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/*
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/errqueue.h>
97#include <linux/types.h>
98#include <linux/socket.h>
99#include <linux/in.h>
100#include <linux/kernel.h>
101#include <linux/module.h>
102#include <linux/proc_fs.h>
103#include <linux/seq_file.h>
104#include <linux/sched.h>
105#include <linux/timer.h>
106#include <linux/string.h>
107#include <linux/sockios.h>
108#include <linux/net.h>
109#include <linux/mm.h>
110#include <linux/slab.h>
111#include <linux/interrupt.h>
112#include <linux/poll.h>
113#include <linux/tcp.h>
114#include <linux/init.h>
115#include <linux/highmem.h>
116#include <linux/user_namespace.h>
117#include <linux/static_key.h>
118#include <linux/memcontrol.h>
119#include <linux/prefetch.h>
120
121#include <asm/uaccess.h>
122
123#include <linux/netdevice.h>
124#include <net/protocol.h>
125#include <linux/skbuff.h>
126#include <net/net_namespace.h>
127#include <net/request_sock.h>
128#include <net/sock.h>
129#include <linux/net_tstamp.h>
130#include <net/xfrm.h>
131#include <linux/ipsec.h>
132#include <net/cls_cgroup.h>
133#include <net/netprio_cgroup.h>
134#include <linux/sock_diag.h>
135
136#include <linux/filter.h>
137#include <net/sock_reuseport.h>
138
139#include <trace/events/sock.h>
140
141#ifdef CONFIG_INET
142#include <net/tcp.h>
143#endif
144
145#include <net/busy_poll.h>
146
147static DEFINE_MUTEX(proto_list_mutex);
148static LIST_HEAD(proto_list);
149
150/**
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
155 *
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
159 */
160bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
162{
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
165}
166EXPORT_SYMBOL(sk_ns_capable);
167
168/**
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
172 *
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
175 * namespaces.
176 */
177bool sk_capable(const struct sock *sk, int cap)
178{
179 return sk_ns_capable(sk, &init_user_ns, cap);
180}
181EXPORT_SYMBOL(sk_capable);
182
183/**
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
187 *
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
191 */
192bool sk_net_capable(const struct sock *sk, int cap)
193{
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195}
196EXPORT_SYMBOL(sk_net_capable);
197
198/*
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
201 */
202static struct lock_class_key af_family_keys[AF_MAX];
203static struct lock_class_key af_family_slock_keys[AF_MAX];
204
205/*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
210static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
225 "sk_lock-AF_MAX"
226};
227static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_MAX"
243};
244static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_MAX"
260};
261
262/*
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
265 */
266static struct lock_class_key af_callback_keys[AF_MAX];
267
268/* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
272 */
273#define _SK_MEM_PACKETS 256
274#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
277
278/* Run time adjustable parameters. */
279__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280EXPORT_SYMBOL(sysctl_wmem_max);
281__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282EXPORT_SYMBOL(sysctl_rmem_max);
283__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
285
286/* Maximal space eaten by iovec or ancillary data plus some space */
287int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288EXPORT_SYMBOL(sysctl_optmem_max);
289
290int sysctl_tstamp_allow_data __read_mostly = 1;
291
292struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293EXPORT_SYMBOL_GPL(memalloc_socks);
294
295/**
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
298 *
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
302 */
303void sk_set_memalloc(struct sock *sk)
304{
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_key_slow_inc(&memalloc_socks);
308}
309EXPORT_SYMBOL_GPL(sk_set_memalloc);
310
311void sk_clear_memalloc(struct sock *sk)
312{
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_key_slow_dec(&memalloc_socks);
316
317 /*
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 */
324 sk_mem_reclaim(sk);
325}
326EXPORT_SYMBOL_GPL(sk_clear_memalloc);
327
328int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
329{
330 int ret;
331 unsigned long pflags = current->flags;
332
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
335
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339
340 return ret;
341}
342EXPORT_SYMBOL(__sk_backlog_rcv);
343
344static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345{
346 struct timeval tv;
347
348 if (optlen < sizeof(tv))
349 return -EINVAL;
350 if (copy_from_user(&tv, optval, sizeof(tv)))
351 return -EFAULT;
352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 return -EDOM;
354
355 if (tv.tv_sec < 0) {
356 static int warned __read_mostly;
357
358 *timeo_p = 0;
359 if (warned < 10 && net_ratelimit()) {
360 warned++;
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
363 }
364 return 0;
365 }
366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
368 return 0;
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 return 0;
372}
373
374static void sock_warn_obsolete_bsdism(const char *name)
375{
376 static int warned;
377 static char warncomm[TASK_COMM_LEN];
378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
381 warncomm, name);
382 warned++;
383 }
384}
385
386static bool sock_needs_netstamp(const struct sock *sk)
387{
388 switch (sk->sk_family) {
389 case AF_UNSPEC:
390 case AF_UNIX:
391 return false;
392 default:
393 return true;
394 }
395}
396
397static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
398{
399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
403 net_disable_timestamp();
404 }
405}
406
407
408int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
409{
410 int err;
411 unsigned long flags;
412 struct sk_buff_head *list = &sk->sk_receive_queue;
413
414 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
415 atomic_inc(&sk->sk_drops);
416 trace_sock_rcvqueue_full(sk, skb);
417 return -ENOMEM;
418 }
419
420 err = sk_filter(sk, skb);
421 if (err)
422 return err;
423
424 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
425 atomic_inc(&sk->sk_drops);
426 return -ENOBUFS;
427 }
428
429 skb->dev = NULL;
430 skb_set_owner_r(skb, sk);
431
432 /* we escape from rcu protected region, make sure we dont leak
433 * a norefcounted dst
434 */
435 skb_dst_force(skb);
436
437 spin_lock_irqsave(&list->lock, flags);
438 sock_skb_set_dropcount(sk, skb);
439 __skb_queue_tail(list, skb);
440 spin_unlock_irqrestore(&list->lock, flags);
441
442 if (!sock_flag(sk, SOCK_DEAD))
443 sk->sk_data_ready(sk);
444 return 0;
445}
446EXPORT_SYMBOL(sock_queue_rcv_skb);
447
448int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
449{
450 int rc = NET_RX_SUCCESS;
451
452 if (sk_filter(sk, skb))
453 goto discard_and_relse;
454
455 skb->dev = NULL;
456
457 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
458 atomic_inc(&sk->sk_drops);
459 goto discard_and_relse;
460 }
461 if (nested)
462 bh_lock_sock_nested(sk);
463 else
464 bh_lock_sock(sk);
465 if (!sock_owned_by_user(sk)) {
466 /*
467 * trylock + unlock semantics:
468 */
469 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
470
471 rc = sk_backlog_rcv(sk, skb);
472
473 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
474 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
475 bh_unlock_sock(sk);
476 atomic_inc(&sk->sk_drops);
477 goto discard_and_relse;
478 }
479
480 bh_unlock_sock(sk);
481out:
482 sock_put(sk);
483 return rc;
484discard_and_relse:
485 kfree_skb(skb);
486 goto out;
487}
488EXPORT_SYMBOL(sk_receive_skb);
489
490struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
491{
492 struct dst_entry *dst = __sk_dst_get(sk);
493
494 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
495 sk_tx_queue_clear(sk);
496 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
497 dst_release(dst);
498 return NULL;
499 }
500
501 return dst;
502}
503EXPORT_SYMBOL(__sk_dst_check);
504
505struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
506{
507 struct dst_entry *dst = sk_dst_get(sk);
508
509 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
510 sk_dst_reset(sk);
511 dst_release(dst);
512 return NULL;
513 }
514
515 return dst;
516}
517EXPORT_SYMBOL(sk_dst_check);
518
519static int sock_setbindtodevice(struct sock *sk, char __user *optval,
520 int optlen)
521{
522 int ret = -ENOPROTOOPT;
523#ifdef CONFIG_NETDEVICES
524 struct net *net = sock_net(sk);
525 char devname[IFNAMSIZ];
526 int index;
527
528 /* Sorry... */
529 ret = -EPERM;
530 if (!ns_capable(net->user_ns, CAP_NET_RAW))
531 goto out;
532
533 ret = -EINVAL;
534 if (optlen < 0)
535 goto out;
536
537 /* Bind this socket to a particular device like "eth0",
538 * as specified in the passed interface name. If the
539 * name is "" or the option length is zero the socket
540 * is not bound.
541 */
542 if (optlen > IFNAMSIZ - 1)
543 optlen = IFNAMSIZ - 1;
544 memset(devname, 0, sizeof(devname));
545
546 ret = -EFAULT;
547 if (copy_from_user(devname, optval, optlen))
548 goto out;
549
550 index = 0;
551 if (devname[0] != '\0') {
552 struct net_device *dev;
553
554 rcu_read_lock();
555 dev = dev_get_by_name_rcu(net, devname);
556 if (dev)
557 index = dev->ifindex;
558 rcu_read_unlock();
559 ret = -ENODEV;
560 if (!dev)
561 goto out;
562 }
563
564 lock_sock(sk);
565 sk->sk_bound_dev_if = index;
566 sk_dst_reset(sk);
567 release_sock(sk);
568
569 ret = 0;
570
571out:
572#endif
573
574 return ret;
575}
576
577static int sock_getbindtodevice(struct sock *sk, char __user *optval,
578 int __user *optlen, int len)
579{
580 int ret = -ENOPROTOOPT;
581#ifdef CONFIG_NETDEVICES
582 struct net *net = sock_net(sk);
583 char devname[IFNAMSIZ];
584
585 if (sk->sk_bound_dev_if == 0) {
586 len = 0;
587 goto zero;
588 }
589
590 ret = -EINVAL;
591 if (len < IFNAMSIZ)
592 goto out;
593
594 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
595 if (ret)
596 goto out;
597
598 len = strlen(devname) + 1;
599
600 ret = -EFAULT;
601 if (copy_to_user(optval, devname, len))
602 goto out;
603
604zero:
605 ret = -EFAULT;
606 if (put_user(len, optlen))
607 goto out;
608
609 ret = 0;
610
611out:
612#endif
613
614 return ret;
615}
616
617static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
618{
619 if (valbool)
620 sock_set_flag(sk, bit);
621 else
622 sock_reset_flag(sk, bit);
623}
624
625bool sk_mc_loop(struct sock *sk)
626{
627 if (dev_recursion_level())
628 return false;
629 if (!sk)
630 return true;
631 switch (sk->sk_family) {
632 case AF_INET:
633 return inet_sk(sk)->mc_loop;
634#if IS_ENABLED(CONFIG_IPV6)
635 case AF_INET6:
636 return inet6_sk(sk)->mc_loop;
637#endif
638 }
639 WARN_ON(1);
640 return true;
641}
642EXPORT_SYMBOL(sk_mc_loop);
643
644/*
645 * This is meant for all protocols to use and covers goings on
646 * at the socket level. Everything here is generic.
