ret = reuseport_detach_prog(sk);
1046 break;
1047
1048 case SO_DETACH_FILTER:
1049 ret = sk_detach_filter(sk);
1050 break;
1051
1052 case SO_LOCK_FILTER:
1053 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1054 ret = -EPERM;
1055 else
1056 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1057 break;
1058
1059 case SO_PASSSEC:
1060 if (valbool)
1061 set_bit(SOCK_PASSSEC, &sock->flags);
1062 else
1063 clear_bit(SOCK_PASSSEC, &sock->flags);
1064 break;
1065 case SO_MARK:
1066 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1067 ret = -EPERM;
1068 } else if (val != sk->sk_mark) {
1069 sk->sk_mark = val;
1070 sk_dst_reset(sk);
1071 }
1072 break;
1073
1074 case SO_RXQ_OVFL:
1075 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1076 break;
1077
1078 case SO_WIFI_STATUS:
1079 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1080 break;
1081
1082 case SO_PEEK_OFF:
1083 if (sock->ops->set_peek_off)
1084 ret = sock->ops->set_peek_off(sk, val);
1085 else
1086 ret = -EOPNOTSUPP;
1087 break;
1088
1089 case SO_NOFCS:
1090 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1091 break;
1092
1093 case SO_SELECT_ERR_QUEUE:
1094 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1095 break;
1096
1097 #ifdef CONFIG_NET_RX_BUSY_POLL
1098 case SO_BUSY_POLL:
1099 /* allow unprivileged users to decrease the value */
1100 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1101 ret = -EPERM;
1102 else {
1103 if (val < 0)
1104 ret = -EINVAL;
1105 else
1106 sk->sk_ll_usec = val;
1107 }
1108 break;
1109 #endif
1110
1111 case SO_MAX_PACING_RATE:
1112 {
1113 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1114
1115 if (sizeof(ulval) != sizeof(val) &&
1116 optlen >= sizeof(ulval) &&
1117 get_user(ulval, (unsigned long __user *)optval)) {
1118 ret = -EFAULT;
1119 break;
1120 }
1121 if (ulval != ~0UL)
1122 cmpxchg(&sk->sk_pacing_status,
1123 SK_PACING_NONE,
1124 SK_PACING_NEEDED);
1125 sk->sk_max_pacing_rate = ulval;
1126 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1127 break;
1128 }
1129 case SO_INCOMING_CPU:
1130 WRITE_ONCE(sk->sk_incoming_cpu, val);
1131 break;
1132
1133 case SO_CNX_ADVICE:
1134 if (val == 1)
1135 dst_negative_advice(sk);
1136 break;
1137
1138 case SO_ZEROCOPY:
1139 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1140 if (!((sk->sk_type == SOCK_STREAM &&
1141 sk->sk_protocol == IPPROTO_TCP) ||
1142 (sk->sk_type == SOCK_DGRAM &&
1143 sk->sk_protocol == IPPROTO_UDP)))
1144 ret = -ENOTSUPP;
1145 } else if (sk->sk_family != PF_RDS) {
1146 ret = -ENOTSUPP;
1147 }
1148 if (!ret) {
1149 if (val < 0 || val > 1)
1150 ret = -EINVAL;
1151 else
1152 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1153 }
1154 break;
1155
1156 case SO_TXTIME:
1157 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1158 ret = -EPERM;
1159 } else if (optlen != sizeof(struct sock_txtime)) {
1160 ret = -EINVAL;
1161 } else if (copy_from_user(&sk_txtime, optval,
1162 sizeof(struct sock_txtime))) {
1163 ret = -EFAULT;
1164 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1165 ret = -EINVAL;
1166 } else {
1167 sock_valbool_flag(sk, SOCK_TXTIME, true);
1168 sk->sk_clockid = sk_txtime.clockid;
1169 sk->sk_txtime_deadline_mode =
1170 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1171 sk->sk_txtime_report_errors =
1172 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1173 }
1174 break;
1175
1176 case SO_BINDTOIFINDEX:
1177 ret = sock_setbindtodevice_locked(sk, val);
1178 break;
1179
1180 default:
1181 ret = -ENOPROTOOPT;
1182 break;
1183 }
1184 release_sock(sk);
1185 return ret;
1186 }
1187 EXPORT_SYMBOL(sock_setsockopt);
1188
1189
1190 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1191 struct ucred *ucred)
1192 {
1193 ucred->pid = pid_vnr(pid);
1194 ucred->uid = ucred->gid = -1;
1195 if (cred) {
1196 struct user_namespace *current_ns = current_user_ns();
1197
1198 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1199 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1200 }
1201 }
1202
1203 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1204 {
1205 struct user_namespace *user_ns = current_user_ns();
1206 int i;
1207
1208 for (i = 0; i < src->ngroups; i++)
1209 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1210 return -EFAULT;
1211
1212 return 0;
1213 }
1214
1215 int sock_getsockopt(struct socket *sock, int level, int optname,
1216 char __user *optval, int __user *optlen)
1217 {
1218 struct sock *sk = sock->sk;
1219
1220 union {
1221 int val;
1222 u64 val64;
1223 unsigned long ulval;
1224 struct linger ling;
1225 struct old_timeval32 tm32;
1226 struct __kernel_old_timeval tm;
1227 struct __kernel_sock_timeval stm;
1228 struct sock_txtime txtime;
1229 } v;
1230
1231 int lv = sizeof(int);
1232 int len;
1233
1234 if (get_user(len, optlen))
1235 return -EFAULT;
1236 if (len < 0)
1237 return -EINVAL;
1238
1239 memset(&v, 0, sizeof(v));
1240
1241 switch (optname) {
1242 case SO_DEBUG:
1243 v.val = sock_flag(sk, SOCK_DBG);
1244 break;
1245
1246 case SO_DONTROUTE:
1247 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1248 break;
1249
1250 case SO_BROADCAST:
1251 v.val = sock_flag(sk, SOCK_BROADCAST);
1252 break;
1253
1254 case SO_SNDBUF:
1255 v.val = sk->sk_sndbuf;
1256 break;
1257
1258 case SO_RCVBUF:
1259 v.val = sk->sk_rcvbuf;
1260 break;
1261
1262 case SO_REUSEADDR:
1263 v.val = sk->sk_reuse;
1264 break;
1265
1266 case SO_REUSEPORT:
1267 v.val = sk->sk_reuseport;
1268 break;
1269
1270 case SO_KEEPALIVE:
1271 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1272 break;
1273
1274 case SO_TYPE:
1275 v.val = sk->sk_type;
1276 break;
1277
1278 case SO_PROTOCOL:
1279 v.val = sk->sk_protocol;
1280 break;
1281
1282 case SO_DOMAIN:
1283 v.val = sk->sk_family;
1284 break;
1285
1286 case SO_ERROR:
1287 v.val = -sock_error(sk);
1288 if (v.val == 0)
1289 v.val = xchg(&sk->sk_err_soft, 0);
1290 break;
1291
1292 case SO_OOBINLINE:
1293 v.val = sock_flag(sk, SOCK_URGINLINE);
1294 break;
1295
1296 case SO_NO_CHECK:
1297 v.val = sk->sk_no_check_tx;
1298 break;
1299
1300 case SO_PRIORITY:
1301 v.val = sk->sk_priority;
1302 break;
1303
1304 case SO_LINGER:
1305 lv = sizeof(v.ling);
1306 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1307 v.ling.l_linger = sk->sk_lingertime / HZ;
1308 break;
1309
1310 case SO_BSDCOMPAT:
1311 sock_warn_obsolete_bsdism("getsockopt");
1312 break;
1313
1314 case SO_TIMESTAMP_OLD:
1315 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1316 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1317 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1318 break;
1319
1320 case SO_TIMESTAMPNS_OLD:
1321 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1322 break;
1323
1324 case SO_TIMESTAMP_NEW:
1325 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1326 break;
1327
1328 case SO_TIMESTAMPNS_NEW:
1329 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1330 break;
1331
1332 case SO_TIMESTAMPING_OLD:
1333 v.val = sk->sk_tsflags;
1334 break;
1335
1336 case SO_RCVTIMEO_OLD:
1337 case SO_RCVTIMEO_NEW:
1338 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1339 break;
1340
1341 case SO_SNDTIMEO_OLD:
1342 case SO_SNDTIMEO_NEW:
1343 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1344 break;
1345
1346 case SO_RCVLOWAT:
1347 v.val = sk->sk_rcvlowat;
1348 break;
1349
1350 case SO_SNDLOWAT:
1351 v.val = 1;
1352 break;
1353
1354 case SO_PASSCRED:
1355 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1356 break;
1357
1358 case SO_PEERCRED:
1359 {
1360 struct ucred peercred;
1361 if (len > sizeof(peercred))
1362 len = sizeof(peercred);
1363 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1364 if (copy_to_user(optval, &peercred, len))
1365 return -EFAULT;
1366 goto lenout;
1367 }
1368
1369 case SO_PEERGROUPS:
1370 {
1371 int ret, n;
1372
1373 if (!sk->sk_peer_cred)
1374 return -ENODATA;
1375
1376 n = sk->sk_peer_cred->group_info->ngroups;
