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#include <linux/capability.h>
93#include <linux/errno.h>
94#include <linux/types.h>
95#include <linux/socket.h>
96#include <linux/in.h>
97#include <linux/kernel.h>
98#include <linux/module.h>
99#include <linux/proc_fs.h>
100#include <linux/seq_file.h>
101#include <linux/sched.h>
102#include <linux/timer.h>
103#include <linux/string.h>
104#include <linux/sockios.h>
105#include <linux/net.h>
106#include <linux/mm.h>
107#include <linux/slab.h>
108#include <linux/interrupt.h>
109#include <linux/poll.h>
110#include <linux/tcp.h>
111#include <linux/init.h>
112#include <linux/highmem.h>
113#include <linux/user_namespace.h>
114
115#include <asm/uaccess.h>
116#include <asm/system.h>
117
118#include <linux/netdevice.h>
119#include <net/protocol.h>
120#include <linux/skbuff.h>
121#include <net/net_namespace.h>
122#include <net/request_sock.h>
123#include <net/sock.h>
124#include <linux/net_tstamp.h>
125#include <net/xfrm.h>
126#include <linux/ipsec.h>
127#include <net/cls_cgroup.h>
128
129#include <linux/filter.h>
130
131#include <trace/events/sock.h>
132
133#ifdef CONFIG_INET
134#include <net/tcp.h>
135#endif
136
137/*
138 * Each address family might have different locking rules, so we have
139 * one slock key per address family:
140 */
141static struct lock_class_key af_family_keys[AF_MAX];
142static struct lock_class_key af_family_slock_keys[AF_MAX];
143
144/*
145 * Make lock validator output more readable. (we pre-construct these
146 * strings build-time, so that runtime initialization of socket
147 * locks is fast):
148 */
149static const char *const af_family_key_strings[AF_MAX+1] = {
150 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
151 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
152 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
153 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
154 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
155 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
156 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
157 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
158 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
159 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
160 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
161 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
162 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
163 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
164};
165static const char *const af_family_slock_key_strings[AF_MAX+1] = {
166 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
167 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
168 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
169 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
170 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
171 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
172 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
173 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
174 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
175 "slock-27" , "slock-28" , "slock-AF_CAN" ,
176 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
177 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
178 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
179 "slock-AF_NFC" , "slock-AF_MAX"
180};
181static const char *const af_family_clock_key_strings[AF_MAX+1] = {
182 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
183 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
184 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
185 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
186 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
187 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
188 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
189 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
190 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
191 "clock-27" , "clock-28" , "clock-AF_CAN" ,
192 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
193 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
194 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
195 "clock-AF_NFC" , "clock-AF_MAX"
196};
197
198/*
199 * sk_callback_lock locking rules are per-address-family,
200 * so split the lock classes by using a per-AF key:
201 */
202static struct lock_class_key af_callback_keys[AF_MAX];
203
204/* Take into consideration the size of the struct sk_buff overhead in the
205 * determination of these values, since that is non-constant across
206 * platforms. This makes socket queueing behavior and performance
207 * not depend upon such differences.
208 */
209#define _SK_MEM_PACKETS 256
210#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
211#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
213
214/* Run time adjustable parameters. */
215__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
216__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
217__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
218__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
219
220/* Maximal space eaten by iovec or ancillary data plus some space */
221int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
222EXPORT_SYMBOL(sysctl_optmem_max);
223
224#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
225int net_cls_subsys_id = -1;
226EXPORT_SYMBOL_GPL(net_cls_subsys_id);
227#endif
228
229static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
230{
231 struct timeval tv;
232
233 if (optlen < sizeof(tv))
234 return -EINVAL;
235 if (copy_from_user(&tv, optval, sizeof(tv)))
236 return -EFAULT;
237 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
238 return -EDOM;
239
240 if (tv.tv_sec < 0) {
241 static int warned __read_mostly;
242
243 *timeo_p = 0;
244 if (warned < 10 && net_ratelimit()) {
245 warned++;
246 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
247 "tries to set negative timeout\n",
248 current->comm, task_pid_nr(current));
249 }
250 return 0;
251 }
252 *timeo_p = MAX_SCHEDULE_TIMEOUT;
253 if (tv.tv_sec == 0 && tv.tv_usec == 0)
254 return 0;
255 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
256 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
257 return 0;
258}
259
260static void sock_warn_obsolete_bsdism(const char *name)
261{
262 static int warned;
263 static char warncomm[TASK_COMM_LEN];
264 if (strcmp(warncomm, current->comm) && warned < 5) {
265 strcpy(warncomm, current->comm);
266 printk(KERN_WARNING "process `%s' is using obsolete "
267 "%s SO_BSDCOMPAT\n", warncomm, name);
268 warned++;
269 }
270}
271
272static void sock_disable_timestamp(struct sock *sk, int flag)
273{
274 if (sock_flag(sk, flag)) {
275 sock_reset_flag(sk, flag);
276 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
277 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
278 net_disable_timestamp();
279 }
280 }
281}
282
283
284int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
285{
286 int err;
287 int skb_len;
288 unsigned long flags;
289 struct sk_buff_head *list = &sk->sk_receive_queue;
290
291 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
292 number of warnings when compiling with -W --ANK
293 */
294 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
295 (unsigned)sk->sk_rcvbuf) {
296 atomic_inc(&sk->sk_drops);
297 trace_sock_rcvqueue_full(sk, skb);
298 return -ENOMEM;
299 }
300
301 err = sk_filter(sk, skb);
302 if (err)
303 return err;
304
305 if (!sk_rmem_schedule(sk, skb->truesize)) {
306 atomic_inc(&sk->sk_drops);
307 return -ENOBUFS;
308 }
309
310 skb->dev = NULL;
311 skb_set_owner_r(skb, sk);
312
313 /* Cache the SKB length before we tack it onto the receive
314 * queue. Once it is added it no longer belongs to us and
315 * may be freed by other threads of control pulling packets
316 * from the queue.