647 */
648
649int sock_setsockopt(struct socket *sock, int level, int optname,
650 char __user *optval, unsigned int optlen)
651{
652 struct sock *sk = sock->sk;
653 int val;
654 int valbool;
655 struct linger ling;
656 int ret = 0;
657
658 /*
659 * Options without arguments
660 */
661
662 if (optname == SO_BINDTODEVICE)
663 return sock_setbindtodevice(sk, optval, optlen);
664
665 if (optlen < sizeof(int))
666 return -EINVAL;
667
668 if (get_user(val, (int __user *)optval))
669 return -EFAULT;
670
671 valbool = val ? 1 : 0;
672
673 lock_sock(sk);
674
675 switch (optname) {
676 case SO_DEBUG:
677 if (val && !capable(CAP_NET_ADMIN))
678 ret = -EACCES;
679 else
680 sock_valbool_flag(sk, SOCK_DBG, valbool);
681 break;
682 case SO_REUSEADDR:
683 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
684 break;
685 case SO_REUSEPORT:
686 sk->sk_reuseport = valbool;
687 break;
688 case SO_TYPE:
689 case SO_PROTOCOL:
690 case SO_DOMAIN:
691 case SO_ERROR:
692 ret = -ENOPROTOOPT;
693 break;
694 case SO_DONTROUTE:
695 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
696 break;
697 case SO_BROADCAST:
698 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
699 break;
700 case SO_SNDBUF:
701 /* Don't error on this BSD doesn't and if you think
702 * about it this is right. Otherwise apps have to
703 * play 'guess the biggest size' games. RCVBUF/SNDBUF
704 * are treated in BSD as hints
705 */
706 val = min_t(u32, val, sysctl_wmem_max);
707set_sndbuf:
708 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
709 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
710 /* Wake up sending tasks if we upped the value. */
711 sk->sk_write_space(sk);
712 break;
713
714 case SO_SNDBUFFORCE:
715 if (!capable(CAP_NET_ADMIN)) {
716 ret = -EPERM;
717 break;
718 }
719 goto set_sndbuf;
720
721 case SO_RCVBUF:
722 /* Don't error on this BSD doesn't and if you think
723 * about it this is right. Otherwise apps have to
724 * play 'guess the biggest size' games. RCVBUF/SNDBUF
725 * are treated in BSD as hints
726 */
727 val = min_t(u32, val, sysctl_rmem_max);
728set_rcvbuf:
729 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
730 /*
731 * We double it on the way in to account for
732 * "struct sk_buff" etc. overhead. Applications
733 * assume that the SO_RCVBUF setting they make will
734 * allow that much actual data to be received on that
735 * socket.
736 *
737 * Applications are unaware that "struct sk_buff" and
738 * other overheads allocate from the receive buffer
739 * during socket buffer allocation.
740 *
741 * And after considering the possible alternatives,
742 * returning the value we actually used in getsockopt
743 * is the most desirable behavior.
744 */
745 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
746 break;
747
748 case SO_RCVBUFFORCE:
749 if (!capable(CAP_NET_ADMIN)) {
750 ret = -EPERM;
751 break;
752 }
753 goto set_rcvbuf;
754
755 case SO_KEEPALIVE:
756#ifdef CONFIG_INET
757 if (sk->sk_protocol == IPPROTO_TCP &&
758 sk->sk_type == SOCK_STREAM)
759 tcp_set_keepalive(sk, valbool);
760#endif
761 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
762 break;
763
764 case SO_OOBINLINE:
765 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
766 break;
767
768 case SO_NO_CHECK:
769 sk->sk_no_check_tx = valbool;
770 break;
771
772 case SO_PRIORITY:
773 if ((val >= 0 && val <= 6) ||
774 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
775 sk->sk_priority = val;
776 else
777 ret = -EPERM;
778 break;
779
780 case SO_LINGER:
781 if (optlen < sizeof(ling)) {
782 ret = -EINVAL; /* 1003.1g */
783 break;
784 }
785 if (copy_from_user(&ling, optval, sizeof(ling))) {
786 ret = -EFAULT;
787 break;
788 }
789 if (!ling.l_onoff)
790 sock_reset_flag(sk, SOCK_LINGER);
791 else {
792#if (BITS_PER_LONG == 32)
793 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
794 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
795 else
796#endif
797 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
798 sock_set_flag(sk, SOCK_LINGER);
799 }
800 break;
801
802 case SO_BSDCOMPAT:
803 sock_warn_obsolete_bsdism("setsockopt");
804 break;
805
806 case SO_PASSCRED:
807 if (valbool)
808 set_bit(SOCK_PASSCRED, &sock->flags);
809 else
810 clear_bit(SOCK_PASSCRED, &sock->flags);
811 break;
812
813 case SO_TIMESTAMP:
814 case SO_TIMESTAMPNS:
815 if (valbool) {
816 if (optname == SO_TIMESTAMP)
817 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
818 else
819 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
820 sock_set_flag(sk, SOCK_RCVTSTAMP);
821 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
822 } else {
823 sock_reset_flag(sk, SOCK_RCVTSTAMP);
824 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
825 }
826 break;
827
828 case SO_TIMESTAMPING:
829 if (val & ~SOF_TIMESTAMPING_MASK) {
830 ret = -EINVAL;
831 break;
832 }
833
834 if (val & SOF_TIMESTAMPING_OPT_ID &&
835 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
836 if (sk->sk_protocol == IPPROTO_TCP &&
837 sk->sk_type == SOCK_STREAM) {
838 if (sk->sk_state != TCP_ESTABLISHED) {
839 ret = -EINVAL;
840 break;
841 }
842 sk->sk_tskey = tcp_sk(sk)->snd_una;
843 } else {
844 sk->sk_tskey = 0;
845 }
846 }
847 sk->sk_tsflags = val;
848 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
849 sock_enable_timestamp(sk,
850 SOCK_TIMESTAMPING_RX_SOFTWARE);
851 else
852 sock_disable_timestamp(sk,
853 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
854 break;
855
856 case SO_RCVLOWAT:
857 if (val < 0)
858 val = INT_MAX;
859 sk->sk_rcvlowat = val ? : 1;
860 break;
861
862 case SO_RCVTIMEO:
863 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
864 break;
865
866 case SO_SNDTIMEO:
867 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
868 break;
869
870 case SO_ATTACH_FILTER:
871 ret = -EINVAL;
872 if (optlen == sizeof(struct sock_fprog)) {
873 struct sock_fprog fprog;
874
875 ret = -EFAULT;
876 if (copy_from_user(&fprog, optval, sizeof(fprog)))
877 break;
878
879 ret = sk_attach_filter(&fprog, sk);
880 }
881 break;
882
883 case SO_ATTACH_BPF:
884 ret = -EINVAL;
885 if (optlen == sizeof(u32)) {
886 u32 ufd;
887
888 ret = -EFAULT;
889 if (copy_from_user(&ufd, optval, sizeof(ufd)))
890 break;
891
892 ret = sk_attach_bpf(ufd, sk);
893 }
894 break;
895
896 case SO_ATTACH_REUSEPORT_CBPF:
897 ret = -EINVAL;
898 if (optlen == sizeof(struct sock_fprog)) {
899 struct sock_fprog fprog;
900
901 ret = -EFAULT;
902 if (copy_from_user(&fprog, optval, sizeof(fprog)))
903 break;
904
905 ret = sk_reuseport_attach_filter(&fprog, sk);
906 }
907 break;
908
909 case SO_ATTACH_REUSEPORT_EBPF:
910 ret = -EINVAL;
911 if (optlen == sizeof(u32)) {
912 u32 ufd;
913
914 ret = -EFAULT;
915 if (copy_from_user(&ufd, optval, sizeof(ufd)))
916 break;
917
918 ret = sk_reuseport_attach_bpf(ufd, sk);
919 }
920 break;
921
922 case SO_DETACH_FILTER:
923 ret = sk_detach_filter(sk);
924 break;
925
926 case SO_LOCK_FILTER:
927 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
928 ret = -EPERM;
929 else
930 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
931 break;
932
933 case SO_PASSSEC:
934 if (valbool)
935 set_bit(SOCK_PASSSEC, &sock->flags);
936 else
937 clear_bit(SOCK_PASSSEC, &sock->flags);
938 break;
939 case SO_MARK:
940 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
941 ret = -EPERM;
942 else
943 sk->sk_mark = val;
944 break;
945
946 case SO_RXQ_OVFL:
947 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
948 break;
949
950 case SO_WIFI_STATUS:
951 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
952 break;
953
954 case SO_PEEK_OFF:
955 if (sock->ops->set_peek_off)
956 ret = sock->ops->set_peek_off(sk, val);
957 else
958 ret = -EOPNOTSUPP;
959 break;
960
961 case SO_NOFCS:
962 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
963 break;
964
965 case SO_SELECT_ERR_QUEUE:
966 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
967 break;
968
969#ifdef CONFIG_NET_RX_BUSY_POLL
970 case SO_BUSY_POLL:
971 /* allow unprivileged users to decrease the value */
972 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
973 ret = -EPERM;
974 else {
975 if (val < 0)
976 ret = -EINVAL;
977 else
978 sk->sk_ll_usec = val;
979 }
980 break;
981#endif
982
983 case SO_MAX_PACING_RATE:
984 sk->sk_max_pacing_rate = val;
985 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
986 sk->sk_max_pacing_rate);
987 break;
988
989 case SO_INCOMING_CPU:
990 sk->sk_incoming_cpu = val;
991 break;
992
993 case SO_CNX_ADVICE:
994 if (val == 1)
995 dst_negative_advice(sk);
996 break;
997 default:
998 ret = -ENOPROTOOPT;
999 break;
1000 }
1001 release_sock(sk);
1002 return ret;
1003}
1004EXPORT_SYMBOL(sock_setsockopt);
1005
1006
1007static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1008 struct ucred *ucred)
1009{
1010 ucred->pid = pid_vnr(pid);
1011 ucred->uid = ucred->gid = -1;
1012 if (cred) {
1013 struct user_namespace *current_ns = current_user_ns();
1014
1015 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1016 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1017 }