1377 if (len < n * sizeof(gid_t)) {
1378 len = n * sizeof(gid_t);
1379 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1380 }
1381 len = n * sizeof(gid_t);
1382
1383 ret = groups_to_user((gid_t __user *)optval,
1384 sk->sk_peer_cred->group_info);
1385 if (ret)
1386 return ret;
1387 goto lenout;
1388 }
1389
1390 case SO_PEERNAME:
1391 {
1392 char address[128];
1393
1394 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1395 if (lv < 0)
1396 return -ENOTCONN;
1397 if (lv < len)
1398 return -EINVAL;
1399 if (copy_to_user(optval, address, len))
1400 return -EFAULT;
1401 goto lenout;
1402 }
1403
1404 /* Dubious BSD thing... Probably nobody even uses it, but
1405 * the UNIX standard wants it for whatever reason... -DaveM
1406 */
1407 case SO_ACCEPTCONN:
1408 v.val = sk->sk_state == TCP_LISTEN;
1409 break;
1410
1411 case SO_PASSSEC:
1412 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1413 break;
1414
1415 case SO_PEERSEC:
1416 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1417
1418 case SO_MARK:
1419 v.val = sk->sk_mark;
1420 break;
1421
1422 case SO_RXQ_OVFL:
1423 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1424 break;
1425
1426 case SO_WIFI_STATUS:
1427 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1428 break;
1429
1430 case SO_PEEK_OFF:
1431 if (!sock->ops->set_peek_off)
1432 return -EOPNOTSUPP;
1433
1434 v.val = sk->sk_peek_off;
1435 break;
1436 case SO_NOFCS:
1437 v.val = sock_flag(sk, SOCK_NOFCS);
1438 break;
1439
1440 case SO_BINDTODEVICE:
1441 return sock_getbindtodevice(sk, optval, optlen, len);
1442
1443 case SO_GET_FILTER:
1444 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1445 if (len < 0)
1446 return len;
1447
1448 goto lenout;
1449
1450 case SO_LOCK_FILTER:
1451 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1452 break;
1453
1454 case SO_BPF_EXTENSIONS:
1455 v.val = bpf_tell_extensions();
1456 break;
1457
1458 case SO_SELECT_ERR_QUEUE:
1459 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1460 break;
1461
1462 #ifdef CONFIG_NET_RX_BUSY_POLL
1463 case SO_BUSY_POLL:
1464 v.val = sk->sk_ll_usec;
1465 break;
1466 #endif
1467
1468 case SO_MAX_PACING_RATE:
1469 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1470 lv = sizeof(v.ulval);
1471 v.ulval = sk->sk_max_pacing_rate;
1472 } else {
1473 /* 32bit version */
1474 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1475 }
1476 break;
1477
1478 case SO_INCOMING_CPU:
1479 v.val = READ_ONCE(sk->sk_incoming_cpu);
1480 break;
1481
1482 case SO_MEMINFO:
1483 {
1484 u32 meminfo[SK_MEMINFO_VARS];
1485
1486 sk_get_meminfo(sk, meminfo);
1487
1488 len = min_t(unsigned int, len, sizeof(meminfo));
1489 if (copy_to_user(optval, &meminfo, len))
1490 return -EFAULT;
1491
1492 goto lenout;
1493 }
1494
1495 #ifdef CONFIG_NET_RX_BUSY_POLL
1496 case SO_INCOMING_NAPI_ID:
1497 v.val = READ_ONCE(sk->sk_napi_id);
1498
1499 /* aggregate non-NAPI IDs down to 0 */
1500 if (v.val < MIN_NAPI_ID)
1501 v.val = 0;
1502
1503 break;
1504 #endif
1505
1506 case SO_COOKIE:
1507 lv = sizeof(u64);
1508 if (len < lv)
1509 return -EINVAL;
1510 v.val64 = sock_gen_cookie(sk);
1511 break;
1512
1513 case SO_ZEROCOPY:
1514 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1515 break;
1516
1517 case SO_TXTIME:
1518 lv = sizeof(v.txtime);
1519 v.txtime.clockid = sk->sk_clockid;
1520 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1521 SOF_TXTIME_DEADLINE_MODE : 0;
1522 v.txtime.flags |= sk->sk_txtime_report_errors ?
1523 SOF_TXTIME_REPORT_ERRORS : 0;
1524 break;
1525
1526 case SO_BINDTOIFINDEX:
1527 v.val = sk->sk_bound_dev_if;
1528 break;
1529
1530 default:
1531 /* We implement the SO_SNDLOWAT etc to not be settable
1532 * (1003.1g 7).
1533 */
1534 return -ENOPROTOOPT;
1535 }
1536
1537 if (len > lv)
1538 len = lv;
1539 if (copy_to_user(optval, &v, len))
1540 return -EFAULT;
1541 lenout:
1542 if (put_user(len, optlen))
1543 return -EFAULT;
1544 return 0;
1545 }
1546
1547 /*
1548 * Initialize an sk_lock.
1549 *
1550 * (We also register the sk_lock with the lock validator.)
1551 */
1552 static inline void sock_lock_init(struct sock *sk)
1553 {
1554 if (sk->sk_kern_sock)
1555 sock_lock_init_class_and_name(
1556 sk,
1557 af_family_kern_slock_key_strings[sk->sk_family],
1558 af_family_kern_slock_keys + sk->sk_family,
1559 af_family_kern_key_strings[sk->sk_family],
1560 af_family_kern_keys + sk->sk_family);
1561 else
1562 sock_lock_init_class_and_name(
1563 sk,
1564 af_family_slock_key_strings[sk->sk_family],
1565 af_family_slock_keys + sk->sk_family,
1566 af_family_key_strings[sk->sk_family],
1567 af_family_keys + sk->sk_family);
1568 }
1569
1570 /*
1571 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1572 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1573 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1574 */
1575 static void sock_copy(struct sock *nsk, const struct sock *osk)
1576 {
1577 #ifdef CONFIG_SECURITY_NETWORK
1578 void *sptr = nsk->sk_security;
1579 #endif
1580 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1581
1582 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1583 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1584
1585 #ifdef CONFIG_SECURITY_NETWORK
1586 nsk->sk_security = sptr;
1587 security_sk_clone(osk, nsk);
1588 #endif
1589 }
1590
1591 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1592 int family)
1593 {
1594 struct sock *sk;
1595 struct kmem_cache *slab;
1596
1597 slab = prot->slab;
1598 if (slab != NULL) {
1599 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1600 if (!sk)
1601 return sk;
1602 if (want_init_on_alloc(priority))
1603 sk_prot_clear_nulls(sk, prot->obj_size);
1604 } else
1605 sk = kmalloc(prot->obj_size, priority);
1606
1607 if (sk != NULL) {
1608 if (security_sk_alloc(sk, family, priority))
1609 goto out_free;
1610
1611 if (!try_module_get(prot->owner))
1612 goto out_free_sec;
1613 sk_tx_queue_clear(sk);
1614 }
1615
1616 return sk;
1617
1618 out_free_sec:
1619 security_sk_free(sk);
1620 out_free:
1621 if (slab != NULL)
1622 kmem_cache_free(slab, sk);
1623 else
1624 kfree(sk);
1625 return NULL;
1626 }
1627
1628 static void sk_prot_free(struct proto *prot, struct sock *sk)
1629 {
1630 struct kmem_cache *slab;
1631 struct module *owner;
1632
1633 owner = prot->owner;
1634 slab = prot->slab;
1635
1636 cgroup_sk_free(&sk->sk_cgrp_data);
1637 mem_cgroup_sk_free(sk);
1638 security_sk_free(sk);
1639 if (slab != NULL)
1640 kmem_cache_free(slab, sk);
1641 else
1642 kfree(sk);
1643 module_put(owner);
1644 }
1645
1646 /**
1647 * sk_alloc - All socket objects are allocated here
1648 * @net: the applicable net namespace
1649 * @family: protocol family
1650 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1651 * @prot: struct proto associated with this new sock instance
1652 * @kern: is this to be a kernel socket?
1653 */
1654 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1655 struct proto *prot, int kern)
1656 {
1657 struct sock *sk;
1658
1659 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1660 if (sk) {
1661 sk->sk_family = family;
1662 /*
1663 * See comment in struct sock definition to understand
1664 * why we need sk_prot_creator -acme
1665 */
1666 sk->sk_prot = sk->sk_prot_creator = prot;
1667 sk->sk_kern_sock = kern;
1668 sock_lock_init(sk);
1669 sk->sk_net_refcnt = kern ? 0 : 1;
1670 if (likely(sk->sk_net_refcnt)) {
1671 get_net(net);
1672 sock_inuse_add(net, 1);
1673 }
1674
1675 sock_net_set(sk, net);
1676 refcount_set(&sk->sk_wmem_alloc, 1);
1677
1678 mem_cgroup_sk_alloc(sk);
1679 cgroup_sk_alloc(&sk->sk_cgrp_data);
1680 sock_update_classid(&sk->sk_cgrp_data);
1681 sock_update_netprioidx(&sk->sk_cgrp_data);
1682 sk_tx_queue_clear(sk);
1683 }
1684
1685 return sk;
1686 }
1687 EXPORT_SYMBOL(sk_alloc);
1688
1689 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1690 * grace period. This is the case for UDP sockets and TCP listeners.