317 */
318 skb_len = skb->len;
319
320 /* we escape from rcu protected region, make sure we dont leak
321 * a norefcounted dst
322 */
323 skb_dst_force(skb);
324
325 spin_lock_irqsave(&list->lock, flags);
326 skb->dropcount = atomic_read(&sk->sk_drops);
327 __skb_queue_tail(list, skb);
328 spin_unlock_irqrestore(&list->lock, flags);
329
330 if (!sock_flag(sk, SOCK_DEAD))
331 sk->sk_data_ready(sk, skb_len);
332 return 0;
333}
334EXPORT_SYMBOL(sock_queue_rcv_skb);
335
336int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
337{
338 int rc = NET_RX_SUCCESS;
339
340 if (sk_filter(sk, skb))
341 goto discard_and_relse;
342
343 skb->dev = NULL;
344
345 if (sk_rcvqueues_full(sk, skb)) {
346 atomic_inc(&sk->sk_drops);
347 goto discard_and_relse;
348 }
349 if (nested)
350 bh_lock_sock_nested(sk);
351 else
352 bh_lock_sock(sk);
353 if (!sock_owned_by_user(sk)) {
354 /*
355 * trylock + unlock semantics:
356 */
357 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
358
359 rc = sk_backlog_rcv(sk, skb);
360
361 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
362 } else if (sk_add_backlog(sk, skb)) {
363 bh_unlock_sock(sk);
364 atomic_inc(&sk->sk_drops);
365 goto discard_and_relse;
366 }
367
368 bh_unlock_sock(sk);
369out:
370 sock_put(sk);
371 return rc;
372discard_and_relse:
373 kfree_skb(skb);
374 goto out;
375}
376EXPORT_SYMBOL(sk_receive_skb);
377
378void sk_reset_txq(struct sock *sk)
379{
380 sk_tx_queue_clear(sk);
381}
382EXPORT_SYMBOL(sk_reset_txq);
383
384struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
385{
386 struct dst_entry *dst = __sk_dst_get(sk);
387
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
389 sk_tx_queue_clear(sk);
390 rcu_assign_pointer(sk->sk_dst_cache, NULL);
391 dst_release(dst);
392 return NULL;
393 }
394
395 return dst;
396}
397EXPORT_SYMBOL(__sk_dst_check);
398
399struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
400{
401 struct dst_entry *dst = sk_dst_get(sk);
402
403 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
404 sk_dst_reset(sk);
405 dst_release(dst);
406 return NULL;
407 }
408
409 return dst;
410}
411EXPORT_SYMBOL(sk_dst_check);
412
413static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
414{
415 int ret = -ENOPROTOOPT;
416#ifdef CONFIG_NETDEVICES
417 struct net *net = sock_net(sk);
418 char devname[IFNAMSIZ];
419 int index;
420
421 /* Sorry... */
422 ret = -EPERM;
423 if (!capable(CAP_NET_RAW))
424 goto out;
425
426 ret = -EINVAL;
427 if (optlen < 0)
428 goto out;
429
430 /* Bind this socket to a particular device like "eth0",
431 * as specified in the passed interface name. If the
432 * name is "" or the option length is zero the socket
433 * is not bound.
434 */
435 if (optlen > IFNAMSIZ - 1)
436 optlen = IFNAMSIZ - 1;
437 memset(devname, 0, sizeof(devname));
438
439 ret = -EFAULT;
440 if (copy_from_user(devname, optval, optlen))
441 goto out;
442
443 index = 0;
444 if (devname[0] != '\0') {
445 struct net_device *dev;
446
447 rcu_read_lock();
448 dev = dev_get_by_name_rcu(net, devname);
449 if (dev)
450 index = dev->ifindex;
451 rcu_read_unlock();
452 ret = -ENODEV;
453 if (!dev)
454 goto out;
455 }
456
457 lock_sock(sk);
458 sk->sk_bound_dev_if = index;
459 sk_dst_reset(sk);
460 release_sock(sk);
461
462 ret = 0;
463
464out:
465#endif
466
467 return ret;
468}
469
470static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
471{
472 if (valbool)
473 sock_set_flag(sk, bit);
474 else
475 sock_reset_flag(sk, bit);
476}
477
478/*
479 * This is meant for all protocols to use and covers goings on
480 * at the socket level. Everything here is generic.
481 */
482
483int sock_setsockopt(struct socket *sock, int level, int optname,
484 char __user *optval, unsigned int optlen)
485{
486 struct sock *sk = sock->sk;
487 int val;
488 int valbool;
489 struct linger ling;
490 int ret = 0;
491
492 /*
493 * Options without arguments
494 */
495
496 if (optname == SO_BINDTODEVICE)
497 return sock_bindtodevice(sk, optval, optlen);
498
499 if (optlen < sizeof(int))
500 return -EINVAL;
501
502 if (get_user(val, (int __user *)optval))
503 return -EFAULT;
504
505 valbool = val ? 1 : 0;
506
507 lock_sock(sk);
508
509 switch (optname) {
510 case SO_DEBUG:
511 if (val && !capable(CAP_NET_ADMIN))
512 ret = -EACCES;
513 else
514 sock_valbool_flag(sk, SOCK_DBG, valbool);
515 break;
516 case SO_REUSEADDR:
517 sk->sk_reuse = valbool;
518 break;
519 case SO_TYPE:
520 case SO_PROTOCOL:
521 case SO_DOMAIN:
522 case SO_ERROR:
523 ret = -ENOPROTOOPT;
524 break;
525 case SO_DONTROUTE:
526 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
527 break;
528 case SO_BROADCAST:
529 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
530 break;
531 case SO_SNDBUF:
532 /* Don't error on this BSD doesn't and if you think
533 about it this is right. Otherwise apps have to
534 play 'guess the biggest size' games. RCVBUF/SNDBUF
535 are treated in BSD as hints */
536
537 if (val > sysctl_wmem_max)
538 val = sysctl_wmem_max;
539set_sndbuf:
540 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
541 if ((val * 2) < SOCK_MIN_SNDBUF)
542 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
543 else
544 sk->sk_sndbuf = val * 2;
545
546 /*
547 * Wake up sending tasks if we
548 * upped the value.