1018}
1019
1020int sock_getsockopt(struct socket *sock, int level, int optname,
1021 char __user *optval, int __user *optlen)
1022{
1023 struct sock *sk = sock->sk;
1024
1025 union {
1026 int val;
1027 struct linger ling;
1028 struct timeval tm;
1029 } v;
1030
1031 int lv = sizeof(int);
1032 int len;
1033
1034 if (get_user(len, optlen))
1035 return -EFAULT;
1036 if (len < 0)
1037 return -EINVAL;
1038
1039 memset(&v, 0, sizeof(v));
1040
1041 switch (optname) {
1042 case SO_DEBUG:
1043 v.val = sock_flag(sk, SOCK_DBG);
1044 break;
1045
1046 case SO_DONTROUTE:
1047 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1048 break;
1049
1050 case SO_BROADCAST:
1051 v.val = sock_flag(sk, SOCK_BROADCAST);
1052 break;
1053
1054 case SO_SNDBUF:
1055 v.val = sk->sk_sndbuf;
1056 break;
1057
1058 case SO_RCVBUF:
1059 v.val = sk->sk_rcvbuf;
1060 break;
1061
1062 case SO_REUSEADDR:
1063 v.val = sk->sk_reuse;
1064 break;
1065
1066 case SO_REUSEPORT:
1067 v.val = sk->sk_reuseport;
1068 break;
1069
1070 case SO_KEEPALIVE:
1071 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1072 break;
1073
1074 case SO_TYPE:
1075 v.val = sk->sk_type;
1076 break;
1077
1078 case SO_PROTOCOL:
1079 v.val = sk->sk_protocol;
1080 break;
1081
1082 case SO_DOMAIN:
1083 v.val = sk->sk_family;
1084 break;
1085
1086 case SO_ERROR:
1087 v.val = -sock_error(sk);
1088 if (v.val == 0)
1089 v.val = xchg(&sk->sk_err_soft, 0);
1090 break;
1091
1092 case SO_OOBINLINE:
1093 v.val = sock_flag(sk, SOCK_URGINLINE);
1094 break;
1095
1096 case SO_NO_CHECK:
1097 v.val = sk->sk_no_check_tx;
1098 break;
1099
1100 case SO_PRIORITY:
1101 v.val = sk->sk_priority;
1102 break;
1103
1104 case SO_LINGER:
1105 lv = sizeof(v.ling);
1106 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1107 v.ling.l_linger = sk->sk_lingertime / HZ;
1108 break;
1109
1110 case SO_BSDCOMPAT:
1111 sock_warn_obsolete_bsdism("getsockopt");
1112 break;
1113
1114 case SO_TIMESTAMP:
1115 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1116 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1117 break;
1118
1119 case SO_TIMESTAMPNS:
1120 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1121 break;
1122
1123 case SO_TIMESTAMPING:
1124 v.val = sk->sk_tsflags;
1125 break;
1126
1127 case SO_RCVTIMEO:
1128 lv = sizeof(struct timeval);
1129 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1130 v.tm.tv_sec = 0;
1131 v.tm.tv_usec = 0;
1132 } else {
1133 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1134 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1135 }
1136 break;
1137
1138 case SO_SNDTIMEO:
1139 lv = sizeof(struct timeval);
1140 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1141 v.tm.tv_sec = 0;
1142 v.tm.tv_usec = 0;
1143 } else {
1144 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1145 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1146 }
1147 break;
1148
1149 case SO_RCVLOWAT:
1150 v.val = sk->sk_rcvlowat;
1151 break;
1152
1153 case SO_SNDLOWAT:
1154 v.val = 1;
1155 break;
1156
1157 case SO_PASSCRED:
1158 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1159 break;
1160
1161 case SO_PEERCRED:
1162 {
1163 struct ucred peercred;
1164 if (len > sizeof(peercred))
1165 len = sizeof(peercred);
1166 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1167 if (copy_to_user(optval, &peercred, len))
1168 return -EFAULT;
1169 goto lenout;
1170 }
1171
1172 case SO_PEERNAME:
1173 {
1174 char address[128];
1175
1176 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1177 return -ENOTCONN;
1178 if (lv < len)
1179 return -EINVAL;
1180 if (copy_to_user(optval, address, len))
1181 return -EFAULT;
1182 goto lenout;
1183 }
1184
1185 /* Dubious BSD thing... Probably nobody even uses it, but
1186 * the UNIX standard wants it for whatever reason... -DaveM
1187 */
1188 case SO_ACCEPTCONN:
1189 v.val = sk->sk_state == TCP_LISTEN;
1190 break;
1191
1192 case SO_PASSSEC:
1193 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1194 break;
1195
1196 case SO_PEERSEC:
1197 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1198
1199 case SO_MARK:
1200 v.val = sk->sk_mark;
1201 break;
1202
1203 case SO_RXQ_OVFL:
1204 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1205 break;
1206
1207 case SO_WIFI_STATUS:
1208 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1209 break;
1210
1211 case SO_PEEK_OFF:
1212 if (!sock->ops->set_peek_off)
1213 return -EOPNOTSUPP;
1214
1215 v.val = sk->sk_peek_off;
1216 break;
1217 case SO_NOFCS:
1218 v.val = sock_flag(sk, SOCK_NOFCS);
1219 break;
1220
1221 case SO_BINDTODEVICE:
1222 return sock_getbindtodevice(sk, optval, optlen, len);
1223
1224 case SO_GET_FILTER:
1225 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1226 if (len < 0)
1227 return len;
1228
1229 goto lenout;
1230
1231 case SO_LOCK_FILTER:
1232 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1233 break;
1234
1235 case SO_BPF_EXTENSIONS:
1236 v.val = bpf_tell_extensions();
1237 break;
1238
1239 case SO_SELECT_ERR_QUEUE:
1240 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1241 break;
1242
1243#ifdef CONFIG_NET_RX_BUSY_POLL
1244 case SO_BUSY_POLL:
1245 v.val = sk->sk_ll_usec;
1246 break;
1247#endif
1248
1249 case SO_MAX_PACING_RATE:
1250 v.val = sk->sk_max_pacing_rate;
1251 break;
1252
1253 case SO_INCOMING_CPU:
1254 v.val = sk->sk_incoming_cpu;
1255 break;
1256
1257 default:
1258 /* We implement the SO_SNDLOWAT etc to not be settable
1259 * (1003.1g 7).
1260 */
1261 return -ENOPROTOOPT;
1262 }
1263
1264 if (len > lv)
1265 len = lv;
1266 if (copy_to_user(optval, &v, len))
1267 return -EFAULT;
1268lenout:
1269 if (put_user(len, optlen))
1270 return -EFAULT;
1271 return 0;
1272}
1273
1274/*
1275 * Initialize an sk_lock.
1276 *
1277 * (We also register the sk_lock with the lock validator.)
1278 */
1279static inline void sock_lock_init(struct sock *sk)
1280{
1281 sock_lock_init_class_and_name(sk,
1282 af_family_slock_key_strings[sk->sk_family],
1283 af_family_slock_keys + sk->sk_family,
1284 af_family_key_strings[sk->sk_family],
1285 af_family_keys + sk->sk_family);
1286}
1287
1288/*
1289 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1290 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1291 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1292 */
1293static void sock_copy(struct sock *nsk, const struct sock *osk)
1294{
1295#ifdef CONFIG_SECURITY_NETWORK
1296 void *sptr = nsk->sk_security;
1297#endif
1298 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1299
1300 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1301 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1302
1303#ifdef CONFIG_SECURITY_NETWORK
1304 nsk->sk_security = sptr;
1305 security_sk_clone(osk, nsk);
1306#endif
1307}
1308
1309void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1310{
1311 unsigned long nulls1, nulls2;
1312
1313 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1314 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1315 if (nulls1 > nulls2)
1316 swap(nulls1, nulls2);
1317
1318 if (nulls1 != 0)
1319 memset((char *)sk, 0, nulls1);
1320 memset((char *)sk + nulls1 + sizeof(void *), 0,
1321 nulls2 - nulls1 - sizeof(void *));
1322 memset((char *)sk + nulls2 + sizeof(void *), 0,
1323 size - nulls2 - sizeof(void *));
1324}
1325EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1326
1327static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1328 int family)
1329{
1330 struct sock *sk;
1331 struct kmem_cache *slab;
1332
1333 slab = prot->slab;
1334 if (slab != NULL) {
1335 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1336 if (!sk)
1337 return sk;
1338 if (priority & __GFP_ZERO) {
1339 if (prot->clear_sk)
1340 prot->clear_sk(sk, prot->obj_size);
1341 else
1342 sk_prot_clear_nulls(sk, prot->obj_size);
1343 }
1344 } else
1345 sk = kmalloc(prot->obj_size, priority);
1346
1347 if (sk != NULL) {
1348 kmemcheck_annotate_bitfield(sk, flags);
1349
1350 if (security_sk_alloc(sk, family, priority))
1351 goto out_free;
1352
1353 if (!try_module_get(prot->owner))
1354 goto out_free_sec;
1355 sk_tx_queue_clear(sk);
1356 cgroup_sk_alloc(&sk->sk_cgrp_data);
1357 }
1358
1359 return sk;
1360
1361out_free_sec:
1362 security_sk_free(sk);
1363out_free:
1364 if (slab != NULL)
1365 kmem_cache_free(slab, sk);
1366 else
1367 kfree(sk);
1368 return NULL;
1369}
1370
1371static void sk_prot_free(struct proto *prot, struct sock *sk)
1372{
1373 struct kmem_cache *slab;
1374 struct module *owner;
1375
1376 owner = prot->owner;
1377 slab = prot->slab;
1378
1379 cgroup_sk_free(&sk->sk_cgrp_data);
1380 security_sk_free(sk);
1381 if (slab != NULL)
1382 kmem_cache_free(slab, sk);
1383 else
1384 kfree(sk);
1385 module_put(owner);
1386}
1387
1388/**
1389 * sk_alloc - All socket objects are allocated here
1390 * @net: the applicable net namespace
1391 * @family: protocol family
1392 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1393 * @prot: struct proto associated with this new sock instance
1394 * @kern: is this to be a kernel socket?