1691 */
1692 static void __sk_destruct(struct rcu_head *head)
1693 {
1694 struct sock *sk = container_of(head, struct sock, sk_rcu);
1695 struct sk_filter *filter;
1696
1697 if (sk->sk_destruct)
1698 sk->sk_destruct(sk);
1699
1700 filter = rcu_dereference_check(sk->sk_filter,
1701 refcount_read(&sk->sk_wmem_alloc) == 0);
1702 if (filter) {
1703 sk_filter_uncharge(sk, filter);
1704 RCU_INIT_POINTER(sk->sk_filter, NULL);
1705 }
1706
1707 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1708
1709 #ifdef CONFIG_BPF_SYSCALL
1710 bpf_sk_storage_free(sk);
1711 #endif
1712
1713 if (atomic_read(&sk->sk_omem_alloc))
1714 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1715 __func__, atomic_read(&sk->sk_omem_alloc));
1716
1717 if (sk->sk_frag.page) {
1718 put_page(sk->sk_frag.page);
1719 sk->sk_frag.page = NULL;
1720 }
1721
1722 if (sk->sk_peer_cred)
1723 put_cred(sk->sk_peer_cred);
1724 put_pid(sk->sk_peer_pid);
1725 if (likely(sk->sk_net_refcnt))
1726 put_net(sock_net(sk));
1727 sk_prot_free(sk->sk_prot_creator, sk);
1728 }
1729
1730 void sk_destruct(struct sock *sk)
1731 {
1732 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1733
1734 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1735 reuseport_detach_sock(sk);
1736 use_call_rcu = true;
1737 }
1738
1739 if (use_call_rcu)
1740 call_rcu(&sk->sk_rcu, __sk_destruct);
1741 else
1742 __sk_destruct(&sk->sk_rcu);
1743 }
1744
1745 static void __sk_free(struct sock *sk)
1746 {
1747 if (likely(sk->sk_net_refcnt))
1748 sock_inuse_add(sock_net(sk), -1);
1749
1750 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1751 sock_diag_broadcast_destroy(sk);
1752 else
1753 sk_destruct(sk);
1754 }
1755
1756 void sk_free(struct sock *sk)
1757 {
1758 /*
1759 * We subtract one from sk_wmem_alloc and can know if
1760 * some packets are still in some tx queue.
1761 * If not null, sock_wfree() will call __sk_free(sk) later
1762 */
1763 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1764 __sk_free(sk);
1765 }
1766 EXPORT_SYMBOL(sk_free);
1767
1768 static void sk_init_common(struct sock *sk)
1769 {
1770 skb_queue_head_init(&sk->sk_receive_queue);
1771 skb_queue_head_init(&sk->sk_write_queue);
1772 skb_queue_head_init(&sk->sk_error_queue);
1773
1774 rwlock_init(&sk->sk_callback_lock);
1775 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1776 af_rlock_keys + sk->sk_family,
1777 af_family_rlock_key_strings[sk->sk_family]);
1778 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1779 af_wlock_keys + sk->sk_family,
1780 af_family_wlock_key_strings[sk->sk_family]);
1781 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1782 af_elock_keys + sk->sk_family,
1783 af_family_elock_key_strings[sk->sk_family]);
1784 lockdep_set_class_and_name(&sk->sk_callback_lock,
1785 af_callback_keys + sk->sk_family,
1786 af_family_clock_key_strings[sk->sk_family]);
1787 }
1788
1789 /**
1790 * sk_clone_lock - clone a socket, and lock its clone
1791 * @sk: the socket to clone
1792 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1793 *
1794 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1795 */
1796 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1797 {
1798 struct sock *newsk;
1799 bool is_charged = true;
1800
1801 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1802 if (newsk != NULL) {
1803 struct sk_filter *filter;
1804
1805 sock_copy(newsk, sk);
1806
1807 newsk->sk_prot_creator = sk->sk_prot;
1808
1809 /* SANITY */
1810 if (likely(newsk->sk_net_refcnt))
1811 get_net(sock_net(newsk));
1812 sk_node_init(&newsk->sk_node);
1813 sock_lock_init(newsk);
1814 bh_lock_sock(newsk);
1815 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1816 newsk->sk_backlog.len = 0;
1817
1818 atomic_set(&newsk->sk_rmem_alloc, 0);
1819 /*
1820 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1821 */
1822 refcount_set(&newsk->sk_wmem_alloc, 1);
1823 atomic_set(&newsk->sk_omem_alloc, 0);
1824 sk_init_common(newsk);
1825
1826 newsk->sk_dst_cache = NULL;
1827 newsk->sk_dst_pending_confirm = 0;
1828 newsk->sk_wmem_queued = 0;
1829 newsk->sk_forward_alloc = 0;
1830 atomic_set(&newsk->sk_drops, 0);
1831 newsk->sk_send_head = NULL;
1832 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1833 atomic_set(&newsk->sk_zckey, 0);
1834
1835 sock_reset_flag(newsk, SOCK_DONE);
1836
1837 /* sk->sk_memcg will be populated at accept() time */
1838 newsk->sk_memcg = NULL;
1839
1840 cgroup_sk_clone(&newsk->sk_cgrp_data);
1841
1842 rcu_read_lock();
1843 filter = rcu_dereference(sk->sk_filter);
1844 if (filter != NULL)
1845 /* though it's an empty new sock, the charging may fail
1846 * if sysctl_optmem_max was changed between creation of
1847 * original socket and cloning
1848 */
1849 is_charged = sk_filter_charge(newsk, filter);
1850 RCU_INIT_POINTER(newsk->sk_filter, filter);
1851 rcu_read_unlock();
1852
1853 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1854 /* We need to make sure that we don't uncharge the new
1855 * socket if we couldn't charge it in the first place
1856 * as otherwise we uncharge the parent's filter.
1857 */
1858 if (!is_charged)
1859 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1860 sk_free_unlock_clone(newsk);
1861 newsk = NULL;
1862 goto out;
1863 }
1864 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1865
1866 if (bpf_sk_storage_clone(sk, newsk)) {
1867 sk_free_unlock_clone(newsk);
1868 newsk = NULL;
1869 goto out;
1870 }
1871
1872 newsk->sk_err = 0;
1873 newsk->sk_err_soft = 0;
1874 newsk->sk_priority = 0;
1875 newsk->sk_incoming_cpu = raw_smp_processor_id();
1876 if (likely(newsk->sk_net_refcnt))
1877 sock_inuse_add(sock_net(newsk), 1);
1878
1879 /*
1880 * Before updating sk_refcnt, we must commit prior changes to memory
1881 * (Documentation/RCU/rculist_nulls.txt for details)
1882 */
1883 smp_wmb();
1884 refcount_set(&newsk->sk_refcnt, 2);
1885
1886 /*
1887 * Increment the counter in the same struct proto as the master
1888 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1889 * is the same as sk->sk_prot->socks, as this field was copied
1890 * with memcpy).
1891 *
1892 * This _changes_ the previous behaviour, where
1893 * tcp_create_openreq_child always was incrementing the
1894 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1895 * to be taken into account in all callers. -acme
1896 */
1897 sk_refcnt_debug_inc(newsk);
1898 sk_set_socket(newsk, NULL);
1899 sk_tx_queue_clear(newsk);
1900 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1901
1902 if (newsk->sk_prot->sockets_allocated)
1903 sk_sockets_allocated_inc(newsk);
1904
1905 if (sock_needs_netstamp(sk) &&
1906 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1907 net_enable_timestamp();
1908 }
1909 out:
1910 return newsk;
1911 }
1912 EXPORT_SYMBOL_GPL(sk_clone_lock);
1913
1914 void sk_free_unlock_clone(struct sock *sk)
1915 {
1916 /* It is still raw copy of parent, so invalidate
1917 * destructor and make plain sk_free() */
1918 sk->sk_destruct = NULL;
1919 bh_unlock_sock(sk);
1920 sk_free(sk);
1921 }
1922 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1923
1924 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1925 {
1926 u32 max_segs = 1;
1927
1928 sk_dst_set(sk, dst);
1929 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1930 if (sk->sk_route_caps & NETIF_F_GSO)
1931 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1932 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1933 if (sk_can_gso(sk)) {
1934 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1935 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1936 } else {
1937 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1938 sk->sk_gso_max_size = dst->dev->gso_max_size;
1939 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1940 }
1941 }
1942 sk->sk_gso_max_segs = max_segs;
1943 }
1944 EXPORT_SYMBOL_GPL(sk_setup_caps);
1945
1946 /*
1947 * Simple resource managers for sockets.
1948 */
1949
1950
1951 /*
1952 * Write buffer destructor automatically called from kfree_skb.
1953 */
1954 void sock_wfree(struct sk_buff *skb)
1955 {
1956 struct sock *sk = skb->sk;
1957 unsigned int len = skb->truesize;
1958
1959 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1960 /*
1961 * Keep a reference on sk_wmem_alloc, this will be released
1962 * after sk_write_space() call
1963 */
1964 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1965 sk->sk_write_space(sk);
1966 len = 1;
1967 }
1968 /*
1969 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1970 * could not do because of in-flight packets
1971 */
1972 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1973 __sk_free(sk);
1974 }
1975 EXPORT_SYMBOL(sock_wfree);
1976
1977 /* This variant of sock_wfree() is used by TCP,
1978 * since it sets SOCK_USE_WRITE_QUEUE.
1979 */
1980 void __sock_wfree(struct sk_buff *skb)
1981 {
1982 struct sock *sk = skb->sk;
1983
1984 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1985 __sk_free(sk);
1986 }
1987
1988 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1989 {
1990 skb_orphan(skb);
1991 skb->sk = sk;
1992 #ifdef CONFIG_INET
1993 if (unlikely(!sk_fullsock(sk))) {
1994 skb->destructor = sock_edemux;
1995 sock_hold(sk);
1996 return;
1997 }
1998 #endif
1999 skb->destructor = sock_wfree;
2000 skb_set_hash_from_sk(skb, sk);
2001 /*
2002 * We used to take a refcount on sk, but following operation
2003 * is enough to guarantee sk_free() wont free this sock until
2004 * all in-flight packets are completed
2005 */
2006 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2007 }
2008 EXPORT_SYMBOL(skb_set_owner_w);
2009
2010 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2011 {
2012 #ifdef CONFIG_TLS_DEVICE
2013 /* Drivers depend on in-order delivery for crypto offload,
2014 * partial orphan breaks out-of-order-OK logic.
