Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94#include <linux/capability.h>
95#include <linux/errno.h>
96#include <linux/types.h>
97#include <linux/socket.h>
98#include <linux/in.h>
99#include <linux/kernel.h>
100#include <linux/module.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
103#include <linux/sched.h>
104#include <linux/timer.h>
105#include <linux/string.h>
106#include <linux/sockios.h>
107#include <linux/net.h>
108#include <linux/mm.h>
109#include <linux/slab.h>
110#include <linux/interrupt.h>
111#include <linux/poll.h>
112#include <linux/tcp.h>
113#include <linux/init.h>
114#include <linux/highmem.h>
115#include <linux/user_namespace.h>
116#include <linux/static_key.h>
117#include <linux/memcontrol.h>
118#include <linux/prefetch.h>
119
120#include <asm/uaccess.h>
121
122#include <linux/netdevice.h>
123#include <net/protocol.h>
124#include <linux/skbuff.h>
125#include <net/net_namespace.h>
126#include <net/request_sock.h>
127#include <net/sock.h>
128#include <linux/net_tstamp.h>
129#include <net/xfrm.h>
130#include <linux/ipsec.h>
131#include <net/cls_cgroup.h>
132#include <net/netprio_cgroup.h>
133
134#include <linux/filter.h>
135
136#include <trace/events/sock.h>
137
138#ifdef CONFIG_INET
139#include <net/tcp.h>
140#endif
141
142static DEFINE_MUTEX(proto_list_mutex);
143static LIST_HEAD(proto_list);
144
145#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
146int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
147{
148 struct proto *proto;
149 int ret = 0;
150
151 mutex_lock(&proto_list_mutex);
152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
154 ret = proto->init_cgroup(memcg, ss);
155 if (ret)
156 goto out;
157 }
158 }
159
160 mutex_unlock(&proto_list_mutex);
161 return ret;
162out:
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
165 proto->destroy_cgroup(memcg);
166 mutex_unlock(&proto_list_mutex);
167 return ret;
168}
169
170void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
171{
172 struct proto *proto;
173
174 mutex_lock(&proto_list_mutex);
175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
177 proto->destroy_cgroup(memcg);
178 mutex_unlock(&proto_list_mutex);
179}
180#endif
181
182/*
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
185 */
186static struct lock_class_key af_family_keys[AF_MAX];
187static struct lock_class_key af_family_slock_keys[AF_MAX];
188
189struct static_key memcg_socket_limit_enabled;
190EXPORT_SYMBOL(memcg_socket_limit_enabled);
191
192/*
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
195 * locks is fast):
196 */
197static const char *const af_family_key_strings[AF_MAX+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
212};
213static const char *const af_family_slock_key_strings[AF_MAX+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
228};
229static const char *const af_family_clock_key_strings[AF_MAX+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
244};
245
246/*
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
249 */
250static struct lock_class_key af_callback_keys[AF_MAX];
251
252/* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
256 */
257#define _SK_MEM_PACKETS 256
258#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
261
262/* Run time adjustable parameters. */
263__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
264EXPORT_SYMBOL(sysctl_wmem_max);
265__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
266EXPORT_SYMBOL(sysctl_rmem_max);
267__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
269
270/* Maximal space eaten by iovec or ancillary data plus some space */
271int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
272EXPORT_SYMBOL(sysctl_optmem_max);
273
274#if defined(CONFIG_CGROUPS)
275#if !defined(CONFIG_NET_CLS_CGROUP)
276int net_cls_subsys_id = -1;
277EXPORT_SYMBOL_GPL(net_cls_subsys_id);
278#endif
279#if !defined(CONFIG_NETPRIO_CGROUP)
280int net_prio_subsys_id = -1;
281EXPORT_SYMBOL_GPL(net_prio_subsys_id);
282#endif
283#endif
284
285static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
286{
287 struct timeval tv;
288
289 if (optlen < sizeof(tv))
290 return -EINVAL;
291 if (copy_from_user(&tv, optval, sizeof(tv)))
292 return -EFAULT;
293 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
294 return -EDOM;
295
296 if (tv.tv_sec < 0) {
297 static int warned __read_mostly;
298
299 *timeo_p = 0;
300 if (warned < 10 && net_ratelimit()) {
301 warned++;
302 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
303 __func__, current->comm, task_pid_nr(current));
304 }
305 return 0;
306 }
307 *timeo_p = MAX_SCHEDULE_TIMEOUT;
308 if (tv.tv_sec == 0 && tv.tv_usec == 0)
309 return 0;
310 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
311 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
312 return 0;
313}
314
315static void sock_warn_obsolete_bsdism(const char *name)
316{
317 static int warned;
318 static char warncomm[TASK_COMM_LEN];
319 if (strcmp(warncomm, current->comm) && warned < 5) {
320 strcpy(warncomm, current->comm);
321 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
322 warncomm, name);
323 warned++;
324 }
325}
326
327#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
328
329static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
330{
331 if (sk->sk_flags & flags) {
332 sk->sk_flags &= ~flags;
333 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
334 net_disable_timestamp();
335 }
336}
337
338
339int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
340{
341 int err;
342 int skb_len;
343 unsigned long flags;
344 struct sk_buff_head *list = &sk->sk_receive_queue;
345
346 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
347 atomic_inc(&sk->sk_drops);
348 trace_sock_rcvqueue_full(sk, skb);
349 return -ENOMEM;
350 }
351
352 err = sk_filter(sk, skb);
353 if (err)
354 return err;
355
356 if (!sk_rmem_schedule(sk, skb->truesize)) {
357 atomic_inc(&sk->sk_drops);
358 return -ENOBUFS;
359 }
360
361 skb->dev = NULL;
362 skb_set_owner_r(skb, sk);
363
364 /* Cache the SKB length before we tack it onto the receive
365 * queue. Once it is added it no longer belongs to us and
366 * may be freed by other threads of control pulling packets
367 * from the queue.
368 */
369 skb_len = skb->len;
370
371 /* we escape from rcu protected region, make sure we dont leak
372 * a norefcounted dst
373 */
374 skb_dst_force(skb);
375
376 spin_lock_irqsave(&list->lock, flags);
377 skb->dropcount = atomic_read(&sk->sk_drops);
378 __skb_queue_tail(list, skb);
379 spin_unlock_irqrestore(&list->lock, flags);
380
381 if (!sock_flag(sk, SOCK_DEAD))
382 sk->sk_data_ready(sk, skb_len);
383 return 0;
384}
385EXPORT_SYMBOL(sock_queue_rcv_skb);
386
387int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
388{
389 int rc = NET_RX_SUCCESS;
390
391 if (sk_filter(sk, skb))
392 goto discard_and_relse;
393
394 skb->dev = NULL;
395
396 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
397 atomic_inc(&sk->sk_drops);
398 goto discard_and_relse;
399 }
400 if (nested)
401 bh_lock_sock_nested(sk);
402 else
403 bh_lock_sock(sk);
404 if (!sock_owned_by_user(sk)) {
405 /*
406 * trylock + unlock semantics:
407 */
408 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
409
410 rc = sk_backlog_rcv(sk, skb);
411
412 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
413 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
414 bh_unlock_sock(sk);
415 atomic_inc(&sk->sk_drops);
416 goto discard_and_relse;
417 }
418
419 bh_unlock_sock(sk);
420out:
421 sock_put(sk);
422 return rc;
423discard_and_relse:
424 kfree_skb(skb);
425 goto out;
426}
427EXPORT_SYMBOL(sk_receive_skb);
428
429void sk_reset_txq(struct sock *sk)
430{
431 sk_tx_queue_clear(sk);
432}
433EXPORT_SYMBOL(sk_reset_txq);
434
435struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
436{
437 struct dst_entry *dst = __sk_dst_get(sk);
438
439 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
440 sk_tx_queue_clear(sk);
441 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
442 dst_release(dst);
443 return NULL;
444 }
445
446 return dst;
447}
448EXPORT_SYMBOL(__sk_dst_check);
449
450struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
451{
452 struct dst_entry *dst = sk_dst_get(sk);
453
454 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
455 sk_dst_reset(sk);
456 dst_release(dst);
457 return NULL;
458 }
459
460 return dst;
461}
462EXPORT_SYMBOL(sk_dst_check);
463
464static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
465{
466 int ret = -ENOPROTOOPT;
467#ifdef CONFIG_NETDEVICES
468 struct net *net = sock_net(sk);
469 char devname[IFNAMSIZ];
470 int index;
471
472 /* Sorry... */
473 ret = -EPERM;
474 if (!capable(CAP_NET_RAW))
475 goto out;
476
477 ret = -EINVAL;
478 if (optlen < 0)
479 goto out;
480
481 /* Bind this socket to a particular device like "eth0",
482 * as specified in the passed interface name. If the
483 * name is "" or the option length is zero the socket
484 * is not bound.
485 */
486 if (optlen > IFNAMSIZ - 1)
487 optlen = IFNAMSIZ - 1;
488 memset(devname, 0, sizeof(devname));
489
490 ret = -EFAULT;
491 if (copy_from_user(devname, optval, optlen))
492 goto out;
493
494 index = 0;
495 if (devname[0] != '\0') {
496 struct net_device *dev;
497
498 rcu_read_lock();
499 dev = dev_get_by_name_rcu(net, devname);
500 if (dev)
501 index = dev->ifindex;
502 rcu_read_unlock();
503 ret = -ENODEV;
504 if (!dev)
505 goto out;
506 }
507
508 lock_sock(sk);
509 sk->sk_bound_dev_if = index;
510 sk_dst_reset(sk);
511 release_sock(sk);
512
513 ret = 0;
514
515out:
516#endif
517
518 return ret;
519}
520
521static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
522{
523 if (valbool)
524 sock_set_flag(sk, bit);
525 else
526 sock_reset_flag(sk, bit);
527}
528
529/*
530 * This is meant for all protocols to use and covers goings on
531 * at the socket level. Everything here is generic.
532 */
533
534int sock_setsockopt(struct socket *sock, int level, int optname,
535 char __user *optval, unsigned int optlen)
536{
537 struct sock *sk = sock->sk;
538 int val;
539 int valbool;
540 struct linger ling;
541 int ret = 0;
542
543 /*
544 * Options without arguments
545 */
546
547 if (optname == SO_BINDTODEVICE)
548 return sock_bindtodevice(sk, optval, optlen);
549
550 if (optlen < sizeof(int))
551 return -EINVAL;
552
553 if (get_user(val, (int __user *)optval))
554 return -EFAULT;
555
556 valbool = val ? 1 : 0;
557
558 lock_sock(sk);
559
560 switch (optname) {
561 case SO_DEBUG:
562 if (val && !capable(CAP_NET_ADMIN))
563 ret = -EACCES;
564 else
565 sock_valbool_flag(sk, SOCK_DBG, valbool);
566 break;
567 case SO_REUSEADDR:
568 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
569 break;
570 case SO_TYPE:
571 case SO_PROTOCOL:
572 case SO_DOMAIN:
573 case SO_ERROR:
574 ret = -ENOPROTOOPT;
575 break;
576 case SO_DONTROUTE:
577 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
578 break;
579 case SO_BROADCAST:
580 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
581 break;
582 case SO_SNDBUF:
583 /* Don't error on this BSD doesn't and if you think
584 * about it this is right. Otherwise apps have to
585 * play 'guess the biggest size' games. RCVBUF/SNDBUF
586 * are treated in BSD as hints
587 */
588 val = min_t(u32, val, sysctl_wmem_max);
589set_sndbuf:
590 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
591 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
592 /* Wake up sending tasks if we upped the value. */
593 sk->sk_write_space(sk);
594 break;
595
596 case SO_SNDBUFFORCE:
597 if (!capable(CAP_NET_ADMIN)) {
598 ret = -EPERM;
599 break;
600 }
601 goto set_sndbuf;
602
603 case SO_RCVBUF:
604 /* Don't error on this BSD doesn't and if you think
605 * about it this is right. Otherwise apps have to
606 * play 'guess the biggest size' games. RCVBUF/SNDBUF
607 * are treated in BSD as hints
608 */
609 val = min_t(u32, val, sysctl_rmem_max);
610set_rcvbuf:
611 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
612 /*
613 * We double it on the way in to account for
614 * "struct sk_buff" etc. overhead. Applications
615 * assume that the SO_RCVBUF setting they make will
616 * allow that much actual data to be received on that
617 * socket.
618 *
619 * Applications are unaware that "struct sk_buff" and
620 * other overheads allocate from the receive buffer
621 * during socket buffer allocation.
622 *
623 * And after considering the possible alternatives,
624 * returning the value we actually used in getsockopt
625 * is the most desirable behavior.
626 */
627 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
628 break;
629
630 case SO_RCVBUFFORCE:
631 if (!capable(CAP_NET_ADMIN)) {
632 ret = -EPERM;
633 break;
634 }
635 goto set_rcvbuf;
636
637 case SO_KEEPALIVE:
638#ifdef CONFIG_INET
639 if (sk->sk_protocol == IPPROTO_TCP)
640 tcp_set_keepalive(sk, valbool);
641#endif
642 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
643 break;
644
645 case SO_OOBINLINE:
646 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
647 break;
648
649 case SO_NO_CHECK:
650 sk->sk_no_check = valbool;
651 break;
652
653 case SO_PRIORITY:
654 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
655 sk->sk_priority = val;
656 else
657 ret = -EPERM;
658 break;
659
660 case SO_LINGER:
661 if (optlen < sizeof(ling)) {
662 ret = -EINVAL; /* 1003.1g */
663 break;
664 }
665 if (copy_from_user(&ling, optval, sizeof(ling))) {
666 ret = -EFAULT;
667 break;
668 }
669 if (!ling.l_onoff)
670 sock_reset_flag(sk, SOCK_LINGER);
671 else {
672#if (BITS_PER_LONG == 32)
673 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
674 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
675 else
676#endif
677 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
678 sock_set_flag(sk, SOCK_LINGER);
679 }
680 break;
681
682 case SO_BSDCOMPAT:
683 sock_warn_obsolete_bsdism("setsockopt");
684 break;
685
686 case SO_PASSCRED:
687 if (valbool)
688 set_bit(SOCK_PASSCRED, &sock->flags);
689 else
690 clear_bit(SOCK_PASSCRED, &sock->flags);
691 break;
692
693 case SO_TIMESTAMP:
694 case SO_TIMESTAMPNS:
695 if (valbool) {
696 if (optname == SO_TIMESTAMP)
697 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
698 else
699 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
700 sock_set_flag(sk, SOCK_RCVTSTAMP);
701 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
702 } else {
703 sock_reset_flag(sk, SOCK_RCVTSTAMP);
704 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
705 }
706 break;
707
708 case SO_TIMESTAMPING:
709 if (val & ~SOF_TIMESTAMPING_MASK) {
710 ret = -EINVAL;
711 break;
712 }
713 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
714 val & SOF_TIMESTAMPING_TX_HARDWARE);
715 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
716 val & SOF_TIMESTAMPING_TX_SOFTWARE);
717 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
718 val & SOF_TIMESTAMPING_RX_HARDWARE);
719 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
720 sock_enable_timestamp(sk,
721 SOCK_TIMESTAMPING_RX_SOFTWARE);
722 else
723 sock_disable_timestamp(sk,
724 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
725 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
726 val & SOF_TIMESTAMPING_SOFTWARE);
727 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
728 val & SOF_TIMESTAMPING_SYS_HARDWARE);
729 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
730 val & SOF_TIMESTAMPING_RAW_HARDWARE);
731 break;
732
733 case SO_RCVLOWAT:
734 if (val < 0)
735 val = INT_MAX;
736 sk->sk_rcvlowat = val ? : 1;
737 break;
738
739 case SO_RCVTIMEO:
740 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
741 break;
742
743 case SO_SNDTIMEO:
744 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
745 break;
746
747 case SO_ATTACH_FILTER:
748 ret = -EINVAL;
749 if (optlen == sizeof(struct sock_fprog)) {
750 struct sock_fprog fprog;
751
752 ret = -EFAULT;
753 if (copy_from_user(&fprog, optval, sizeof(fprog)))
754 break;
755
756 ret = sk_attach_filter(&fprog, sk);
757 }
758 break;
759
760 case SO_DETACH_FILTER:
761 ret = sk_detach_filter(sk);
762 break;
763
764 case SO_PASSSEC:
765 if (valbool)
766 set_bit(SOCK_PASSSEC, &sock->flags);
767 else
768 clear_bit(SOCK_PASSSEC, &sock->flags);
769 break;
770 case SO_MARK:
771 if (!capable(CAP_NET_ADMIN))
772 ret = -EPERM;
773 else
774 sk->sk_mark = val;
775 break;
776
777 /* We implement the SO_SNDLOWAT etc to
778 not be settable (1003.1g 5.3) */
779 case SO_RXQ_OVFL:
780 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
781 break;
782
783 case SO_WIFI_STATUS:
784 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
785 break;
786
787 case SO_PEEK_OFF:
788 if (sock->ops->set_peek_off)
789 sock->ops->set_peek_off(sk, val);
790 else
791 ret = -EOPNOTSUPP;
792 break;
793
794 case SO_NOFCS:
795 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
796 break;
797
798 default:
799 ret = -ENOPROTOOPT;
800 break;
801 }
802 release_sock(sk);
803 return ret;
804}
805EXPORT_SYMBOL(sock_setsockopt);
806
807
808void cred_to_ucred(struct pid *pid, const struct cred *cred,
809 struct ucred *ucred)
810{
811 ucred->pid = pid_vnr(pid);
812 ucred->uid = ucred->gid = -1;
813 if (cred) {
814 struct user_namespace *current_ns = current_user_ns();
815
816 ucred->uid = from_kuid(current_ns, cred->euid);
817 ucred->gid = from_kgid(current_ns, cred->egid);
818 }
819}
820EXPORT_SYMBOL_GPL(cred_to_ucred);
821
822int sock_getsockopt(struct socket *sock, int level, int optname,
823 char __user *optval, int __user *optlen)
824{
825 struct sock *sk = sock->sk;
826
827 union {
828 int val;
829 struct linger ling;
830 struct timeval tm;
831 } v;
832
833 int lv = sizeof(int);
834 int len;
835
836 if (get_user(len, optlen))
837 return -EFAULT;
838 if (len < 0)
839 return -EINVAL;
840
841 memset(&v, 0, sizeof(v));
842
843 switch (optname) {
844 case SO_DEBUG:
845 v.val = sock_flag(sk, SOCK_DBG);
846 break;
847
848 case SO_DONTROUTE:
849 v.val = sock_flag(sk, SOCK_LOCALROUTE);
850 break;
851
852 case SO_BROADCAST:
853 v.val = sock_flag(sk, SOCK_BROADCAST);
854 break;
855
856 case SO_SNDBUF:
857 v.val = sk->sk_sndbuf;
858 break;
859
860 case SO_RCVBUF:
861 v.val = sk->sk_rcvbuf;
862 break;
863
864 case SO_REUSEADDR:
865 v.val = sk->sk_reuse;
866 break;
867
868 case SO_KEEPALIVE:
869 v.val = sock_flag(sk, SOCK_KEEPOPEN);
870 break;
871
872 case SO_TYPE:
873 v.val = sk->sk_type;
874 break;
875
876 case SO_PROTOCOL:
877 v.val = sk->sk_protocol;
878 break;
879
880 case SO_DOMAIN:
881 v.val = sk->sk_family;
882 break;
883
884 case SO_ERROR:
885 v.val = -sock_error(sk);
886 if (v.val == 0)
887 v.val = xchg(&sk->sk_err_soft, 0);
888 break;
889
890 case SO_OOBINLINE:
891 v.val = sock_flag(sk, SOCK_URGINLINE);
892 break;
893
894 case SO_NO_CHECK:
895 v.val = sk->sk_no_check;
896 break;
897
898 case SO_PRIORITY:
899 v.val = sk->sk_priority;
900 break;
901
902 case SO_LINGER:
903 lv = sizeof(v.ling);
904 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
905 v.ling.l_linger = sk->sk_lingertime / HZ;
906 break;
907
908 case SO_BSDCOMPAT:
909 sock_warn_obsolete_bsdism("getsockopt");
910 break;
911
912 case SO_TIMESTAMP:
913 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
914 !sock_flag(sk, SOCK_RCVTSTAMPNS);
915 break;
916
917 case SO_TIMESTAMPNS:
918 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
919 break;
920
921 case SO_TIMESTAMPING:
922 v.val = 0;
923 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
924 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
925 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
926 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
927 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
928 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
929 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
930 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
931 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
932 v.val |= SOF_TIMESTAMPING_SOFTWARE;
933 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
934 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
935 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
936 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
937 break;
938
939 case SO_RCVTIMEO:
940 lv = sizeof(struct timeval);
941 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
942 v.tm.tv_sec = 0;
943 v.tm.tv_usec = 0;
944 } else {
945 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
946 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
947 }
948 break;
949
950 case SO_SNDTIMEO:
951 lv = sizeof(struct timeval);
952 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
953 v.tm.tv_sec = 0;
954 v.tm.tv_usec = 0;
955 } else {
956 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
957 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
958 }
959 break;
960
961 case SO_RCVLOWAT:
962 v.val = sk->sk_rcvlowat;
963 break;
964
965 case SO_SNDLOWAT:
966 v.val = 1;
967 break;
968
969 case SO_PASSCRED:
970 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
971 break;
972
973 case SO_PEERCRED:
974 {
975 struct ucred peercred;
976 if (len > sizeof(peercred))
977 len = sizeof(peercred);
978 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
979 if (copy_to_user(optval, &peercred, len))
980 return -EFAULT;
981 goto lenout;
982 }
983
984 case SO_PEERNAME:
985 {
986 char address[128];
987
988 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
989 return -ENOTCONN;
990 if (lv < len)
991 return -EINVAL;
992 if (copy_to_user(optval, address, len))
993 return -EFAULT;
994 goto lenout;
995 }
996
997 /* Dubious BSD thing... Probably nobody even uses it, but
998 * the UNIX standard wants it for whatever reason... -DaveM
999 */
1000 case SO_ACCEPTCONN:
1001 v.val = sk->sk_state == TCP_LISTEN;
1002 break;
1003
1004 case SO_PASSSEC:
1005 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1006 break;
1007
1008 case SO_PEERSEC:
1009 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1010
1011 case SO_MARK:
1012 v.val = sk->sk_mark;
1013 break;
1014
1015 case SO_RXQ_OVFL:
1016 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1017 break;
1018
1019 case SO_WIFI_STATUS:
1020 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1021 break;
1022
1023 case SO_PEEK_OFF:
1024 if (!sock->ops->set_peek_off)
1025 return -EOPNOTSUPP;
1026
1027 v.val = sk->sk_peek_off;
1028 break;
1029 case SO_NOFCS:
1030 v.val = sock_flag(sk, SOCK_NOFCS);
1031 break;
1032 default:
1033 return -ENOPROTOOPT;
1034 }
1035
1036 if (len > lv)
1037 len = lv;
1038 if (copy_to_user(optval, &v, len))
1039 return -EFAULT;
1040lenout:
1041 if (put_user(len, optlen))
1042 return -EFAULT;
1043 return 0;
1044}
1045
1046/*
1047 * Initialize an sk_lock.