549 */
550 sk->sk_write_space(sk);
551 break;
552
553 case SO_SNDBUFFORCE:
554 if (!capable(CAP_NET_ADMIN)) {
555 ret = -EPERM;
556 break;
557 }
558 goto set_sndbuf;
559
560 case SO_RCVBUF:
561 /* Don't error on this BSD doesn't and if you think
562 about it this is right. Otherwise apps have to
563 play 'guess the biggest size' games. RCVBUF/SNDBUF
564 are treated in BSD as hints */
565
566 if (val > sysctl_rmem_max)
567 val = sysctl_rmem_max;
568set_rcvbuf:
569 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
570 /*
571 * We double it on the way in to account for
572 * "struct sk_buff" etc. overhead. Applications
573 * assume that the SO_RCVBUF setting they make will
574 * allow that much actual data to be received on that
575 * socket.
576 *
577 * Applications are unaware that "struct sk_buff" and
578 * other overheads allocate from the receive buffer
579 * during socket buffer allocation.
580 *
581 * And after considering the possible alternatives,
582 * returning the value we actually used in getsockopt
583 * is the most desirable behavior.
584 */
585 if ((val * 2) < SOCK_MIN_RCVBUF)
586 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
587 else
588 sk->sk_rcvbuf = val * 2;
589 break;
590
591 case SO_RCVBUFFORCE:
592 if (!capable(CAP_NET_ADMIN)) {
593 ret = -EPERM;
594 break;
595 }
596 goto set_rcvbuf;
597
598 case SO_KEEPALIVE:
599#ifdef CONFIG_INET
600 if (sk->sk_protocol == IPPROTO_TCP)
601 tcp_set_keepalive(sk, valbool);
602#endif
603 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
604 break;
605
606 case SO_OOBINLINE:
607 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
608 break;
609
610 case SO_NO_CHECK:
611 sk->sk_no_check = valbool;
612 break;
613
614 case SO_PRIORITY:
615 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
616 sk->sk_priority = val;
617 else
618 ret = -EPERM;
619 break;
620
621 case SO_LINGER:
622 if (optlen < sizeof(ling)) {
623 ret = -EINVAL; /* 1003.1g */
624 break;
625 }
626 if (copy_from_user(&ling, optval, sizeof(ling))) {
627 ret = -EFAULT;
628 break;
629 }
630 if (!ling.l_onoff)
631 sock_reset_flag(sk, SOCK_LINGER);
632 else {
633#if (BITS_PER_LONG == 32)
634 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
635 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
636 else
637#endif
638 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
639 sock_set_flag(sk, SOCK_LINGER);
640 }
641 break;
642
643 case SO_BSDCOMPAT:
644 sock_warn_obsolete_bsdism("setsockopt");
645 break;
646
647 case SO_PASSCRED:
648 if (valbool)
649 set_bit(SOCK_PASSCRED, &sock->flags);
650 else
651 clear_bit(SOCK_PASSCRED, &sock->flags);
652 break;
653
654 case SO_TIMESTAMP:
655 case SO_TIMESTAMPNS:
656 if (valbool) {
657 if (optname == SO_TIMESTAMP)
658 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
659 else
660 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
661 sock_set_flag(sk, SOCK_RCVTSTAMP);
662 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
663 } else {
664 sock_reset_flag(sk, SOCK_RCVTSTAMP);
665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
666 }
667 break;
668
669 case SO_TIMESTAMPING:
670 if (val & ~SOF_TIMESTAMPING_MASK) {
671 ret = -EINVAL;
672 break;
673 }
674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
675 val & SOF_TIMESTAMPING_TX_HARDWARE);
676 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
677 val & SOF_TIMESTAMPING_TX_SOFTWARE);
678 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
679 val & SOF_TIMESTAMPING_RX_HARDWARE);
680 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
681 sock_enable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 else
684 sock_disable_timestamp(sk,
685 SOCK_TIMESTAMPING_RX_SOFTWARE);
686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
687 val & SOF_TIMESTAMPING_SOFTWARE);
688 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
689 val & SOF_TIMESTAMPING_SYS_HARDWARE);
690 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
691 val & SOF_TIMESTAMPING_RAW_HARDWARE);
692 break;
693
694 case SO_RCVLOWAT:
695 if (val < 0)
696 val = INT_MAX;
697 sk->sk_rcvlowat = val ? : 1;
698 break;
699
700 case SO_RCVTIMEO:
701 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
702 break;
703
704 case SO_SNDTIMEO:
705 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
706 break;
707
708 case SO_ATTACH_FILTER:
709 ret = -EINVAL;
710 if (optlen == sizeof(struct sock_fprog)) {
711 struct sock_fprog fprog;
712
713 ret = -EFAULT;
714 if (copy_from_user(&fprog, optval, sizeof(fprog)))
715 break;
716
717 ret = sk_attach_filter(&fprog, sk);
718 }
719 break;
720
721 case SO_DETACH_FILTER:
722 ret = sk_detach_filter(sk);
723 break;
724
725 case SO_PASSSEC:
726 if (valbool)
727 set_bit(SOCK_PASSSEC, &sock->flags);
728 else
729 clear_bit(SOCK_PASSSEC, &sock->flags);
730 break;
731 case SO_MARK:
732 if (!capable(CAP_NET_ADMIN))
733 ret = -EPERM;
734 else
735 sk->sk_mark = val;
736 break;
737
738 /* We implement the SO_SNDLOWAT etc to
739 not be settable (1003.1g 5.3) */
740 case SO_RXQ_OVFL:
741 if (valbool)
742 sock_set_flag(sk, SOCK_RXQ_OVFL);
743 else
744 sock_reset_flag(sk, SOCK_RXQ_OVFL);
745 break;
746 default:
747 ret = -ENOPROTOOPT;
748 break;
749 }
750 release_sock(sk);
751 return ret;
752}
753EXPORT_SYMBOL(sock_setsockopt);
754
755
756void cred_to_ucred(struct pid *pid, const struct cred *cred,
757 struct ucred *ucred)
758{
759 ucred->pid = pid_vnr(pid);
760 ucred->uid = ucred->gid = -1;
761 if (cred) {
762 struct user_namespace *current_ns = current_user_ns();
763
764 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
765 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
766 }
767}
768EXPORT_SYMBOL_GPL(cred_to_ucred);
769
770int sock_getsockopt(struct socket *sock, int level, int optname,
771 char __user *optval, int __user *optlen)
772{
773 struct sock *sk = sock->sk;
774
775 union {
776 int val;
777 struct linger ling;
778 struct timeval tm;
779 } v;