2015 */
2016 if (skb->decrypted)
2017 return false;
2018 #endif
2019 return (skb->destructor == sock_wfree ||
2020 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2021 }
2022
2023 /* This helper is used by netem, as it can hold packets in its
2024 * delay queue. We want to allow the owner socket to send more
2025 * packets, as if they were already TX completed by a typical driver.
2026 * But we also want to keep skb->sk set because some packet schedulers
2027 * rely on it (sch_fq for example).
2028 */
2029 void skb_orphan_partial(struct sk_buff *skb)
2030 {
2031 if (skb_is_tcp_pure_ack(skb))
2032 return;
2033
2034 if (can_skb_orphan_partial(skb)) {
2035 struct sock *sk = skb->sk;
2036
2037 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2038 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2039 skb->destructor = sock_efree;
2040 }
2041 } else {
2042 skb_orphan(skb);
2043 }
2044 }
2045 EXPORT_SYMBOL(skb_orphan_partial);
2046
2047 /*
2048 * Read buffer destructor automatically called from kfree_skb.
2049 */
2050 void sock_rfree(struct sk_buff *skb)
2051 {
2052 struct sock *sk = skb->sk;
2053 unsigned int len = skb->truesize;
2054
2055 atomic_sub(len, &sk->sk_rmem_alloc);
2056 sk_mem_uncharge(sk, len);
2057 }
2058 EXPORT_SYMBOL(sock_rfree);
2059
2060 /*
2061 * Buffer destructor for skbs that are not used directly in read or write
2062 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2063 */
2064 void sock_efree(struct sk_buff *skb)
2065 {
2066 sock_put(skb->sk);
2067 }
2068 EXPORT_SYMBOL(sock_efree);
2069
2070 kuid_t sock_i_uid(struct sock *sk)
2071 {
2072 kuid_t uid;
2073
2074 read_lock_bh(&sk->sk_callback_lock);
2075 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2076 read_unlock_bh(&sk->sk_callback_lock);
2077 return uid;
2078 }
2079 EXPORT_SYMBOL(sock_i_uid);
2080
2081 unsigned long sock_i_ino(struct sock *sk)
2082 {
2083 unsigned long ino;
2084
2085 read_lock_bh(&sk->sk_callback_lock);
2086 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2087 read_unlock_bh(&sk->sk_callback_lock);
2088 return ino;
2089 }
2090 EXPORT_SYMBOL(sock_i_ino);
2091
2092 /*
2093 * Allocate a skb from the socket's send buffer.
2094 */
2095 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2096 gfp_t priority)
2097 {
2098 if (force ||
2099 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2100 struct sk_buff *skb = alloc_skb(size, priority);
2101
2102 if (skb) {
2103 skb_set_owner_w(skb, sk);
2104 return skb;
2105 }
2106 }
2107 return NULL;
2108 }
2109 EXPORT_SYMBOL(sock_wmalloc);
2110
2111 static void sock_ofree(struct sk_buff *skb)
2112 {
2113 struct sock *sk = skb->sk;
2114
2115 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2116 }
2117
2118 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2119 gfp_t priority)
2120 {
2121 struct sk_buff *skb;
2122
2123 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2124 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2125 sysctl_optmem_max)
2126 return NULL;
2127
2128 skb = alloc_skb(size, priority);
2129 if (!skb)
2130 return NULL;
2131
2132 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2133 skb->sk = sk;
2134 skb->destructor = sock_ofree;
2135 return skb;
2136 }
2137
2138 /*
2139 * Allocate a memory block from the socket's option memory buffer.
2140 */
2141 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2142 {
2143 if ((unsigned int)size <= sysctl_optmem_max &&
2144 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2145 void *mem;
2146 /* First do the add, to avoid the race if kmalloc
2147 * might sleep.
2148 */
2149 atomic_add(size, &sk->sk_omem_alloc);
2150 mem = kmalloc(size, priority);
2151 if (mem)
2152 return mem;
2153 atomic_sub(size, &sk->sk_omem_alloc);
2154 }
2155 return NULL;
2156 }
2157 EXPORT_SYMBOL(sock_kmalloc);
2158
2159 /* Free an option memory block. Note, we actually want the inline
2160 * here as this allows gcc to detect the nullify and fold away the
2161 * condition entirely.
2162 */
2163 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2164 const bool nullify)
2165 {
2166 if (WARN_ON_ONCE(!mem))
2167 return;
2168 if (nullify)
2169 kzfree(mem);
2170 else
2171 kfree(mem);
2172 atomic_sub(size, &sk->sk_omem_alloc);
2173 }
2174
2175 void sock_kfree_s(struct sock *sk, void *mem, int size)
2176 {
2177 __sock_kfree_s(sk, mem, size, false);
2178 }
2179 EXPORT_SYMBOL(sock_kfree_s);
2180
2181 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2182 {
2183 __sock_kfree_s(sk, mem, size, true);
2184 }
2185 EXPORT_SYMBOL(sock_kzfree_s);
2186
2187 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2188 I think, these locks should be removed for datagram sockets.
2189 */
2190 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2191 {
2192 DEFINE_WAIT(wait);
2193
2194 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2195 for (;;) {
2196 if (!timeo)
2197 break;
2198 if (signal_pending(current))
2199 break;
2200 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2201 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2202 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2203 break;
2204 if (sk->sk_shutdown & SEND_SHUTDOWN)
2205 break;
2206 if (sk->sk_err)
2207 break;
2208 timeo = schedule_timeout(timeo);
2209 }
2210 finish_wait(sk_sleep(sk), &wait);
2211 return timeo;
2212 }
2213
2214
2215 /*
2216 * Generic send/receive buffer handlers
2217 */
2218
2219 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2220 unsigned long data_len, int noblock,
2221 int *errcode, int max_page_order)
2222 {
2223 struct sk_buff *skb;
2224 long timeo;
2225 int err;
2226
2227 timeo = sock_sndtimeo(sk, noblock);
2228 for (;;) {
2229 err = sock_error(sk);
2230 if (err != 0)
2231 goto failure;
2232
2233 err = -EPIPE;
2234 if (sk->sk_shutdown & SEND_SHUTDOWN)
2235 goto failure;
2236
2237 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2238 break;
2239
2240 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2241 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2242 err = -EAGAIN;
2243 if (!timeo)
2244 goto failure;
2245 if (signal_pending(current))
2246 goto interrupted;
2247 timeo = sock_wait_for_wmem(sk, timeo);
2248 }
2249 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2250 errcode, sk->sk_allocation);
2251 if (skb)
2252 skb_set_owner_w(skb, sk);
2253 return skb;
2254
2255 interrupted:
2256 err = sock_intr_errno(timeo);
2257 failure:
2258 *errcode = err;
2259 return NULL;
2260 }
2261 EXPORT_SYMBOL(sock_alloc_send_pskb);
2262
2263 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2264 int noblock, int *errcode)
2265 {
2266 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2267 }
2268 EXPORT_SYMBOL(sock_alloc_send_skb);
2269
2270 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2271 struct sockcm_cookie *sockc)
2272 {
2273 u32 tsflags;
2274
2275 switch (cmsg->cmsg_type) {
2276 case SO_MARK:
2277 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2278 return -EPERM;
2279 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2280 return -EINVAL;
2281 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2282 break;
2283 case SO_TIMESTAMPING_OLD:
2284 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2285 return -EINVAL;
2286
2287 tsflags = *(u32 *)CMSG_DATA(cmsg);
2288 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2289 return -EINVAL;
2290
2291 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2292 sockc->tsflags |= tsflags;
2293 break;
2294 case SCM_TXTIME:
2295 if (!sock_flag(sk, SOCK_TXTIME))
2296 return -EINVAL;
2297 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2298 return -EINVAL;
2299 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2300 break;
2301 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2302 case SCM_RIGHTS:
2303 case SCM_CREDENTIALS:
2304 break;
2305 default:
2306 return -EINVAL;
2307 }
2308 return 0;
2309 }
2310 EXPORT_SYMBOL(__sock_cmsg_send);
2311
2312 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2313 struct sockcm_cookie *sockc)
2314 {
2315 struct cmsghdr *cmsg;
2316 int ret;
2317
2318 for_each_cmsghdr(cmsg, msg) {
2319 if (!CMSG_OK(msg, cmsg))
2320 return -EINVAL;
2321 if (cmsg->cmsg_level != SOL_SOCKET)
2322 continue;
2323 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2324 if (ret)
2325 return ret;
2326 }
2327 return 0;
2328 }
2329 EXPORT_SYMBOL(sock_cmsg_send);
2330
2331 static void sk_enter_memory_pressure(struct sock *sk)
2332 {
2333 if (!sk->sk_prot->enter_memory_pressure)
2334 return;
2335
2336 sk->sk_prot->enter_memory_pressure(sk);
2337 }
2338
2339 static void sk_leave_memory_pressure(struct sock *sk)
2340 {
2341 if (sk->sk_prot->leave_memory_pressure) {
2342 sk->sk_prot->leave_memory_pressure(sk);
2343 } else {
2344 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2345
2346 if (memory_pressure && READ_ONCE(*memory_pressure))
2347 WRITE_ONCE(*memory_pressure, 0);
2348 }
2349 }
2350
2351 /* On 32bit arches, an skb frag is limited to 2^15 */
2352 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2353 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2354
2355 /**
2356 * skb_page_frag_refill - check that a page_frag contains enough room
2357 * @sz: minimum size of the fragment we want to get
2358 * @pfrag: pointer to page_frag
2359 * @gfp: priority for memory allocation
2360 *
2361 * Note: While this allocator tries to use high order pages, there is
2362 * no guarantee that allocations succeed. Therefore, @sz MUST be
2363 * less or equal than PAGE_SIZE.