1048 *
1049 * (We also register the sk_lock with the lock validator.)
1050 */
1051static inline void sock_lock_init(struct sock *sk)
1052{
1053 sock_lock_init_class_and_name(sk,
1054 af_family_slock_key_strings[sk->sk_family],
1055 af_family_slock_keys + sk->sk_family,
1056 af_family_key_strings[sk->sk_family],
1057 af_family_keys + sk->sk_family);
1058}
1059
1060/*
1061 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1062 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1063 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1064 */
1065static void sock_copy(struct sock *nsk, const struct sock *osk)
1066{
1067#ifdef CONFIG_SECURITY_NETWORK
1068 void *sptr = nsk->sk_security;
1069#endif
1070 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1071
1072 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1073 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1074
1075#ifdef CONFIG_SECURITY_NETWORK
1076 nsk->sk_security = sptr;
1077 security_sk_clone(osk, nsk);
1078#endif
1079}
1080
1081/*
1082 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1083 * un-modified. Special care is taken when initializing object to zero.
1084 */
1085static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1086{
1087 if (offsetof(struct sock, sk_node.next) != 0)
1088 memset(sk, 0, offsetof(struct sock, sk_node.next));
1089 memset(&sk->sk_node.pprev, 0,
1090 size - offsetof(struct sock, sk_node.pprev));
1091}
1092
1093void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1094{
1095 unsigned long nulls1, nulls2;
1096
1097 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1098 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1099 if (nulls1 > nulls2)
1100 swap(nulls1, nulls2);
1101
1102 if (nulls1 != 0)
1103 memset((char *)sk, 0, nulls1);
1104 memset((char *)sk + nulls1 + sizeof(void *), 0,
1105 nulls2 - nulls1 - sizeof(void *));
1106 memset((char *)sk + nulls2 + sizeof(void *), 0,
1107 size - nulls2 - sizeof(void *));
1108}
1109EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1110
1111static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1112 int family)
1113{
1114 struct sock *sk;
1115 struct kmem_cache *slab;
1116
1117 slab = prot->slab;
1118 if (slab != NULL) {
1119 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1120 if (!sk)
1121 return sk;
1122 if (priority & __GFP_ZERO) {
1123 if (prot->clear_sk)
1124 prot->clear_sk(sk, prot->obj_size);
1125 else
1126 sk_prot_clear_nulls(sk, prot->obj_size);
1127 }
1128 } else
1129 sk = kmalloc(prot->obj_size, priority);
1130
1131 if (sk != NULL) {
1132 kmemcheck_annotate_bitfield(sk, flags);
1133
1134 if (security_sk_alloc(sk, family, priority))
1135 goto out_free;
1136
1137 if (!try_module_get(prot->owner))
1138 goto out_free_sec;
1139 sk_tx_queue_clear(sk);
1140 }
1141
1142 return sk;
1143
1144out_free_sec:
1145 security_sk_free(sk);
1146out_free:
1147 if (slab != NULL)
1148 kmem_cache_free(slab, sk);
1149 else
1150 kfree(sk);
1151 return NULL;
1152}
1153
1154static void sk_prot_free(struct proto *prot, struct sock *sk)
1155{
1156 struct kmem_cache *slab;
1157 struct module *owner;
1158
1159 owner = prot->owner;
1160 slab = prot->slab;
1161
1162 security_sk_free(sk);
1163 if (slab != NULL)
1164 kmem_cache_free(slab, sk);
1165 else
1166 kfree(sk);
1167 module_put(owner);
1168}
1169
1170#ifdef CONFIG_CGROUPS
1171void sock_update_classid(struct sock *sk)
1172{
1173 u32 classid;
1174
1175 rcu_read_lock(); /* doing current task, which cannot vanish. */
1176 classid = task_cls_classid(current);
1177 rcu_read_unlock();
1178 if (classid && classid != sk->sk_classid)
1179 sk->sk_classid = classid;
1180}
1181EXPORT_SYMBOL(sock_update_classid);
1182
1183void sock_update_netprioidx(struct sock *sk)
1184{
1185 if (in_interrupt())
1186 return;
1187
1188 sk->sk_cgrp_prioidx = task_netprioidx(current);
1189}
1190EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1191#endif
1192
1193/**
1194 * sk_alloc - All socket objects are allocated here
1195 * @net: the applicable net namespace
1196 * @family: protocol family
1197 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1198 * @prot: struct proto associated with this new sock instance
1199 */
1200struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1201 struct proto *prot)
1202{
1203 struct sock *sk;
1204
1205 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1206 if (sk) {
1207 sk->sk_family = family;
1208 /*
1209 * See comment in struct sock definition to understand
1210 * why we need sk_prot_creator -acme
1211 */
1212 sk->sk_prot = sk->sk_prot_creator = prot;
1213 sock_lock_init(sk);
1214 sock_net_set(sk, get_net(net));
1215 atomic_set(&sk->sk_wmem_alloc, 1);
1216
1217 sock_update_classid(sk);
1218 sock_update_netprioidx(sk);
1219 }
1220
1221 return sk;
1222}
1223EXPORT_SYMBOL(sk_alloc);
1224
1225static void __sk_free(struct sock *sk)
1226{
1227 struct sk_filter *filter;
1228
1229 if (sk->sk_destruct)
1230 sk->sk_destruct(sk);
1231
1232 filter = rcu_dereference_check(sk->sk_filter,
1233 atomic_read(&sk->sk_wmem_alloc) == 0);
1234 if (filter) {
1235 sk_filter_uncharge(sk, filter);
1236 RCU_INIT_POINTER(sk->sk_filter, NULL);
1237 }
1238
1239 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1240
1241 if (atomic_read(&sk->sk_omem_alloc))
1242 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1243 __func__, atomic_read(&sk->sk_omem_alloc));
1244
1245 if (sk->sk_peer_cred)
1246 put_cred(sk->sk_peer_cred);
1247 put_pid(sk->sk_peer_pid);
1248 put_net(sock_net(sk));
1249 sk_prot_free(sk->sk_prot_creator, sk);
1250}
1251
1252void sk_free(struct sock *sk)
1253{
1254 /*
1255 * We subtract one from sk_wmem_alloc and can know if
1256 * some packets are still in some tx queue.
1257 * If not null, sock_wfree() will call __sk_free(sk) later
1258 */
1259 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1260 __sk_free(sk);
1261}
1262EXPORT_SYMBOL(sk_free);
1263
1264/*
1265 * Last sock_put should drop reference to sk->sk_net. It has already
1266 * been dropped in sk_change_net. Taking reference to stopping namespace
1267 * is not an option.
1268 * Take reference to a socket to remove it from hash _alive_ and after that
1269 * destroy it in the context of init_net.
1270 */
1271void sk_release_kernel(struct sock *sk)
1272{
1273 if (sk == NULL || sk->sk_socket == NULL)
1274 return;
1275
1276 sock_hold(sk);
1277 sock_release(sk->sk_socket);
1278 release_net(sock_net(sk));
1279 sock_net_set(sk, get_net(&init_net));
1280 sock_put(sk);
1281}
1282EXPORT_SYMBOL(sk_release_kernel);
1283
1284static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1285{
1286 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1287 sock_update_memcg(newsk);
1288}
1289
1290/**
1291 * sk_clone_lock - clone a socket, and lock its clone
1292 * @sk: the socket to clone
1293 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1294 *
1295 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1296 */
1297struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1298{
1299 struct sock *newsk;
1300
1301 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1302 if (newsk != NULL) {
1303 struct sk_filter *filter;
1304
1305 sock_copy(newsk, sk);
1306
1307 /* SANITY */
1308 get_net(sock_net(newsk));
1309 sk_node_init(&newsk->sk_node);
1310 sock_lock_init(newsk);
1311 bh_lock_sock(newsk);
1312 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1313 newsk->sk_backlog.len = 0;
1314
1315 atomic_set(&newsk->sk_rmem_alloc, 0);
1316 /*
1317 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1318 */
1319 atomic_set(&newsk->sk_wmem_alloc, 1);
1320 atomic_set(&newsk->sk_omem_alloc, 0);
1321 skb_queue_head_init(&newsk->sk_receive_queue);
1322 skb_queue_head_init(&newsk->sk_write_queue);
1323#ifdef CONFIG_NET_DMA
1324 skb_queue_head_init(&newsk->sk_async_wait_queue);
1325#endif
1326
1327 spin_lock_init(&newsk->sk_dst_lock);
1328 rwlock_init(&newsk->sk_callback_lock);
1329 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1330 af_callback_keys + newsk->sk_family,
1331 af_family_clock_key_strings[newsk->sk_family]);
1332
1333 newsk->sk_dst_cache = NULL;
1334 newsk->sk_wmem_queued = 0;
1335 newsk->sk_forward_alloc = 0;
1336 newsk->sk_send_head = NULL;
1337 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1338
1339 sock_reset_flag(newsk, SOCK_DONE);
1340 skb_queue_head_init(&newsk->sk_error_queue);
1341
1342 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1343 if (filter != NULL)
1344 sk_filter_charge(newsk, filter);
1345
1346 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1347 /* It is still raw copy of parent, so invalidate
1348 * destructor and make plain sk_free() */
1349 newsk->sk_destruct = NULL;
1350 bh_unlock_sock(newsk);
1351 sk_free(newsk);
1352 newsk = NULL;
1353 goto out;
1354 }
1355
1356 newsk->sk_err = 0;
1357 newsk->sk_priority = 0;
1358 /*
1359 * Before updating sk_refcnt, we must commit prior changes to memory
1360 * (Documentation/RCU/rculist_nulls.txt for details)
1361 */
1362 smp_wmb();
1363 atomic_set(&newsk->sk_refcnt, 2);
1364
1365 /*
1366 * Increment the counter in the same struct proto as the master
1367 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1368 * is the same as sk->sk_prot->socks, as this field was copied
1369 * with memcpy).
1370 *
1371 * This _changes_ the previous behaviour, where
1372 * tcp_create_openreq_child always was incrementing the
1373 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1374 * to be taken into account in all callers. -acme
1375 */
1376 sk_refcnt_debug_inc(newsk);
1377 sk_set_socket(newsk, NULL);
1378 newsk->sk_wq = NULL;
1379
1380 sk_update_clone(sk, newsk);
1381
1382 if (newsk->sk_prot->sockets_allocated)
1383 sk_sockets_allocated_inc(newsk);
1384
1385 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1386 net_enable_timestamp();
1387 }
1388out:
1389 return newsk;
1390}
1391EXPORT_SYMBOL_GPL(sk_clone_lock);
1392
1393void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1394{
1395 __sk_dst_set(sk, dst);
1396 sk->sk_route_caps = dst->dev->features;
1397 if (sk->sk_route_caps & NETIF_F_GSO)
1398 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1399 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1400 if (sk_can_gso(sk)) {
1401 if (dst->header_len) {
1402 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1403 } else {
1404 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1405 sk->sk_gso_max_size = dst->dev->gso_max_size;
1406 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1407 }
1408 }
1409}
1410EXPORT_SYMBOL_GPL(sk_setup_caps);
1411
1412void __init sk_init(void)
1413{
1414 if (totalram_pages <= 4096) {
1415 sysctl_wmem_max = 32767;
1416 sysctl_rmem_max = 32767;
1417 sysctl_wmem_default = 32767;
1418 sysctl_rmem_default = 32767;
1419 } else if (totalram_pages >= 131072) {
1420 sysctl_wmem_max = 131071;
1421 sysctl_rmem_max = 131071;
1422 }
1423}
1424
1425/*
1426 * Simple resource managers for sockets.
1427 */
1428
1429
1430/*
1431 * Write buffer destructor automatically called from kfree_skb.
1432 */
1433void sock_wfree(struct sk_buff *skb)
1434{
1435 struct sock *sk = skb->sk;
1436 unsigned int len = skb->truesize;
1437
1438 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1439 /*
1440 * Keep a reference on sk_wmem_alloc, this will be released
1441 * after sk_write_space() call
1442 */
1443 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1444 sk->sk_write_space(sk);
1445 len = 1;
1446 }
1447 /*
1448 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1449 * could not do because of in-flight packets
1450 */
1451 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1452 __sk_free(sk);
1453}
1454EXPORT_SYMBOL(sock_wfree);
1455
1456/*
1457 * Read buffer destructor automatically called from kfree_skb.
1458 */
1459void sock_rfree(struct sk_buff *skb)
1460{
1461 struct sock *sk = skb->sk;
1462 unsigned int len = skb->truesize;
1463
1464 atomic_sub(len, &sk->sk_rmem_alloc);
1465 sk_mem_uncharge(sk, len);
1466}
1467EXPORT_SYMBOL(sock_rfree);
1468
1469
1470int sock_i_uid(struct sock *sk)
1471{
1472 int uid;
1473
1474 read_lock_bh(&sk->sk_callback_lock);
1475 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1476 read_unlock_bh(&sk->sk_callback_lock);
1477 return uid;
1478}
1479EXPORT_SYMBOL(sock_i_uid);
1480
1481unsigned long sock_i_ino(struct sock *sk)
1482{
1483 unsigned long ino;
1484
1485 read_lock_bh(&sk->sk_callback_lock);
1486 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1487 read_unlock_bh(&sk->sk_callback_lock);
1488 return ino;
1489}
1490EXPORT_SYMBOL(sock_i_ino);
1491
1492/*
1493 * Allocate a skb from the socket's send buffer.
1494 */
1495struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1496 gfp_t priority)
1497{
1498 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1499 struct sk_buff *skb = alloc_skb(size, priority);
1500 if (skb) {
1501 skb_set_owner_w(skb, sk);
1502 return skb;
1503 }
1504 }
1505 return NULL;
1506}
1507EXPORT_SYMBOL(sock_wmalloc);
1508
1509/*
1510 * Allocate a skb from the socket's receive buffer.
1511 */
1512struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1513 gfp_t priority)
1514{
1515 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1516 struct sk_buff *skb = alloc_skb(size, priority);
1517 if (skb) {
1518 skb_set_owner_r(skb, sk);
1519 return skb;
1520 }
1521 }
1522 return NULL;
1523}
1524
1525/*
1526 * Allocate a memory block from the socket's option memory buffer.
1527 */
1528void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1529{
1530 if ((unsigned int)size <= sysctl_optmem_max &&
1531 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1532 void *mem;
1533 /* First do the add, to avoid the race if kmalloc
1534 * might sleep.
1535 */
1536 atomic_add(size, &sk->sk_omem_alloc);
1537 mem = kmalloc(size, priority);
1538 if (mem)
1539 return mem;
1540 atomic_sub(size, &sk->sk_omem_alloc);
1541 }
1542 return NULL;
1543}
1544EXPORT_SYMBOL(sock_kmalloc);
1545
1546/*
1547 * Free an option memory block.
1548 */
1549void sock_kfree_s(struct sock *sk, void *mem, int size)
1550{
1551 kfree(mem);
1552 atomic_sub(size, &sk->sk_omem_alloc);
1553}
1554EXPORT_SYMBOL(sock_kfree_s);
1555
1556/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1557 I think, these locks should be removed for datagram sockets.
1558 */
1559static long sock_wait_for_wmem(struct sock *sk, long timeo)
1560{
1561 DEFINE_WAIT(wait);
1562
1563 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1564 for (;;) {
1565 if (!timeo)
1566 break;
1567 if (signal_pending(current))
1568 break;
1569 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1570 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1571 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1572 break;
1573 if (sk->sk_shutdown & SEND_SHUTDOWN)
1574 break;
1575 if (sk->sk_err)
1576 break;
1577 timeo = schedule_timeout(timeo);
1578 }
1579 finish_wait(sk_sleep(sk), &wait);
1580 return timeo;
1581}
1582
1583
1584/*
1585 * Generic send/receive buffer handlers
1586 */
1587
1588struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1589 unsigned long data_len, int noblock,
1590 int *errcode)
1591{
1592 struct sk_buff *skb;
1593 gfp_t gfp_mask;
1594 long timeo;
1595 int err;
1596 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1597
1598 err = -EMSGSIZE;
1599 if (npages > MAX_SKB_FRAGS)
1600 goto failure;
1601
1602 gfp_mask = sk->sk_allocation;
1603 if (gfp_mask & __GFP_WAIT)
1604 gfp_mask |= __GFP_REPEAT;
1605
1606 timeo = sock_sndtimeo(sk, noblock);
1607 while (1) {
1608 err = sock_error(sk);
1609 if (err != 0)
1610 goto failure;
1611
1612 err = -EPIPE;
1613 if (sk->sk_shutdown & SEND_SHUTDOWN)
1614 goto failure;
1615
1616 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1617 skb = alloc_skb(header_len, gfp_mask);
1618 if (skb) {
1619 int i;
1620
1621 /* No pages, we're done... */
1622 if (!data_len)
1623 break;
1624
1625 skb->truesize += data_len;
1626 skb_shinfo(skb)->nr_frags = npages;
1627 for (i = 0; i < npages; i++) {
1628 struct page *page;
1629
1630 page = alloc_pages(sk->sk_allocation, 0);
1631 if (!page) {
1632 err = -ENOBUFS;
1633 skb_shinfo(skb)->nr_frags = i;
1634 kfree_skb(skb);
1635 goto failure;
1636 }
1637
1638 __skb_fill_page_desc(skb, i,
1639 page, 0,
1640 (data_len >= PAGE_SIZE ?
1641 PAGE_SIZE :
1642 data_len));
1643 data_len -= PAGE_SIZE;
1644 }
1645
1646 /* Full success... */
1647 break;
1648 }
1649 err = -ENOBUFS;
1650 goto failure;
1651 }
1652 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1653 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1654 err = -EAGAIN;
1655 if (!timeo)
1656 goto failure;
1657 if (signal_pending(current))
1658 goto interrupted;
1659 timeo = sock_wait_for_wmem(sk, timeo);
1660 }
1661
1662 skb_set_owner_w(skb, sk);
1663 return skb;
1664
1665interrupted:
1666 err = sock_intr_errno(timeo);
1667failure:
1668 *errcode = err;
1669 return NULL;
1670}
1671EXPORT_SYMBOL(sock_alloc_send_pskb);
1672
1673struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1674 int noblock, int *errcode)
1675{
1676 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1677}
1678EXPORT_SYMBOL(sock_alloc_send_skb);
1679
1680static void __lock_sock(struct sock *sk)
1681 __releases(&sk->sk_lock.slock)
1682 __acquires(&sk->sk_lock.slock)
1683{
1684 DEFINE_WAIT(wait);
1685
1686 for (;;) {
1687 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1688 TASK_UNINTERRUPTIBLE);
1689 spin_unlock_bh(&sk->sk_lock.slock);
1690 schedule();
1691 spin_lock_bh(&sk->sk_lock.slock);
1692 if (!sock_owned_by_user(sk))
1693 break;
1694 }
1695 finish_wait(&sk->sk_lock.wq, &wait);
1696}
1697
1698static void __release_sock(struct sock *sk)
1699 __releases(&sk->sk_lock.slock)
1700 __acquires(&sk->sk_lock.slock)
1701{
1702 struct sk_buff *skb = sk->sk_backlog.head;
1703
1704 do {
1705 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1706 bh_unlock_sock(sk);
1707
1708 do {
1709 struct sk_buff *next = skb->next;
1710
1711 prefetch(next);
1712 WARN_ON_ONCE(skb_dst_is_noref(skb));
1713 skb->next = NULL;
1714 sk_backlog_rcv(sk, skb);
1715
1716 /*
1717 * We are in process context here with softirqs
1718 * disabled, use cond_resched_softirq() to preempt.
1719 * This is safe to do because we've taken the backlog
1720 * queue private:
1721 */
1722 cond_resched_softirq();
1723
1724 skb = next;
1725 } while (skb != NULL);
1726
1727 bh_lock_sock(sk);
1728 } while ((skb = sk->sk_backlog.head) != NULL);
1729
1730 /*
1731 * Doing the zeroing here guarantee we can not loop forever
1732 * while a wild producer attempts to flood us.
1733 */
1734 sk->sk_backlog.len = 0;
1735}
1736
1737/**
1738 * sk_wait_data - wait for data to arrive at sk_receive_queue
1739 * @sk: sock to wait on
1740 * @timeo: for how long
1741 *
1742 * Now socket state including sk->sk_err is changed only under lock,
1743 * hence we may omit checks after joining wait queue.
1744 * We check receive queue before schedule() only as optimization;
1745 * it is very likely that release_sock() added new data.