780
781 int lv = sizeof(int);
782 int len;
783
784 if (get_user(len, optlen))
785 return -EFAULT;
786 if (len < 0)
787 return -EINVAL;
788
789 memset(&v, 0, sizeof(v));
790
791 switch (optname) {
792 case SO_DEBUG:
793 v.val = sock_flag(sk, SOCK_DBG);
794 break;
795
796 case SO_DONTROUTE:
797 v.val = sock_flag(sk, SOCK_LOCALROUTE);
798 break;
799
800 case SO_BROADCAST:
801 v.val = !!sock_flag(sk, SOCK_BROADCAST);
802 break;
803
804 case SO_SNDBUF:
805 v.val = sk->sk_sndbuf;
806 break;
807
808 case SO_RCVBUF:
809 v.val = sk->sk_rcvbuf;
810 break;
811
812 case SO_REUSEADDR:
813 v.val = sk->sk_reuse;
814 break;
815
816 case SO_KEEPALIVE:
817 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
818 break;
819
820 case SO_TYPE:
821 v.val = sk->sk_type;
822 break;
823
824 case SO_PROTOCOL:
825 v.val = sk->sk_protocol;
826 break;
827
828 case SO_DOMAIN:
829 v.val = sk->sk_family;
830 break;
831
832 case SO_ERROR:
833 v.val = -sock_error(sk);
834 if (v.val == 0)
835 v.val = xchg(&sk->sk_err_soft, 0);
836 break;
837
838 case SO_OOBINLINE:
839 v.val = !!sock_flag(sk, SOCK_URGINLINE);
840 break;
841
842 case SO_NO_CHECK:
843 v.val = sk->sk_no_check;
844 break;
845
846 case SO_PRIORITY:
847 v.val = sk->sk_priority;
848 break;
849
850 case SO_LINGER:
851 lv = sizeof(v.ling);
852 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
853 v.ling.l_linger = sk->sk_lingertime / HZ;
854 break;
855
856 case SO_BSDCOMPAT:
857 sock_warn_obsolete_bsdism("getsockopt");
858 break;
859
860 case SO_TIMESTAMP:
861 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
862 !sock_flag(sk, SOCK_RCVTSTAMPNS);
863 break;
864
865 case SO_TIMESTAMPNS:
866 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
867 break;
868
869 case SO_TIMESTAMPING:
870 v.val = 0;
871 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
872 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
873 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
874 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
875 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
876 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
877 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
878 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
879 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
880 v.val |= SOF_TIMESTAMPING_SOFTWARE;
881 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
882 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
883 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
884 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
885 break;
886
887 case SO_RCVTIMEO:
888 lv = sizeof(struct timeval);
889 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
890 v.tm.tv_sec = 0;
891 v.tm.tv_usec = 0;
892 } else {
893 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
894 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
895 }
896 break;
897
898 case SO_SNDTIMEO:
899 lv = sizeof(struct timeval);
900 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
901 v.tm.tv_sec = 0;
902 v.tm.tv_usec = 0;
903 } else {
904 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
905 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
906 }
907 break;
908
909 case SO_RCVLOWAT:
910 v.val = sk->sk_rcvlowat;
911 break;
912
913 case SO_SNDLOWAT:
914 v.val = 1;
915 break;
916
917 case SO_PASSCRED:
918 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
919 break;
920
921 case SO_PEERCRED:
922 {
923 struct ucred peercred;
924 if (len > sizeof(peercred))
925 len = sizeof(peercred);
926 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
927 if (copy_to_user(optval, &peercred, len))
928 return -EFAULT;
929 goto lenout;
930 }
931
932 case SO_PEERNAME:
933 {
934 char address[128];
935
936 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
937 return -ENOTCONN;
938 if (lv < len)
939 return -EINVAL;
940 if (copy_to_user(optval, address, len))
941 return -EFAULT;
942 goto lenout;
943 }
944
945 /* Dubious BSD thing... Probably nobody even uses it, but
946 * the UNIX standard wants it for whatever reason... -DaveM
947 */
948 case SO_ACCEPTCONN:
949 v.val = sk->sk_state == TCP_LISTEN;
950 break;
951
952 case SO_PASSSEC:
953 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
954 break;
955
956 case SO_PEERSEC:
957 return security_socket_getpeersec_stream(sock, optval, optlen, len);
958
959 case SO_MARK:
960 v.val = sk->sk_mark;
961 break;
962
963 case SO_RXQ_OVFL:
964 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
965 break;
966
967 default:
968 return -ENOPROTOOPT;
969 }
970
971 if (len > lv)
972 len = lv;
973 if (copy_to_user(optval, &v, len))
974 return -EFAULT;
975lenout:
976 if (put_user(len, optlen))
977 return -EFAULT;
978 return 0;
979}
980
981/*
982 * Initialize an sk_lock.
983 *
984 * (We also register the sk_lock with the lock validator.)
985 */
986static inline void sock_lock_init(struct sock *sk)
987{
988 sock_lock_init_class_and_name(sk,
989 af_family_slock_key_strings[sk->sk_family],
990 af_family_slock_keys + sk->sk_family,
991 af_family_key_strings[sk->sk_family],
992 af_family_keys + sk->sk_family);
993}
994
995/*
996 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
997 * even temporarly, because of RCU lookups. sk_node should also be left as is.
998 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
999 */
1000static void sock_copy(struct sock *nsk, const struct sock *osk)
1001{
1002#ifdef CONFIG_SECURITY_NETWORK
1003 void *sptr = nsk->sk_security;
1004#endif
1005 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1006
1007 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1008 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1009
1010#ifdef CONFIG_SECURITY_NETWORK
1011 nsk->sk_security = sptr;
1012 security_sk_clone(osk, nsk);
1013#endif
1014}
1015
1016/*
1017 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1018 * un-modified. Special care is taken when initializing object to zero.