2364 */
2365 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2366 {
2367 if (pfrag->page) {
2368 if (page_ref_count(pfrag->page) == 1) {
2369 pfrag->offset = 0;
2370 return true;
2371 }
2372 if (pfrag->offset + sz <= pfrag->size)
2373 return true;
2374 put_page(pfrag->page);
2375 }
2376
2377 pfrag->offset = 0;
2378 if (SKB_FRAG_PAGE_ORDER &&
2379 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2380 /* Avoid direct reclaim but allow kswapd to wake */
2381 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2382 __GFP_COMP | __GFP_NOWARN |
2383 __GFP_NORETRY,
2384 SKB_FRAG_PAGE_ORDER);
2385 if (likely(pfrag->page)) {
2386 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2387 return true;
2388 }
2389 }
2390 pfrag->page = alloc_page(gfp);
2391 if (likely(pfrag->page)) {
2392 pfrag->size = PAGE_SIZE;
2393 return true;
2394 }
2395 return false;
2396 }
2397 EXPORT_SYMBOL(skb_page_frag_refill);
2398
2399 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2400 {
2401 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2402 return true;
2403
2404 sk_enter_memory_pressure(sk);
2405 sk_stream_moderate_sndbuf(sk);
2406 return false;
2407 }
2408 EXPORT_SYMBOL(sk_page_frag_refill);
2409
2410 static void __lock_sock(struct sock *sk)
2411 __releases(&sk->sk_lock.slock)
2412 __acquires(&sk->sk_lock.slock)
2413 {
2414 DEFINE_WAIT(wait);
2415
2416 for (;;) {
2417 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2418 TASK_UNINTERRUPTIBLE);
2419 spin_unlock_bh(&sk->sk_lock.slock);
2420 schedule();
2421 spin_lock_bh(&sk->sk_lock.slock);
2422 if (!sock_owned_by_user(sk))
2423 break;
2424 }
2425 finish_wait(&sk->sk_lock.wq, &wait);
2426 }
2427
2428 void __release_sock(struct sock *sk)
2429 __releases(&sk->sk_lock.slock)
2430 __acquires(&sk->sk_lock.slock)
2431 {
2432 struct sk_buff *skb, *next;
2433
2434 while ((skb = sk->sk_backlog.head) != NULL) {
2435 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2436
2437 spin_unlock_bh(&sk->sk_lock.slock);
2438
2439 do {
2440 next = skb->next;
2441 prefetch(next);
2442 WARN_ON_ONCE(skb_dst_is_noref(skb));
2443 skb_mark_not_on_list(skb);
2444 sk_backlog_rcv(sk, skb);
2445
2446 cond_resched();
2447
2448 skb = next;
2449 } while (skb != NULL);
2450
2451 spin_lock_bh(&sk->sk_lock.slock);
2452 }
2453
2454 /*
2455 * Doing the zeroing here guarantee we can not loop forever
2456 * while a wild producer attempts to flood us.
2457 */
2458 sk->sk_backlog.len = 0;
2459 }
2460
2461 void __sk_flush_backlog(struct sock *sk)
2462 {
2463 spin_lock_bh(&sk->sk_lock.slock);
2464 __release_sock(sk);
2465 spin_unlock_bh(&sk->sk_lock.slock);
2466 }
2467
2468 /**
2469 * sk_wait_data - wait for data to arrive at sk_receive_queue
2470 * @sk: sock to wait on
2471 * @timeo: for how long
2472 * @skb: last skb seen on sk_receive_queue
2473 *
2474 * Now socket state including sk->sk_err is changed only under lock,
2475 * hence we may omit checks after joining wait queue.
2476 * We check receive queue before schedule() only as optimization;
2477 * it is very likely that release_sock() added new data.
2478 */
2479 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2480 {
2481 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2482 int rc;
2483
2484 add_wait_queue(sk_sleep(sk), &wait);
2485 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2486 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2487 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2488 remove_wait_queue(sk_sleep(sk), &wait);
2489 return rc;
2490 }
2491 EXPORT_SYMBOL(sk_wait_data);
2492
2493 /**
2494 * __sk_mem_raise_allocated - increase memory_allocated
2495 * @sk: socket
2496 * @size: memory size to allocate
2497 * @amt: pages to allocate
2498 * @kind: allocation type
2499 *
2500 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2501 */
2502 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2503 {
2504 struct proto *prot = sk->sk_prot;
2505 long allocated = sk_memory_allocated_add(sk, amt);
2506 bool charged = true;
2507
2508 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2509 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2510 goto suppress_allocation;
2511
2512 /* Under limit. */
2513 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2514 sk_leave_memory_pressure(sk);
2515 return 1;
2516 }
2517
2518 /* Under pressure. */
2519 if (allocated > sk_prot_mem_limits(sk, 1))
2520 sk_enter_memory_pressure(sk);
2521
2522 /* Over hard limit. */
2523 if (allocated > sk_prot_mem_limits(sk, 2))
2524 goto suppress_allocation;
2525
2526 /* guarantee minimum buffer size under pressure */
2527 if (kind == SK_MEM_RECV) {
2528 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2529 return 1;
2530
2531 } else { /* SK_MEM_SEND */
2532 int wmem0 = sk_get_wmem0(sk, prot);
2533
2534 if (sk->sk_type == SOCK_STREAM) {
2535 if (sk->sk_wmem_queued < wmem0)
2536 return 1;
2537 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2538 return 1;
2539 }
2540 }
2541
2542 if (sk_has_memory_pressure(sk)) {
2543 u64 alloc;
2544
2545 if (!sk_under_memory_pressure(sk))
2546 return 1;
2547 alloc = sk_sockets_allocated_read_positive(sk);
2548 if (sk_prot_mem_limits(sk, 2) > alloc *
2549 sk_mem_pages(sk->sk_wmem_queued +
2550 atomic_read(&sk->sk_rmem_alloc) +
2551 sk->sk_forward_alloc))
2552 return 1;
2553 }
2554
2555 suppress_allocation:
2556
2557 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2558 sk_stream_moderate_sndbuf(sk);
2559
2560 /* Fail only if socket is _under_ its sndbuf.
2561 * In this case we cannot block, so that we have to fail.
2562 */
2563 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2564 return 1;
2565 }
2566
2567 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2568 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2569
2570 sk_memory_allocated_sub(sk, amt);
2571
2572 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2573 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2574
2575 return 0;
2576 }
2577 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2578
2579 /**
2580 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2581 * @sk: socket
2582 * @size: memory size to allocate
2583 * @kind: allocation type
2584 *
2585 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2586 * rmem allocation. This function assumes that protocols which have
2587 * memory_pressure use sk_wmem_queued as write buffer accounting.
2588 */
2589 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2590 {
2591 int ret, amt = sk_mem_pages(size);
2592
2593 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2594 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2595 if (!ret)
2596 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2597 return ret;
2598 }
2599 EXPORT_SYMBOL(__sk_mem_schedule);
2600
2601 /**
2602 * __sk_mem_reduce_allocated - reclaim memory_allocated
2603 * @sk: socket
2604 * @amount: number of quanta
2605 *
2606 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2607 */
2608 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2609 {
2610 sk_memory_allocated_sub(sk, amount);
2611
2612 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2613 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2614
2615 if (sk_under_memory_pressure(sk) &&
2616 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2617 sk_leave_memory_pressure(sk);
2618 }
2619 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2620
2621 /**
2622 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2623 * @sk: socket
2624 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2625 */
2626 void __sk_mem_reclaim(struct sock *sk, int amount)
2627 {
2628 amount >>= SK_MEM_QUANTUM_SHIFT;
2629 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2630 __sk_mem_reduce_allocated(sk, amount);
2631 }
2632 EXPORT_SYMBOL(__sk_mem_reclaim);
2633
2634 int sk_set_peek_off(struct sock *sk, int val)
2635 {
2636 sk->sk_peek_off = val;
2637 return 0;
2638 }
2639 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2640
2641 /*
2642 * Set of default routines for initialising struct proto_ops when
2643 * the protocol does not support a particular function. In certain
2644 * cases where it makes no sense for a protocol to have a "do nothing"
2645 * function, some default processing is provided.