1746 */
1747int sk_wait_data(struct sock *sk, long *timeo)
1748{
1749 int rc;
1750 DEFINE_WAIT(wait);
1751
1752 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1753 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1754 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1755 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1756 finish_wait(sk_sleep(sk), &wait);
1757 return rc;
1758}
1759EXPORT_SYMBOL(sk_wait_data);
1760
1761/**
1762 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1763 * @sk: socket
1764 * @size: memory size to allocate
1765 * @kind: allocation type
1766 *
1767 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1768 * rmem allocation. This function assumes that protocols which have
1769 * memory_pressure use sk_wmem_queued as write buffer accounting.
1770 */
1771int __sk_mem_schedule(struct sock *sk, int size, int kind)
1772{
1773 struct proto *prot = sk->sk_prot;
1774 int amt = sk_mem_pages(size);
1775 long allocated;
1776 int parent_status = UNDER_LIMIT;
1777
1778 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1779
1780 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1781
1782 /* Under limit. */
1783 if (parent_status == UNDER_LIMIT &&
1784 allocated <= sk_prot_mem_limits(sk, 0)) {
1785 sk_leave_memory_pressure(sk);
1786 return 1;
1787 }
1788
1789 /* Under pressure. (we or our parents) */
1790 if ((parent_status > SOFT_LIMIT) ||
1791 allocated > sk_prot_mem_limits(sk, 1))
1792 sk_enter_memory_pressure(sk);
1793
1794 /* Over hard limit (we or our parents) */
1795 if ((parent_status == OVER_LIMIT) ||
1796 (allocated > sk_prot_mem_limits(sk, 2)))
1797 goto suppress_allocation;
1798
1799 /* guarantee minimum buffer size under pressure */
1800 if (kind == SK_MEM_RECV) {
1801 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1802 return 1;
1803
1804 } else { /* SK_MEM_SEND */
1805 if (sk->sk_type == SOCK_STREAM) {
1806 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1807 return 1;
1808 } else if (atomic_read(&sk->sk_wmem_alloc) <
1809 prot->sysctl_wmem[0])
1810 return 1;
1811 }
1812
1813 if (sk_has_memory_pressure(sk)) {
1814 int alloc;
1815
1816 if (!sk_under_memory_pressure(sk))
1817 return 1;
1818 alloc = sk_sockets_allocated_read_positive(sk);
1819 if (sk_prot_mem_limits(sk, 2) > alloc *
1820 sk_mem_pages(sk->sk_wmem_queued +
1821 atomic_read(&sk->sk_rmem_alloc) +
1822 sk->sk_forward_alloc))
1823 return 1;
1824 }
1825
1826suppress_allocation:
1827
1828 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1829 sk_stream_moderate_sndbuf(sk);
1830
1831 /* Fail only if socket is _under_ its sndbuf.
1832 * In this case we cannot block, so that we have to fail.
1833 */
1834 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1835 return 1;
1836 }
1837
1838 trace_sock_exceed_buf_limit(sk, prot, allocated);
1839
1840 /* Alas. Undo changes. */
1841 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1842
1843 sk_memory_allocated_sub(sk, amt);
1844
1845 return 0;
1846}
1847EXPORT_SYMBOL(__sk_mem_schedule);
1848
1849/**
1850 * __sk_reclaim - reclaim memory_allocated
1851 * @sk: socket
1852 */
1853void __sk_mem_reclaim(struct sock *sk)
1854{
1855 sk_memory_allocated_sub(sk,
1856 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1857 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1858
1859 if (sk_under_memory_pressure(sk) &&
1860 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1861 sk_leave_memory_pressure(sk);
1862}
1863EXPORT_SYMBOL(__sk_mem_reclaim);
1864
1865
1866/*
1867 * Set of default routines for initialising struct proto_ops when
1868 * the protocol does not support a particular function. In certain
1869 * cases where it makes no sense for a protocol to have a "do nothing"
1870 * function, some default processing is provided.
1871 */
1872
1873int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1874{
1875 return -EOPNOTSUPP;
1876}
1877EXPORT_SYMBOL(sock_no_bind);
1878
1879int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1880 int len, int flags)
1881{
1882 return -EOPNOTSUPP;
1883}
1884EXPORT_SYMBOL(sock_no_connect);
1885
1886int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1887{
1888 return -EOPNOTSUPP;
1889}
1890EXPORT_SYMBOL(sock_no_socketpair);
1891
1892int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1893{
1894 return -EOPNOTSUPP;
1895}
1896EXPORT_SYMBOL(sock_no_accept);
1897
1898int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1899 int *len, int peer)
1900{
1901 return -EOPNOTSUPP;
1902}
1903EXPORT_SYMBOL(sock_no_getname);
1904
1905unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1906{
1907 return 0;
1908}
1909EXPORT_SYMBOL(sock_no_poll);
1910
1911int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1912{
1913 return -EOPNOTSUPP;
1914}
1915EXPORT_SYMBOL(sock_no_ioctl);
1916
1917int sock_no_listen(struct socket *sock, int backlog)
1918{
1919 return -EOPNOTSUPP;
1920}
1921EXPORT_SYMBOL(sock_no_listen);
1922
1923int sock_no_shutdown(struct socket *sock, int how)
1924{
1925 return -EOPNOTSUPP;
1926}
1927EXPORT_SYMBOL(sock_no_shutdown);
1928
1929int sock_no_setsockopt(struct socket *sock, int level, int optname,
1930 char __user *optval, unsigned int optlen)
1931{
1932 return -EOPNOTSUPP;
1933}
1934EXPORT_SYMBOL(sock_no_setsockopt);
1935
1936int sock_no_getsockopt(struct socket *sock, int level, int optname,
1937 char __user *optval, int __user *optlen)
1938{
1939 return -EOPNOTSUPP;
1940}
1941EXPORT_SYMBOL(sock_no_getsockopt);
1942
1943int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1944 size_t len)
1945{
1946 return -EOPNOTSUPP;
1947}
1948EXPORT_SYMBOL(sock_no_sendmsg);
1949
1950int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1951 size_t len, int flags)
1952{
1953 return -EOPNOTSUPP;
1954}
1955EXPORT_SYMBOL(sock_no_recvmsg);
1956
1957int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1958{
1959 /* Mirror missing mmap method error code */
1960 return -ENODEV;
1961}
1962EXPORT_SYMBOL(sock_no_mmap);
1963
1964ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1965{
1966 ssize_t res;
1967 struct msghdr msg = {.msg_flags = flags};
1968 struct kvec iov;
1969 char *kaddr = kmap(page);
1970 iov.iov_base = kaddr + offset;
1971 iov.iov_len = size;
1972 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1973 kunmap(page);
1974 return res;
1975}
1976EXPORT_SYMBOL(sock_no_sendpage);
1977
1978/*
1979 * Default Socket Callbacks
1980 */
1981
1982static void sock_def_wakeup(struct sock *sk)
1983{
1984 struct socket_wq *wq;
1985
1986 rcu_read_lock();
1987 wq = rcu_dereference(sk->sk_wq);
1988 if (wq_has_sleeper(wq))
1989 wake_up_interruptible_all(&wq->wait);
1990 rcu_read_unlock();
1991}
1992
1993static void sock_def_error_report(struct sock *sk)
1994{
1995 struct socket_wq *wq;
1996
1997 rcu_read_lock();
1998 wq = rcu_dereference(sk->sk_wq);
1999 if (wq_has_sleeper(wq))
2000 wake_up_interruptible_poll(&wq->wait, POLLERR);
2001 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2002 rcu_read_unlock();
2003}
2004
2005static void sock_def_readable(struct sock *sk, int len)
2006{
2007 struct socket_wq *wq;
2008
2009 rcu_read_lock();
2010 wq = rcu_dereference(sk->sk_wq);
2011 if (wq_has_sleeper(wq))
2012 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2013 POLLRDNORM | POLLRDBAND);
2014 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2015 rcu_read_unlock();
2016}
2017
2018static void sock_def_write_space(struct sock *sk)
2019{
2020 struct socket_wq *wq;
2021
2022 rcu_read_lock();
2023
2024 /* Do not wake up a writer until he can make "significant"
2025 * progress. --DaveM
2026 */
2027 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2028 wq = rcu_dereference(sk->sk_wq);
2029 if (wq_has_sleeper(wq))
2030 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2031 POLLWRNORM | POLLWRBAND);
2032
2033 /* Should agree with poll, otherwise some programs break */
2034 if (sock_writeable(sk))
2035 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2036 }
2037
2038 rcu_read_unlock();
2039}
2040
2041static void sock_def_destruct(struct sock *sk)
2042{
2043 kfree(sk->sk_protinfo);
2044}
2045
2046void sk_send_sigurg(struct sock *sk)
2047{
2048 if (sk->sk_socket && sk->sk_socket->file)
2049 if (send_sigurg(&sk->sk_socket->file->f_owner))
2050 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2051}
2052EXPORT_SYMBOL(sk_send_sigurg);
2053
2054void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2055 unsigned long expires)
2056{
2057 if (!mod_timer(timer, expires))
2058 sock_hold(sk);
2059}
2060EXPORT_SYMBOL(sk_reset_timer);
2061
2062void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2063{
2064 if (timer_pending(timer) && del_timer(timer))
2065 __sock_put(sk);
2066}
2067EXPORT_SYMBOL(sk_stop_timer);
2068
2069void sock_init_data(struct socket *sock, struct sock *sk)
2070{
2071 skb_queue_head_init(&sk->sk_receive_queue);
2072 skb_queue_head_init(&sk->sk_write_queue);
2073 skb_queue_head_init(&sk->sk_error_queue);
2074#ifdef CONFIG_NET_DMA
2075 skb_queue_head_init(&sk->sk_async_wait_queue);
2076#endif
2077
2078 sk->sk_send_head = NULL;
2079
2080 init_timer(&sk->sk_timer);
2081
2082 sk->sk_allocation = GFP_KERNEL;
2083 sk->sk_rcvbuf = sysctl_rmem_default;
2084 sk->sk_sndbuf = sysctl_wmem_default;
2085 sk->sk_state = TCP_CLOSE;
2086 sk_set_socket(sk, sock);
2087
2088 sock_set_flag(sk, SOCK_ZAPPED);
2089
2090 if (sock) {
2091 sk->sk_type = sock->type;
2092 sk->sk_wq = sock->wq;
2093 sock->sk = sk;
2094 } else
2095 sk->sk_wq = NULL;
2096
2097 spin_lock_init(&sk->sk_dst_lock);
2098 rwlock_init(&sk->sk_callback_lock);
2099 lockdep_set_class_and_name(&sk->sk_callback_lock,
2100 af_callback_keys + sk->sk_family,
2101 af_family_clock_key_strings[sk->sk_family]);
2102
2103 sk->sk_state_change = sock_def_wakeup;
2104 sk->sk_data_ready = sock_def_readable;
2105 sk->sk_write_space = sock_def_write_space;
2106 sk->sk_error_report = sock_def_error_report;
2107 sk->sk_destruct = sock_def_destruct;
2108
2109 sk->sk_sndmsg_page = NULL;
2110 sk->sk_sndmsg_off = 0;
2111 sk->sk_peek_off = -1;
2112
2113 sk->sk_peer_pid = NULL;
2114 sk->sk_peer_cred = NULL;
2115 sk->sk_write_pending = 0;
2116 sk->sk_rcvlowat = 1;
2117 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2118 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2119
2120 sk->sk_stamp = ktime_set(-1L, 0);
2121
2122 /*
2123 * Before updating sk_refcnt, we must commit prior changes to memory
2124 * (Documentation/RCU/rculist_nulls.txt for details)
2125 */
2126 smp_wmb();
2127 atomic_set(&sk->sk_refcnt, 1);
2128 atomic_set(&sk->sk_drops, 0);
2129}
2130EXPORT_SYMBOL(sock_init_data);
2131
2132void lock_sock_nested(struct sock *sk, int subclass)
2133{
2134 might_sleep();
2135 spin_lock_bh(&sk->sk_lock.slock);
2136 if (sk->sk_lock.owned)
2137 __lock_sock(sk);
2138 sk->sk_lock.owned = 1;
2139 spin_unlock(&sk->sk_lock.slock);
2140 /*
2141 * The sk_lock has mutex_lock() semantics here:
2142 */
2143 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2144 local_bh_enable();
2145}
2146EXPORT_SYMBOL(lock_sock_nested);
2147
2148void release_sock(struct sock *sk)
2149{
2150 /*
2151 * The sk_lock has mutex_unlock() semantics:
2152 */
2153 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2154
2155 spin_lock_bh(&sk->sk_lock.slock);
2156 if (sk->sk_backlog.tail)
2157 __release_sock(sk);
2158 sk->sk_lock.owned = 0;
2159 if (waitqueue_active(&sk->sk_lock.wq))
2160 wake_up(&sk->sk_lock.wq);
2161 spin_unlock_bh(&sk->sk_lock.slock);
2162}
2163EXPORT_SYMBOL(release_sock);
2164
2165/**
2166 * lock_sock_fast - fast version of lock_sock
2167 * @sk: socket
2168 *
2169 * This version should be used for very small section, where process wont block
2170 * return false if fast path is taken
2171 * sk_lock.slock locked, owned = 0, BH disabled
2172 * return true if slow path is taken
2173 * sk_lock.slock unlocked, owned = 1, BH enabled
2174 */
2175bool lock_sock_fast(struct sock *sk)
2176{
2177 might_sleep();
2178 spin_lock_bh(&sk->sk_lock.slock);
2179
2180 if (!sk->sk_lock.owned)
2181 /*
2182 * Note : We must disable BH
2183 */
2184 return false;
2185
2186 __lock_sock(sk);
2187 sk->sk_lock.owned = 1;
2188 spin_unlock(&sk->sk_lock.slock);
2189 /*
2190 * The sk_lock has mutex_lock() semantics here:
2191 */
2192 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2193 local_bh_enable();
2194 return true;
2195}
2196EXPORT_SYMBOL(lock_sock_fast);
2197
2198int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2199{
2200 struct timeval tv;
2201 if (!sock_flag(sk, SOCK_TIMESTAMP))
2202 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2203 tv = ktime_to_timeval(sk->sk_stamp);
2204 if (tv.tv_sec == -1)
2205 return -ENOENT;
2206 if (tv.tv_sec == 0) {
2207 sk->sk_stamp = ktime_get_real();
2208 tv = ktime_to_timeval(sk->sk_stamp);
2209 }
2210 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2211}
2212EXPORT_SYMBOL(sock_get_timestamp);
2213
2214int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2215{
2216 struct timespec ts;
2217 if (!sock_flag(sk, SOCK_TIMESTAMP))
2218 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2219 ts = ktime_to_timespec(sk->sk_stamp);
2220 if (ts.tv_sec == -1)
2221 return -ENOENT;
2222 if (ts.tv_sec == 0) {
2223 sk->sk_stamp = ktime_get_real();
2224 ts = ktime_to_timespec(sk->sk_stamp);
2225 }
2226 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2227}
2228EXPORT_SYMBOL(sock_get_timestampns);
2229
2230void sock_enable_timestamp(struct sock *sk, int flag)
2231{
2232 if (!sock_flag(sk, flag)) {
2233 unsigned long previous_flags = sk->sk_flags;
2234
2235 sock_set_flag(sk, flag);
2236 /*
2237 * we just set one of the two flags which require net
2238 * time stamping, but time stamping might have been on
2239 * already because of the other one
2240 */
2241 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2242 net_enable_timestamp();
2243 }
2244}
2245
2246/*
2247 * Get a socket option on an socket.
2248 *
2249 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2250 * asynchronous errors should be reported by getsockopt. We assume
2251 * this means if you specify SO_ERROR (otherwise whats the point of it).
2252 */
2253int sock_common_getsockopt(struct socket *sock, int level, int optname,
2254 char __user *optval, int __user *optlen)
2255{
2256 struct sock *sk = sock->sk;
2257
2258 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2259}
2260EXPORT_SYMBOL(sock_common_getsockopt);
2261
2262#ifdef CONFIG_COMPAT
2263int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2264 char __user *optval, int __user *optlen)
2265{
2266 struct sock *sk = sock->sk;
2267
2268 if (sk->sk_prot->compat_getsockopt != NULL)
2269 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2270 optval, optlen);
2271 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2272}
2273EXPORT_SYMBOL(compat_sock_common_getsockopt);
2274#endif
2275
2276int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2277 struct msghdr *msg, size_t size, int flags)
2278{
2279 struct sock *sk = sock->sk;
2280 int addr_len = 0;
2281 int err;
2282
2283 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2284 flags & ~MSG_DONTWAIT, &addr_len);
2285 if (err >= 0)
2286 msg->msg_namelen = addr_len;
2287 return err;
2288}
2289EXPORT_SYMBOL(sock_common_recvmsg);
2290
2291/*
2292 * Set socket options on an inet socket.
2293 */
2294int sock_common_setsockopt(struct socket *sock, int level, int optname,
2295 char __user *optval, unsigned int optlen)
2296{
2297 struct sock *sk = sock->sk;
2298
2299 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2300}
2301EXPORT_SYMBOL(sock_common_setsockopt);
2302
2303#ifdef CONFIG_COMPAT
2304int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2305 char __user *optval, unsigned int optlen)
2306{
2307 struct sock *sk = sock->sk;
2308
2309 if (sk->sk_prot->compat_setsockopt != NULL)
2310 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2311 optval, optlen);
2312 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2313}
2314EXPORT_SYMBOL(compat_sock_common_setsockopt);
2315#endif
2316
2317void sk_common_release(struct sock *sk)
2318{
2319 if (sk->sk_prot->destroy)
2320 sk->sk_prot->destroy(sk);
2321
2322 /*
2323 * Observation: when sock_common_release is called, processes have
2324 * no access to socket. But net still has.
2325 * Step one, detach it from networking:
2326 *
2327 * A. Remove from hash tables.
2328 */
2329
2330 sk->sk_prot->unhash(sk);
2331
2332 /*
2333 * In this point socket cannot receive new packets, but it is possible
2334 * that some packets are in flight because some CPU runs receiver and
2335 * did hash table lookup before we unhashed socket. They will achieve
2336 * receive queue and will be purged by socket destructor.
2337 *
2338 * Also we still have packets pending on receive queue and probably,
2339 * our own packets waiting in device queues. sock_destroy will drain
2340 * receive queue, but transmitted packets will delay socket destruction
2341 * until the last reference will be released.
2342 */
2343
2344 sock_orphan(sk);
2345
2346 xfrm_sk_free_policy(sk);
2347
2348 sk_refcnt_debug_release(sk);
2349 sock_put(sk);
2350}
2351EXPORT_SYMBOL(sk_common_release);
2352
2353#ifdef CONFIG_PROC_FS
2354#define PROTO_INUSE_NR 64 /* should be enough for the first time */
2355struct prot_inuse {
2356 int val[PROTO_INUSE_NR];
2357};
2358
2359static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2360
2361#ifdef CONFIG_NET_NS
2362void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2363{
2364 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2365}
2366EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2367
2368int sock_prot_inuse_get(struct net *net, struct proto *prot)
2369{
2370 int cpu, idx = prot->inuse_idx;
2371 int res = 0;
2372
2373 for_each_possible_cpu(cpu)
2374 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2375
2376 return res >= 0 ? res : 0;
2377}
2378EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2379
2380static int __net_init sock_inuse_init_net(struct net *net)
2381{
2382 net->core.inuse = alloc_percpu(struct prot_inuse);
2383 return net->core.inuse ? 0 : -ENOMEM;
2384}
2385
2386static void __net_exit sock_inuse_exit_net(struct net *net)
2387{
2388 free_percpu(net->core.inuse);
2389}
2390
2391static struct pernet_operations net_inuse_ops = {
2392 .init = sock_inuse_init_net,
2393 .exit = sock_inuse_exit_net,
2394};
2395
2396static __init int net_inuse_init(void)
2397{
2398 if (register_pernet_subsys(&net_inuse_ops))
2399 panic("Cannot initialize net inuse counters");
2400
2401 return 0;
2402}
2403
2404core_initcall(net_inuse_init);
2405#else
2406static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2407
2408void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2409{
2410 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2411}
2412EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2413
2414int sock_prot_inuse_get(struct net *net, struct proto *prot)
2415{
2416 int cpu, idx = prot->inuse_idx;
2417 int res = 0;
2418
2419 for_each_possible_cpu(cpu)
2420 res += per_cpu(prot_inuse, cpu).val[idx];
2421
2422 return res >= 0 ? res : 0;
2423}
2424EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2425#endif
2426
2427static void assign_proto_idx(struct proto *prot)
2428{
2429 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2430
2431 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2432 pr_err("PROTO_INUSE_NR exhausted\n");
2433 return;
2434 }
2435
2436 set_bit(prot->inuse_idx, proto_inuse_idx);
2437}
2438
2439static void release_proto_idx(struct proto *prot)
2440{
2441 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2442 clear_bit(prot->inuse_idx, proto_inuse_idx);
2443}
2444#else
2445static inline void assign_proto_idx(struct proto *prot)
2446{
2447}
2448
2449static inline void release_proto_idx(struct proto *prot)
2450{
2451}
2452#endif
2453
2454int proto_register(struct proto *prot, int alloc_slab)
2455{
2456 if (alloc_slab) {
2457 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2458 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2459 NULL);
2460
2461 if (prot->slab == NULL) {
2462 pr_crit("%s: Can't create sock SLAB cache!\n",
2463 prot->name);
2464 goto out;
2465 }
2466
2467 if (prot->rsk_prot != NULL) {
2468 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2469 if (prot->rsk_prot->slab_name == NULL)
2470 goto out_free_sock_slab;
2471
2472 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2473 prot->rsk_prot->obj_size, 0,
2474 SLAB_HWCACHE_ALIGN, NULL);
2475
2476 if (prot->rsk_prot->slab == NULL) {
2477 pr_crit("%s: Can't create request sock SLAB cache!\n",
2478 prot->name);
2479 goto out_free_request_sock_slab_name;
2480 }
2481 }
2482
2483 if (prot->twsk_prot != NULL) {
2484 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2485
2486 if (prot->twsk_prot->twsk_slab_name == NULL)
2487 goto out_free_request_sock_slab;
2488
2489 prot->twsk_prot->twsk_slab =
2490 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2491 prot->twsk_prot->twsk_obj_size,
2492 0,
2493 SLAB_HWCACHE_ALIGN |
2494 prot->slab_flags,
2495 NULL);
2496 if (prot->twsk_prot->twsk_slab == NULL)
2497 goto out_free_timewait_sock_slab_name;
2498 }
2499 }
2500
2501 mutex_lock(&proto_list_mutex);
2502 list_add(&prot->node, &proto_list);
2503 assign_proto_idx(prot);
2504 mutex_unlock(&proto_list_mutex);
2505 return 0;
2506
2507out_free_timewait_sock_slab_name:
2508 kfree(prot->twsk_prot->twsk_slab_name);
2509out_free_request_sock_slab:
2510 if (prot->rsk_prot && prot->rsk_prot->slab) {
2511 kmem_cache_destroy(prot->rsk_prot->slab);
2512 prot->rsk_prot->slab = NULL;
2513 }
2514out_free_request_sock_slab_name:
2515 if (prot->rsk_prot)
2516 kfree(prot->rsk_prot->slab_name);
2517out_free_sock_slab:
2518 kmem_cache_destroy(prot->slab);
2519 prot->slab = NULL;
2520out:
2521 return -ENOBUFS;
2522}
2523EXPORT_SYMBOL(proto_register);
2524
2525void proto_unregister(struct proto *prot)
2526{
2527 mutex_lock(&proto_list_mutex);
2528 release_proto_idx(prot);
2529 list_del(&prot->node);
2530 mutex_unlock(&proto_list_mutex);
2531
2532 if (prot->slab != NULL) {
2533 kmem_cache_destroy(prot->slab);
2534 prot->slab = NULL;
2535 }
2536
2537 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2538 kmem_cache_destroy(prot->rsk_prot->slab);
2539 kfree(prot->rsk_prot->slab_name);
2540 prot->rsk_prot->slab = NULL;
2541 }
2542
2543 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2544 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2545 kfree(prot->twsk_prot->twsk_slab_name);
2546 prot->twsk_prot->twsk_slab = NULL;
2547 }
2548}
2549EXPORT_SYMBOL(proto_unregister);
2550
2551#ifdef CONFIG_PROC_FS
2552static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2553 __acquires(proto_list_mutex)
2554{
2555 mutex_lock(&proto_list_mutex);
2556 return seq_list_start_head(&proto_list, *pos);
2557}
2558
2559static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2560{
2561 return seq_list_next(v, &proto_list, pos);
2562}
2563
2564static void proto_seq_stop(struct seq_file *seq, void *v)
2565 __releases(proto_list_mutex)
2566{
2567 mutex_unlock(&proto_list_mutex);
2568}
2569
2570static char proto_method_implemented(const void *method)
2571{
2572 return method == NULL ? 'n' : 'y';
2573}
2574static long sock_prot_memory_allocated(struct proto *proto)
2575{
2576 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2577}
2578
2579static char *sock_prot_memory_pressure(struct proto *proto)
2580{
2581 return proto->memory_pressure != NULL ?