1395 */
1396struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1397 struct proto *prot, int kern)
1398{
1399 struct sock *sk;
1400
1401 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1402 if (sk) {
1403 sk->sk_family = family;
1404 /*
1405 * See comment in struct sock definition to understand
1406 * why we need sk_prot_creator -acme
1407 */
1408 sk->sk_prot = sk->sk_prot_creator = prot;
1409 sock_lock_init(sk);
1410 sk->sk_net_refcnt = kern ? 0 : 1;
1411 if (likely(sk->sk_net_refcnt))
1412 get_net(net);
1413 sock_net_set(sk, net);
1414 atomic_set(&sk->sk_wmem_alloc, 1);
1415
1416 sock_update_classid(&sk->sk_cgrp_data);
1417 sock_update_netprioidx(&sk->sk_cgrp_data);
1418 }
1419
1420 return sk;
1421}
1422EXPORT_SYMBOL(sk_alloc);
1423
1424void sk_destruct(struct sock *sk)
1425{
1426 struct sk_filter *filter;
1427
1428 if (sk->sk_destruct)
1429 sk->sk_destruct(sk);
1430
1431 filter = rcu_dereference_check(sk->sk_filter,
1432 atomic_read(&sk->sk_wmem_alloc) == 0);
1433 if (filter) {
1434 sk_filter_uncharge(sk, filter);
1435 RCU_INIT_POINTER(sk->sk_filter, NULL);
1436 }
1437 if (rcu_access_pointer(sk->sk_reuseport_cb))
1438 reuseport_detach_sock(sk);
1439
1440 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1441
1442 if (atomic_read(&sk->sk_omem_alloc))
1443 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1444 __func__, atomic_read(&sk->sk_omem_alloc));
1445
1446 if (sk->sk_peer_cred)
1447 put_cred(sk->sk_peer_cred);
1448 put_pid(sk->sk_peer_pid);
1449 if (likely(sk->sk_net_refcnt))
1450 put_net(sock_net(sk));
1451 sk_prot_free(sk->sk_prot_creator, sk);
1452}
1453
1454static void __sk_free(struct sock *sk)
1455{
1456 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1457 sock_diag_broadcast_destroy(sk);
1458 else
1459 sk_destruct(sk);
1460}
1461
1462void sk_free(struct sock *sk)
1463{
1464 /*
1465 * We subtract one from sk_wmem_alloc and can know if
1466 * some packets are still in some tx queue.
1467 * If not null, sock_wfree() will call __sk_free(sk) later
1468 */
1469 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1470 __sk_free(sk);
1471}
1472EXPORT_SYMBOL(sk_free);
1473
1474/**
1475 * sk_clone_lock - clone a socket, and lock its clone
1476 * @sk: the socket to clone
1477 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1478 *
1479 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1480 */
1481struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1482{
1483 struct sock *newsk;
1484 bool is_charged = true;
1485
1486 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1487 if (newsk != NULL) {
1488 struct sk_filter *filter;
1489
1490 sock_copy(newsk, sk);
1491
1492 /* SANITY */
1493 if (likely(newsk->sk_net_refcnt))
1494 get_net(sock_net(newsk));
1495 sk_node_init(&newsk->sk_node);
1496 sock_lock_init(newsk);
1497 bh_lock_sock(newsk);
1498 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1499 newsk->sk_backlog.len = 0;
1500
1501 atomic_set(&newsk->sk_rmem_alloc, 0);
1502 /*
1503 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1504 */
1505 atomic_set(&newsk->sk_wmem_alloc, 1);
1506 atomic_set(&newsk->sk_omem_alloc, 0);
1507 skb_queue_head_init(&newsk->sk_receive_queue);
1508 skb_queue_head_init(&newsk->sk_write_queue);
1509
1510 rwlock_init(&newsk->sk_callback_lock);
1511 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1512 af_callback_keys + newsk->sk_family,
1513 af_family_clock_key_strings[newsk->sk_family]);
1514
1515 newsk->sk_dst_cache = NULL;
1516 newsk->sk_wmem_queued = 0;
1517 newsk->sk_forward_alloc = 0;
1518 newsk->sk_send_head = NULL;
1519 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1520
1521 sock_reset_flag(newsk, SOCK_DONE);
1522 skb_queue_head_init(&newsk->sk_error_queue);
1523
1524 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1525 if (filter != NULL)
1526 /* though it's an empty new sock, the charging may fail
1527 * if sysctl_optmem_max was changed between creation of
1528 * original socket and cloning
1529 */
1530 is_charged = sk_filter_charge(newsk, filter);
1531
1532 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1533 /* It is still raw copy of parent, so invalidate
1534 * destructor and make plain sk_free() */
1535 newsk->sk_destruct = NULL;
1536 bh_unlock_sock(newsk);
1537 sk_free(newsk);
1538 newsk = NULL;
1539 goto out;
1540 }
1541 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1542
1543 newsk->sk_err = 0;
1544 newsk->sk_priority = 0;
1545 newsk->sk_incoming_cpu = raw_smp_processor_id();
1546 atomic64_set(&newsk->sk_cookie, 0);
1547 /*
1548 * Before updating sk_refcnt, we must commit prior changes to memory
1549 * (Documentation/RCU/rculist_nulls.txt for details)
1550 */
1551 smp_wmb();
1552 atomic_set(&newsk->sk_refcnt, 2);
1553
1554 /*
1555 * Increment the counter in the same struct proto as the master
1556 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1557 * is the same as sk->sk_prot->socks, as this field was copied
1558 * with memcpy).
1559 *
1560 * This _changes_ the previous behaviour, where
1561 * tcp_create_openreq_child always was incrementing the
1562 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1563 * to be taken into account in all callers. -acme
1564 */
1565 sk_refcnt_debug_inc(newsk);
1566 sk_set_socket(newsk, NULL);
1567 newsk->sk_wq = NULL;
1568
1569 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1570 sock_update_memcg(newsk);
1571
1572 if (newsk->sk_prot->sockets_allocated)
1573 sk_sockets_allocated_inc(newsk);
1574
1575 if (sock_needs_netstamp(sk) &&
1576 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1577 net_enable_timestamp();
1578 }
1579out:
1580 return newsk;
1581}
1582EXPORT_SYMBOL_GPL(sk_clone_lock);
1583
1584void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1585{
1586 u32 max_segs = 1;
1587
1588 sk_dst_set(sk, dst);
1589 sk->sk_route_caps = dst->dev->features;
1590 if (sk->sk_route_caps & NETIF_F_GSO)
1591 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1592 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1593 if (sk_can_gso(sk)) {
1594 if (dst->header_len) {
1595 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1596 } else {
1597 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1598 sk->sk_gso_max_size = dst->dev->gso_max_size;
1599 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1600 }
1601 }
1602 sk->sk_gso_max_segs = max_segs;
1603}
1604EXPORT_SYMBOL_GPL(sk_setup_caps);
1605
1606/*
1607 * Simple resource managers for sockets.
1608 */
1609
1610
1611/*
1612 * Write buffer destructor automatically called from kfree_skb.
1613 */
1614void sock_wfree(struct sk_buff *skb)
1615{
1616 struct sock *sk = skb->sk;
1617 unsigned int len = skb->truesize;
1618
1619 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1620 /*
1621 * Keep a reference on sk_wmem_alloc, this will be released
1622 * after sk_write_space() call
1623 */
1624 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1625 sk->sk_write_space(sk);
1626 len = 1;
1627 }
1628 /*
1629 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1630 * could not do because of in-flight packets
1631 */
1632 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1633 __sk_free(sk);
1634}
1635EXPORT_SYMBOL(sock_wfree);
1636
1637void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1638{
1639 skb_orphan(skb);
1640 skb->sk = sk;
1641#ifdef CONFIG_INET
1642 if (unlikely(!sk_fullsock(sk))) {
1643 skb->destructor = sock_edemux;
1644 sock_hold(sk);
1645 return;
1646 }
1647#endif
1648 skb->destructor = sock_wfree;
1649 skb_set_hash_from_sk(skb, sk);
1650 /*
1651 * We used to take a refcount on sk, but following operation
1652 * is enough to guarantee sk_free() wont free this sock until
1653 * all in-flight packets are completed
1654 */
1655 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1656}
1657EXPORT_SYMBOL(skb_set_owner_w);
1658
1659void skb_orphan_partial(struct sk_buff *skb)
1660{
1661 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1662 * so we do not completely orphan skb, but transfert all
1663 * accounted bytes but one, to avoid unexpected reorders.
1664 */
1665 if (skb->destructor == sock_wfree
1666#ifdef CONFIG_INET
1667 || skb->destructor == tcp_wfree
1668#endif
1669 ) {
1670 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1671 skb->truesize = 1;
1672 } else {
1673 skb_orphan(skb);
1674 }
1675}
1676EXPORT_SYMBOL(skb_orphan_partial);
1677
1678/*
1679 * Read buffer destructor automatically called from kfree_skb.