1019 */
1020static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1021{
1022 if (offsetof(struct sock, sk_node.next) != 0)
1023 memset(sk, 0, offsetof(struct sock, sk_node.next));
1024 memset(&sk->sk_node.pprev, 0,
1025 size - offsetof(struct sock, sk_node.pprev));
1026}
1027
1028void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1029{
1030 unsigned long nulls1, nulls2;
1031
1032 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1033 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1034 if (nulls1 > nulls2)
1035 swap(nulls1, nulls2);
1036
1037 if (nulls1 != 0)
1038 memset((char *)sk, 0, nulls1);
1039 memset((char *)sk + nulls1 + sizeof(void *), 0,
1040 nulls2 - nulls1 - sizeof(void *));
1041 memset((char *)sk + nulls2 + sizeof(void *), 0,
1042 size - nulls2 - sizeof(void *));
1043}
1044EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1045
1046static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1047 int family)
1048{
1049 struct sock *sk;
1050 struct kmem_cache *slab;
1051
1052 slab = prot->slab;
1053 if (slab != NULL) {
1054 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1055 if (!sk)
1056 return sk;
1057 if (priority & __GFP_ZERO) {
1058 if (prot->clear_sk)
1059 prot->clear_sk(sk, prot->obj_size);
1060 else
1061 sk_prot_clear_nulls(sk, prot->obj_size);
1062 }
1063 } else
1064 sk = kmalloc(prot->obj_size, priority);
1065
1066 if (sk != NULL) {
1067 kmemcheck_annotate_bitfield(sk, flags);
1068
1069 if (security_sk_alloc(sk, family, priority))
1070 goto out_free;
1071
1072 if (!try_module_get(prot->owner))
1073 goto out_free_sec;
1074 sk_tx_queue_clear(sk);
1075 }
1076
1077 return sk;
1078
1079out_free_sec:
1080 security_sk_free(sk);
1081out_free:
1082 if (slab != NULL)
1083 kmem_cache_free(slab, sk);
1084 else
1085 kfree(sk);
1086 return NULL;
1087}
1088
1089static void sk_prot_free(struct proto *prot, struct sock *sk)
1090{
1091 struct kmem_cache *slab;
1092 struct module *owner;
1093
1094 owner = prot->owner;
1095 slab = prot->slab;
1096
1097 security_sk_free(sk);
1098 if (slab != NULL)
1099 kmem_cache_free(slab, sk);
1100 else
1101 kfree(sk);
1102 module_put(owner);
1103}
1104
1105#ifdef CONFIG_CGROUPS
1106void sock_update_classid(struct sock *sk)
1107{
1108 u32 classid;
1109
1110 rcu_read_lock(); /* doing current task, which cannot vanish. */
1111 classid = task_cls_classid(current);
1112 rcu_read_unlock();
1113 if (classid && classid != sk->sk_classid)
1114 sk->sk_classid = classid;
1115}
1116EXPORT_SYMBOL(sock_update_classid);
1117#endif
1118
1119/**
1120 * sk_alloc - All socket objects are allocated here
1121 * @net: the applicable net namespace
1122 * @family: protocol family
1123 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1124 * @prot: struct proto associated with this new sock instance
1125 */
1126struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1127 struct proto *prot)
1128{
1129 struct sock *sk;
1130
1131 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1132 if (sk) {
1133 sk->sk_family = family;
1134 /*
1135 * See comment in struct sock definition to understand
1136 * why we need sk_prot_creator -acme
1137 */
1138 sk->sk_prot = sk->sk_prot_creator = prot;
1139 sock_lock_init(sk);
1140 sock_net_set(sk, get_net(net));
1141 atomic_set(&sk->sk_wmem_alloc, 1);
1142
1143 sock_update_classid(sk);
1144 }
1145
1146 return sk;
1147}
1148EXPORT_SYMBOL(sk_alloc);
1149
1150static void __sk_free(struct sock *sk)
1151{
1152 struct sk_filter *filter;
1153
1154 if (sk->sk_destruct)
1155 sk->sk_destruct(sk);
1156
1157 filter = rcu_dereference_check(sk->sk_filter,
1158 atomic_read(&sk->sk_wmem_alloc) == 0);
1159 if (filter) {
1160 sk_filter_uncharge(sk, filter);
1161 rcu_assign_pointer(sk->sk_filter, NULL);
1162 }
1163
1164 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1165 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1166
1167 if (atomic_read(&sk->sk_omem_alloc))
1168 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1169 __func__, atomic_read(&sk->sk_omem_alloc));
1170
1171 if (sk->sk_peer_cred)
1172 put_cred(sk->sk_peer_cred);
1173 put_pid(sk->sk_peer_pid);
1174 put_net(sock_net(sk));
1175 sk_prot_free(sk->sk_prot_creator, sk);
1176}
1177
1178void sk_free(struct sock *sk)
1179{
1180 /*
1181 * We subtract one from sk_wmem_alloc and can know if
1182 * some packets are still in some tx queue.
1183 * If not null, sock_wfree() will call __sk_free(sk) later
1184 */
1185 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1186 __sk_free(sk);
1187}
1188EXPORT_SYMBOL(sk_free);
1189
1190/*
1191 * Last sock_put should drop reference to sk->sk_net. It has already
1192 * been dropped in sk_change_net. Taking reference to stopping namespace
1193 * is not an option.
1194 * Take reference to a socket to remove it from hash _alive_ and after that
1195 * destroy it in the context of init_net.