2582 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2583}
2584
2585static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2586{
2587
2588 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2589 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2590 proto->name,
2591 proto->obj_size,
2592 sock_prot_inuse_get(seq_file_net(seq), proto),
2593 sock_prot_memory_allocated(proto),
2594 sock_prot_memory_pressure(proto),
2595 proto->max_header,
2596 proto->slab == NULL ? "no" : "yes",
2597 module_name(proto->owner),
2598 proto_method_implemented(proto->close),
2599 proto_method_implemented(proto->connect),
2600 proto_method_implemented(proto->disconnect),
2601 proto_method_implemented(proto->accept),
2602 proto_method_implemented(proto->ioctl),
2603 proto_method_implemented(proto->init),
2604 proto_method_implemented(proto->destroy),
2605 proto_method_implemented(proto->shutdown),
2606 proto_method_implemented(proto->setsockopt),
2607 proto_method_implemented(proto->getsockopt),
2608 proto_method_implemented(proto->sendmsg),
2609 proto_method_implemented(proto->recvmsg),
2610 proto_method_implemented(proto->sendpage),
2611 proto_method_implemented(proto->bind),
2612 proto_method_implemented(proto->backlog_rcv),
2613 proto_method_implemented(proto->hash),
2614 proto_method_implemented(proto->unhash),
2615 proto_method_implemented(proto->get_port),
2616 proto_method_implemented(proto->enter_memory_pressure));
2617}
2618
2619static int proto_seq_show(struct seq_file *seq, void *v)
2620{
2621 if (v == &proto_list)
2622 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2623 "protocol",
2624 "size",
2625 "sockets",
2626 "memory",
2627 "press",
2628 "maxhdr",
2629 "slab",
2630 "module",
2631 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2632 else
2633 proto_seq_printf(seq, list_entry(v, struct proto, node));
2634 return 0;
2635}
2636
2637static const struct seq_operations proto_seq_ops = {
2638 .start = proto_seq_start,
2639 .next = proto_seq_next,
2640 .stop = proto_seq_stop,
2641 .show = proto_seq_show,
2642};
2643
2644static int proto_seq_open(struct inode *inode, struct file *file)
2645{
2646 return seq_open_net(inode, file, &proto_seq_ops,
2647 sizeof(struct seq_net_private));
2648}
2649
2650static const struct file_operations proto_seq_fops = {
2651 .owner = THIS_MODULE,
2652 .open = proto_seq_open,
2653 .read = seq_read,
2654 .llseek = seq_lseek,
2655 .release = seq_release_net,
2656};
2657
2658static __net_init int proto_init_net(struct net *net)
2659{
2660 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2661 return -ENOMEM;
2662
2663 return 0;
2664}
2665
2666static __net_exit void proto_exit_net(struct net *net)
2667{
2668 proc_net_remove(net, "protocols");
2669}
2670
2671
2672static __net_initdata struct pernet_operations proto_net_ops = {
2673 .init = proto_init_net,
2674 .exit = proto_exit_net,
2675};
2676
2677static int __init proto_init(void)
2678{
2679 return register_pernet_subsys(&proto_net_ops);
2680}
2681
2682subsys_initcall(proto_init);
2683
2684#endif /* PROC_FS */
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic socket support routines. Memory allocators, socket lock/release
8 * handler for protocols to use and generic option handler.
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 */
85
86#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87
88#include <asm/unaligned.h>
89#include <linux/capability.h>
90#include <linux/errno.h>
91#include <linux/errqueue.h>
92#include <linux/types.h>
93#include <linux/socket.h>
94#include <linux/in.h>
95#include <linux/kernel.h>
96#include <linux/module.h>
97#include <linux/proc_fs.h>
98#include <linux/seq_file.h>
99#include <linux/sched.h>
100#include <linux/sched/mm.h>
101#include <linux/timer.h>
102#include <linux/string.h>
103#include <linux/sockios.h>
104#include <linux/net.h>
105#include <linux/mm.h>
106#include <linux/slab.h>
107#include <linux/interrupt.h>
108#include <linux/poll.h>
109#include <linux/tcp.h>
110#include <linux/init.h>
111#include <linux/highmem.h>
112#include <linux/user_namespace.h>
113#include <linux/static_key.h>
114#include <linux/memcontrol.h>
115#include <linux/prefetch.h>
116
117#include <linux/uaccess.h>
118
119#include <linux/netdevice.h>
120#include <net/protocol.h>
121#include <linux/skbuff.h>
122#include <net/net_namespace.h>
123#include <net/request_sock.h>
124#include <net/sock.h>
125#include <linux/net_tstamp.h>
126#include <net/xfrm.h>
127#include <linux/ipsec.h>
128#include <net/cls_cgroup.h>
129#include <net/netprio_cgroup.h>
130#include <linux/sock_diag.h>
131
132#include <linux/filter.h>
133#include <net/sock_reuseport.h>
134#include <net/bpf_sk_storage.h>
135
136#include <trace/events/sock.h>
137
138#include <net/tcp.h>
139#include <net/busy_poll.h>
140
141static DEFINE_MUTEX(proto_list_mutex);
142static LIST_HEAD(proto_list);
143
144static void sock_inuse_add(struct net *net, int val);
145
146/**
147 * sk_ns_capable - General socket capability test
148 * @sk: Socket to use a capability on or through
149 * @user_ns: The user namespace of the capability to use
150 * @cap: The capability to use
151 *
152 * Test to see if the opener of the socket had when the socket was
153 * created and the current process has the capability @cap in the user
154 * namespace @user_ns.
155 */
156bool sk_ns_capable(const struct sock *sk,
157 struct user_namespace *user_ns, int cap)
158{
159 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
160 ns_capable(user_ns, cap);
161}
162EXPORT_SYMBOL(sk_ns_capable);
163
164/**
165 * sk_capable - Socket global capability test
166 * @sk: Socket to use a capability on or through
167 * @cap: The global capability to use
168 *
169 * Test to see if the opener of the socket had when the socket was
170 * created and the current process has the capability @cap in all user
171 * namespaces.
172 */
173bool sk_capable(const struct sock *sk, int cap)
174{
175 return sk_ns_capable(sk, &init_user_ns, cap);
176}
177EXPORT_SYMBOL(sk_capable);
178
179/**
180 * sk_net_capable - Network namespace socket capability test
181 * @sk: Socket to use a capability on or through
182 * @cap: The capability to use
183 *
184 * Test to see if the opener of the socket had when the socket was created
185 * and the current process has the capability @cap over the network namespace
186 * the socket is a member of.
187 */
188bool sk_net_capable(const struct sock *sk, int cap)
189{
190 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
191}
192EXPORT_SYMBOL(sk_net_capable);
193
194/*
195 * Each address family might have different locking rules, so we have
196 * one slock key per address family and separate keys for internal and
197 * userspace sockets.
198 */
199static struct lock_class_key af_family_keys[AF_MAX];
200static struct lock_class_key af_family_kern_keys[AF_MAX];
201static struct lock_class_key af_family_slock_keys[AF_MAX];
202static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
203
204/*
205 * Make lock validator output more readable. (we pre-construct these
206 * strings build-time, so that runtime initialization of socket
207 * locks is fast):
208 */
209
210#define _sock_locks(x) \
211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \
212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \
213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \
214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \
215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \
216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \
217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \
218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \
219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \
220 x "27" , x "28" , x "AF_CAN" , \
221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \
222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \
223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \
224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \
225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \
226 x "AF_MAX"
227
228static const char *const af_family_key_strings[AF_MAX+1] = {
229 _sock_locks("sk_lock-")
230};
231static const char *const af_family_slock_key_strings[AF_MAX+1] = {
232 _sock_locks("slock-")
233};
234static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 _sock_locks("clock-")
236};
237
238static const char *const af_family_kern_key_strings[AF_MAX+1] = {
239 _sock_locks("k-sk_lock-")
240};
241static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
242 _sock_locks("k-slock-")
243};
244static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
245 _sock_locks("k-clock-")
246};
247static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
248 _sock_locks("rlock-")
249};
250static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
251 _sock_locks("wlock-")
252};
253static const char *const af_family_elock_key_strings[AF_MAX+1] = {
254 _sock_locks("elock-")
255};
256
257/*
258 * sk_callback_lock and sk queues locking rules are per-address-family,
259 * so split the lock classes by using a per-AF key:
260 */
261static struct lock_class_key af_callback_keys[AF_MAX];
262static struct lock_class_key af_rlock_keys[AF_MAX];
263static struct lock_class_key af_wlock_keys[AF_MAX];
264static struct lock_class_key af_elock_keys[AF_MAX];
265static struct lock_class_key af_kern_callback_keys[AF_MAX];
266
267/* Run time adjustable parameters. */
268__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269EXPORT_SYMBOL(sysctl_wmem_max);
270__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271EXPORT_SYMBOL(sysctl_rmem_max);
272__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
274
275/* Maximal space eaten by iovec or ancillary data plus some space */
276int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277EXPORT_SYMBOL(sysctl_optmem_max);
278
279int sysctl_tstamp_allow_data __read_mostly = 1;
280
281DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
282EXPORT_SYMBOL_GPL(memalloc_socks_key);
283
284/**
285 * sk_set_memalloc - sets %SOCK_MEMALLOC
286 * @sk: socket to set it on
287 *
288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
289 * It's the responsibility of the admin to adjust min_free_kbytes
290 * to meet the requirements
291 */
292void sk_set_memalloc(struct sock *sk)
293{
294 sock_set_flag(sk, SOCK_MEMALLOC);
295 sk->sk_allocation |= __GFP_MEMALLOC;
296 static_branch_inc(&memalloc_socks_key);
297}
298EXPORT_SYMBOL_GPL(sk_set_memalloc);
299
300void sk_clear_memalloc(struct sock *sk)
301{
302 sock_reset_flag(sk, SOCK_MEMALLOC);
303 sk->sk_allocation &= ~__GFP_MEMALLOC;
304 static_branch_dec(&memalloc_socks_key);
305
306 /*
307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
308 * progress of swapping. SOCK_MEMALLOC may be cleared while
309 * it has rmem allocations due to the last swapfile being deactivated
310 * but there is a risk that the socket is unusable due to exceeding
311 * the rmem limits. Reclaim the reserves and obey rmem limits again.
312 */
313 sk_mem_reclaim(sk);
314}
315EXPORT_SYMBOL_GPL(sk_clear_memalloc);
316
317int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
318{
319 int ret;
320 unsigned int noreclaim_flag;
321
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
324
325 noreclaim_flag = memalloc_noreclaim_save();
326 ret = sk->sk_backlog_rcv(sk, skb);
327 memalloc_noreclaim_restore(noreclaim_flag);
328
329 return ret;
330}
331EXPORT_SYMBOL(__sk_backlog_rcv);
332
333static int sock_get_timeout(long timeo, void *optval, bool old_timeval)
334{
335 struct __kernel_sock_timeval tv;
336 int size;
337
338 if (timeo == MAX_SCHEDULE_TIMEOUT) {
339 tv.tv_sec = 0;
340 tv.tv_usec = 0;
341 } else {
342 tv.tv_sec = timeo / HZ;
343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
344 }
345
346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
348 *(struct old_timeval32 *)optval = tv32;
349 return sizeof(tv32);
350 }
351
352 if (old_timeval) {
353 struct __kernel_old_timeval old_tv;
354 old_tv.tv_sec = tv.tv_sec;
355 old_tv.tv_usec = tv.tv_usec;
356 *(struct __kernel_old_timeval *)optval = old_tv;
357 size = sizeof(old_tv);
358 } else {
359 *(struct __kernel_sock_timeval *)optval = tv;
360 size = sizeof(tv);
361 }
362
363 return size;
364}
365
366static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval)
367{
368 struct __kernel_sock_timeval tv;
369
370 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
371 struct old_timeval32 tv32;
372
373 if (optlen < sizeof(tv32))
374 return -EINVAL;
375
376 if (copy_from_user(&tv32, optval, sizeof(tv32)))
377 return -EFAULT;
378 tv.tv_sec = tv32.tv_sec;
379 tv.tv_usec = tv32.tv_usec;
380 } else if (old_timeval) {
381 struct __kernel_old_timeval old_tv;
382
383 if (optlen < sizeof(old_tv))
384 return -EINVAL;
385 if (copy_from_user(&old_tv, optval, sizeof(old_tv)))
386 return -EFAULT;
387 tv.tv_sec = old_tv.tv_sec;
388 tv.tv_usec = old_tv.tv_usec;
389 } else {
390 if (optlen < sizeof(tv))
391 return -EINVAL;
392 if (copy_from_user(&tv, optval, sizeof(tv)))
393 return -EFAULT;
394 }
395 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
396 return -EDOM;
397
398 if (tv.tv_sec < 0) {
399 static int warned __read_mostly;
400
401 *timeo_p = 0;
402 if (warned < 10 && net_ratelimit()) {
403 warned++;
404 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
405 __func__, current->comm, task_pid_nr(current));
406 }
407 return 0;
408 }
409 *timeo_p = MAX_SCHEDULE_TIMEOUT;
410 if (tv.tv_sec == 0 && tv.tv_usec == 0)
411 return 0;
412 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))
413 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ);
414 return 0;
415}
416
417static void sock_warn_obsolete_bsdism(const char *name)
418{
419 static int warned;
420 static char warncomm[TASK_COMM_LEN];
421 if (strcmp(warncomm, current->comm) && warned < 5) {
422 strcpy(warncomm, current->comm);
423 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
424 warncomm, name);
425 warned++;
426 }
427}
428
429static bool sock_needs_netstamp(const struct sock *sk)
430{
431 switch (sk->sk_family) {
432 case AF_UNSPEC:
433 case AF_UNIX:
434 return false;
435 default:
436 return true;
437 }
438}
439
440static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
441{
442 if (sk->sk_flags & flags) {
443 sk->sk_flags &= ~flags;
444 if (sock_needs_netstamp(sk) &&
445 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
446 net_disable_timestamp();
447 }
448}
449
450
451int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
452{
453 unsigned long flags;
454 struct sk_buff_head *list = &sk->sk_receive_queue;
455
456 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
457 atomic_inc(&sk->sk_drops);
458 trace_sock_rcvqueue_full(sk, skb);
459 return -ENOMEM;
460 }
461
462 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
463 atomic_inc(&sk->sk_drops);
464 return -ENOBUFS;
465 }
466
467 skb->dev = NULL;
468 skb_set_owner_r(skb, sk);
469
470 /* we escape from rcu protected region, make sure we dont leak
471 * a norefcounted dst
472 */
473 skb_dst_force(skb);
474
475 spin_lock_irqsave(&list->lock, flags);
476 sock_skb_set_dropcount(sk, skb);
477 __skb_queue_tail(list, skb);
478 spin_unlock_irqrestore(&list->lock, flags);
479
480 if (!sock_flag(sk, SOCK_DEAD))
481 sk->sk_data_ready(sk);
482 return 0;
483}
484EXPORT_SYMBOL(__sock_queue_rcv_skb);
485
486int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
487{
488 int err;
489
490 err = sk_filter(sk, skb);
491 if (err)
492 return err;
493
494 return __sock_queue_rcv_skb(sk, skb);
495}
496EXPORT_SYMBOL(sock_queue_rcv_skb);
497
498int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
499 const int nested, unsigned int trim_cap, bool refcounted)
500{
501 int rc = NET_RX_SUCCESS;
502
503 if (sk_filter_trim_cap(sk, skb, trim_cap))
504 goto discard_and_relse;
505
506 skb->dev = NULL;
507
508 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
509 atomic_inc(&sk->sk_drops);
510 goto discard_and_relse;
511 }
512 if (nested)
513 bh_lock_sock_nested(sk);
514 else
515 bh_lock_sock(sk);
516 if (!sock_owned_by_user(sk)) {
517 /*
518 * trylock + unlock semantics:
519 */
520 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
521
522 rc = sk_backlog_rcv(sk, skb);
523
524 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
525 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) {
526 bh_unlock_sock(sk);
527 atomic_inc(&sk->sk_drops);
528 goto discard_and_relse;
529 }
530
531 bh_unlock_sock(sk);
532out:
533 if (refcounted)
534 sock_put(sk);
535 return rc;
536discard_and_relse:
537 kfree_skb(skb);
538 goto out;
539}
540EXPORT_SYMBOL(__sk_receive_skb);
541
542struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
543{
544 struct dst_entry *dst = __sk_dst_get(sk);
545
546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
547 sk_tx_queue_clear(sk);
548 sk->sk_dst_pending_confirm = 0;
549 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
550 dst_release(dst);
551 return NULL;
552 }
553
554 return dst;
555}
556EXPORT_SYMBOL(__sk_dst_check);
557
558struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
559{
560 struct dst_entry *dst = sk_dst_get(sk);
561
562 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
563 sk_dst_reset(sk);
564 dst_release(dst);
565 return NULL;
566 }
567
568 return dst;
569}
570EXPORT_SYMBOL(sk_dst_check);
571
572static int sock_setbindtodevice_locked(struct sock *sk, int ifindex)
573{
574 int ret = -ENOPROTOOPT;
575#ifdef CONFIG_NETDEVICES
576 struct net *net = sock_net(sk);
577
578 /* Sorry... */
579 ret = -EPERM;
580 if (!ns_capable(net->user_ns, CAP_NET_RAW))
581 goto out;
582
583 ret = -EINVAL;
584 if (ifindex < 0)
585 goto out;
586
587 sk->sk_bound_dev_if = ifindex;
588 if (sk->sk_prot->rehash)
589 sk->sk_prot->rehash(sk);
590 sk_dst_reset(sk);
591
592 ret = 0;
593
594out:
595#endif
596
597 return ret;
598}
599
600static int sock_setbindtodevice(struct sock *sk, char __user *optval,
601 int optlen)
602{
603 int ret = -ENOPROTOOPT;
604#ifdef CONFIG_NETDEVICES
605 struct net *net = sock_net(sk);
606 char devname[IFNAMSIZ];
607 int index;
608
609 ret = -EINVAL;
610 if (optlen < 0)
611 goto out;
612
613 /* Bind this socket to a particular device like "eth0",
614 * as specified in the passed interface name. If the
615 * name is "" or the option length is zero the socket
616 * is not bound.