1680 */
1681void sock_rfree(struct sk_buff *skb)
1682{
1683 struct sock *sk = skb->sk;
1684 unsigned int len = skb->truesize;
1685
1686 atomic_sub(len, &sk->sk_rmem_alloc);
1687 sk_mem_uncharge(sk, len);
1688}
1689EXPORT_SYMBOL(sock_rfree);
1690
1691/*
1692 * Buffer destructor for skbs that are not used directly in read or write
1693 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1694 */
1695void sock_efree(struct sk_buff *skb)
1696{
1697 sock_put(skb->sk);
1698}
1699EXPORT_SYMBOL(sock_efree);
1700
1701kuid_t sock_i_uid(struct sock *sk)
1702{
1703 kuid_t uid;
1704
1705 read_lock_bh(&sk->sk_callback_lock);
1706 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1707 read_unlock_bh(&sk->sk_callback_lock);
1708 return uid;
1709}
1710EXPORT_SYMBOL(sock_i_uid);
1711
1712unsigned long sock_i_ino(struct sock *sk)
1713{
1714 unsigned long ino;
1715
1716 read_lock_bh(&sk->sk_callback_lock);
1717 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1718 read_unlock_bh(&sk->sk_callback_lock);
1719 return ino;
1720}
1721EXPORT_SYMBOL(sock_i_ino);
1722
1723/*
1724 * Allocate a skb from the socket's send buffer.
1725 */
1726struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1727 gfp_t priority)
1728{
1729 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1730 struct sk_buff *skb = alloc_skb(size, priority);
1731 if (skb) {
1732 skb_set_owner_w(skb, sk);
1733 return skb;
1734 }
1735 }
1736 return NULL;
1737}
1738EXPORT_SYMBOL(sock_wmalloc);
1739
1740/*
1741 * Allocate a memory block from the socket's option memory buffer.
1742 */
1743void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1744{
1745 if ((unsigned int)size <= sysctl_optmem_max &&
1746 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1747 void *mem;
1748 /* First do the add, to avoid the race if kmalloc
1749 * might sleep.
1750 */
1751 atomic_add(size, &sk->sk_omem_alloc);
1752 mem = kmalloc(size, priority);
1753 if (mem)
1754 return mem;
1755 atomic_sub(size, &sk->sk_omem_alloc);
1756 }
1757 return NULL;
1758}
1759EXPORT_SYMBOL(sock_kmalloc);
1760
1761/* Free an option memory block. Note, we actually want the inline
1762 * here as this allows gcc to detect the nullify and fold away the
1763 * condition entirely.
1764 */
1765static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1766 const bool nullify)
1767{
1768 if (WARN_ON_ONCE(!mem))
1769 return;
1770 if (nullify)
1771 kzfree(mem);
1772 else
1773 kfree(mem);
1774 atomic_sub(size, &sk->sk_omem_alloc);
1775}
1776
1777void sock_kfree_s(struct sock *sk, void *mem, int size)
1778{
1779 __sock_kfree_s(sk, mem, size, false);
1780}
1781EXPORT_SYMBOL(sock_kfree_s);
1782
1783void sock_kzfree_s(struct sock *sk, void *mem, int size)
1784{
1785 __sock_kfree_s(sk, mem, size, true);
1786}
1787EXPORT_SYMBOL(sock_kzfree_s);
1788
1789/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1790 I think, these locks should be removed for datagram sockets.
1791 */
1792static long sock_wait_for_wmem(struct sock *sk, long timeo)
1793{
1794 DEFINE_WAIT(wait);
1795
1796 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1797 for (;;) {
1798 if (!timeo)
1799 break;
1800 if (signal_pending(current))
1801 break;
1802 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1803 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1804 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1805 break;
1806 if (sk->sk_shutdown & SEND_SHUTDOWN)
1807 break;
1808 if (sk->sk_err)
1809 break;
1810 timeo = schedule_timeout(timeo);
1811 }
1812 finish_wait(sk_sleep(sk), &wait);
1813 return timeo;
1814}
1815
1816
1817/*
1818 * Generic send/receive buffer handlers
1819 */
1820
1821struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1822 unsigned long data_len, int noblock,
1823 int *errcode, int max_page_order)
1824{
1825 struct sk_buff *skb;
1826 long timeo;
1827 int err;
1828
1829 timeo = sock_sndtimeo(sk, noblock);
1830 for (;;) {
1831 err = sock_error(sk);
1832 if (err != 0)
1833 goto failure;
1834
1835 err = -EPIPE;
1836 if (sk->sk_shutdown & SEND_SHUTDOWN)
1837 goto failure;
1838
1839 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1840 break;
1841
1842 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1843 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1844 err = -EAGAIN;
1845 if (!timeo)
1846 goto failure;
1847 if (signal_pending(current))
1848 goto interrupted;
1849 timeo = sock_wait_for_wmem(sk, timeo);
1850 }
1851 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1852 errcode, sk->sk_allocation);
1853 if (skb)
1854 skb_set_owner_w(skb, sk);
1855 return skb;
1856
1857interrupted:
1858 err = sock_intr_errno(timeo);
1859failure:
1860 *errcode = err;
1861 return NULL;
1862}
1863EXPORT_SYMBOL(sock_alloc_send_pskb);
1864
1865struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1866 int noblock, int *errcode)
1867{
1868 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1869}
1870EXPORT_SYMBOL(sock_alloc_send_skb);
1871
1872int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1873 struct sockcm_cookie *sockc)
1874{
1875 struct cmsghdr *cmsg;
1876
1877 for_each_cmsghdr(cmsg, msg) {
1878 if (!CMSG_OK(msg, cmsg))
1879 return -EINVAL;
1880 if (cmsg->cmsg_level != SOL_SOCKET)
1881 continue;
1882 switch (cmsg->cmsg_type) {
1883 case SO_MARK:
1884 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1885 return -EPERM;
1886 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1887 return -EINVAL;
1888 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1889 break;
1890 default:
1891 return -EINVAL;
1892 }
1893 }
1894 return 0;
1895}
1896EXPORT_SYMBOL(sock_cmsg_send);
1897
1898/* On 32bit arches, an skb frag is limited to 2^15 */
1899#define SKB_FRAG_PAGE_ORDER get_order(32768)
1900
1901/**
1902 * skb_page_frag_refill - check that a page_frag contains enough room
1903 * @sz: minimum size of the fragment we want to get
1904 * @pfrag: pointer to page_frag
1905 * @gfp: priority for memory allocation
1906 *
1907 * Note: While this allocator tries to use high order pages, there is
1908 * no guarantee that allocations succeed. Therefore, @sz MUST be
1909 * less or equal than PAGE_SIZE.
1910 */
1911bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1912{
1913 if (pfrag->page) {
1914 if (page_ref_count(pfrag->page) == 1) {
1915 pfrag->offset = 0;
1916 return true;
1917 }
1918 if (pfrag->offset + sz <= pfrag->size)
1919 return true;
1920 put_page(pfrag->page);
1921 }
1922
1923 pfrag->offset = 0;
1924 if (SKB_FRAG_PAGE_ORDER) {
1925 /* Avoid direct reclaim but allow kswapd to wake */
1926 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1927 __GFP_COMP | __GFP_NOWARN |
1928 __GFP_NORETRY,
1929 SKB_FRAG_PAGE_ORDER);
1930 if (likely(pfrag->page)) {
1931 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1932 return true;
1933 }
1934 }
1935 pfrag->page = alloc_page(gfp);
1936 if (likely(pfrag->page)) {
1937 pfrag->size = PAGE_SIZE;
1938 return true;
1939 }
1940 return false;
1941}
1942EXPORT_SYMBOL(skb_page_frag_refill);
1943
1944bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1945{
1946 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1947 return true;
1948
1949 sk_enter_memory_pressure(sk);
1950 sk_stream_moderate_sndbuf(sk);
1951 return false;
1952}
1953EXPORT_SYMBOL(sk_page_frag_refill);
1954
1955static void __lock_sock(struct sock *sk)
1956 __releases(&sk->sk_lock.slock)
1957 __acquires(&sk->sk_lock.slock)
1958{
1959 DEFINE_WAIT(wait);
1960
1961 for (;;) {
1962 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1963 TASK_UNINTERRUPTIBLE);
1964 spin_unlock_bh(&sk->sk_lock.slock);
1965 schedule();
1966 spin_lock_bh(&sk->sk_lock.slock);
1967 if (!sock_owned_by_user(sk))
1968 break;
1969 }
1970 finish_wait(&sk->sk_lock.wq, &wait);
1971}
1972
1973static void __release_sock(struct sock *sk)
1974 __releases(&sk->sk_lock.slock)
1975 __acquires(&sk->sk_lock.slock)
1976{
1977 struct sk_buff *skb = sk->sk_backlog.head;
1978
1979 do {
1980 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1981 bh_unlock_sock(sk);
1982
1983 do {
1984 struct sk_buff *next = skb->next;
1985
1986 prefetch(next);
1987 WARN_ON_ONCE(skb_dst_is_noref(skb));
1988 skb->next = NULL;
1989 sk_backlog_rcv(sk, skb);
1990
1991 /*
1992 * We are in process context here with softirqs
1993 * disabled, use cond_resched_softirq() to preempt.
1994 * This is safe to do because we've taken the backlog
1995 * queue private:
1996 */
1997 cond_resched_softirq();
1998
1999 skb = next;
2000 } while (skb != NULL);
2001
2002 bh_lock_sock(sk);
2003 } while ((skb = sk->sk_backlog.head) != NULL);
2004
2005 /*
2006 * Doing the zeroing here guarantee we can not loop forever
2007 * while a wild producer attempts to flood us.
2008 */
2009 sk->sk_backlog.len = 0;
2010}
2011
2012/**
2013 * sk_wait_data - wait for data to arrive at sk_receive_queue
2014 * @sk: sock to wait on
2015 * @timeo: for how long
2016 * @skb: last skb seen on sk_receive_queue
2017 *
2018 * Now socket state including sk->sk_err is changed only under lock,
2019 * hence we may omit checks after joining wait queue.