2646 */
2647
2648 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2649 {
2650 return -EOPNOTSUPP;
2651 }
2652 EXPORT_SYMBOL(sock_no_bind);
2653
2654 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2655 int len, int flags)
2656 {
2657 return -EOPNOTSUPP;
2658 }
2659 EXPORT_SYMBOL(sock_no_connect);
2660
2661 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2662 {
2663 return -EOPNOTSUPP;
2664 }
2665 EXPORT_SYMBOL(sock_no_socketpair);
2666
2667 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2668 bool kern)
2669 {
2670 return -EOPNOTSUPP;
2671 }
2672 EXPORT_SYMBOL(sock_no_accept);
2673
2674 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2675 int peer)
2676 {
2677 return -EOPNOTSUPP;
2678 }
2679 EXPORT_SYMBOL(sock_no_getname);
2680
2681 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2682 {
2683 return -EOPNOTSUPP;
2684 }
2685 EXPORT_SYMBOL(sock_no_ioctl);
2686
2687 int sock_no_listen(struct socket *sock, int backlog)
2688 {
2689 return -EOPNOTSUPP;
2690 }
2691 EXPORT_SYMBOL(sock_no_listen);
2692
2693 int sock_no_shutdown(struct socket *sock, int how)
2694 {
2695 return -EOPNOTSUPP;
2696 }
2697 EXPORT_SYMBOL(sock_no_shutdown);
2698
2699 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2700 char __user *optval, unsigned int optlen)
2701 {
2702 return -EOPNOTSUPP;
2703 }
2704 EXPORT_SYMBOL(sock_no_setsockopt);
2705
2706 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2707 char __user *optval, int __user *optlen)
2708 {
2709 return -EOPNOTSUPP;
2710 }
2711 EXPORT_SYMBOL(sock_no_getsockopt);
2712
2713 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2714 {
2715 return -EOPNOTSUPP;
2716 }
2717 EXPORT_SYMBOL(sock_no_sendmsg);
2718
2719 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2720 {
2721 return -EOPNOTSUPP;
2722 }
2723 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2724
2725 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2726 int flags)
2727 {
2728 return -EOPNOTSUPP;
2729 }
2730 EXPORT_SYMBOL(sock_no_recvmsg);
2731
2732 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2733 {
2734 /* Mirror missing mmap method error code */
2735 return -ENODEV;
2736 }
2737 EXPORT_SYMBOL(sock_no_mmap);
2738
2739 /*
2740 * When a file is received (via SCM_RIGHTS, etc), we must bump the
2741 * various sock-based usage counts.
2742 */
2743 void __receive_sock(struct file *file)
2744 {
2745 struct socket *sock;
2746 int error;
2747
2748 /*
2749 * The resulting value of "error" is ignored here since we only
2750 * need to take action when the file is a socket and testing
2751 * "sock" for NULL is sufficient.
2752 */
2753 sock = sock_from_file(file, &error);
2754 if (sock) {
2755 sock_update_netprioidx(&sock->sk->sk_cgrp_data);
2756 sock_update_classid(&sock->sk->sk_cgrp_data);
2757 }
2758 }
2759
2760 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2761 {
2762 ssize_t res;
2763 struct msghdr msg = {.msg_flags = flags};
2764 struct kvec iov;
2765 char *kaddr = kmap(page);
2766 iov.iov_base = kaddr + offset;
2767 iov.iov_len = size;
2768 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2769 kunmap(page);
2770 return res;
2771 }
2772 EXPORT_SYMBOL(sock_no_sendpage);
2773
2774 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2775 int offset, size_t size, int flags)
2776 {
2777 ssize_t res;
2778 struct msghdr msg = {.msg_flags = flags};
2779 struct kvec iov;
2780 char *kaddr = kmap(page);
2781
2782 iov.iov_base = kaddr + offset;
2783 iov.iov_len = size;
2784 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2785 kunmap(page);
2786 return res;
2787 }
2788 EXPORT_SYMBOL(sock_no_sendpage_locked);
2789
2790 /*
2791 * Default Socket Callbacks
2792 */
2793
2794 static void sock_def_wakeup(struct sock *sk)
2795 {
2796 struct socket_wq *wq;
2797
2798 rcu_read_lock();
2799 wq = rcu_dereference(sk->sk_wq);
2800 if (skwq_has_sleeper(wq))
2801 wake_up_interruptible_all(&wq->wait);
2802 rcu_read_unlock();
2803 }
2804
2805 static void sock_def_error_report(struct sock *sk)
2806 {
2807 struct socket_wq *wq;
2808
2809 rcu_read_lock();
2810 wq = rcu_dereference(sk->sk_wq);
2811 if (skwq_has_sleeper(wq))
2812 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2813 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2814 rcu_read_unlock();
2815 }
2816
2817 static void sock_def_readable(struct sock *sk)
2818 {
2819 struct socket_wq *wq;
2820
2821 rcu_read_lock();
2822 wq = rcu_dereference(sk->sk_wq);
2823 if (skwq_has_sleeper(wq))
2824 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2825 EPOLLRDNORM | EPOLLRDBAND);
2826 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2827 rcu_read_unlock();
2828 }
2829
2830 static void sock_def_write_space(struct sock *sk)
2831 {
2832 struct socket_wq *wq;
2833
2834 rcu_read_lock();
2835
2836 /* Do not wake up a writer until he can make "significant"
2837 * progress. --DaveM
2838 */
2839 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2840 wq = rcu_dereference(sk->sk_wq);
2841 if (skwq_has_sleeper(wq))
2842 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2843 EPOLLWRNORM | EPOLLWRBAND);
2844
2845 /* Should agree with poll, otherwise some programs break */
2846 if (sock_writeable(sk))
2847 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2848 }
2849
2850 rcu_read_unlock();
2851 }
2852
2853 static void sock_def_destruct(struct sock *sk)
2854 {
2855 }
2856
2857 void sk_send_sigurg(struct sock *sk)
2858 {
2859 if (sk->sk_socket && sk->sk_socket->file)
2860 if (send_sigurg(&sk->sk_socket->file->f_owner))
2861 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2862 }
2863 EXPORT_SYMBOL(sk_send_sigurg);
2864
2865 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2866 unsigned long expires)
2867 {
2868 if (!mod_timer(timer, expires))
2869 sock_hold(sk);
2870 }
2871 EXPORT_SYMBOL(sk_reset_timer);
2872
2873 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2874 {
2875 if (del_timer(timer))
2876 __sock_put(sk);
2877 }
2878 EXPORT_SYMBOL(sk_stop_timer);
2879
2880 void sock_init_data(struct socket *sock, struct sock *sk)
2881 {
2882 sk_init_common(sk);
2883 sk->sk_send_head = NULL;
2884
2885 timer_setup(&sk->sk_timer, NULL, 0);
2886
2887 sk->sk_allocation = GFP_KERNEL;
2888 sk->sk_rcvbuf = sysctl_rmem_default;
2889 sk->sk_sndbuf = sysctl_wmem_default;
2890 sk->sk_state = TCP_CLOSE;
2891 sk_set_socket(sk, sock);
2892
2893 sock_set_flag(sk, SOCK_ZAPPED);
2894
2895 if (sock) {
2896 sk->sk_type = sock->type;
2897 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2898 sock->sk = sk;
2899 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2900 } else {
2901 RCU_INIT_POINTER(sk->sk_wq, NULL);
2902 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2903 }
2904
2905 rwlock_init(&sk->sk_callback_lock);
2906 if (sk->sk_kern_sock)
2907 lockdep_set_class_and_name(
2908 &sk->sk_callback_lock,
2909 af_kern_callback_keys + sk->sk_family,
2910 af_family_kern_clock_key_strings[sk->sk_family]);
2911 else
2912 lockdep_set_class_and_name(
2913 &sk->sk_callback_lock,
2914 af_callback_keys + sk->sk_family,
2915 af_family_clock_key_strings[sk->sk_family]);
2916
2917 sk->sk_state_change = sock_def_wakeup;
2918 sk->sk_data_ready = sock_def_readable;
2919 sk->sk_write_space = sock_def_write_space;
2920 sk->sk_error_report = sock_def_error_report;
2921 sk->sk_destruct = sock_def_destruct;
2922
2923 sk->sk_frag.page = NULL;
2924 sk->sk_frag.offset = 0;
2925 sk->sk_peek_off = -1;
2926
2927 sk->sk_peer_pid = NULL;
2928 sk->sk_peer_cred = NULL;
2929 sk->sk_write_pending = 0;
2930 sk->sk_rcvlowat = 1;
2931 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2932 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2933
2934 sk->sk_stamp = SK_DEFAULT_STAMP;
2935 #if BITS_PER_LONG==32
2936 seqlock_init(&sk->sk_stamp_seq);
2937 #endif
2938 atomic_set(&sk->sk_zckey, 0);
2939
2940 #ifdef CONFIG_NET_RX_BUSY_POLL
2941 sk->sk_napi_id = 0;
2942 sk->sk_ll_usec = sysctl_net_busy_read;
2943 #endif
2944
2945 sk->sk_max_pacing_rate = ~0UL;
2946 sk->sk_pacing_rate = ~0UL;
2947 WRITE_ONCE(sk->sk_pacing_shift, 10);
2948 sk->sk_incoming_cpu = -1;
2949
2950 sk_rx_queue_clear(sk);
2951 /*
2952 * Before updating sk_refcnt, we must commit prior changes to memory
2953 * (Documentation/RCU/rculist_nulls.txt for details)
2954 */
2955 smp_wmb();
2956 refcount_set(&sk->sk_refcnt, 1);
2957 atomic_set(&sk->sk_drops, 0);
2958 }
2959 EXPORT_SYMBOL(sock_init_data);
2960
2961 void lock_sock_nested(struct sock *sk, int subclass)
2962 {
2963 might_sleep();
2964 spin_lock_bh(&sk->sk_lock.slock);
2965 if (sk->sk_lock.owned)
2966 __lock_sock(sk);
2967 sk->sk_lock.owned = 1;
2968 spin_unlock(&sk->sk_lock.slock);
2969 /*
2970 * The sk_lock has mutex_lock() semantics here:
2971 */
2972 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2973 local_bh_enable();
2974 }
2975 EXPORT_SYMBOL(lock_sock_nested);
2976
2977 void release_sock(struct sock *sk)
2978 {
2979 spin_lock_bh(&sk->sk_lock.slock);
2980 if (sk->sk_backlog.tail)
2981 __release_sock(sk);
2982
2983 /* Warning : release_cb() might need to release sk ownership,
2984 * ie call sock_release_ownership(sk) before us.