617 */
618 if (optlen > IFNAMSIZ - 1)
619 optlen = IFNAMSIZ - 1;
620 memset(devname, 0, sizeof(devname));
621
622 ret = -EFAULT;
623 if (copy_from_user(devname, optval, optlen))
624 goto out;
625
626 index = 0;
627 if (devname[0] != '\0') {
628 struct net_device *dev;
629
630 rcu_read_lock();
631 dev = dev_get_by_name_rcu(net, devname);
632 if (dev)
633 index = dev->ifindex;
634 rcu_read_unlock();
635 ret = -ENODEV;
636 if (!dev)
637 goto out;
638 }
639
640 lock_sock(sk);
641 ret = sock_setbindtodevice_locked(sk, index);
642 release_sock(sk);
643
644out:
645#endif
646
647 return ret;
648}
649
650static int sock_getbindtodevice(struct sock *sk, char __user *optval,
651 int __user *optlen, int len)
652{
653 int ret = -ENOPROTOOPT;
654#ifdef CONFIG_NETDEVICES
655 struct net *net = sock_net(sk);
656 char devname[IFNAMSIZ];
657
658 if (sk->sk_bound_dev_if == 0) {
659 len = 0;
660 goto zero;
661 }
662
663 ret = -EINVAL;
664 if (len < IFNAMSIZ)
665 goto out;
666
667 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
668 if (ret)
669 goto out;
670
671 len = strlen(devname) + 1;
672
673 ret = -EFAULT;
674 if (copy_to_user(optval, devname, len))
675 goto out;
676
677zero:
678 ret = -EFAULT;
679 if (put_user(len, optlen))
680 goto out;
681
682 ret = 0;
683
684out:
685#endif
686
687 return ret;
688}
689
690static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
691{
692 if (valbool)
693 sock_set_flag(sk, bit);
694 else
695 sock_reset_flag(sk, bit);
696}
697
698bool sk_mc_loop(struct sock *sk)
699{
700 if (dev_recursion_level())
701 return false;
702 if (!sk)
703 return true;
704 switch (sk->sk_family) {
705 case AF_INET:
706 return inet_sk(sk)->mc_loop;
707#if IS_ENABLED(CONFIG_IPV6)
708 case AF_INET6:
709 return inet6_sk(sk)->mc_loop;
710#endif
711 }
712 WARN_ON(1);
713 return true;
714}
715EXPORT_SYMBOL(sk_mc_loop);
716
717/*
718 * This is meant for all protocols to use and covers goings on
719 * at the socket level. Everything here is generic.
720 */
721
722int sock_setsockopt(struct socket *sock, int level, int optname,
723 char __user *optval, unsigned int optlen)
724{
725 struct sock_txtime sk_txtime;
726 struct sock *sk = sock->sk;
727 int val;
728 int valbool;
729 struct linger ling;
730 int ret = 0;
731
732 /*
733 * Options without arguments
734 */
735
736 if (optname == SO_BINDTODEVICE)
737 return sock_setbindtodevice(sk, optval, optlen);
738
739 if (optlen < sizeof(int))
740 return -EINVAL;
741
742 if (get_user(val, (int __user *)optval))
743 return -EFAULT;
744
745 valbool = val ? 1 : 0;
746
747 lock_sock(sk);
748
749 switch (optname) {
750 case SO_DEBUG:
751 if (val && !capable(CAP_NET_ADMIN))
752 ret = -EACCES;
753 else
754 sock_valbool_flag(sk, SOCK_DBG, valbool);
755 break;
756 case SO_REUSEADDR:
757 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
758 break;
759 case SO_REUSEPORT:
760 sk->sk_reuseport = valbool;
761 break;
762 case SO_TYPE:
763 case SO_PROTOCOL:
764 case SO_DOMAIN:
765 case SO_ERROR:
766 ret = -ENOPROTOOPT;
767 break;
768 case SO_DONTROUTE:
769 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
770 sk_dst_reset(sk);
771 break;
772 case SO_BROADCAST:
773 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
774 break;
775 case SO_SNDBUF:
776 /* Don't error on this BSD doesn't and if you think
777 * about it this is right. Otherwise apps have to
778 * play 'guess the biggest size' games. RCVBUF/SNDBUF
779 * are treated in BSD as hints
780 */
781 val = min_t(u32, val, sysctl_wmem_max);
782set_sndbuf:
783 /* Ensure val * 2 fits into an int, to prevent max_t()
784 * from treating it as a negative value.
785 */
786 val = min_t(int, val, INT_MAX / 2);
787 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
788 WRITE_ONCE(sk->sk_sndbuf,
789 max_t(int, val * 2, SOCK_MIN_SNDBUF));
790 /* Wake up sending tasks if we upped the value. */
791 sk->sk_write_space(sk);
792 break;
793
794 case SO_SNDBUFFORCE:
795 if (!capable(CAP_NET_ADMIN)) {
796 ret = -EPERM;
797 break;
798 }
799
800 /* No negative values (to prevent underflow, as val will be
801 * multiplied by 2).
802 */
803 if (val < 0)
804 val = 0;
805 goto set_sndbuf;
806
807 case SO_RCVBUF:
808 /* Don't error on this BSD doesn't and if you think
809 * about it this is right. Otherwise apps have to
810 * play 'guess the biggest size' games. RCVBUF/SNDBUF
811 * are treated in BSD as hints
812 */
813 val = min_t(u32, val, sysctl_rmem_max);
814set_rcvbuf:
815 /* Ensure val * 2 fits into an int, to prevent max_t()
816 * from treating it as a negative value.
817 */
818 val = min_t(int, val, INT_MAX / 2);
819 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
820 /*
821 * We double it on the way in to account for
822 * "struct sk_buff" etc. overhead. Applications
823 * assume that the SO_RCVBUF setting they make will
824 * allow that much actual data to be received on that
825 * socket.
826 *
827 * Applications are unaware that "struct sk_buff" and
828 * other overheads allocate from the receive buffer
829 * during socket buffer allocation.
830 *
831 * And after considering the possible alternatives,
832 * returning the value we actually used in getsockopt
833 * is the most desirable behavior.
834 */
835 WRITE_ONCE(sk->sk_rcvbuf,
836 max_t(int, val * 2, SOCK_MIN_RCVBUF));
837 break;
838
839 case SO_RCVBUFFORCE:
840 if (!capable(CAP_NET_ADMIN)) {
841 ret = -EPERM;
842 break;
843 }
844
845 /* No negative values (to prevent underflow, as val will be
846 * multiplied by 2).
847 */
848 if (val < 0)
849 val = 0;
850 goto set_rcvbuf;
851
852 case SO_KEEPALIVE:
853 if (sk->sk_prot->keepalive)
854 sk->sk_prot->keepalive(sk, valbool);
855 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
856 break;
857
858 case SO_OOBINLINE:
859 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
860 break;
861
862 case SO_NO_CHECK:
863 sk->sk_no_check_tx = valbool;
864 break;
865
866 case SO_PRIORITY:
867 if ((val >= 0 && val <= 6) ||
868 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
869 sk->sk_priority = val;
870 else
871 ret = -EPERM;
872 break;
873
874 case SO_LINGER:
875 if (optlen < sizeof(ling)) {
876 ret = -EINVAL; /* 1003.1g */
877 break;
878 }
879 if (copy_from_user(&ling, optval, sizeof(ling))) {
880 ret = -EFAULT;
881 break;
882 }
883 if (!ling.l_onoff)
884 sock_reset_flag(sk, SOCK_LINGER);
885 else {
886#if (BITS_PER_LONG == 32)
887 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
888 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
889 else
890#endif
891 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
892 sock_set_flag(sk, SOCK_LINGER);
893 }
894 break;
895
896 case SO_BSDCOMPAT:
897 sock_warn_obsolete_bsdism("setsockopt");
898 break;
899
900 case SO_PASSCRED:
901 if (valbool)
902 set_bit(SOCK_PASSCRED, &sock->flags);
903 else
904 clear_bit(SOCK_PASSCRED, &sock->flags);
905 break;
906
907 case SO_TIMESTAMP_OLD:
908 case SO_TIMESTAMP_NEW:
909 case SO_TIMESTAMPNS_OLD:
910 case SO_TIMESTAMPNS_NEW:
911 if (valbool) {
912 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW)
913 sock_set_flag(sk, SOCK_TSTAMP_NEW);
914 else
915 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
916
917 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW)
918 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
919 else
920 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
921 sock_set_flag(sk, SOCK_RCVTSTAMP);
922 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
923 } else {
924 sock_reset_flag(sk, SOCK_RCVTSTAMP);
925 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
926 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
927 }
928 break;
929
930 case SO_TIMESTAMPING_NEW:
931 sock_set_flag(sk, SOCK_TSTAMP_NEW);
932 /* fall through */
933 case SO_TIMESTAMPING_OLD:
934 if (val & ~SOF_TIMESTAMPING_MASK) {
935 ret = -EINVAL;
936 break;
937 }
938
939 if (val & SOF_TIMESTAMPING_OPT_ID &&
940 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
941 if (sk->sk_protocol == IPPROTO_TCP &&
942 sk->sk_type == SOCK_STREAM) {
943 if ((1 << sk->sk_state) &
944 (TCPF_CLOSE | TCPF_LISTEN)) {
945 ret = -EINVAL;
946 break;
947 }
948 sk->sk_tskey = tcp_sk(sk)->snd_una;
949 } else {
950 sk->sk_tskey = 0;
951 }
952 }
953
954 if (val & SOF_TIMESTAMPING_OPT_STATS &&
955 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
956 ret = -EINVAL;
957 break;
958 }
959
960 sk->sk_tsflags = val;
961 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
962 sock_enable_timestamp(sk,
963 SOCK_TIMESTAMPING_RX_SOFTWARE);
964 else {
965 if (optname == SO_TIMESTAMPING_NEW)
966 sock_reset_flag(sk, SOCK_TSTAMP_NEW);
967
968 sock_disable_timestamp(sk,
969 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
970 }
971 break;
972
973 case SO_RCVLOWAT:
974 if (val < 0)
975 val = INT_MAX;
976 if (sock->ops->set_rcvlowat)
977 ret = sock->ops->set_rcvlowat(sk, val);
978 else
979 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
980 break;
981
982 case SO_RCVTIMEO_OLD:
983 case SO_RCVTIMEO_NEW:
984 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD);
985 break;
986
987 case SO_SNDTIMEO_OLD:
988 case SO_SNDTIMEO_NEW:
989 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD);
990 break;
991
992 case SO_ATTACH_FILTER:
993 ret = -EINVAL;
994 if (optlen == sizeof(struct sock_fprog)) {
995 struct sock_fprog fprog;
996
997 ret = -EFAULT;
998 if (copy_from_user(&fprog, optval, sizeof(fprog)))
999 break;
1000
1001 ret = sk_attach_filter(&fprog, sk);
1002 }
1003 break;
1004
1005 case SO_ATTACH_BPF:
1006 ret = -EINVAL;
1007 if (optlen == sizeof(u32)) {
1008 u32 ufd;
1009
1010 ret = -EFAULT;
1011 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1012 break;
1013
1014 ret = sk_attach_bpf(ufd, sk);
1015 }
1016 break;
1017
1018 case SO_ATTACH_REUSEPORT_CBPF:
1019 ret = -EINVAL;
1020 if (optlen == sizeof(struct sock_fprog)) {
1021 struct sock_fprog fprog;
1022
1023 ret = -EFAULT;
1024 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1025 break;
1026
1027 ret = sk_reuseport_attach_filter(&fprog, sk);
1028 }
1029 break;
1030
1031 case SO_ATTACH_REUSEPORT_EBPF:
1032 ret = -EINVAL;
1033 if (optlen == sizeof(u32)) {
1034 u32 ufd;
1035
1036 ret = -EFAULT;
1037 if (copy_from_user(&ufd, optval, sizeof(ufd)))
1038 break;
1039
1040 ret = sk_reuseport_attach_bpf(ufd, sk);
1041 }
1042 break;
1043
1044 case SO_DETACH_REUSEPORT_BPF:
1045 ret = reuseport_detach_prog(sk);
1046 break;
1047
1048 case SO_DETACH_FILTER:
1049 ret = sk_detach_filter(sk);
1050 break;
1051
1052 case SO_LOCK_FILTER:
1053 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1054 ret = -EPERM;
1055 else
1056 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1057 break;
1058
1059 case SO_PASSSEC:
1060 if (valbool)
1061 set_bit(SOCK_PASSSEC, &sock->flags);
1062 else
1063 clear_bit(SOCK_PASSSEC, &sock->flags);
1064 break;
1065 case SO_MARK:
1066 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1067 ret = -EPERM;
1068 } else if (val != sk->sk_mark) {
1069 sk->sk_mark = val;
1070 sk_dst_reset(sk);
1071 }
1072 break;
1073
1074 case SO_RXQ_OVFL:
1075 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1076 break;
1077
1078 case SO_WIFI_STATUS:
1079 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1080 break;
1081
1082 case SO_PEEK_OFF:
1083 if (sock->ops->set_peek_off)
1084 ret = sock->ops->set_peek_off(sk, val);
1085 else
1086 ret = -EOPNOTSUPP;
1087 break;
1088
1089 case SO_NOFCS:
1090 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1091 break;
1092
1093 case SO_SELECT_ERR_QUEUE:
1094 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1095 break;
1096
1097#ifdef CONFIG_NET_RX_BUSY_POLL
1098 case SO_BUSY_POLL:
1099 /* allow unprivileged users to decrease the value */
1100 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1101 ret = -EPERM;
1102 else {
1103 if (val < 0)
1104 ret = -EINVAL;
1105 else
1106 sk->sk_ll_usec = val;
1107 }
1108 break;
1109#endif
1110
1111 case SO_MAX_PACING_RATE:
1112 {
1113 unsigned long ulval = (val == ~0U) ? ~0UL : val;
1114
1115 if (sizeof(ulval) != sizeof(val) &&
1116 optlen >= sizeof(ulval) &&
1117 get_user(ulval, (unsigned long __user *)optval)) {
1118 ret = -EFAULT;
1119 break;
1120 }
1121 if (ulval != ~0UL)
1122 cmpxchg(&sk->sk_pacing_status,
1123 SK_PACING_NONE,
1124 SK_PACING_NEEDED);
1125 sk->sk_max_pacing_rate = ulval;
1126 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval);
1127 break;
1128 }
1129 case SO_INCOMING_CPU:
1130 WRITE_ONCE(sk->sk_incoming_cpu, val);
1131 break;
1132
1133 case SO_CNX_ADVICE:
1134 if (val == 1)
1135 dst_negative_advice(sk);
1136 break;
1137
1138 case SO_ZEROCOPY:
1139 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1140 if (!((sk->sk_type == SOCK_STREAM &&
1141 sk->sk_protocol == IPPROTO_TCP) ||
1142 (sk->sk_type == SOCK_DGRAM &&
1143 sk->sk_protocol == IPPROTO_UDP)))
1144 ret = -ENOTSUPP;
1145 } else if (sk->sk_family != PF_RDS) {
1146 ret = -ENOTSUPP;
1147 }
1148 if (!ret) {
1149 if (val < 0 || val > 1)
1150 ret = -EINVAL;
1151 else
1152 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1153 }
1154 break;
1155
1156 case SO_TXTIME:
1157 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1158 ret = -EPERM;
1159 } else if (optlen != sizeof(struct sock_txtime)) {
1160 ret = -EINVAL;
1161 } else if (copy_from_user(&sk_txtime, optval,
1162 sizeof(struct sock_txtime))) {
1163 ret = -EFAULT;
1164 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1165 ret = -EINVAL;
1166 } else {
1167 sock_valbool_flag(sk, SOCK_TXTIME, true);
1168 sk->sk_clockid = sk_txtime.clockid;
1169 sk->sk_txtime_deadline_mode =
1170 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1171 sk->sk_txtime_report_errors =
1172 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1173 }
1174 break;
1175
1176 case SO_BINDTOIFINDEX:
1177 ret = sock_setbindtodevice_locked(sk, val);
1178 break;
1179
1180 default:
1181 ret = -ENOPROTOOPT;
1182 break;
1183 }
1184 release_sock(sk);
1185 return ret;
1186}
1187EXPORT_SYMBOL(sock_setsockopt);
1188
1189
1190static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1191 struct ucred *ucred)
1192{
1193 ucred->pid = pid_vnr(pid);
1194 ucred->uid = ucred->gid = -1;
1195 if (cred) {
1196 struct user_namespace *current_ns = current_user_ns();
1197
1198 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1199 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1200 }
1201}
1202
1203static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1204{
1205 struct user_namespace *user_ns = current_user_ns();
1206 int i;
1207
1208 for (i = 0; i < src->ngroups; i++)
1209 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1210 return -EFAULT;
1211
1212 return 0;
1213}
1214
1215int sock_getsockopt(struct socket *sock, int level, int optname,
1216 char __user *optval, int __user *optlen)
1217{
1218 struct sock *sk = sock->sk;
1219
1220 union {
1221 int val;
1222 u64 val64;
1223 unsigned long ulval;
1224 struct linger ling;
1225 struct old_timeval32 tm32;
1226 struct __kernel_old_timeval tm;
1227 struct __kernel_sock_timeval stm;
1228 struct sock_txtime txtime;
1229 } v;
1230
1231 int lv = sizeof(int);
1232 int len;
1233
1234 if (get_user(len, optlen))
1235 return -EFAULT;
1236 if (len < 0)
1237 return -EINVAL;
1238
1239 memset(&v, 0, sizeof(v));
1240
1241 switch (optname) {
1242 case SO_DEBUG:
1243 v.val = sock_flag(sk, SOCK_DBG);
1244 break;
1245
1246 case SO_DONTROUTE:
1247 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1248 break;
1249
1250 case SO_BROADCAST:
1251 v.val = sock_flag(sk, SOCK_BROADCAST);
1252 break;
1253
1254 case SO_SNDBUF:
1255 v.val = sk->sk_sndbuf;
1256 break;
1257
1258 case SO_RCVBUF:
1259 v.val = sk->sk_rcvbuf;
1260 break;
1261
1262 case SO_REUSEADDR:
1263 v.val = sk->sk_reuse;
1264 break;
1265
1266 case SO_REUSEPORT:
1267 v.val = sk->sk_reuseport;
1268 break;
1269
1270 case SO_KEEPALIVE:
1271 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1272 break;
1273
1274 case SO_TYPE:
1275 v.val = sk->sk_type;
1276 break;
1277
1278 case SO_PROTOCOL:
1279 v.val = sk->sk_protocol;
1280 break;
1281
1282 case SO_DOMAIN:
1283 v.val = sk->sk_family;
1284 break;
1285
1286 case SO_ERROR:
1287 v.val = -sock_error(sk);
1288 if (v.val == 0)
1289 v.val = xchg(&sk->sk_err_soft, 0);
1290 break;
1291
1292 case SO_OOBINLINE:
1293 v.val = sock_flag(sk, SOCK_URGINLINE);
1294 break;
1295
1296 case SO_NO_CHECK:
1297 v.val = sk->sk_no_check_tx;
1298 break;
1299
1300 case SO_PRIORITY:
1301 v.val = sk->sk_priority;
1302 break;
1303
1304 case SO_LINGER:
1305 lv = sizeof(v.ling);
1306 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1307 v.ling.l_linger = sk->sk_lingertime / HZ;
1308 break;
1309
1310 case SO_BSDCOMPAT:
1311 sock_warn_obsolete_bsdism("getsockopt");
1312 break;
1313
1314 case SO_TIMESTAMP_OLD:
1315 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1316 !sock_flag(sk, SOCK_TSTAMP_NEW) &&
1317 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1318 break;
1319
1320 case SO_TIMESTAMPNS_OLD:
1321 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1322 break;
1323
1324 case SO_TIMESTAMP_NEW:
1325 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1326 break;
1327
1328 case SO_TIMESTAMPNS_NEW:
1329 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1330 break;
1331
1332 case SO_TIMESTAMPING_OLD:
1333 v.val = sk->sk_tsflags;
1334 break;
1335
1336 case SO_RCVTIMEO_OLD:
1337 case SO_RCVTIMEO_NEW:
1338 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname);
1339 break;
1340
1341 case SO_SNDTIMEO_OLD:
1342 case SO_SNDTIMEO_NEW:
1343 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname);
1344 break;
1345
1346 case SO_RCVLOWAT:
1347 v.val = sk->sk_rcvlowat;
1348 break;
1349
1350 case SO_SNDLOWAT:
1351 v.val = 1;
1352 break;
1353
1354 case SO_PASSCRED:
1355 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1356 break;
1357
1358 case SO_PEERCRED:
1359 {
1360 struct ucred peercred;
1361 if (len > sizeof(peercred))
1362 len = sizeof(peercred);
1363 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1364 if (copy_to_user(optval, &peercred, len))
1365 return -EFAULT;
1366 goto lenout;
1367 }
1368
1369 case SO_PEERGROUPS:
1370 {
1371 int ret, n;
1372
1373 if (!sk->sk_peer_cred)
1374 return -ENODATA;
1375
1376 n = sk->sk_peer_cred->group_info->ngroups;
1377 if (len < n * sizeof(gid_t)) {
1378 len = n * sizeof(gid_t);
1379 return put_user(len, optlen) ? -EFAULT : -ERANGE;
1380 }
1381 len = n * sizeof(gid_t);
1382
1383 ret = groups_to_user((gid_t __user *)optval,
1384 sk->sk_peer_cred->group_info);
1385 if (ret)
1386 return ret;
1387 goto lenout;
1388 }
1389
1390 case SO_PEERNAME:
1391 {
1392 char address[128];
1393
1394 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
1395 if (lv < 0)
1396 return -ENOTCONN;
1397 if (lv < len)
1398 return -EINVAL;
1399 if (copy_to_user(optval, address, len))
1400 return -EFAULT;
1401 goto lenout;
1402 }
1403
1404 /* Dubious BSD thing... Probably nobody even uses it, but
1405 * the UNIX standard wants it for whatever reason... -DaveM
1406 */
1407 case SO_ACCEPTCONN:
1408 v.val = sk->sk_state == TCP_LISTEN;
1409 break;
1410
1411 case SO_PASSSEC:
1412 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1413 break;
1414
1415 case SO_PEERSEC:
1416 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1417
1418 case SO_MARK:
1419 v.val = sk->sk_mark;
1420 break;
1421
1422 case SO_RXQ_OVFL:
1423 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1424 break;
1425
1426 case SO_WIFI_STATUS:
1427 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1428 break;
1429
1430 case SO_PEEK_OFF:
1431 if (!sock->ops->set_peek_off)
1432 return -EOPNOTSUPP;
1433
1434 v.val = sk->sk_peek_off;
1435 break;
1436 case SO_NOFCS:
1437 v.val = sock_flag(sk, SOCK_NOFCS);
1438 break;
1439
1440 case SO_BINDTODEVICE:
1441 return sock_getbindtodevice(sk, optval, optlen, len);
1442
1443 case SO_GET_FILTER:
1444 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1445 if (len < 0)
1446 return len;
1447
1448 goto lenout;
1449
1450 case SO_LOCK_FILTER:
1451 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1452 break;
1453
1454 case SO_BPF_EXTENSIONS:
1455 v.val = bpf_tell_extensions();
1456 break;
1457
1458 case SO_SELECT_ERR_QUEUE:
1459 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1460 break;
1461
1462#ifdef CONFIG_NET_RX_BUSY_POLL
1463 case SO_BUSY_POLL:
1464 v.val = sk->sk_ll_usec;
1465 break;
1466#endif
1467
1468 case SO_MAX_PACING_RATE:
1469 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
1470 lv = sizeof(v.ulval);
1471 v.ulval = sk->sk_max_pacing_rate;
1472 } else {
1473 /* 32bit version */
1474 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U);
1475 }
1476 break;
1477
1478 case SO_INCOMING_CPU:
1479 v.val = READ_ONCE(sk->sk_incoming_cpu);
1480 break;
1481
1482 case SO_MEMINFO:
1483 {
1484 u32 meminfo[SK_MEMINFO_VARS];
1485
1486 sk_get_meminfo(sk, meminfo);
1487
1488 len = min_t(unsigned int, len, sizeof(meminfo));
1489 if (copy_to_user(optval, &meminfo, len))
1490 return -EFAULT;
1491
1492 goto lenout;
1493 }
1494
1495#ifdef CONFIG_NET_RX_BUSY_POLL
1496 case SO_INCOMING_NAPI_ID:
1497 v.val = READ_ONCE(sk->sk_napi_id);
1498
1499 /* aggregate non-NAPI IDs down to 0 */
1500 if (v.val < MIN_NAPI_ID)
1501 v.val = 0;
1502
1503 break;
1504#endif
1505
1506 case SO_COOKIE:
1507 lv = sizeof(u64);
1508 if (len < lv)
1509 return -EINVAL;
1510 v.val64 = sock_gen_cookie(sk);
1511 break;
1512
1513 case SO_ZEROCOPY:
1514 v.val = sock_flag(sk, SOCK_ZEROCOPY);
1515 break;
1516
1517 case SO_TXTIME:
1518 lv = sizeof(v.txtime);
1519 v.txtime.clockid = sk->sk_clockid;
1520 v.txtime.flags |= sk->sk_txtime_deadline_mode ?