2020 * We check receive queue before schedule() only as optimization;
2021 * it is very likely that release_sock() added new data.
2022 */
2023int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2024{
2025 int rc;
2026 DEFINE_WAIT(wait);
2027
2028 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2029 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2030 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2031 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2032 finish_wait(sk_sleep(sk), &wait);
2033 return rc;
2034}
2035EXPORT_SYMBOL(sk_wait_data);
2036
2037/**
2038 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2039 * @sk: socket
2040 * @size: memory size to allocate
2041 * @kind: allocation type
2042 *
2043 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2044 * rmem allocation. This function assumes that protocols which have
2045 * memory_pressure use sk_wmem_queued as write buffer accounting.
2046 */
2047int __sk_mem_schedule(struct sock *sk, int size, int kind)
2048{
2049 struct proto *prot = sk->sk_prot;
2050 int amt = sk_mem_pages(size);
2051 long allocated;
2052
2053 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2054
2055 allocated = sk_memory_allocated_add(sk, amt);
2056
2057 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2058 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2059 goto suppress_allocation;
2060
2061 /* Under limit. */
2062 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2063 sk_leave_memory_pressure(sk);
2064 return 1;
2065 }
2066
2067 /* Under pressure. */
2068 if (allocated > sk_prot_mem_limits(sk, 1))
2069 sk_enter_memory_pressure(sk);
2070
2071 /* Over hard limit. */
2072 if (allocated > sk_prot_mem_limits(sk, 2))
2073 goto suppress_allocation;
2074
2075 /* guarantee minimum buffer size under pressure */
2076 if (kind == SK_MEM_RECV) {
2077 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2078 return 1;
2079
2080 } else { /* SK_MEM_SEND */
2081 if (sk->sk_type == SOCK_STREAM) {
2082 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2083 return 1;
2084 } else if (atomic_read(&sk->sk_wmem_alloc) <
2085 prot->sysctl_wmem[0])
2086 return 1;
2087 }
2088
2089 if (sk_has_memory_pressure(sk)) {
2090 int alloc;
2091
2092 if (!sk_under_memory_pressure(sk))
2093 return 1;
2094 alloc = sk_sockets_allocated_read_positive(sk);
2095 if (sk_prot_mem_limits(sk, 2) > alloc *
2096 sk_mem_pages(sk->sk_wmem_queued +
2097 atomic_read(&sk->sk_rmem_alloc) +
2098 sk->sk_forward_alloc))
2099 return 1;
2100 }
2101
2102suppress_allocation:
2103
2104 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2105 sk_stream_moderate_sndbuf(sk);
2106
2107 /* Fail only if socket is _under_ its sndbuf.
2108 * In this case we cannot block, so that we have to fail.
2109 */
2110 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2111 return 1;
2112 }
2113
2114 trace_sock_exceed_buf_limit(sk, prot, allocated);
2115
2116 /* Alas. Undo changes. */
2117 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2118
2119 sk_memory_allocated_sub(sk, amt);
2120
2121 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2122 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2123
2124 return 0;
2125}
2126EXPORT_SYMBOL(__sk_mem_schedule);
2127
2128/**
2129 * __sk_mem_reclaim - reclaim memory_allocated
2130 * @sk: socket
2131 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2132 */
2133void __sk_mem_reclaim(struct sock *sk, int amount)
2134{
2135 amount >>= SK_MEM_QUANTUM_SHIFT;
2136 sk_memory_allocated_sub(sk, amount);
2137 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2138
2139 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2140 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2141
2142 if (sk_under_memory_pressure(sk) &&
2143 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2144 sk_leave_memory_pressure(sk);
2145}
2146EXPORT_SYMBOL(__sk_mem_reclaim);
2147
2148
2149/*
2150 * Set of default routines for initialising struct proto_ops when
2151 * the protocol does not support a particular function. In certain
2152 * cases where it makes no sense for a protocol to have a "do nothing"
2153 * function, some default processing is provided.
2154 */
2155
2156int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2157{
2158 return -EOPNOTSUPP;
2159}
2160EXPORT_SYMBOL(sock_no_bind);
2161
2162int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2163 int len, int flags)
2164{
2165 return -EOPNOTSUPP;
2166}
2167EXPORT_SYMBOL(sock_no_connect);
2168
2169int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2170{
2171 return -EOPNOTSUPP;
2172}
2173EXPORT_SYMBOL(sock_no_socketpair);
2174
2175int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2176{
2177 return -EOPNOTSUPP;
2178}
2179EXPORT_SYMBOL(sock_no_accept);
2180
2181int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2182 int *len, int peer)
2183{
2184 return -EOPNOTSUPP;
2185}
2186EXPORT_SYMBOL(sock_no_getname);
2187
2188unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2189{
2190 return 0;
2191}
2192EXPORT_SYMBOL(sock_no_poll);
2193
2194int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2195{
2196 return -EOPNOTSUPP;
2197}
2198EXPORT_SYMBOL(sock_no_ioctl);
2199
2200int sock_no_listen(struct socket *sock, int backlog)
2201{
2202 return -EOPNOTSUPP;
2203}
2204EXPORT_SYMBOL(sock_no_listen);
2205
2206int sock_no_shutdown(struct socket *sock, int how)
2207{
2208 return -EOPNOTSUPP;
2209}
2210EXPORT_SYMBOL(sock_no_shutdown);
2211
2212int sock_no_setsockopt(struct socket *sock, int level, int optname,
2213 char __user *optval, unsigned int optlen)
2214{
2215 return -EOPNOTSUPP;
2216}
2217EXPORT_SYMBOL(sock_no_setsockopt);
2218
2219int sock_no_getsockopt(struct socket *sock, int level, int optname,
2220 char __user *optval, int __user *optlen)
2221{
2222 return -EOPNOTSUPP;
2223}
2224EXPORT_SYMBOL(sock_no_getsockopt);
2225
2226int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2227{
2228 return -EOPNOTSUPP;
2229}
2230EXPORT_SYMBOL(sock_no_sendmsg);
2231
2232int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2233 int flags)
2234{
2235 return -EOPNOTSUPP;
2236}
2237EXPORT_SYMBOL(sock_no_recvmsg);
2238
2239int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2240{
2241 /* Mirror missing mmap method error code */
2242 return -ENODEV;
2243}
2244EXPORT_SYMBOL(sock_no_mmap);
2245
2246ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2247{
2248 ssize_t res;
2249 struct msghdr msg = {.msg_flags = flags};
2250 struct kvec iov;
2251 char *kaddr = kmap(page);
2252 iov.iov_base = kaddr + offset;
2253 iov.iov_len = size;
2254 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2255 kunmap(page);
2256 return res;
2257}
2258EXPORT_SYMBOL(sock_no_sendpage);
2259
2260/*
2261 * Default Socket Callbacks
2262 */
2263
2264static void sock_def_wakeup(struct sock *sk)
2265{
2266 struct socket_wq *wq;
2267
2268 rcu_read_lock();
2269 wq = rcu_dereference(sk->sk_wq);
2270 if (skwq_has_sleeper(wq))
2271 wake_up_interruptible_all(&wq->wait);
2272 rcu_read_unlock();
2273}
2274
2275static void sock_def_error_report(struct sock *sk)
2276{
2277 struct socket_wq *wq;
2278
2279 rcu_read_lock();
2280 wq = rcu_dereference(sk->sk_wq);
2281 if (skwq_has_sleeper(wq))
2282 wake_up_interruptible_poll(&wq->wait, POLLERR);
2283 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2284 rcu_read_unlock();
2285}
2286
2287static void sock_def_readable(struct sock *sk)
2288{
2289 struct socket_wq *wq;
2290
2291 rcu_read_lock();
2292 wq = rcu_dereference(sk->sk_wq);
2293 if (skwq_has_sleeper(wq))
2294 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2295 POLLRDNORM | POLLRDBAND);
2296 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2297 rcu_read_unlock();
2298}
2299
2300static void sock_def_write_space(struct sock *sk)
2301{
2302 struct socket_wq *wq;
2303
2304 rcu_read_lock();
2305
2306 /* Do not wake up a writer until he can make "significant"
2307 * progress. --DaveM
2308 */
2309 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2310 wq = rcu_dereference(sk->sk_wq);
2311 if (skwq_has_sleeper(wq))
2312 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2313 POLLWRNORM | POLLWRBAND);
2314
2315 /* Should agree with poll, otherwise some programs break */
2316 if (sock_writeable(sk))
2317 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2318 }
2319
2320 rcu_read_unlock();
2321}
2322
2323static void sock_def_destruct(struct sock *sk)
2324{
2325}
2326
2327void sk_send_sigurg(struct sock *sk)
2328{
2329 if (sk->sk_socket && sk->sk_socket->file)
2330 if (send_sigurg(&sk->sk_socket->file->f_owner))
2331 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2332}
2333EXPORT_SYMBOL(sk_send_sigurg);
2334
2335void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2336 unsigned long expires)
2337{
2338 if (!mod_timer(timer, expires))
2339 sock_hold(sk);
2340}
2341EXPORT_SYMBOL(sk_reset_timer);
2342
2343void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2344{
2345 if (del_timer(timer))
2346 __sock_put(sk);
2347}
2348EXPORT_SYMBOL(sk_stop_timer);
2349
2350void sock_init_data(struct socket *sock, struct sock *sk)
2351{
2352 skb_queue_head_init(&sk->sk_receive_queue);
2353 skb_queue_head_init(&sk->sk_write_queue);
2354 skb_queue_head_init(&sk->sk_error_queue);
2355
2356 sk->sk_send_head = NULL;
2357
2358 init_timer(&sk->sk_timer);
2359
2360 sk->sk_allocation = GFP_KERNEL;
2361 sk->sk_rcvbuf = sysctl_rmem_default;
2362 sk->sk_sndbuf = sysctl_wmem_default;
2363 sk->sk_state = TCP_CLOSE;
2364 sk_set_socket(sk, sock);
2365
2366 sock_set_flag(sk, SOCK_ZAPPED);
2367
2368 if (sock) {
2369 sk->sk_type = sock->type;
2370 sk->sk_wq = sock->wq;
2371 sock->sk = sk;
2372 } else
2373 sk->sk_wq = NULL;
2374
2375 rwlock_init(&sk->sk_callback_lock);
2376 lockdep_set_class_and_name(&sk->sk_callback_lock,
2377 af_callback_keys + sk->sk_family,
2378 af_family_clock_key_strings[sk->sk_family]);
2379
2380 sk->sk_state_change = sock_def_wakeup;
2381 sk->sk_data_ready = sock_def_readable;
2382 sk->sk_write_space = sock_def_write_space;
2383 sk->sk_error_report = sock_def_error_report;
2384 sk->sk_destruct = sock_def_destruct;
2385
2386 sk->sk_frag.page = NULL;
2387 sk->sk_frag.offset = 0;
2388 sk->sk_peek_off = -1;
2389
2390 sk->sk_peer_pid = NULL;
2391 sk->sk_peer_cred = NULL;
2392 sk->sk_write_pending = 0;
2393 sk->sk_rcvlowat = 1;
2394 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2395 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2396
2397 sk->sk_stamp = ktime_set(-1L, 0);
2398
2399#ifdef CONFIG_NET_RX_BUSY_POLL
2400 sk->sk_napi_id = 0;
2401 sk->sk_ll_usec = sysctl_net_busy_read;
2402#endif
2403
2404 sk->sk_max_pacing_rate = ~0U;
2405 sk->sk_pacing_rate = ~0U;
2406 sk->sk_incoming_cpu = -1;
2407 /*
2408 * Before updating sk_refcnt, we must commit prior changes to memory
2409 * (Documentation/RCU/rculist_nulls.txt for details)
2410 */
2411 smp_wmb();
2412 atomic_set(&sk->sk_refcnt, 1);
2413 atomic_set(&sk->sk_drops, 0);
2414}
2415EXPORT_SYMBOL(sock_init_data);
2416
2417void lock_sock_nested(struct sock *sk, int subclass)
2418{
2419 might_sleep();
2420 spin_lock_bh(&sk->sk_lock.slock);
2421 if (sk->sk_lock.owned)
2422 __lock_sock(sk);
2423 sk->sk_lock.owned = 1;
2424 spin_unlock(&sk->sk_lock.slock);
2425 /*
2426 * The sk_lock has mutex_lock() semantics here:
2427 */
2428 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2429 local_bh_enable();
2430}
2431EXPORT_SYMBOL(lock_sock_nested);
2432
2433void release_sock(struct sock *sk)
2434{
2435 /*
2436 * The sk_lock has mutex_unlock() semantics:
2437 */
2438 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2439
2440 spin_lock_bh(&sk->sk_lock.slock);
2441 if (sk->sk_backlog.tail)
2442 __release_sock(sk);
2443
2444 /* Warning : release_cb() might need to release sk ownership,
2445 * ie call sock_release_ownership(sk) before us.