2985 */
2986 if (sk->sk_prot->release_cb)
2987 sk->sk_prot->release_cb(sk);
2988
2989 sock_release_ownership(sk);
2990 if (waitqueue_active(&sk->sk_lock.wq))
2991 wake_up(&sk->sk_lock.wq);
2992 spin_unlock_bh(&sk->sk_lock.slock);
2993 }
2994 EXPORT_SYMBOL(release_sock);
2995
2996 /**
2997 * lock_sock_fast - fast version of lock_sock
2998 * @sk: socket
2999 *
3000 * This version should be used for very small section, where process wont block
3001 * return false if fast path is taken:
3002 *
3003 * sk_lock.slock locked, owned = 0, BH disabled
3004 *
3005 * return true if slow path is taken:
3006 *
3007 * sk_lock.slock unlocked, owned = 1, BH enabled
3008 */
3009 bool lock_sock_fast(struct sock *sk)
3010 {
3011 might_sleep();
3012 spin_lock_bh(&sk->sk_lock.slock);
3013
3014 if (!sk->sk_lock.owned)
3015 /*
3016 * Note : We must disable BH
3017 */
3018 return false;
3019
3020 __lock_sock(sk);
3021 sk->sk_lock.owned = 1;
3022 spin_unlock(&sk->sk_lock.slock);
3023 /*
3024 * The sk_lock has mutex_lock() semantics here:
3025 */
3026 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3027 local_bh_enable();
3028 return true;
3029 }
3030 EXPORT_SYMBOL(lock_sock_fast);
3031
3032 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3033 bool timeval, bool time32)
3034 {
3035 struct sock *sk = sock->sk;
3036 struct timespec64 ts;
3037
3038 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3039 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3040 if (ts.tv_sec == -1)
3041 return -ENOENT;
3042 if (ts.tv_sec == 0) {
3043 ktime_t kt = ktime_get_real();
3044 sock_write_timestamp(sk, kt);;
3045 ts = ktime_to_timespec64(kt);
3046 }
3047
3048 if (timeval)
3049 ts.tv_nsec /= 1000;
3050
3051 #ifdef CONFIG_COMPAT_32BIT_TIME
3052 if (time32)
3053 return put_old_timespec32(&ts, userstamp);
3054 #endif
3055 #ifdef CONFIG_SPARC64
3056 /* beware of padding in sparc64 timeval */
3057 if (timeval && !in_compat_syscall()) {
3058 struct __kernel_old_timeval __user tv = {
3059 .tv_sec = ts.tv_sec,
3060 .tv_usec = ts.tv_nsec,
3061 };
3062 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3063 return -EFAULT;
3064 return 0;
3065 }
3066 #endif
3067 return put_timespec64(&ts, userstamp);
3068 }
3069 EXPORT_SYMBOL(sock_gettstamp);
3070
3071 void sock_enable_timestamp(struct sock *sk, int flag)
3072 {
3073 if (!sock_flag(sk, flag)) {
3074 unsigned long previous_flags = sk->sk_flags;
3075
3076 sock_set_flag(sk, flag);
3077 /*
3078 * we just set one of the two flags which require net
3079 * time stamping, but time stamping might have been on
3080 * already because of the other one
3081 */
3082 if (sock_needs_netstamp(sk) &&
3083 !(previous_flags & SK_FLAGS_TIMESTAMP))
3084 net_enable_timestamp();
3085 }
3086 }
3087
3088 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3089 int level, int type)
3090 {
3091 struct sock_exterr_skb *serr;
3092 struct sk_buff *skb;
3093 int copied, err;
3094
3095 err = -EAGAIN;
3096 skb = sock_dequeue_err_skb(sk);
3097 if (skb == NULL)
3098 goto out;
3099
3100 copied = skb->len;
3101 if (copied > len) {
3102 msg->msg_flags |= MSG_TRUNC;
3103 copied = len;
3104 }
3105 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3106 if (err)
3107 goto out_free_skb;
3108
3109 sock_recv_timestamp(msg, sk, skb);
3110
3111 serr = SKB_EXT_ERR(skb);
3112 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3113
3114 msg->msg_flags |= MSG_ERRQUEUE;
3115 err = copied;
3116
3117 out_free_skb:
3118 kfree_skb(skb);
3119 out:
3120 return err;
3121 }
3122 EXPORT_SYMBOL(sock_recv_errqueue);
3123
3124 /*
3125 * Get a socket option on an socket.
3126 *
3127 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3128 * asynchronous errors should be reported by getsockopt. We assume
3129 * this means if you specify SO_ERROR (otherwise whats the point of it).
3130 */
3131 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3132 char __user *optval, int __user *optlen)
3133 {
3134 struct sock *sk = sock->sk;
3135
3136 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3137 }
3138 EXPORT_SYMBOL(sock_common_getsockopt);
3139
3140 #ifdef CONFIG_COMPAT
3141 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3142 char __user *optval, int __user *optlen)
3143 {
3144 struct sock *sk = sock->sk;
3145
3146 if (sk->sk_prot->compat_getsockopt != NULL)
3147 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3148 optval, optlen);
3149 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3150 }
3151 EXPORT_SYMBOL(compat_sock_common_getsockopt);
3152 #endif
3153
3154 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3155 int flags)
3156 {
3157 struct sock *sk = sock->sk;
3158 int addr_len = 0;
3159 int err;
3160
3161 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3162 flags & ~MSG_DONTWAIT, &addr_len);
3163 if (err >= 0)
3164 msg->msg_namelen = addr_len;
3165 return err;
3166 }
3167 EXPORT_SYMBOL(sock_common_recvmsg);
3168
3169 /*
3170 * Set socket options on an inet socket.
3171 */
3172 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3173 char __user *optval, unsigned int optlen)
3174 {
3175 struct sock *sk = sock->sk;
3176
3177 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3178 }
3179 EXPORT_SYMBOL(sock_common_setsockopt);
3180
3181 #ifdef CONFIG_COMPAT
3182 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3183 char __user *optval, unsigned int optlen)
3184 {
3185 struct sock *sk = sock->sk;
3186
3187 if (sk->sk_prot->compat_setsockopt != NULL)
3188 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3189 optval, optlen);
3190 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3191 }
3192 EXPORT_SYMBOL(compat_sock_common_setsockopt);
3193 #endif
3194
3195 void sk_common_release(struct sock *sk)
3196 {
3197 if (sk->sk_prot->destroy)
3198 sk->sk_prot->destroy(sk);
3199
3200 /*
3201 * Observation: when sock_common_release is called, processes have
3202 * no access to socket. But net still has.
3203 * Step one, detach it from networking:
3204 *
3205 * A. Remove from hash tables.
3206 */
3207
3208 sk->sk_prot->unhash(sk);
3209
3210 /*
3211 * In this point socket cannot receive new packets, but it is possible
3212 * that some packets are in flight because some CPU runs receiver and
3213 * did hash table lookup before we unhashed socket. They will achieve
3214 * receive queue and will be purged by socket destructor.
3215 *
3216 * Also we still have packets pending on receive queue and probably,
3217 * our own packets waiting in device queues. sock_destroy will drain
3218 * receive queue, but transmitted packets will delay socket destruction
3219 * until the last reference will be released.