1521 SOF_TXTIME_DEADLINE_MODE : 0;
1522 v.txtime.flags |= sk->sk_txtime_report_errors ?
1523 SOF_TXTIME_REPORT_ERRORS : 0;
1524 break;
1525
1526 case SO_BINDTOIFINDEX:
1527 v.val = sk->sk_bound_dev_if;
1528 break;
1529
1530 default:
1531 /* We implement the SO_SNDLOWAT etc to not be settable
1532 * (1003.1g 7).
1533 */
1534 return -ENOPROTOOPT;
1535 }
1536
1537 if (len > lv)
1538 len = lv;
1539 if (copy_to_user(optval, &v, len))
1540 return -EFAULT;
1541lenout:
1542 if (put_user(len, optlen))
1543 return -EFAULT;
1544 return 0;
1545}
1546
1547/*
1548 * Initialize an sk_lock.
1549 *
1550 * (We also register the sk_lock with the lock validator.)
1551 */
1552static inline void sock_lock_init(struct sock *sk)
1553{
1554 if (sk->sk_kern_sock)
1555 sock_lock_init_class_and_name(
1556 sk,
1557 af_family_kern_slock_key_strings[sk->sk_family],
1558 af_family_kern_slock_keys + sk->sk_family,
1559 af_family_kern_key_strings[sk->sk_family],
1560 af_family_kern_keys + sk->sk_family);
1561 else
1562 sock_lock_init_class_and_name(
1563 sk,
1564 af_family_slock_key_strings[sk->sk_family],
1565 af_family_slock_keys + sk->sk_family,
1566 af_family_key_strings[sk->sk_family],
1567 af_family_keys + sk->sk_family);
1568}
1569
1570/*
1571 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1572 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1573 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1574 */
1575static void sock_copy(struct sock *nsk, const struct sock *osk)
1576{
1577#ifdef CONFIG_SECURITY_NETWORK
1578 void *sptr = nsk->sk_security;
1579#endif
1580 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1581
1582 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1583 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1584
1585#ifdef CONFIG_SECURITY_NETWORK
1586 nsk->sk_security = sptr;
1587 security_sk_clone(osk, nsk);
1588#endif
1589}
1590
1591static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1592 int family)
1593{
1594 struct sock *sk;
1595 struct kmem_cache *slab;
1596
1597 slab = prot->slab;
1598 if (slab != NULL) {
1599 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1600 if (!sk)
1601 return sk;
1602 if (want_init_on_alloc(priority))
1603 sk_prot_clear_nulls(sk, prot->obj_size);
1604 } else
1605 sk = kmalloc(prot->obj_size, priority);
1606
1607 if (sk != NULL) {
1608 if (security_sk_alloc(sk, family, priority))
1609 goto out_free;
1610
1611 if (!try_module_get(prot->owner))
1612 goto out_free_sec;
1613 sk_tx_queue_clear(sk);
1614 }
1615
1616 return sk;
1617
1618out_free_sec:
1619 security_sk_free(sk);
1620out_free:
1621 if (slab != NULL)
1622 kmem_cache_free(slab, sk);
1623 else
1624 kfree(sk);
1625 return NULL;
1626}
1627
1628static void sk_prot_free(struct proto *prot, struct sock *sk)
1629{
1630 struct kmem_cache *slab;
1631 struct module *owner;
1632
1633 owner = prot->owner;
1634 slab = prot->slab;
1635
1636 cgroup_sk_free(&sk->sk_cgrp_data);
1637 mem_cgroup_sk_free(sk);
1638 security_sk_free(sk);
1639 if (slab != NULL)
1640 kmem_cache_free(slab, sk);
1641 else
1642 kfree(sk);
1643 module_put(owner);
1644}
1645
1646/**
1647 * sk_alloc - All socket objects are allocated here
1648 * @net: the applicable net namespace
1649 * @family: protocol family
1650 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1651 * @prot: struct proto associated with this new sock instance
1652 * @kern: is this to be a kernel socket?
1653 */
1654struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1655 struct proto *prot, int kern)
1656{
1657 struct sock *sk;
1658
1659 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1660 if (sk) {
1661 sk->sk_family = family;
1662 /*
1663 * See comment in struct sock definition to understand
1664 * why we need sk_prot_creator -acme
1665 */
1666 sk->sk_prot = sk->sk_prot_creator = prot;
1667 sk->sk_kern_sock = kern;
1668 sock_lock_init(sk);
1669 sk->sk_net_refcnt = kern ? 0 : 1;
1670 if (likely(sk->sk_net_refcnt)) {
1671 get_net(net);
1672 sock_inuse_add(net, 1);
1673 }
1674
1675 sock_net_set(sk, net);
1676 refcount_set(&sk->sk_wmem_alloc, 1);
1677
1678 mem_cgroup_sk_alloc(sk);
1679 cgroup_sk_alloc(&sk->sk_cgrp_data);
1680 sock_update_classid(&sk->sk_cgrp_data);
1681 sock_update_netprioidx(&sk->sk_cgrp_data);
1682 }
1683
1684 return sk;
1685}
1686EXPORT_SYMBOL(sk_alloc);
1687
1688/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1689 * grace period. This is the case for UDP sockets and TCP listeners.
1690 */
1691static void __sk_destruct(struct rcu_head *head)
1692{
1693 struct sock *sk = container_of(head, struct sock, sk_rcu);
1694 struct sk_filter *filter;
1695
1696 if (sk->sk_destruct)
1697 sk->sk_destruct(sk);
1698
1699 filter = rcu_dereference_check(sk->sk_filter,
1700 refcount_read(&sk->sk_wmem_alloc) == 0);
1701 if (filter) {
1702 sk_filter_uncharge(sk, filter);
1703 RCU_INIT_POINTER(sk->sk_filter, NULL);
1704 }
1705
1706 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1707
1708#ifdef CONFIG_BPF_SYSCALL
1709 bpf_sk_storage_free(sk);
1710#endif
1711
1712 if (atomic_read(&sk->sk_omem_alloc))
1713 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1714 __func__, atomic_read(&sk->sk_omem_alloc));
1715
1716 if (sk->sk_frag.page) {
1717 put_page(sk->sk_frag.page);
1718 sk->sk_frag.page = NULL;
1719 }
1720
1721 if (sk->sk_peer_cred)
1722 put_cred(sk->sk_peer_cred);
1723 put_pid(sk->sk_peer_pid);
1724 if (likely(sk->sk_net_refcnt))
1725 put_net(sock_net(sk));
1726 sk_prot_free(sk->sk_prot_creator, sk);
1727}
1728
1729void sk_destruct(struct sock *sk)
1730{
1731 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
1732
1733 if (rcu_access_pointer(sk->sk_reuseport_cb)) {
1734 reuseport_detach_sock(sk);
1735 use_call_rcu = true;
1736 }
1737
1738 if (use_call_rcu)
1739 call_rcu(&sk->sk_rcu, __sk_destruct);
1740 else
1741 __sk_destruct(&sk->sk_rcu);
1742}
1743
1744static void __sk_free(struct sock *sk)
1745{
1746 if (likely(sk->sk_net_refcnt))
1747 sock_inuse_add(sock_net(sk), -1);
1748
1749 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
1750 sock_diag_broadcast_destroy(sk);
1751 else
1752 sk_destruct(sk);
1753}
1754
1755void sk_free(struct sock *sk)
1756{
1757 /*
1758 * We subtract one from sk_wmem_alloc and can know if
1759 * some packets are still in some tx queue.
1760 * If not null, sock_wfree() will call __sk_free(sk) later
1761 */
1762 if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1763 __sk_free(sk);
1764}
1765EXPORT_SYMBOL(sk_free);
1766
1767static void sk_init_common(struct sock *sk)
1768{
1769 skb_queue_head_init(&sk->sk_receive_queue);
1770 skb_queue_head_init(&sk->sk_write_queue);
1771 skb_queue_head_init(&sk->sk_error_queue);
1772
1773 rwlock_init(&sk->sk_callback_lock);
1774 lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1775 af_rlock_keys + sk->sk_family,
1776 af_family_rlock_key_strings[sk->sk_family]);
1777 lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1778 af_wlock_keys + sk->sk_family,
1779 af_family_wlock_key_strings[sk->sk_family]);
1780 lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1781 af_elock_keys + sk->sk_family,
1782 af_family_elock_key_strings[sk->sk_family]);
1783 lockdep_set_class_and_name(&sk->sk_callback_lock,
1784 af_callback_keys + sk->sk_family,
1785 af_family_clock_key_strings[sk->sk_family]);
1786}
1787
1788/**
1789 * sk_clone_lock - clone a socket, and lock its clone
1790 * @sk: the socket to clone
1791 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1792 *
1793 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1794 */
1795struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1796{
1797 struct sock *newsk;
1798 bool is_charged = true;
1799
1800 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1801 if (newsk != NULL) {
1802 struct sk_filter *filter;
1803
1804 sock_copy(newsk, sk);
1805
1806 newsk->sk_prot_creator = sk->sk_prot;
1807
1808 /* SANITY */
1809 if (likely(newsk->sk_net_refcnt))
1810 get_net(sock_net(newsk));
1811 sk_node_init(&newsk->sk_node);
1812 sock_lock_init(newsk);
1813 bh_lock_sock(newsk);
1814 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1815 newsk->sk_backlog.len = 0;
1816
1817 atomic_set(&newsk->sk_rmem_alloc, 0);
1818 /*
1819 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1820 */
1821 refcount_set(&newsk->sk_wmem_alloc, 1);
1822 atomic_set(&newsk->sk_omem_alloc, 0);
1823 sk_init_common(newsk);
1824
1825 newsk->sk_dst_cache = NULL;
1826 newsk->sk_dst_pending_confirm = 0;
1827 newsk->sk_wmem_queued = 0;
1828 newsk->sk_forward_alloc = 0;
1829 atomic_set(&newsk->sk_drops, 0);
1830 newsk->sk_send_head = NULL;
1831 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1832 atomic_set(&newsk->sk_zckey, 0);
1833
1834 sock_reset_flag(newsk, SOCK_DONE);
1835 mem_cgroup_sk_alloc(newsk);
1836 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1837
1838 rcu_read_lock();
1839 filter = rcu_dereference(sk->sk_filter);
1840 if (filter != NULL)
1841 /* though it's an empty new sock, the charging may fail
1842 * if sysctl_optmem_max was changed between creation of
1843 * original socket and cloning
1844 */
1845 is_charged = sk_filter_charge(newsk, filter);
1846 RCU_INIT_POINTER(newsk->sk_filter, filter);
1847 rcu_read_unlock();
1848
1849 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1850 /* We need to make sure that we don't uncharge the new
1851 * socket if we couldn't charge it in the first place
1852 * as otherwise we uncharge the parent's filter.
1853 */
1854 if (!is_charged)
1855 RCU_INIT_POINTER(newsk->sk_filter, NULL);
1856 sk_free_unlock_clone(newsk);
1857 newsk = NULL;
1858 goto out;
1859 }
1860 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1861
1862 if (bpf_sk_storage_clone(sk, newsk)) {
1863 sk_free_unlock_clone(newsk);
1864 newsk = NULL;
1865 goto out;
1866 }
1867
1868 newsk->sk_err = 0;
1869 newsk->sk_err_soft = 0;
1870 newsk->sk_priority = 0;
1871 newsk->sk_incoming_cpu = raw_smp_processor_id();
1872 if (likely(newsk->sk_net_refcnt))
1873 sock_inuse_add(sock_net(newsk), 1);
1874
1875 /*
1876 * Before updating sk_refcnt, we must commit prior changes to memory
1877 * (Documentation/RCU/rculist_nulls.txt for details)
1878 */
1879 smp_wmb();
1880 refcount_set(&newsk->sk_refcnt, 2);
1881
1882 /*
1883 * Increment the counter in the same struct proto as the master
1884 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1885 * is the same as sk->sk_prot->socks, as this field was copied
1886 * with memcpy).
1887 *
1888 * This _changes_ the previous behaviour, where
1889 * tcp_create_openreq_child always was incrementing the
1890 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1891 * to be taken into account in all callers. -acme
1892 */
1893 sk_refcnt_debug_inc(newsk);
1894 sk_set_socket(newsk, NULL);
1895 RCU_INIT_POINTER(newsk->sk_wq, NULL);
1896
1897 if (newsk->sk_prot->sockets_allocated)
1898 sk_sockets_allocated_inc(newsk);
1899
1900 if (sock_needs_netstamp(sk) &&
1901 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1902 net_enable_timestamp();
1903 }
1904out:
1905 return newsk;
1906}
1907EXPORT_SYMBOL_GPL(sk_clone_lock);
1908
1909void sk_free_unlock_clone(struct sock *sk)
1910{
1911 /* It is still raw copy of parent, so invalidate
1912 * destructor and make plain sk_free() */
1913 sk->sk_destruct = NULL;
1914 bh_unlock_sock(sk);
1915 sk_free(sk);
1916}
1917EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1918
1919void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1920{
1921 u32 max_segs = 1;
1922
1923 sk_dst_set(sk, dst);
1924 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps;
1925 if (sk->sk_route_caps & NETIF_F_GSO)
1926 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1927 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1928 if (sk_can_gso(sk)) {
1929 if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1930 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1931 } else {
1932 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1933 sk->sk_gso_max_size = dst->dev->gso_max_size;
1934 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1935 }
1936 }
1937 sk->sk_gso_max_segs = max_segs;
1938}
1939EXPORT_SYMBOL_GPL(sk_setup_caps);
1940
1941/*
1942 * Simple resource managers for sockets.
1943 */
1944
1945
1946/*
1947 * Write buffer destructor automatically called from kfree_skb.
1948 */
1949void sock_wfree(struct sk_buff *skb)
1950{
1951 struct sock *sk = skb->sk;
1952 unsigned int len = skb->truesize;
1953
1954 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1955 /*
1956 * Keep a reference on sk_wmem_alloc, this will be released
1957 * after sk_write_space() call
1958 */
1959 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1960 sk->sk_write_space(sk);
1961 len = 1;
1962 }
1963 /*
1964 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1965 * could not do because of in-flight packets
1966 */
1967 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1968 __sk_free(sk);
1969}
1970EXPORT_SYMBOL(sock_wfree);
1971
1972/* This variant of sock_wfree() is used by TCP,
1973 * since it sets SOCK_USE_WRITE_QUEUE.
1974 */
1975void __sock_wfree(struct sk_buff *skb)
1976{
1977 struct sock *sk = skb->sk;
1978
1979 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1980 __sk_free(sk);
1981}
1982
1983void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1984{
1985 skb_orphan(skb);
1986 skb->sk = sk;
1987#ifdef CONFIG_INET
1988 if (unlikely(!sk_fullsock(sk))) {
1989 skb->destructor = sock_edemux;
1990 sock_hold(sk);
1991 return;
1992 }
1993#endif
1994 skb->destructor = sock_wfree;
1995 skb_set_hash_from_sk(skb, sk);
1996 /*
1997 * We used to take a refcount on sk, but following operation
1998 * is enough to guarantee sk_free() wont free this sock until
1999 * all in-flight packets are completed
2000 */
2001 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2002}
2003EXPORT_SYMBOL(skb_set_owner_w);
2004
2005static bool can_skb_orphan_partial(const struct sk_buff *skb)
2006{
2007#ifdef CONFIG_TLS_DEVICE
2008 /* Drivers depend on in-order delivery for crypto offload,
2009 * partial orphan breaks out-of-order-OK logic.
2010 */
2011 if (skb->decrypted)
2012 return false;
2013#endif
2014 return (skb->destructor == sock_wfree ||
2015 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2016}
2017
2018/* This helper is used by netem, as it can hold packets in its
2019 * delay queue. We want to allow the owner socket to send more
2020 * packets, as if they were already TX completed by a typical driver.
2021 * But we also want to keep skb->sk set because some packet schedulers
2022 * rely on it (sch_fq for example).
2023 */
2024void skb_orphan_partial(struct sk_buff *skb)
2025{
2026 if (skb_is_tcp_pure_ack(skb))
2027 return;
2028
2029 if (can_skb_orphan_partial(skb)) {
2030 struct sock *sk = skb->sk;
2031
2032 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
2033 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
2034 skb->destructor = sock_efree;
2035 }
2036 } else {
2037 skb_orphan(skb);
2038 }
2039}
2040EXPORT_SYMBOL(skb_orphan_partial);
2041
2042/*
2043 * Read buffer destructor automatically called from kfree_skb.
2044 */
2045void sock_rfree(struct sk_buff *skb)
2046{
2047 struct sock *sk = skb->sk;
2048 unsigned int len = skb->truesize;
2049
2050 atomic_sub(len, &sk->sk_rmem_alloc);
2051 sk_mem_uncharge(sk, len);
2052}
2053EXPORT_SYMBOL(sock_rfree);
2054
2055/*
2056 * Buffer destructor for skbs that are not used directly in read or write
2057 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2058 */
2059void sock_efree(struct sk_buff *skb)
2060{
2061 sock_put(skb->sk);
2062}
2063EXPORT_SYMBOL(sock_efree);
2064
2065kuid_t sock_i_uid(struct sock *sk)
2066{
2067 kuid_t uid;
2068
2069 read_lock_bh(&sk->sk_callback_lock);
2070 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
2071 read_unlock_bh(&sk->sk_callback_lock);
2072 return uid;
2073}
2074EXPORT_SYMBOL(sock_i_uid);
2075
2076unsigned long sock_i_ino(struct sock *sk)
2077{
2078 unsigned long ino;
2079
2080 read_lock_bh(&sk->sk_callback_lock);
2081 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
2082 read_unlock_bh(&sk->sk_callback_lock);
2083 return ino;
2084}
2085EXPORT_SYMBOL(sock_i_ino);
2086
2087/*
2088 * Allocate a skb from the socket's send buffer.
2089 */
2090struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2091 gfp_t priority)
2092{
2093 if (force ||
2094 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2095 struct sk_buff *skb = alloc_skb(size, priority);
2096
2097 if (skb) {
2098 skb_set_owner_w(skb, sk);
2099 return skb;
2100 }
2101 }
2102 return NULL;
2103}
2104EXPORT_SYMBOL(sock_wmalloc);
2105
2106static void sock_ofree(struct sk_buff *skb)
2107{
2108 struct sock *sk = skb->sk;
2109
2110 atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2111}
2112
2113struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2114 gfp_t priority)
2115{
2116 struct sk_buff *skb;
2117
2118 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2119 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2120 sysctl_optmem_max)
2121 return NULL;
2122
2123 skb = alloc_skb(size, priority);
2124 if (!skb)
2125 return NULL;
2126
2127 atomic_add(skb->truesize, &sk->sk_omem_alloc);
2128 skb->sk = sk;
2129 skb->destructor = sock_ofree;
2130 return skb;
2131}
2132
2133/*
2134 * Allocate a memory block from the socket's option memory buffer.