1196 */
1197void sk_release_kernel(struct sock *sk)
1198{
1199 if (sk == NULL || sk->sk_socket == NULL)
1200 return;
1201
1202 sock_hold(sk);
1203 sock_release(sk->sk_socket);
1204 release_net(sock_net(sk));
1205 sock_net_set(sk, get_net(&init_net));
1206 sock_put(sk);
1207}
1208EXPORT_SYMBOL(sk_release_kernel);
1209
1210struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1211{
1212 struct sock *newsk;
1213
1214 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1215 if (newsk != NULL) {
1216 struct sk_filter *filter;
1217
1218 sock_copy(newsk, sk);
1219
1220 /* SANITY */
1221 get_net(sock_net(newsk));
1222 sk_node_init(&newsk->sk_node);
1223 sock_lock_init(newsk);
1224 bh_lock_sock(newsk);
1225 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1226 newsk->sk_backlog.len = 0;
1227
1228 atomic_set(&newsk->sk_rmem_alloc, 0);
1229 /*
1230 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1231 */
1232 atomic_set(&newsk->sk_wmem_alloc, 1);
1233 atomic_set(&newsk->sk_omem_alloc, 0);
1234 skb_queue_head_init(&newsk->sk_receive_queue);
1235 skb_queue_head_init(&newsk->sk_write_queue);
1236#ifdef CONFIG_NET_DMA
1237 skb_queue_head_init(&newsk->sk_async_wait_queue);
1238#endif
1239
1240 spin_lock_init(&newsk->sk_dst_lock);
1241 rwlock_init(&newsk->sk_callback_lock);
1242 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1243 af_callback_keys + newsk->sk_family,
1244 af_family_clock_key_strings[newsk->sk_family]);
1245
1246 newsk->sk_dst_cache = NULL;
1247 newsk->sk_wmem_queued = 0;
1248 newsk->sk_forward_alloc = 0;
1249 newsk->sk_send_head = NULL;
1250 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1251
1252 sock_reset_flag(newsk, SOCK_DONE);
1253 skb_queue_head_init(&newsk->sk_error_queue);
1254
1255 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1256 if (filter != NULL)
1257 sk_filter_charge(newsk, filter);
1258
1259 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1260 /* It is still raw copy of parent, so invalidate
1261 * destructor and make plain sk_free() */
1262 newsk->sk_destruct = NULL;
1263 sk_free(newsk);
1264 newsk = NULL;
1265 goto out;
1266 }
1267
1268 newsk->sk_err = 0;
1269 newsk->sk_priority = 0;
1270 /*
1271 * Before updating sk_refcnt, we must commit prior changes to memory
1272 * (Documentation/RCU/rculist_nulls.txt for details)
1273 */
1274 smp_wmb();
1275 atomic_set(&newsk->sk_refcnt, 2);
1276
1277 /*
1278 * Increment the counter in the same struct proto as the master
1279 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1280 * is the same as sk->sk_prot->socks, as this field was copied
1281 * with memcpy).
1282 *
1283 * This _changes_ the previous behaviour, where
1284 * tcp_create_openreq_child always was incrementing the
1285 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1286 * to be taken into account in all callers. -acme
1287 */
1288 sk_refcnt_debug_inc(newsk);
1289 sk_set_socket(newsk, NULL);
1290 newsk->sk_wq = NULL;
1291
1292 if (newsk->sk_prot->sockets_allocated)
1293 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1294
1295 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1296 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1297 net_enable_timestamp();
1298 }
1299out:
1300 return newsk;
1301}
1302EXPORT_SYMBOL_GPL(sk_clone);
1303
1304void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1305{
1306 __sk_dst_set(sk, dst);
1307 sk->sk_route_caps = dst->dev->features;
1308 if (sk->sk_route_caps & NETIF_F_GSO)
1309 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1310 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1311 if (sk_can_gso(sk)) {
1312 if (dst->header_len) {
1313 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1314 } else {
1315 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1316 sk->sk_gso_max_size = dst->dev->gso_max_size;
1317 }
1318 }
1319}
1320EXPORT_SYMBOL_GPL(sk_setup_caps);
1321
1322void __init sk_init(void)
1323{
1324 if (totalram_pages <= 4096) {
1325 sysctl_wmem_max = 32767;
1326 sysctl_rmem_max = 32767;
1327 sysctl_wmem_default = 32767;
1328 sysctl_rmem_default = 32767;
1329 } else if (totalram_pages >= 131072) {
1330 sysctl_wmem_max = 131071;
1331 sysctl_rmem_max = 131071;
1332 }
1333}
1334
1335/*
1336 * Simple resource managers for sockets.
1337 */
1338
1339
1340/*
1341 * Write buffer destructor automatically called from kfree_skb.
1342 */
1343void sock_wfree(struct sk_buff *skb)
1344{
1345 struct sock *sk = skb->sk;
1346 unsigned int len = skb->truesize;
1347
1348 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1349 /*
1350 * Keep a reference on sk_wmem_alloc, this will be released
1351 * after sk_write_space() call
1352 */
1353 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1354 sk->sk_write_space(sk);
1355 len = 1;
1356 }
1357 /*
1358 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1359 * could not do because of in-flight packets
1360 */
1361 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1362 __sk_free(sk);
1363}
1364EXPORT_SYMBOL(sock_wfree);
1365
1366/*
1367 * Read buffer destructor automatically called from kfree_skb.
1368 */
1369void sock_rfree(struct sk_buff *skb)
1370{
1371 struct sock *sk = skb->sk;
1372 unsigned int len = skb->truesize;
1373
1374 atomic_sub(len, &sk->sk_rmem_alloc);
1375 sk_mem_uncharge(sk, len);
1376}
1377EXPORT_SYMBOL(sock_rfree);
1378
1379
1380int sock_i_uid(struct sock *sk)
1381{
1382 int uid;
1383
1384 read_lock_bh(&sk->sk_callback_lock);
1385 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1386 read_unlock_bh(&sk->sk_callback_lock);
1387 return uid;
1388}
1389EXPORT_SYMBOL(sock_i_uid);
1390
1391unsigned long sock_i_ino(struct sock *sk)
1392{
1393 unsigned long ino;
1394
1395 read_lock_bh(&sk->sk_callback_lock);
1396 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1397 read_unlock_bh(&sk->sk_callback_lock);
1398 return ino;
1399}
1400EXPORT_SYMBOL(sock_i_ino);
1401
1402/*
1403 * Allocate a skb from the socket's send buffer.
1404 */
1405struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1406 gfp_t priority)
1407{
1408 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1409 struct sk_buff *skb = alloc_skb(size, priority);
1410 if (skb) {
1411 skb_set_owner_w(skb, sk);
1412 return skb;
1413 }
1414 }
1415 return NULL;
1416}
1417EXPORT_SYMBOL(sock_wmalloc);
1418
1419/*
1420 * Allocate a skb from the socket's receive buffer.