2446 */
2447 if (sk->sk_prot->release_cb)
2448 sk->sk_prot->release_cb(sk);
2449
2450 sock_release_ownership(sk);
2451 if (waitqueue_active(&sk->sk_lock.wq))
2452 wake_up(&sk->sk_lock.wq);
2453 spin_unlock_bh(&sk->sk_lock.slock);
2454}
2455EXPORT_SYMBOL(release_sock);
2456
2457/**
2458 * lock_sock_fast - fast version of lock_sock
2459 * @sk: socket
2460 *
2461 * This version should be used for very small section, where process wont block
2462 * return false if fast path is taken
2463 * sk_lock.slock locked, owned = 0, BH disabled
2464 * return true if slow path is taken
2465 * sk_lock.slock unlocked, owned = 1, BH enabled
2466 */
2467bool lock_sock_fast(struct sock *sk)
2468{
2469 might_sleep();
2470 spin_lock_bh(&sk->sk_lock.slock);
2471
2472 if (!sk->sk_lock.owned)
2473 /*
2474 * Note : We must disable BH
2475 */
2476 return false;
2477
2478 __lock_sock(sk);
2479 sk->sk_lock.owned = 1;
2480 spin_unlock(&sk->sk_lock.slock);
2481 /*
2482 * The sk_lock has mutex_lock() semantics here:
2483 */
2484 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2485 local_bh_enable();
2486 return true;
2487}
2488EXPORT_SYMBOL(lock_sock_fast);
2489
2490int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2491{
2492 struct timeval tv;
2493 if (!sock_flag(sk, SOCK_TIMESTAMP))
2494 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2495 tv = ktime_to_timeval(sk->sk_stamp);
2496 if (tv.tv_sec == -1)
2497 return -ENOENT;
2498 if (tv.tv_sec == 0) {
2499 sk->sk_stamp = ktime_get_real();
2500 tv = ktime_to_timeval(sk->sk_stamp);
2501 }
2502 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2503}
2504EXPORT_SYMBOL(sock_get_timestamp);
2505
2506int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2507{
2508 struct timespec ts;
2509 if (!sock_flag(sk, SOCK_TIMESTAMP))
2510 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2511 ts = ktime_to_timespec(sk->sk_stamp);
2512 if (ts.tv_sec == -1)
2513 return -ENOENT;
2514 if (ts.tv_sec == 0) {
2515 sk->sk_stamp = ktime_get_real();
2516 ts = ktime_to_timespec(sk->sk_stamp);
2517 }
2518 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2519}
2520EXPORT_SYMBOL(sock_get_timestampns);
2521
2522void sock_enable_timestamp(struct sock *sk, int flag)
2523{
2524 if (!sock_flag(sk, flag)) {
2525 unsigned long previous_flags = sk->sk_flags;
2526
2527 sock_set_flag(sk, flag);
2528 /*
2529 * we just set one of the two flags which require net
2530 * time stamping, but time stamping might have been on
2531 * already because of the other one
2532 */
2533 if (sock_needs_netstamp(sk) &&
2534 !(previous_flags & SK_FLAGS_TIMESTAMP))
2535 net_enable_timestamp();
2536 }
2537}
2538
2539int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2540 int level, int type)
2541{
2542 struct sock_exterr_skb *serr;
2543 struct sk_buff *skb;
2544 int copied, err;
2545
2546 err = -EAGAIN;
2547 skb = sock_dequeue_err_skb(sk);
2548 if (skb == NULL)
2549 goto out;
2550
2551 copied = skb->len;
2552 if (copied > len) {
2553 msg->msg_flags |= MSG_TRUNC;
2554 copied = len;
2555 }
2556 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2557 if (err)
2558 goto out_free_skb;
2559
2560 sock_recv_timestamp(msg, sk, skb);
2561
2562 serr = SKB_EXT_ERR(skb);
2563 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2564
2565 msg->msg_flags |= MSG_ERRQUEUE;
2566 err = copied;
2567
2568out_free_skb:
2569 kfree_skb(skb);
2570out:
2571 return err;
2572}
2573EXPORT_SYMBOL(sock_recv_errqueue);
2574
2575/*
2576 * Get a socket option on an socket.
2577 *
2578 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2579 * asynchronous errors should be reported by getsockopt. We assume
2580 * this means if you specify SO_ERROR (otherwise whats the point of it).
2581 */
2582int sock_common_getsockopt(struct socket *sock, int level, int optname,
2583 char __user *optval, int __user *optlen)
2584{
2585 struct sock *sk = sock->sk;
2586
2587 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2588}
2589EXPORT_SYMBOL(sock_common_getsockopt);
2590
2591#ifdef CONFIG_COMPAT
2592int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2593 char __user *optval, int __user *optlen)
2594{
2595 struct sock *sk = sock->sk;
2596
2597 if (sk->sk_prot->compat_getsockopt != NULL)
2598 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2599 optval, optlen);
2600 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2601}
2602EXPORT_SYMBOL(compat_sock_common_getsockopt);
2603#endif
2604
2605int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2606 int flags)
2607{
2608 struct sock *sk = sock->sk;
2609 int addr_len = 0;
2610 int err;
2611
2612 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2613 flags & ~MSG_DONTWAIT, &addr_len);
2614 if (err >= 0)
2615 msg->msg_namelen = addr_len;
2616 return err;
2617}
2618EXPORT_SYMBOL(sock_common_recvmsg);
2619
2620/*
2621 * Set socket options on an inet socket.
2622 */
2623int sock_common_setsockopt(struct socket *sock, int level, int optname,
2624 char __user *optval, unsigned int optlen)
2625{
2626 struct sock *sk = sock->sk;
2627
2628 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2629}
2630EXPORT_SYMBOL(sock_common_setsockopt);
2631
2632#ifdef CONFIG_COMPAT
2633int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2634 char __user *optval, unsigned int optlen)
2635{
2636 struct sock *sk = sock->sk;
2637
2638 if (sk->sk_prot->compat_setsockopt != NULL)
2639 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2640 optval, optlen);
2641 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2642}
2643EXPORT_SYMBOL(compat_sock_common_setsockopt);
2644#endif
2645
2646void sk_common_release(struct sock *sk)
2647{
2648 if (sk->sk_prot->destroy)
2649 sk->sk_prot->destroy(sk);
2650
2651 /*
2652 * Observation: when sock_common_release is called, processes have
2653 * no access to socket. But net still has.
2654 * Step one, detach it from networking:
2655 *
2656 * A. Remove from hash tables.
2657 */
2658
2659 sk->sk_prot->unhash(sk);
2660
2661 /*
2662 * In this point socket cannot receive new packets, but it is possible
2663 * that some packets are in flight because some CPU runs receiver and
2664 * did hash table lookup before we unhashed socket. They will achieve
2665 * receive queue and will be purged by socket destructor.
2666 *
2667 * Also we still have packets pending on receive queue and probably,
2668 * our own packets waiting in device queues. sock_destroy will drain
2669 * receive queue, but transmitted packets will delay socket destruction
2670 * until the last reference will be released.