2135 */
2136void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2137{
2138 if ((unsigned int)size <= sysctl_optmem_max &&
2139 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
2140 void *mem;
2141 /* First do the add, to avoid the race if kmalloc
2142 * might sleep.
2143 */
2144 atomic_add(size, &sk->sk_omem_alloc);
2145 mem = kmalloc(size, priority);
2146 if (mem)
2147 return mem;
2148 atomic_sub(size, &sk->sk_omem_alloc);
2149 }
2150 return NULL;
2151}
2152EXPORT_SYMBOL(sock_kmalloc);
2153
2154/* Free an option memory block. Note, we actually want the inline
2155 * here as this allows gcc to detect the nullify and fold away the
2156 * condition entirely.
2157 */
2158static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2159 const bool nullify)
2160{
2161 if (WARN_ON_ONCE(!mem))
2162 return;
2163 if (nullify)
2164 kzfree(mem);
2165 else
2166 kfree(mem);
2167 atomic_sub(size, &sk->sk_omem_alloc);
2168}
2169
2170void sock_kfree_s(struct sock *sk, void *mem, int size)
2171{
2172 __sock_kfree_s(sk, mem, size, false);
2173}
2174EXPORT_SYMBOL(sock_kfree_s);
2175
2176void sock_kzfree_s(struct sock *sk, void *mem, int size)
2177{
2178 __sock_kfree_s(sk, mem, size, true);
2179}
2180EXPORT_SYMBOL(sock_kzfree_s);
2181
2182/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2183 I think, these locks should be removed for datagram sockets.
2184 */
2185static long sock_wait_for_wmem(struct sock *sk, long timeo)
2186{
2187 DEFINE_WAIT(wait);
2188
2189 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2190 for (;;) {
2191 if (!timeo)
2192 break;
2193 if (signal_pending(current))
2194 break;
2195 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2196 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2197 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2198 break;
2199 if (sk->sk_shutdown & SEND_SHUTDOWN)
2200 break;
2201 if (sk->sk_err)
2202 break;
2203 timeo = schedule_timeout(timeo);
2204 }
2205 finish_wait(sk_sleep(sk), &wait);
2206 return timeo;
2207}
2208
2209
2210/*
2211 * Generic send/receive buffer handlers
2212 */
2213
2214struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2215 unsigned long data_len, int noblock,
2216 int *errcode, int max_page_order)
2217{
2218 struct sk_buff *skb;
2219 long timeo;
2220 int err;
2221
2222 timeo = sock_sndtimeo(sk, noblock);
2223 for (;;) {
2224 err = sock_error(sk);
2225 if (err != 0)
2226 goto failure;
2227
2228 err = -EPIPE;
2229 if (sk->sk_shutdown & SEND_SHUTDOWN)
2230 goto failure;
2231
2232 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2233 break;
2234
2235 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2236 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2237 err = -EAGAIN;
2238 if (!timeo)
2239 goto failure;
2240 if (signal_pending(current))
2241 goto interrupted;
2242 timeo = sock_wait_for_wmem(sk, timeo);
2243 }
2244 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2245 errcode, sk->sk_allocation);
2246 if (skb)
2247 skb_set_owner_w(skb, sk);
2248 return skb;
2249
2250interrupted:
2251 err = sock_intr_errno(timeo);
2252failure:
2253 *errcode = err;
2254 return NULL;
2255}
2256EXPORT_SYMBOL(sock_alloc_send_pskb);
2257
2258struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2259 int noblock, int *errcode)
2260{
2261 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2262}
2263EXPORT_SYMBOL(sock_alloc_send_skb);
2264
2265int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2266 struct sockcm_cookie *sockc)
2267{
2268 u32 tsflags;
2269
2270 switch (cmsg->cmsg_type) {
2271 case SO_MARK:
2272 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2273 return -EPERM;
2274 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2275 return -EINVAL;
2276 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2277 break;
2278 case SO_TIMESTAMPING_OLD:
2279 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2280 return -EINVAL;
2281
2282 tsflags = *(u32 *)CMSG_DATA(cmsg);
2283 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2284 return -EINVAL;
2285
2286 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2287 sockc->tsflags |= tsflags;
2288 break;
2289 case SCM_TXTIME:
2290 if (!sock_flag(sk, SOCK_TXTIME))
2291 return -EINVAL;
2292 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
2293 return -EINVAL;
2294 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
2295 break;
2296 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2297 case SCM_RIGHTS:
2298 case SCM_CREDENTIALS:
2299 break;
2300 default:
2301 return -EINVAL;
2302 }
2303 return 0;
2304}
2305EXPORT_SYMBOL(__sock_cmsg_send);
2306
2307int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2308 struct sockcm_cookie *sockc)
2309{
2310 struct cmsghdr *cmsg;
2311 int ret;
2312
2313 for_each_cmsghdr(cmsg, msg) {
2314 if (!CMSG_OK(msg, cmsg))
2315 return -EINVAL;
2316 if (cmsg->cmsg_level != SOL_SOCKET)
2317 continue;
2318 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2319 if (ret)
2320 return ret;
2321 }
2322 return 0;
2323}
2324EXPORT_SYMBOL(sock_cmsg_send);
2325
2326static void sk_enter_memory_pressure(struct sock *sk)
2327{
2328 if (!sk->sk_prot->enter_memory_pressure)
2329 return;
2330
2331 sk->sk_prot->enter_memory_pressure(sk);
2332}
2333
2334static void sk_leave_memory_pressure(struct sock *sk)
2335{
2336 if (sk->sk_prot->leave_memory_pressure) {
2337 sk->sk_prot->leave_memory_pressure(sk);
2338 } else {
2339 unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2340
2341 if (memory_pressure && READ_ONCE(*memory_pressure))
2342 WRITE_ONCE(*memory_pressure, 0);
2343 }
2344}
2345
2346/* On 32bit arches, an skb frag is limited to 2^15 */
2347#define SKB_FRAG_PAGE_ORDER get_order(32768)
2348DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2349
2350/**
2351 * skb_page_frag_refill - check that a page_frag contains enough room
2352 * @sz: minimum size of the fragment we want to get
2353 * @pfrag: pointer to page_frag
2354 * @gfp: priority for memory allocation
2355 *
2356 * Note: While this allocator tries to use high order pages, there is
2357 * no guarantee that allocations succeed. Therefore, @sz MUST be
2358 * less or equal than PAGE_SIZE.
2359 */
2360bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2361{
2362 if (pfrag->page) {
2363 if (page_ref_count(pfrag->page) == 1) {
2364 pfrag->offset = 0;
2365 return true;
2366 }
2367 if (pfrag->offset + sz <= pfrag->size)
2368 return true;
2369 put_page(pfrag->page);
2370 }
2371
2372 pfrag->offset = 0;
2373 if (SKB_FRAG_PAGE_ORDER &&
2374 !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
2375 /* Avoid direct reclaim but allow kswapd to wake */
2376 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2377 __GFP_COMP | __GFP_NOWARN |
2378 __GFP_NORETRY,
2379 SKB_FRAG_PAGE_ORDER);
2380 if (likely(pfrag->page)) {
2381 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2382 return true;
2383 }
2384 }
2385 pfrag->page = alloc_page(gfp);
2386 if (likely(pfrag->page)) {
2387 pfrag->size = PAGE_SIZE;
2388 return true;
2389 }
2390 return false;
2391}
2392EXPORT_SYMBOL(skb_page_frag_refill);
2393
2394bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2395{
2396 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2397 return true;
2398
2399 sk_enter_memory_pressure(sk);
2400 sk_stream_moderate_sndbuf(sk);
2401 return false;
2402}
2403EXPORT_SYMBOL(sk_page_frag_refill);
2404
2405static void __lock_sock(struct sock *sk)
2406 __releases(&sk->sk_lock.slock)
2407 __acquires(&sk->sk_lock.slock)
2408{
2409 DEFINE_WAIT(wait);
2410
2411 for (;;) {
2412 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2413 TASK_UNINTERRUPTIBLE);
2414 spin_unlock_bh(&sk->sk_lock.slock);
2415 schedule();
2416 spin_lock_bh(&sk->sk_lock.slock);
2417 if (!sock_owned_by_user(sk))
2418 break;
2419 }
2420 finish_wait(&sk->sk_lock.wq, &wait);
2421}
2422
2423void __release_sock(struct sock *sk)
2424 __releases(&sk->sk_lock.slock)
2425 __acquires(&sk->sk_lock.slock)
2426{
2427 struct sk_buff *skb, *next;
2428
2429 while ((skb = sk->sk_backlog.head) != NULL) {
2430 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2431
2432 spin_unlock_bh(&sk->sk_lock.slock);
2433
2434 do {
2435 next = skb->next;
2436 prefetch(next);
2437 WARN_ON_ONCE(skb_dst_is_noref(skb));
2438 skb_mark_not_on_list(skb);
2439 sk_backlog_rcv(sk, skb);
2440
2441 cond_resched();
2442
2443 skb = next;
2444 } while (skb != NULL);
2445
2446 spin_lock_bh(&sk->sk_lock.slock);
2447 }
2448
2449 /*
2450 * Doing the zeroing here guarantee we can not loop forever
2451 * while a wild producer attempts to flood us.
2452 */
2453 sk->sk_backlog.len = 0;
2454}
2455
2456void __sk_flush_backlog(struct sock *sk)
2457{
2458 spin_lock_bh(&sk->sk_lock.slock);
2459 __release_sock(sk);
2460 spin_unlock_bh(&sk->sk_lock.slock);
2461}
2462
2463/**
2464 * sk_wait_data - wait for data to arrive at sk_receive_queue
2465 * @sk: sock to wait on
2466 * @timeo: for how long
2467 * @skb: last skb seen on sk_receive_queue
2468 *
2469 * Now socket state including sk->sk_err is changed only under lock,
2470 * hence we may omit checks after joining wait queue.
2471 * We check receive queue before schedule() only as optimization;
2472 * it is very likely that release_sock() added new data.
2473 */
2474int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2475{
2476 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2477 int rc;
2478
2479 add_wait_queue(sk_sleep(sk), &wait);
2480 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2481 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2482 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2483 remove_wait_queue(sk_sleep(sk), &wait);
2484 return rc;
2485}
2486EXPORT_SYMBOL(sk_wait_data);
2487
2488/**
2489 * __sk_mem_raise_allocated - increase memory_allocated
2490 * @sk: socket
2491 * @size: memory size to allocate
2492 * @amt: pages to allocate
2493 * @kind: allocation type
2494 *
2495 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2496 */
2497int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2498{
2499 struct proto *prot = sk->sk_prot;
2500 long allocated = sk_memory_allocated_add(sk, amt);
2501 bool charged = true;
2502
2503 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2504 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt)))
2505 goto suppress_allocation;
2506
2507 /* Under limit. */
2508 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2509 sk_leave_memory_pressure(sk);
2510 return 1;
2511 }
2512
2513 /* Under pressure. */
2514 if (allocated > sk_prot_mem_limits(sk, 1))
2515 sk_enter_memory_pressure(sk);
2516
2517 /* Over hard limit. */
2518 if (allocated > sk_prot_mem_limits(sk, 2))
2519 goto suppress_allocation;
2520
2521 /* guarantee minimum buffer size under pressure */
2522 if (kind == SK_MEM_RECV) {
2523 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2524 return 1;
2525
2526 } else { /* SK_MEM_SEND */
2527 int wmem0 = sk_get_wmem0(sk, prot);
2528
2529 if (sk->sk_type == SOCK_STREAM) {
2530 if (sk->sk_wmem_queued < wmem0)
2531 return 1;
2532 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2533 return 1;
2534 }
2535 }
2536
2537 if (sk_has_memory_pressure(sk)) {
2538 u64 alloc;
2539
2540 if (!sk_under_memory_pressure(sk))
2541 return 1;
2542 alloc = sk_sockets_allocated_read_positive(sk);
2543 if (sk_prot_mem_limits(sk, 2) > alloc *
2544 sk_mem_pages(sk->sk_wmem_queued +
2545 atomic_read(&sk->sk_rmem_alloc) +
2546 sk->sk_forward_alloc))
2547 return 1;
2548 }
2549
2550suppress_allocation:
2551
2552 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2553 sk_stream_moderate_sndbuf(sk);
2554
2555 /* Fail only if socket is _under_ its sndbuf.
2556 * In this case we cannot block, so that we have to fail.
2557 */
2558 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2559 return 1;
2560 }
2561
2562 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
2563 trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
2564
2565 sk_memory_allocated_sub(sk, amt);
2566
2567 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2568 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2569
2570 return 0;
2571}
2572EXPORT_SYMBOL(__sk_mem_raise_allocated);
2573
2574/**
2575 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2576 * @sk: socket
2577 * @size: memory size to allocate
2578 * @kind: allocation type
2579 *
2580 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2581 * rmem allocation. This function assumes that protocols which have
2582 * memory_pressure use sk_wmem_queued as write buffer accounting.
2583 */
2584int __sk_mem_schedule(struct sock *sk, int size, int kind)
2585{
2586 int ret, amt = sk_mem_pages(size);
2587
2588 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2589 ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2590 if (!ret)
2591 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2592 return ret;
2593}
2594EXPORT_SYMBOL(__sk_mem_schedule);
2595
2596/**
2597 * __sk_mem_reduce_allocated - reclaim memory_allocated
2598 * @sk: socket
2599 * @amount: number of quanta
2600 *
2601 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2602 */
2603void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2604{
2605 sk_memory_allocated_sub(sk, amount);
2606
2607 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2608 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2609
2610 if (sk_under_memory_pressure(sk) &&
2611 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2612 sk_leave_memory_pressure(sk);
2613}
2614EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2615
2616/**
2617 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2618 * @sk: socket
2619 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2620 */
2621void __sk_mem_reclaim(struct sock *sk, int amount)
2622{
2623 amount >>= SK_MEM_QUANTUM_SHIFT;
2624 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2625 __sk_mem_reduce_allocated(sk, amount);
2626}
2627EXPORT_SYMBOL(__sk_mem_reclaim);
2628
2629int sk_set_peek_off(struct sock *sk, int val)
2630{
2631 sk->sk_peek_off = val;
2632 return 0;
2633}
2634EXPORT_SYMBOL_GPL(sk_set_peek_off);
2635
2636/*
2637 * Set of default routines for initialising struct proto_ops when
2638 * the protocol does not support a particular function. In certain
2639 * cases where it makes no sense for a protocol to have a "do nothing"
2640 * function, some default processing is provided.
2641 */
2642
2643int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2644{
2645 return -EOPNOTSUPP;
2646}
2647EXPORT_SYMBOL(sock_no_bind);
2648
2649int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2650 int len, int flags)
2651{
2652 return -EOPNOTSUPP;
2653}
2654EXPORT_SYMBOL(sock_no_connect);
2655
2656int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2657{
2658 return -EOPNOTSUPP;
2659}
2660EXPORT_SYMBOL(sock_no_socketpair);
2661
2662int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2663 bool kern)
2664{
2665 return -EOPNOTSUPP;
2666}
2667EXPORT_SYMBOL(sock_no_accept);
2668
2669int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2670 int peer)
2671{
2672 return -EOPNOTSUPP;
2673}
2674EXPORT_SYMBOL(sock_no_getname);
2675
2676int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2677{
2678 return -EOPNOTSUPP;
2679}
2680EXPORT_SYMBOL(sock_no_ioctl);
2681
2682int sock_no_listen(struct socket *sock, int backlog)
2683{
2684 return -EOPNOTSUPP;
2685}
2686EXPORT_SYMBOL(sock_no_listen);
2687
2688int sock_no_shutdown(struct socket *sock, int how)
2689{
2690 return -EOPNOTSUPP;
2691}
2692EXPORT_SYMBOL(sock_no_shutdown);
2693
2694int sock_no_setsockopt(struct socket *sock, int level, int optname,
2695 char __user *optval, unsigned int optlen)
2696{
2697 return -EOPNOTSUPP;
2698}
2699EXPORT_SYMBOL(sock_no_setsockopt);
2700
2701int sock_no_getsockopt(struct socket *sock, int level, int optname,
2702 char __user *optval, int __user *optlen)
2703{
2704 return -EOPNOTSUPP;
2705}
2706EXPORT_SYMBOL(sock_no_getsockopt);
2707
2708int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2709{
2710 return -EOPNOTSUPP;
2711}
2712EXPORT_SYMBOL(sock_no_sendmsg);
2713
2714int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2715{
2716 return -EOPNOTSUPP;
2717}
2718EXPORT_SYMBOL(sock_no_sendmsg_locked);
2719
2720int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2721 int flags)
2722{
2723 return -EOPNOTSUPP;
2724}
2725EXPORT_SYMBOL(sock_no_recvmsg);
2726
2727int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2728{
2729 /* Mirror missing mmap method error code */
2730 return -ENODEV;
2731}
2732EXPORT_SYMBOL(sock_no_mmap);
2733
2734ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2735{
2736 ssize_t res;
2737 struct msghdr msg = {.msg_flags = flags};
2738 struct kvec iov;
2739 char *kaddr = kmap(page);
2740 iov.iov_base = kaddr + offset;
2741 iov.iov_len = size;
2742 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2743 kunmap(page);
2744 return res;
2745}
2746EXPORT_SYMBOL(sock_no_sendpage);
2747
2748ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2749 int offset, size_t size, int flags)
2750{
2751 ssize_t res;
2752 struct msghdr msg = {.msg_flags = flags};
2753 struct kvec iov;
2754 char *kaddr = kmap(page);
2755
2756 iov.iov_base = kaddr + offset;
2757 iov.iov_len = size;
2758 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2759 kunmap(page);
2760 return res;
2761}
2762EXPORT_SYMBOL(sock_no_sendpage_locked);
2763
2764/*
2765 * Default Socket Callbacks
2766 */
2767
2768static void sock_def_wakeup(struct sock *sk)
2769{
2770 struct socket_wq *wq;
2771
2772 rcu_read_lock();
2773 wq = rcu_dereference(sk->sk_wq);
2774 if (skwq_has_sleeper(wq))
2775 wake_up_interruptible_all(&wq->wait);
2776 rcu_read_unlock();
2777}
2778
2779static void sock_def_error_report(struct sock *sk)
2780{
2781 struct socket_wq *wq;
2782
2783 rcu_read_lock();
2784 wq = rcu_dereference(sk->sk_wq);
2785 if (skwq_has_sleeper(wq))
2786 wake_up_interruptible_poll(&wq->wait, EPOLLERR);
2787 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2788 rcu_read_unlock();
2789}
2790
2791static void sock_def_readable(struct sock *sk)
2792{
2793 struct socket_wq *wq;
2794
2795 rcu_read_lock();
2796 wq = rcu_dereference(sk->sk_wq);
2797 if (skwq_has_sleeper(wq))
2798 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
2799 EPOLLRDNORM | EPOLLRDBAND);
2800 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2801 rcu_read_unlock();
2802}
2803
2804static void sock_def_write_space(struct sock *sk)
2805{
2806 struct socket_wq *wq;
2807
2808 rcu_read_lock();
2809
2810 /* Do not wake up a writer until he can make "significant"
2811 * progress. --DaveM
2812 */
2813 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
2814 wq = rcu_dereference(sk->sk_wq);
2815 if (skwq_has_sleeper(wq))
2816 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
2817 EPOLLWRNORM | EPOLLWRBAND);
2818
2819 /* Should agree with poll, otherwise some programs break */
2820 if (sock_writeable(sk))
2821 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2822 }
2823
2824 rcu_read_unlock();
2825}
2826
2827static void sock_def_destruct(struct sock *sk)
2828{
2829}
2830
2831void sk_send_sigurg(struct sock *sk)
2832{
2833 if (sk->sk_socket && sk->sk_socket->file)
2834 if (send_sigurg(&sk->sk_socket->file->f_owner))
2835 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2836}
2837EXPORT_SYMBOL(sk_send_sigurg);
2838
2839void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2840 unsigned long expires)
2841{
2842 if (!mod_timer(timer, expires))
2843 sock_hold(sk);
2844}
2845EXPORT_SYMBOL(sk_reset_timer);
2846
2847void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2848{
2849 if (del_timer(timer))
2850 __sock_put(sk);
2851}
2852EXPORT_SYMBOL(sk_stop_timer);
2853
2854void sock_init_data(struct socket *sock, struct sock *sk)
2855{
2856 sk_init_common(sk);
2857 sk->sk_send_head = NULL;
2858
2859 timer_setup(&sk->sk_timer, NULL, 0);
2860
2861 sk->sk_allocation = GFP_KERNEL;
2862 sk->sk_rcvbuf = sysctl_rmem_default;
2863 sk->sk_sndbuf = sysctl_wmem_default;
2864 sk->sk_state = TCP_CLOSE;
2865 sk_set_socket(sk, sock);
2866
2867 sock_set_flag(sk, SOCK_ZAPPED);
2868
2869 if (sock) {
2870 sk->sk_type = sock->type;
2871 RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
2872 sock->sk = sk;
2873 sk->sk_uid = SOCK_INODE(sock)->i_uid;
2874 } else {
2875 RCU_INIT_POINTER(sk->sk_wq, NULL);
2876 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0);
2877 }
2878
2879 rwlock_init(&sk->sk_callback_lock);
2880 if (sk->sk_kern_sock)
2881 lockdep_set_class_and_name(
2882 &sk->sk_callback_lock,
2883 af_kern_callback_keys + sk->sk_family,
2884 af_family_kern_clock_key_strings[sk->sk_family]);
2885 else
2886 lockdep_set_class_and_name(
2887 &sk->sk_callback_lock,
2888 af_callback_keys + sk->sk_family,
2889 af_family_clock_key_strings[sk->sk_family]);
2890
2891 sk->sk_state_change = sock_def_wakeup;
2892 sk->sk_data_ready = sock_def_readable;
2893 sk->sk_write_space = sock_def_write_space;
2894 sk->sk_error_report = sock_def_error_report;
2895 sk->sk_destruct = sock_def_destruct;
2896
2897 sk->sk_frag.page = NULL;
2898 sk->sk_frag.offset = 0;
2899 sk->sk_peek_off = -1;
2900
2901 sk->sk_peer_pid = NULL;
2902 sk->sk_peer_cred = NULL;
2903 sk->sk_write_pending = 0;
2904 sk->sk_rcvlowat = 1;
2905 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2906 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2907
2908 sk->sk_stamp = SK_DEFAULT_STAMP;
2909#if BITS_PER_LONG==32
2910 seqlock_init(&sk->sk_stamp_seq);
2911#endif
2912 atomic_set(&sk->sk_zckey, 0);
2913
2914#ifdef CONFIG_NET_RX_BUSY_POLL
2915 sk->sk_napi_id = 0;
2916 sk->sk_ll_usec = sysctl_net_busy_read;
2917#endif
2918
2919 sk->sk_max_pacing_rate = ~0UL;
2920 sk->sk_pacing_rate = ~0UL;
2921 sk->sk_pacing_shift = 10;
2922 sk->sk_incoming_cpu = -1;
2923
2924 sk_rx_queue_clear(sk);
2925 /*
2926 * Before updating sk_refcnt, we must commit prior changes to memory
2927 * (Documentation/RCU/rculist_nulls.txt for details)
2928 */
2929 smp_wmb();
2930 refcount_set(&sk->sk_refcnt, 1);
2931 atomic_set(&sk->sk_drops, 0);
2932}
2933EXPORT_SYMBOL(sock_init_data);
2934
2935void lock_sock_nested(struct sock *sk, int subclass)
2936{
2937 might_sleep();
2938 spin_lock_bh(&sk->sk_lock.slock);
2939 if (sk->sk_lock.owned)
2940 __lock_sock(sk);
2941 sk->sk_lock.owned = 1;
2942 spin_unlock(&sk->sk_lock.slock);
2943 /*
2944 * The sk_lock has mutex_lock() semantics here:
2945 */
2946 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2947 local_bh_enable();
2948}
2949EXPORT_SYMBOL(lock_sock_nested);
2950
2951void release_sock(struct sock *sk)
2952{
2953 spin_lock_bh(&sk->sk_lock.slock);
2954 if (sk->sk_backlog.tail)
2955 __release_sock(sk);
2956
2957 /* Warning : release_cb() might need to release sk ownership,
2958 * ie call sock_release_ownership(sk) before us.