1421 */
1422struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1423 gfp_t priority)
1424{
1425 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1426 struct sk_buff *skb = alloc_skb(size, priority);
1427 if (skb) {
1428 skb_set_owner_r(skb, sk);
1429 return skb;
1430 }
1431 }
1432 return NULL;
1433}
1434
1435/*
1436 * Allocate a memory block from the socket's option memory buffer.
1437 */
1438void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1439{
1440 if ((unsigned)size <= sysctl_optmem_max &&
1441 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1442 void *mem;
1443 /* First do the add, to avoid the race if kmalloc
1444 * might sleep.
1445 */
1446 atomic_add(size, &sk->sk_omem_alloc);
1447 mem = kmalloc(size, priority);
1448 if (mem)
1449 return mem;
1450 atomic_sub(size, &sk->sk_omem_alloc);
1451 }
1452 return NULL;
1453}
1454EXPORT_SYMBOL(sock_kmalloc);
1455
1456/*
1457 * Free an option memory block.
1458 */
1459void sock_kfree_s(struct sock *sk, void *mem, int size)
1460{
1461 kfree(mem);
1462 atomic_sub(size, &sk->sk_omem_alloc);
1463}
1464EXPORT_SYMBOL(sock_kfree_s);
1465
1466/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1467 I think, these locks should be removed for datagram sockets.
1468 */
1469static long sock_wait_for_wmem(struct sock *sk, long timeo)
1470{
1471 DEFINE_WAIT(wait);
1472
1473 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1474 for (;;) {
1475 if (!timeo)
1476 break;
1477 if (signal_pending(current))
1478 break;
1479 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1480 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1481 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1482 break;
1483 if (sk->sk_shutdown & SEND_SHUTDOWN)
1484 break;
1485 if (sk->sk_err)
1486 break;
1487 timeo = schedule_timeout(timeo);
1488 }
1489 finish_wait(sk_sleep(sk), &wait);
1490 return timeo;
1491}
1492
1493
1494/*
1495 * Generic send/receive buffer handlers
1496 */
1497
1498struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1499 unsigned long data_len, int noblock,
1500 int *errcode)
1501{
1502 struct sk_buff *skb;
1503 gfp_t gfp_mask;
1504 long timeo;
1505 int err;
1506
1507 gfp_mask = sk->sk_allocation;
1508 if (gfp_mask & __GFP_WAIT)
1509 gfp_mask |= __GFP_REPEAT;
1510
1511 timeo = sock_sndtimeo(sk, noblock);
1512 while (1) {
1513 err = sock_error(sk);
1514 if (err != 0)
1515 goto failure;
1516
1517 err = -EPIPE;
1518 if (sk->sk_shutdown & SEND_SHUTDOWN)
1519 goto failure;
1520
1521 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1522 skb = alloc_skb(header_len, gfp_mask);
1523 if (skb) {
1524 int npages;
1525 int i;
1526
1527 /* No pages, we're done... */
1528 if (!data_len)
1529 break;
1530
1531 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1532 skb->truesize += data_len;
1533 skb_shinfo(skb)->nr_frags = npages;
1534 for (i = 0; i < npages; i++) {
1535 struct page *page;
1536 skb_frag_t *frag;
1537
1538 page = alloc_pages(sk->sk_allocation, 0);
1539 if (!page) {
1540 err = -ENOBUFS;
1541 skb_shinfo(skb)->nr_frags = i;
1542 kfree_skb(skb);
1543 goto failure;
1544 }
1545
1546 frag = &skb_shinfo(skb)->frags[i];
1547 frag->page = page;
1548 frag->page_offset = 0;
1549 frag->size = (data_len >= PAGE_SIZE ?
1550 PAGE_SIZE :
1551 data_len);
1552 data_len -= PAGE_SIZE;
1553 }
1554
1555 /* Full success... */
1556 break;
1557 }
1558 err = -ENOBUFS;
1559 goto failure;
1560 }
1561 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1562 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1563 err = -EAGAIN;
1564 if (!timeo)
1565 goto failure;
1566 if (signal_pending(current))
1567 goto interrupted;
1568 timeo = sock_wait_for_wmem(sk, timeo);
1569 }
1570
1571 skb_set_owner_w(skb, sk);
1572 return skb;
1573
1574interrupted:
1575 err = sock_intr_errno(timeo);
1576failure:
1577 *errcode = err;
1578 return NULL;
1579}
1580EXPORT_SYMBOL(sock_alloc_send_pskb);
1581
1582struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1583 int noblock, int *errcode)
1584{
1585 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1586}
1587EXPORT_SYMBOL(sock_alloc_send_skb);
1588
1589static void __lock_sock(struct sock *sk)
1590 __releases(&sk->sk_lock.slock)
1591 __acquires(&sk->sk_lock.slock)
1592{
1593 DEFINE_WAIT(wait);
1594
1595 for (;;) {
1596 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1597 TASK_UNINTERRUPTIBLE);
1598 spin_unlock_bh(&sk->sk_lock.slock);
1599 schedule();
1600 spin_lock_bh(&sk->sk_lock.slock);
1601 if (!sock_owned_by_user(sk))
1602 break;
1603 }
1604 finish_wait(&sk->sk_lock.wq, &wait);
1605}
1606
1607static void __release_sock(struct sock *sk)
1608 __releases(&sk->sk_lock.slock)
1609 __acquires(&sk->sk_lock.slock)
1610{
1611 struct sk_buff *skb = sk->sk_backlog.head;
1612
1613 do {
1614 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1615 bh_unlock_sock(sk);
1616
1617 do {
1618 struct sk_buff *next = skb->next;
1619
1620 WARN_ON_ONCE(skb_dst_is_noref(skb));
1621 skb->next = NULL;
1622 sk_backlog_rcv(sk, skb);
1623
1624 /*
1625 * We are in process context here with softirqs
1626 * disabled, use cond_resched_softirq() to preempt.
1627 * This is safe to do because we've taken the backlog
1628 * queue private:
1629 */
1630 cond_resched_softirq();
1631
1632 skb = next;
1633 } while (skb != NULL);
1634
1635 bh_lock_sock(sk);
1636 } while ((skb = sk->sk_backlog.head) != NULL);
1637
1638 /*
1639 * Doing the zeroing here guarantee we can not loop forever
1640 * while a wild producer attempts to flood us.