2671 */
2672
2673 sock_orphan(sk);
2674
2675 xfrm_sk_free_policy(sk);
2676
2677 sk_refcnt_debug_release(sk);
2678
2679 if (sk->sk_frag.page) {
2680 put_page(sk->sk_frag.page);
2681 sk->sk_frag.page = NULL;
2682 }
2683
2684 sock_put(sk);
2685}
2686EXPORT_SYMBOL(sk_common_release);
2687
2688#ifdef CONFIG_PROC_FS
2689#define PROTO_INUSE_NR 64 /* should be enough for the first time */
2690struct prot_inuse {
2691 int val[PROTO_INUSE_NR];
2692};
2693
2694static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2695
2696#ifdef CONFIG_NET_NS
2697void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2698{
2699 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2700}
2701EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2702
2703int sock_prot_inuse_get(struct net *net, struct proto *prot)
2704{
2705 int cpu, idx = prot->inuse_idx;
2706 int res = 0;
2707
2708 for_each_possible_cpu(cpu)
2709 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2710
2711 return res >= 0 ? res : 0;
2712}
2713EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2714
2715static int __net_init sock_inuse_init_net(struct net *net)
2716{
2717 net->core.inuse = alloc_percpu(struct prot_inuse);
2718 return net->core.inuse ? 0 : -ENOMEM;
2719}
2720
2721static void __net_exit sock_inuse_exit_net(struct net *net)
2722{
2723 free_percpu(net->core.inuse);
2724}
2725
2726static struct pernet_operations net_inuse_ops = {
2727 .init = sock_inuse_init_net,
2728 .exit = sock_inuse_exit_net,
2729};
2730
2731static __init int net_inuse_init(void)
2732{
2733 if (register_pernet_subsys(&net_inuse_ops))
2734 panic("Cannot initialize net inuse counters");
2735
2736 return 0;
2737}
2738
2739core_initcall(net_inuse_init);
2740#else
2741static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2742
2743void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2744{
2745 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2746}
2747EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2748
2749int sock_prot_inuse_get(struct net *net, struct proto *prot)
2750{
2751 int cpu, idx = prot->inuse_idx;
2752 int res = 0;
2753
2754 for_each_possible_cpu(cpu)
2755 res += per_cpu(prot_inuse, cpu).val[idx];
2756
2757 return res >= 0 ? res : 0;
2758}
2759EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2760#endif
2761
2762static void assign_proto_idx(struct proto *prot)
2763{
2764 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2765
2766 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2767 pr_err("PROTO_INUSE_NR exhausted\n");
2768 return;
2769 }
2770
2771 set_bit(prot->inuse_idx, proto_inuse_idx);
2772}
2773
2774static void release_proto_idx(struct proto *prot)
2775{
2776 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2777 clear_bit(prot->inuse_idx, proto_inuse_idx);
2778}
2779#else
2780static inline void assign_proto_idx(struct proto *prot)
2781{
2782}
2783
2784static inline void release_proto_idx(struct proto *prot)
2785{
2786}
2787#endif
2788
2789static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2790{
2791 if (!rsk_prot)
2792 return;
2793 kfree(rsk_prot->slab_name);
2794 rsk_prot->slab_name = NULL;
2795 kmem_cache_destroy(rsk_prot->slab);
2796 rsk_prot->slab = NULL;
2797}
2798
2799static int req_prot_init(const struct proto *prot)
2800{
2801 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2802
2803 if (!rsk_prot)
2804 return 0;
2805
2806 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2807 prot->name);
2808 if (!rsk_prot->slab_name)
2809 return -ENOMEM;
2810
2811 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2812 rsk_prot->obj_size, 0,
2813 prot->slab_flags, NULL);
2814
2815 if (!rsk_prot->slab) {
2816 pr_crit("%s: Can't create request sock SLAB cache!\n",
2817 prot->name);
2818 return -ENOMEM;
2819 }
2820 return 0;
2821}
2822
2823int proto_register(struct proto *prot, int alloc_slab)
2824{
2825 if (alloc_slab) {
2826 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2827 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2828 NULL);
2829
2830 if (prot->slab == NULL) {
2831 pr_crit("%s: Can't create sock SLAB cache!\n",
2832 prot->name);
2833 goto out;
2834 }
2835
2836 if (req_prot_init(prot))
2837 goto out_free_request_sock_slab;
2838
2839 if (prot->twsk_prot != NULL) {
2840 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2841
2842 if (prot->twsk_prot->twsk_slab_name == NULL)
2843 goto out_free_request_sock_slab;
2844
2845 prot->twsk_prot->twsk_slab =
2846 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2847 prot->twsk_prot->twsk_obj_size,
2848 0,
2849 prot->slab_flags,
2850 NULL);
2851 if (prot->twsk_prot->twsk_slab == NULL)
2852 goto out_free_timewait_sock_slab_name;
2853 }
2854 }
2855
2856 mutex_lock(&proto_list_mutex);
2857 list_add(&prot->node, &proto_list);
2858 assign_proto_idx(prot);
2859 mutex_unlock(&proto_list_mutex);
2860 return 0;
2861
2862out_free_timewait_sock_slab_name:
2863 kfree(prot->twsk_prot->twsk_slab_name);
2864out_free_request_sock_slab:
2865 req_prot_cleanup(prot->rsk_prot);
2866
2867 kmem_cache_destroy(prot->slab);
2868 prot->slab = NULL;
2869out:
2870 return -ENOBUFS;
2871}
2872EXPORT_SYMBOL(proto_register);
2873
2874void proto_unregister(struct proto *prot)
2875{
2876 mutex_lock(&proto_list_mutex);
2877 release_proto_idx(prot);
2878 list_del(&prot->node);
2879 mutex_unlock(&proto_list_mutex);
2880
2881 kmem_cache_destroy(prot->slab);
2882 prot->slab = NULL;
2883
2884 req_prot_cleanup(prot->rsk_prot);
2885
2886 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2887 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2888 kfree(prot->twsk_prot->twsk_slab_name);
2889 prot->twsk_prot->twsk_slab = NULL;
2890 }
2891}
2892EXPORT_SYMBOL(proto_unregister);
2893
2894#ifdef CONFIG_PROC_FS
2895static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2896 __acquires(proto_list_mutex)
2897{
2898 mutex_lock(&proto_list_mutex);
2899 return seq_list_start_head(&proto_list, *pos);
2900}
2901
2902static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2903{
2904 return seq_list_next(v, &proto_list, pos);
2905}
2906
2907static void proto_seq_stop(struct seq_file *seq, void *v)
2908 __releases(proto_list_mutex)
2909{
2910 mutex_unlock(&proto_list_mutex);
2911}
2912
2913static char proto_method_implemented(const void *method)
2914{
2915 return method == NULL ? 'n' : 'y';
2916}
2917static long sock_prot_memory_allocated(struct proto *proto)
2918{
2919 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2920}
2921
2922static char *sock_prot_memory_pressure(struct proto *proto)
2923{
2924 return proto->memory_pressure != NULL ?
2925 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2926}
2927
2928static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2929{
2930
2931 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2932 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2933 proto->name,
2934 proto->obj_size,
2935 sock_prot_inuse_get(seq_file_net(seq), proto),
2936 sock_prot_memory_allocated(proto),
2937 sock_prot_memory_pressure(proto),
2938 proto->max_header,
2939 proto->slab == NULL ? "no" : "yes",
2940 module_name(proto->owner),
2941 proto_method_implemented(proto->close),
2942 proto_method_implemented(proto->connect),
2943 proto_method_implemented(proto->disconnect),
2944 proto_method_implemented(proto->accept),
2945 proto_method_implemented(proto->ioctl),
2946 proto_method_implemented(proto->init),
2947 proto_method_implemented(proto->destroy),
2948 proto_method_implemented(proto->shutdown),
2949 proto_method_implemented(proto->setsockopt),
2950 proto_method_implemented(proto->getsockopt),
2951 proto_method_implemented(proto->sendmsg),
2952 proto_method_implemented(proto->recvmsg),
2953 proto_method_implemented(proto->sendpage),
2954 proto_method_implemented(proto->bind),
2955 proto_method_implemented(proto->backlog_rcv),
2956 proto_method_implemented(proto->hash),
2957 proto_method_implemented(proto->unhash),
2958 proto_method_implemented(proto->get_port),
2959 proto_method_implemented(proto->enter_memory_pressure));
2960}
2961
2962static int proto_seq_show(struct seq_file *seq, void *v)
2963{
2964 if (v == &proto_list)
2965 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2966 "protocol",
2967 "size",
2968 "sockets",
2969 "memory",
2970 "press",
2971 "maxhdr",
2972 "slab",
2973 "module",
2974 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2975 else
2976 proto_seq_printf(seq, list_entry(v, struct proto, node));
2977 return 0;
2978}
2979
2980static const struct seq_operations proto_seq_ops = {
2981 .start = proto_seq_start,
2982 .next = proto_seq_next,
2983 .stop = proto_seq_stop,
2984 .show = proto_seq_show,
2985};
2986
2987static int proto_seq_open(struct inode *inode, struct file *file)
2988{
2989 return seq_open_net(inode, file, &proto_seq_ops,
2990 sizeof(struct seq_net_private));
2991}
2992
2993static const struct file_operations proto_seq_fops = {
2994 .owner = THIS_MODULE,
2995 .open = proto_seq_open,
2996 .read = seq_read,
2997 .llseek = seq_lseek,
2998 .release = seq_release_net,
2999};
3000
3001static __net_init int proto_init_net(struct net *net)
3002{
3003 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3004 return -ENOMEM;
3005
3006 return 0;
3007}
3008
3009static __net_exit void proto_exit_net(struct net *net)
3010{
3011 remove_proc_entry("protocols", net->proc_net);
3012}
3013
3014
3015static __net_initdata struct pernet_operations proto_net_ops = {
3016 .init = proto_init_net,
3017 .exit = proto_exit_net,
3018};
3019
3020static int __init proto_init(void)
3021{
3022 return register_pernet_subsys(&proto_net_ops);
3023}
3024
3025subsys_initcall(proto_init);
3026
3027#endif /* PROC_FS */