2959 */
2960 if (sk->sk_prot->release_cb)
2961 sk->sk_prot->release_cb(sk);
2962
2963 sock_release_ownership(sk);
2964 if (waitqueue_active(&sk->sk_lock.wq))
2965 wake_up(&sk->sk_lock.wq);
2966 spin_unlock_bh(&sk->sk_lock.slock);
2967}
2968EXPORT_SYMBOL(release_sock);
2969
2970/**
2971 * lock_sock_fast - fast version of lock_sock
2972 * @sk: socket
2973 *
2974 * This version should be used for very small section, where process wont block
2975 * return false if fast path is taken:
2976 *
2977 * sk_lock.slock locked, owned = 0, BH disabled
2978 *
2979 * return true if slow path is taken:
2980 *
2981 * sk_lock.slock unlocked, owned = 1, BH enabled
2982 */
2983bool lock_sock_fast(struct sock *sk)
2984{
2985 might_sleep();
2986 spin_lock_bh(&sk->sk_lock.slock);
2987
2988 if (!sk->sk_lock.owned)
2989 /*
2990 * Note : We must disable BH
2991 */
2992 return false;
2993
2994 __lock_sock(sk);
2995 sk->sk_lock.owned = 1;
2996 spin_unlock(&sk->sk_lock.slock);
2997 /*
2998 * The sk_lock has mutex_lock() semantics here:
2999 */
3000 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
3001 local_bh_enable();
3002 return true;
3003}
3004EXPORT_SYMBOL(lock_sock_fast);
3005
3006int sock_gettstamp(struct socket *sock, void __user *userstamp,
3007 bool timeval, bool time32)
3008{
3009 struct sock *sk = sock->sk;
3010 struct timespec64 ts;
3011
3012 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3013 ts = ktime_to_timespec64(sock_read_timestamp(sk));
3014 if (ts.tv_sec == -1)
3015 return -ENOENT;
3016 if (ts.tv_sec == 0) {
3017 ktime_t kt = ktime_get_real();
3018 sock_write_timestamp(sk, kt);;
3019 ts = ktime_to_timespec64(kt);
3020 }
3021
3022 if (timeval)
3023 ts.tv_nsec /= 1000;
3024
3025#ifdef CONFIG_COMPAT_32BIT_TIME
3026 if (time32)
3027 return put_old_timespec32(&ts, userstamp);
3028#endif
3029#ifdef CONFIG_SPARC64
3030 /* beware of padding in sparc64 timeval */
3031 if (timeval && !in_compat_syscall()) {
3032 struct __kernel_old_timeval __user tv = {
3033 .tv_sec = ts.tv_sec,
3034 .tv_usec = ts.tv_nsec,
3035 };
3036 if (copy_to_user(userstamp, &tv, sizeof(tv)))
3037 return -EFAULT;
3038 return 0;
3039 }
3040#endif
3041 return put_timespec64(&ts, userstamp);
3042}
3043EXPORT_SYMBOL(sock_gettstamp);
3044
3045void sock_enable_timestamp(struct sock *sk, int flag)
3046{
3047 if (!sock_flag(sk, flag)) {
3048 unsigned long previous_flags = sk->sk_flags;
3049
3050 sock_set_flag(sk, flag);
3051 /*
3052 * we just set one of the two flags which require net
3053 * time stamping, but time stamping might have been on
3054 * already because of the other one
3055 */
3056 if (sock_needs_netstamp(sk) &&
3057 !(previous_flags & SK_FLAGS_TIMESTAMP))
3058 net_enable_timestamp();
3059 }
3060}
3061
3062int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3063 int level, int type)
3064{
3065 struct sock_exterr_skb *serr;
3066 struct sk_buff *skb;
3067 int copied, err;
3068
3069 err = -EAGAIN;
3070 skb = sock_dequeue_err_skb(sk);
3071 if (skb == NULL)
3072 goto out;
3073
3074 copied = skb->len;
3075 if (copied > len) {
3076 msg->msg_flags |= MSG_TRUNC;
3077 copied = len;
3078 }
3079 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3080 if (err)
3081 goto out_free_skb;
3082
3083 sock_recv_timestamp(msg, sk, skb);
3084
3085 serr = SKB_EXT_ERR(skb);
3086 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3087
3088 msg->msg_flags |= MSG_ERRQUEUE;
3089 err = copied;
3090
3091out_free_skb:
3092 kfree_skb(skb);
3093out:
3094 return err;
3095}
3096EXPORT_SYMBOL(sock_recv_errqueue);
3097
3098/*
3099 * Get a socket option on an socket.
3100 *
3101 * FIX: POSIX 1003.1g is very ambiguous here. It states that
3102 * asynchronous errors should be reported by getsockopt. We assume
3103 * this means if you specify SO_ERROR (otherwise whats the point of it).
3104 */
3105int sock_common_getsockopt(struct socket *sock, int level, int optname,
3106 char __user *optval, int __user *optlen)
3107{
3108 struct sock *sk = sock->sk;
3109
3110 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3111}
3112EXPORT_SYMBOL(sock_common_getsockopt);
3113
3114#ifdef CONFIG_COMPAT
3115int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
3116 char __user *optval, int __user *optlen)
3117{
3118 struct sock *sk = sock->sk;
3119
3120 if (sk->sk_prot->compat_getsockopt != NULL)
3121 return sk->sk_prot->compat_getsockopt(sk, level, optname,
3122 optval, optlen);
3123 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
3124}
3125EXPORT_SYMBOL(compat_sock_common_getsockopt);
3126#endif
3127
3128int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3129 int flags)
3130{
3131 struct sock *sk = sock->sk;
3132 int addr_len = 0;
3133 int err;
3134
3135 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
3136 flags & ~MSG_DONTWAIT, &addr_len);
3137 if (err >= 0)
3138 msg->msg_namelen = addr_len;
3139 return err;
3140}
3141EXPORT_SYMBOL(sock_common_recvmsg);
3142
3143/*
3144 * Set socket options on an inet socket.
3145 */
3146int sock_common_setsockopt(struct socket *sock, int level, int optname,
3147 char __user *optval, unsigned int optlen)
3148{
3149 struct sock *sk = sock->sk;
3150
3151 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3152}
3153EXPORT_SYMBOL(sock_common_setsockopt);
3154
3155#ifdef CONFIG_COMPAT
3156int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
3157 char __user *optval, unsigned int optlen)
3158{
3159 struct sock *sk = sock->sk;
3160
3161 if (sk->sk_prot->compat_setsockopt != NULL)
3162 return sk->sk_prot->compat_setsockopt(sk, level, optname,
3163 optval, optlen);
3164 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
3165}
3166EXPORT_SYMBOL(compat_sock_common_setsockopt);
3167#endif
3168
3169void sk_common_release(struct sock *sk)
3170{
3171 if (sk->sk_prot->destroy)
3172 sk->sk_prot->destroy(sk);
3173
3174 /*
3175 * Observation: when sock_common_release is called, processes have
3176 * no access to socket. But net still has.
3177 * Step one, detach it from networking:
3178 *
3179 * A. Remove from hash tables.
3180 */
3181
3182 sk->sk_prot->unhash(sk);
3183
3184 /*
3185 * In this point socket cannot receive new packets, but it is possible
3186 * that some packets are in flight because some CPU runs receiver and
3187 * did hash table lookup before we unhashed socket. They will achieve
3188 * receive queue and will be purged by socket destructor.
3189 *
3190 * Also we still have packets pending on receive queue and probably,
3191 * our own packets waiting in device queues. sock_destroy will drain
3192 * receive queue, but transmitted packets will delay socket destruction
3193 * until the last reference will be released.
3194 */
3195
3196 sock_orphan(sk);
3197
3198 xfrm_sk_free_policy(sk);
3199
3200 sk_refcnt_debug_release(sk);
3201
3202 sock_put(sk);
3203}
3204EXPORT_SYMBOL(sk_common_release);
3205
3206void sk_get_meminfo(const struct sock *sk, u32 *mem)
3207{
3208 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3209
3210 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3211 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3212 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3213 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3214 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3215 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3216 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3217 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3218 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3219}
3220
3221#ifdef CONFIG_PROC_FS
3222#define PROTO_INUSE_NR 64 /* should be enough for the first time */
3223struct prot_inuse {
3224 int val[PROTO_INUSE_NR];
3225};
3226
3227static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3228
3229void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3230{
3231 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3232}
3233EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3234
3235int sock_prot_inuse_get(struct net *net, struct proto *prot)
3236{
3237 int cpu, idx = prot->inuse_idx;
3238 int res = 0;
3239
3240 for_each_possible_cpu(cpu)
3241 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3242
3243 return res >= 0 ? res : 0;
3244}
3245EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3246
3247static void sock_inuse_add(struct net *net, int val)
3248{
3249 this_cpu_add(*net->core.sock_inuse, val);
3250}
3251
3252int sock_inuse_get(struct net *net)
3253{
3254 int cpu, res = 0;
3255
3256 for_each_possible_cpu(cpu)
3257 res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3258
3259 return res;
3260}
3261
3262EXPORT_SYMBOL_GPL(sock_inuse_get);
3263
3264static int __net_init sock_inuse_init_net(struct net *net)
3265{
3266 net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3267 if (net->core.prot_inuse == NULL)
3268 return -ENOMEM;
3269
3270 net->core.sock_inuse = alloc_percpu(int);
3271 if (net->core.sock_inuse == NULL)
3272 goto out;
3273
3274 return 0;
3275
3276out:
3277 free_percpu(net->core.prot_inuse);
3278 return -ENOMEM;
3279}
3280
3281static void __net_exit sock_inuse_exit_net(struct net *net)
3282{
3283 free_percpu(net->core.prot_inuse);
3284 free_percpu(net->core.sock_inuse);
3285}
3286
3287static struct pernet_operations net_inuse_ops = {
3288 .init = sock_inuse_init_net,
3289 .exit = sock_inuse_exit_net,
3290};
3291
3292static __init int net_inuse_init(void)
3293{
3294 if (register_pernet_subsys(&net_inuse_ops))
3295 panic("Cannot initialize net inuse counters");
3296
3297 return 0;
3298}
3299
3300core_initcall(net_inuse_init);
3301
3302static int assign_proto_idx(struct proto *prot)
3303{
3304 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3305
3306 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3307 pr_err("PROTO_INUSE_NR exhausted\n");
3308 return -ENOSPC;
3309 }
3310
3311 set_bit(prot->inuse_idx, proto_inuse_idx);
3312 return 0;
3313}
3314
3315static void release_proto_idx(struct proto *prot)
3316{
3317 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3318 clear_bit(prot->inuse_idx, proto_inuse_idx);
3319}
3320#else
3321static inline int assign_proto_idx(struct proto *prot)
3322{
3323 return 0;
3324}
3325
3326static inline void release_proto_idx(struct proto *prot)
3327{
3328}
3329
3330static void sock_inuse_add(struct net *net, int val)
3331{
3332}
3333#endif
3334
3335static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3336{
3337 if (!rsk_prot)
3338 return;
3339 kfree(rsk_prot->slab_name);
3340 rsk_prot->slab_name = NULL;
3341 kmem_cache_destroy(rsk_prot->slab);
3342 rsk_prot->slab = NULL;
3343}
3344
3345static int req_prot_init(const struct proto *prot)
3346{
3347 struct request_sock_ops *rsk_prot = prot->rsk_prot;
3348
3349 if (!rsk_prot)
3350 return 0;
3351
3352 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3353 prot->name);
3354 if (!rsk_prot->slab_name)
3355 return -ENOMEM;
3356
3357 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3358 rsk_prot->obj_size, 0,
3359 SLAB_ACCOUNT | prot->slab_flags,
3360 NULL);
3361
3362 if (!rsk_prot->slab) {
3363 pr_crit("%s: Can't create request sock SLAB cache!\n",
3364 prot->name);
3365 return -ENOMEM;
3366 }
3367 return 0;
3368}
3369
3370int proto_register(struct proto *prot, int alloc_slab)
3371{
3372 int ret = -ENOBUFS;
3373
3374 if (alloc_slab) {
3375 prot->slab = kmem_cache_create_usercopy(prot->name,
3376 prot->obj_size, 0,
3377 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
3378 prot->slab_flags,
3379 prot->useroffset, prot->usersize,
3380 NULL);
3381
3382 if (prot->slab == NULL) {
3383 pr_crit("%s: Can't create sock SLAB cache!\n",
3384 prot->name);
3385 goto out;
3386 }
3387
3388 if (req_prot_init(prot))
3389 goto out_free_request_sock_slab;
3390
3391 if (prot->twsk_prot != NULL) {
3392 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3393
3394 if (prot->twsk_prot->twsk_slab_name == NULL)
3395 goto out_free_request_sock_slab;
3396
3397 prot->twsk_prot->twsk_slab =
3398 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3399 prot->twsk_prot->twsk_obj_size,
3400 0,
3401 SLAB_ACCOUNT |
3402 prot->slab_flags,
3403 NULL);
3404 if (prot->twsk_prot->twsk_slab == NULL)
3405 goto out_free_timewait_sock_slab_name;
3406 }
3407 }
3408
3409 mutex_lock(&proto_list_mutex);
3410 ret = assign_proto_idx(prot);
3411 if (ret) {
3412 mutex_unlock(&proto_list_mutex);
3413 goto out_free_timewait_sock_slab_name;
3414 }
3415 list_add(&prot->node, &proto_list);
3416 mutex_unlock(&proto_list_mutex);
3417 return ret;
3418
3419out_free_timewait_sock_slab_name:
3420 if (alloc_slab && prot->twsk_prot)
3421 kfree(prot->twsk_prot->twsk_slab_name);
3422out_free_request_sock_slab:
3423 if (alloc_slab) {
3424 req_prot_cleanup(prot->rsk_prot);
3425
3426 kmem_cache_destroy(prot->slab);
3427 prot->slab = NULL;
3428 }
3429out:
3430 return ret;
3431}
3432EXPORT_SYMBOL(proto_register);
3433
3434void proto_unregister(struct proto *prot)
3435{
3436 mutex_lock(&proto_list_mutex);
3437 release_proto_idx(prot);
3438 list_del(&prot->node);
3439 mutex_unlock(&proto_list_mutex);
3440
3441 kmem_cache_destroy(prot->slab);
3442 prot->slab = NULL;
3443
3444 req_prot_cleanup(prot->rsk_prot);
3445
3446 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3447 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3448 kfree(prot->twsk_prot->twsk_slab_name);
3449 prot->twsk_prot->twsk_slab = NULL;
3450 }
3451}
3452EXPORT_SYMBOL(proto_unregister);
3453
3454int sock_load_diag_module(int family, int protocol)
3455{
3456 if (!protocol) {
3457 if (!sock_is_registered(family))
3458 return -ENOENT;
3459
3460 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
3461 NETLINK_SOCK_DIAG, family);
3462 }
3463
3464#ifdef CONFIG_INET
3465 if (family == AF_INET &&
3466 protocol != IPPROTO_RAW &&
3467 !rcu_access_pointer(inet_protos[protocol]))
3468 return -ENOENT;
3469#endif
3470
3471 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
3472 NETLINK_SOCK_DIAG, family, protocol);
3473}
3474EXPORT_SYMBOL(sock_load_diag_module);
3475
3476#ifdef CONFIG_PROC_FS
3477static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3478 __acquires(proto_list_mutex)
3479{
3480 mutex_lock(&proto_list_mutex);
3481 return seq_list_start_head(&proto_list, *pos);
3482}
3483
3484static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3485{
3486 return seq_list_next(v, &proto_list, pos);
3487}
3488
3489static void proto_seq_stop(struct seq_file *seq, void *v)
3490 __releases(proto_list_mutex)
3491{
3492 mutex_unlock(&proto_list_mutex);
3493}
3494
3495static char proto_method_implemented(const void *method)
3496{
3497 return method == NULL ? 'n' : 'y';
3498}
3499static long sock_prot_memory_allocated(struct proto *proto)
3500{
3501 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3502}
3503
3504static const char *sock_prot_memory_pressure(struct proto *proto)
3505{
3506 return proto->memory_pressure != NULL ?
3507 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3508}
3509
3510static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3511{
3512
3513 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
3514 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3515 proto->name,
3516 proto->obj_size,
3517 sock_prot_inuse_get(seq_file_net(seq), proto),
3518 sock_prot_memory_allocated(proto),
3519 sock_prot_memory_pressure(proto),
3520 proto->max_header,
3521 proto->slab == NULL ? "no" : "yes",
3522 module_name(proto->owner),
3523 proto_method_implemented(proto->close),
3524 proto_method_implemented(proto->connect),
3525 proto_method_implemented(proto->disconnect),
3526 proto_method_implemented(proto->accept),
3527 proto_method_implemented(proto->ioctl),
3528 proto_method_implemented(proto->init),
3529 proto_method_implemented(proto->destroy),
3530 proto_method_implemented(proto->shutdown),
3531 proto_method_implemented(proto->setsockopt),
3532 proto_method_implemented(proto->getsockopt),
3533 proto_method_implemented(proto->sendmsg),
3534 proto_method_implemented(proto->recvmsg),
3535 proto_method_implemented(proto->sendpage),
3536 proto_method_implemented(proto->bind),
3537 proto_method_implemented(proto->backlog_rcv),
3538 proto_method_implemented(proto->hash),
3539 proto_method_implemented(proto->unhash),
3540 proto_method_implemented(proto->get_port),
3541 proto_method_implemented(proto->enter_memory_pressure));
3542}
3543
3544static int proto_seq_show(struct seq_file *seq, void *v)
3545{
3546 if (v == &proto_list)
3547 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3548 "protocol",
3549 "size",
3550 "sockets",
3551 "memory",
3552 "press",
3553 "maxhdr",
3554 "slab",
3555 "module",
3556 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3557 else
3558 proto_seq_printf(seq, list_entry(v, struct proto, node));
3559 return 0;
3560}
3561
3562static const struct seq_operations proto_seq_ops = {
3563 .start = proto_seq_start,
3564 .next = proto_seq_next,
3565 .stop = proto_seq_stop,
3566 .show = proto_seq_show,
3567};
3568
3569static __net_init int proto_init_net(struct net *net)
3570{
3571 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
3572 sizeof(struct seq_net_private)))
3573 return -ENOMEM;
3574
3575 return 0;
3576}
3577
3578static __net_exit void proto_exit_net(struct net *net)
3579{
3580 remove_proc_entry("protocols", net->proc_net);
3581}
3582
3583
3584static __net_initdata struct pernet_operations proto_net_ops = {
3585 .init = proto_init_net,
3586 .exit = proto_exit_net,
3587};
3588
3589static int __init proto_init(void)
3590{
3591 return register_pernet_subsys(&proto_net_ops);
3592}
3593
3594subsys_initcall(proto_init);
3595
3596#endif /* PROC_FS */
3597
3598#ifdef CONFIG_NET_RX_BUSY_POLL
3599bool sk_busy_loop_end(void *p, unsigned long start_time)
3600{
3601 struct sock *sk = p;
3602
3603 return !skb_queue_empty_lockless(&sk->sk_receive_queue) ||
3604 sk_busy_loop_timeout(sk, start_time);
3605}
3606EXPORT_SYMBOL(sk_busy_loop_end);
3607#endif /* CONFIG_NET_RX_BUSY_POLL */