3220 */
3221
3222 sock_orphan(sk);
3223
3224 xfrm_sk_free_policy(sk);
3225
3226 sk_refcnt_debug_release(sk);
3227
3228 sock_put(sk);
3229 }
3230 EXPORT_SYMBOL(sk_common_release);
3231
3232 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3233 {
3234 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3235
3236 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3237 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3238 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3239 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3240 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3241 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3242 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3243 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3244 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3245 }
3246
3247 #ifdef CONFIG_PROC_FS
3248 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
3249 struct prot_inuse {
3250 int val[PROTO_INUSE_NR];
3251 };
3252
3253 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3254
3255 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3256 {
3257 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3258 }
3259 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3260
3261 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3262 {
3263 int cpu, idx = prot->inuse_idx;
3264 int res = 0;
3265
3266 for_each_possible_cpu(cpu)
3267 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3268
3269 return res >= 0 ? res : 0;
3270 }
3271 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3272
3273 static void sock_inuse_add(struct net *net, int val)
3274 {
3275 this_cpu_add(*net->core.sock_inuse, val);
3276 }
3277
3278 int sock_inuse_get(struct net *net)
3279 {
3280 int cpu, res = 0;
3281
3282 for_each_possible_cpu(cpu)
3283 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3284
3285 return res;
3286 }
3287
3288 EXPORT_SYMBOL_GPL(sock_inuse_get);
3289
3290 static int __net_init sock_inuse_init_net(struct net *net)
3291 {
3292 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3293 if (net->core.prot_inuse == NULL)
3294 return -ENOMEM;
3295
3296 net->core.sock_inuse = alloc_percpu(int);
3297 if (net->core.sock_inuse == NULL)
3298 goto out;
3299
3300 return 0;
3301
3302 out:
3303 free_percpu(net->core.prot_inuse);
3304 return -ENOMEM;
3305 }
3306
3307 static void __net_exit sock_inuse_exit_net(struct net *net)
3308 {
3309 free_percpu(net->core.prot_inuse);
3310 free_percpu(net->core.sock_inuse);
3311 }
3312
3313 static struct pernet_operations net_inuse_ops = {
3314 .init = sock_inuse_init_net,
3315 .exit = sock_inuse_exit_net,
3316 };
3317
3318 static __init int net_inuse_init(void)
3319 {
3320 if (register_pernet_subsys(&net_inuse_ops))
3321 panic("Cannot initialize net inuse counters");
3322
3323 return 0;
3324 }
3325
3326 core_initcall(net_inuse_init);
3327
3328 static int assign_proto_idx(struct proto *prot)
3329 {
3330 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3331
3332 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3333 pr_err("PROTO_INUSE_NR exhausted\n");
3334 return -ENOSPC;
3335 }
3336
3337 set_bit(prot->inuse_idx, proto_inuse_idx);
3338 return 0;
3339 }
3340
3341 static void release_proto_idx(struct proto *prot)
3342 {
3343 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3344 clear_bit(prot->inuse_idx, proto_inuse_idx);
3345 }
3346 #else
3347 static inline int assign_proto_idx(struct proto *prot)
3348 {
3349 return 0;
3350 }
3351
3352 static inline void release_proto_idx(struct proto *prot)
3353 {
3354 }
3355
3356 static void sock_inuse_add(struct net *net, int val)
3357 {
3358 }
3359 #endif
3360
3361 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
3362 {
3363 if (!twsk_prot)
3364 return;
3365 kfree(twsk_prot->twsk_slab_name);
3366 twsk_prot->twsk_slab_name = NULL;
3367 kmem_cache_destroy(twsk_prot->twsk_slab);
3368 twsk_prot->twsk_slab = NULL;
3369 }
3370
3371 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3372 {
3373 if (!rsk_prot)
3374 return;
3375 kfree(rsk_prot->slab_name);
3376 rsk_prot->slab_name = NULL;
3377 kmem_cache_destroy(rsk_prot->slab);
3378 rsk_prot->slab = NULL;
3379 }
3380
3381 static int req_prot_init(const struct proto *prot)
3382 {
3383 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3384
3385 if (!rsk_prot)
3386 return 0;
3387
3388 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3389 prot->name);
3390 if (!rsk_prot->slab_name)
3391 return -ENOMEM;
3392
3393 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3394 rsk_prot->obj_size, 0,
3395 SLAB_ACCOUNT | prot->slab_flags,
3396 NULL);
3397
3398 if (!rsk_prot->slab) {
3399 pr_crit("%s: Can't create request sock SLAB cache!\n",
3400 prot->name);
3401 return -ENOMEM;
3402 }
3403 return 0;
3404 }
3405
3406 int proto_register(struct proto *prot, int alloc_slab)
3407 {
3408 int ret = -ENOBUFS;
3409
3410 if (alloc_slab) {
3411 prot->slab = kmem_cache_create_usercopy(prot->name,
3412 prot->obj_size, 0,
3413 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3414 prot->slab_flags,
3415 prot->useroffset, prot->usersize,
3416 NULL);
3417
3418 if (prot->slab == NULL) {
3419 pr_crit("%s: Can't create sock SLAB cache!\n",
3420 prot->name);
3421 goto out;
3422 }
3423
3424 if (req_prot_init(prot))
3425 goto out_free_request_sock_slab;
3426
3427 if (prot->twsk_prot != NULL) {
3428 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3429
3430 if (prot->twsk_prot->twsk_slab_name == NULL)
3431 goto out_free_request_sock_slab;
3432
3433 prot->twsk_prot->twsk_slab =
3434 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3435 prot->twsk_prot->twsk_obj_size,
3436 0,
3437 SLAB_ACCOUNT |
3438 prot->slab_flags,
3439 NULL);
3440 if (prot->twsk_prot->twsk_slab == NULL)
3441 goto out_free_timewait_sock_slab;
3442 }
3443 }
3444
3445 mutex_lock(&proto_list_mutex);
3446 ret = assign_proto_idx(prot);
3447 if (ret) {
3448 mutex_unlock(&proto_list_mutex);
3449 goto out_free_timewait_sock_slab;
3450 }
3451 list_add(&prot->node, &proto_list);
3452 mutex_unlock(&proto_list_mutex);
3453 return ret;
3454
3455 out_free_timewait_sock_slab:
3456 if (alloc_slab && prot->twsk_prot)
3457 tw_prot_cleanup(prot->twsk_prot);
3458 out_free_request_sock_slab:
3459 if (alloc_slab) {
3460 req_prot_cleanup(prot->rsk_prot);
3461
3462 kmem_cache_destroy(prot->slab);
3463 prot->slab = NULL;
3464 }
3465 out:
3466 return ret;
3467 }
3468 EXPORT_SYMBOL(proto_register);
3469
3470 void proto_unregister(struct proto *prot)
3471 {
3472 mutex_lock(&proto_list_mutex);
3473 release_proto_idx(prot);
3474 list_del(&prot->node);
3475 mutex_unlock(&proto_list_mutex);
3476
3477 kmem_cache_destroy(prot->slab);
3478 prot->slab = NULL;
3479
3480 req_prot_cleanup(prot->rsk_prot);
3481 tw_prot_cleanup(prot->twsk_prot);
3482 }
3483 EXPORT_SYMBOL(proto_unregister);
3484
3485 int sock_load_diag_module(int family, int protocol)
3486 {
3487 if (!protocol) {
3488 if (!sock_is_registered(family))
3489 return -ENOENT;
3490
3491 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3492 NETLINK_SOCK_DIAG, family);
3493 }
3494
3495 #ifdef CONFIG_INET
3496 if (family == AF_INET &&
3497 protocol != IPPROTO_RAW &&
3498 !rcu_access_pointer(inet_protos[protocol]))
3499 return -ENOENT;
3500 #endif
3501
3502 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3503 NETLINK_SOCK_DIAG, family, protocol);
3504 }
3505 EXPORT_SYMBOL(sock_load_diag_module);
3506
3507 #ifdef CONFIG_PROC_FS
3508 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3509 __acquires(proto_list_mutex)
3510 {
3511 mutex_lock(&proto_list_mutex);
3512 return seq_list_start_head(&proto_list, *pos);
3513 }
3514
3515 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3516 {
3517 return seq_list_next(v, &proto_list, pos);
3518 }
3519
3520 static void proto_seq_stop(struct seq_file *seq, void *v)
3521 __releases(proto_list_mutex)
3522 {
3523 mutex_unlock(&proto_list_mutex);
3524 }
3525
3526 static char proto_method_implemented(const void *method)
3527 {
3528 return method == NULL ? 'n' : 'y';
3529 }
3530 static long sock_prot_memory_allocated(struct proto *proto)
3531 {
3532 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3533 }
3534
3535 static const char *sock_prot_memory_pressure(struct proto *proto)
3536 {
3537 return proto->memory_pressure != NULL ?
3538 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3539 }
3540
3541 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3542 {
3543
3544 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3545 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3546 proto->name,
3547 proto->obj_size,
3548 sock_prot_inuse_get(seq_file_net(seq), proto),
3549 sock_prot_memory_allocated(proto),
3550 sock_prot_memory_pressure(proto),
3551 proto->max_header,
3552 proto->slab == NULL ? "no" : "yes",
3553 module_name(proto->owner),
3554 proto_method_implemented(proto->close),
3555 proto_method_implemented(proto->connect),
3556 proto_method_implemented(proto->disconnect),
3557 proto_method_implemented(proto->accept),
3558 proto_method_implemented(proto->ioctl),
3559 proto_method_implemented(proto->init),
3560 proto_method_implemented(proto->destroy),
3561 proto_method_implemented(proto->shutdown),
3562 proto_method_implemented(proto->setsockopt),
3563 proto_method_implemented(proto->getsockopt),
3564 proto_method_implemented(proto->sendmsg),
3565 proto_method_implemented(proto->recvmsg),
3566 proto_method_implemented(proto->sendpage),
3567 proto_method_implemented(proto->bind),
3568 proto_method_implemented(proto->backlog_rcv),
3569 proto_method_implemented(proto->hash),
3570 proto_method_implemented(proto->unhash),
3571 proto_method_implemented(proto->get_port),
3572 proto_method_implemented(proto->enter_memory_pressure));
3573 }
3574
3575 static int proto_seq_show(struct seq_file *seq, void *v)
3576 {
3577 if (v == &proto_list)
3578 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3579 "protocol",
3580 "size",
3581 "sockets",
3582 "memory",
3583 "press",
3584 "maxhdr",
3585 "slab",
3586 "module",
3587 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3588 else
3589 proto_seq_printf(seq, list_entry(v, struct proto, node));
3590 return 0;
3591 }
3592
3593 static const struct seq_operations proto_seq_ops = {
3594 .start = proto_seq_start,
3595 .next = proto_seq_next,
3596 .stop = proto_seq_stop,
3597 .show = proto_seq_show,
3598 };
3599
3600 static __net_init int proto_init_net(struct net *net)
3601 {
3602 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3603 sizeof(struct seq_net_private)))
3604 return -ENOMEM;
3605
3606 return 0;
3607 }
3608
3609 static __net_exit void proto_exit_net(struct net *net)
3610 {
3611 remove_proc_entry("protocols", net->proc_net);
3612 }
3613
3614
3615 static __net_initdata struct pernet_operations proto_net_ops = {
3616 .init = proto_init_net,
3617 .exit = proto_exit_net,
3618 };
3619
3620 static int __init proto_init(void)
3621 {
3622 return register_pernet_subsys(&proto_net_ops);
3623 }
3624
3625 subsys_initcall(proto_init);
3626
3627 #endif /* PROC_FS */
3628
3629 #ifdef CONFIG_NET_RX_BUSY_POLL
3630 bool sk_busy_loop_end(void *p, unsigned long start_time)
3631 {
3632 struct sock *sk = p;
3633
3634 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3635 sk_busy_loop_timeout(sk, start_time);
3636 }
3637 EXPORT_SYMBOL(sk_busy_loop_end);
3638 #endif /* CONFIG_NET_RX_BUSY_POLL */
Linux with added FPC-III support