1641 */
1642 sk->sk_backlog.len = 0;
1643}
1644
1645/**
1646 * sk_wait_data - wait for data to arrive at sk_receive_queue
1647 * @sk: sock to wait on
1648 * @timeo: for how long
1649 *
1650 * Now socket state including sk->sk_err is changed only under lock,
1651 * hence we may omit checks after joining wait queue.
1652 * We check receive queue before schedule() only as optimization;
1653 * it is very likely that release_sock() added new data.
1654 */
1655int sk_wait_data(struct sock *sk, long *timeo)
1656{
1657 int rc;
1658 DEFINE_WAIT(wait);
1659
1660 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1661 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1662 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1663 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1664 finish_wait(sk_sleep(sk), &wait);
1665 return rc;
1666}
1667EXPORT_SYMBOL(sk_wait_data);
1668
1669/**
1670 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1671 * @sk: socket
1672 * @size: memory size to allocate
1673 * @kind: allocation type
1674 *
1675 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1676 * rmem allocation. This function assumes that protocols which have
1677 * memory_pressure use sk_wmem_queued as write buffer accounting.
1678 */
1679int __sk_mem_schedule(struct sock *sk, int size, int kind)
1680{
1681 struct proto *prot = sk->sk_prot;
1682 int amt = sk_mem_pages(size);
1683 long allocated;
1684
1685 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1686 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1687
1688 /* Under limit. */
1689 if (allocated <= prot->sysctl_mem[0]) {
1690 if (prot->memory_pressure && *prot->memory_pressure)
1691 *prot->memory_pressure = 0;
1692 return 1;
1693 }
1694
1695 /* Under pressure. */
1696 if (allocated > prot->sysctl_mem[1])
1697 if (prot->enter_memory_pressure)
1698 prot->enter_memory_pressure(sk);
1699
1700 /* Over hard limit. */
1701 if (allocated > prot->sysctl_mem[2])
1702 goto suppress_allocation;
1703
1704 /* guarantee minimum buffer size under pressure */
1705 if (kind == SK_MEM_RECV) {
1706 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1707 return 1;
1708 } else { /* SK_MEM_SEND */
1709 if (sk->sk_type == SOCK_STREAM) {
1710 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1711 return 1;
1712 } else if (atomic_read(&sk->sk_wmem_alloc) <
1713 prot->sysctl_wmem[0])
1714 return 1;
1715 }
1716
1717 if (prot->memory_pressure) {
1718 int alloc;
1719
1720 if (!*prot->memory_pressure)
1721 return 1;
1722 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1723 if (prot->sysctl_mem[2] > alloc *
1724 sk_mem_pages(sk->sk_wmem_queued +
1725 atomic_read(&sk->sk_rmem_alloc) +
1726 sk->sk_forward_alloc))
1727 return 1;
1728 }
1729
1730suppress_allocation:
1731
1732 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1733 sk_stream_moderate_sndbuf(sk);
1734
1735 /* Fail only if socket is _under_ its sndbuf.
1736 * In this case we cannot block, so that we have to fail.
1737 */
1738 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1739 return 1;
1740 }
1741
1742 trace_sock_exceed_buf_limit(sk, prot, allocated);
1743
1744 /* Alas. Undo changes. */
1745 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1746 atomic_long_sub(amt, prot->memory_allocated);
1747 return 0;
1748}
1749EXPORT_SYMBOL(__sk_mem_schedule);
1750
1751/**
1752 * __sk_reclaim - reclaim memory_allocated
1753 * @sk: socket
1754 */
1755void __sk_mem_reclaim(struct sock *sk)
1756{
1757 struct proto *prot = sk->sk_prot;
1758
1759 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1760 prot->memory_allocated);
1761 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1762
1763 if (prot->memory_pressure && *prot->memory_pressure &&
1764 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1765 *prot->memory_pressure = 0;
1766}
1767EXPORT_SYMBOL(__sk_mem_reclaim);
1768
1769
1770/*
1771 * Set of default routines for initialising struct proto_ops when
1772 * the protocol does not support a particular function. In certain
1773 * cases where it makes no sense for a protocol to have a "do nothing"
1774 * function, some default processing is provided.
1775 */
1776
1777int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1778{
1779 return -EOPNOTSUPP;
1780}
1781EXPORT_SYMBOL(sock_no_bind);
1782
1783int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1784 int len, int flags)
1785{
1786 return -EOPNOTSUPP;
1787}
1788EXPORT_SYMBOL(sock_no_connect);
1789
1790int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1791{
1792 return -EOPNOTSUPP;
1793}
1794EXPORT_SYMBOL(sock_no_socketpair);
1795
1796int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1797{
1798 return -EOPNOTSUPP;
1799}
1800EXPORT_SYMBOL(sock_no_accept);
1801
1802int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1803 int *len, int peer)
1804{
1805 return -EOPNOTSUPP;
1806}
1807EXPORT_SYMBOL(sock_no_getname);
1808
1809unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1810{
1811 return 0;
1812}
1813EXPORT_SYMBOL(sock_no_poll);
1814
1815int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1816{
1817 return -EOPNOTSUPP;
1818}
1819EXPORT_SYMBOL(sock_no_ioctl);
1820
1821int sock_no_listen(struct socket *sock, int backlog)
1822{
1823 return -EOPNOTSUPP;
1824}
1825EXPORT_SYMBOL(sock_no_listen);
1826
1827int sock_no_shutdown(struct socket *sock, int how)
1828{
1829 return -EOPNOTSUPP;
1830}
1831EXPORT_SYMBOL(sock_no_shutdown);
1832
1833int sock_no_setsockopt(struct socket *sock, int level, int optname,
1834 char __user *optval, unsigned int optlen)
1835{
1836 return -EOPNOTSUPP;
1837}
1838EXPORT_SYMBOL(sock_no_setsockopt);
1839
1840int sock_no_getsockopt(struct socket *sock, int level, int optname,
1841 char __user *optval, int __user *optlen)
1842{
1843 return -EOPNOTSUPP;
1844}
1845EXPORT_SYMBOL(sock_no_getsockopt);
1846
1847int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1848 size_t len)
1849{
1850 return -EOPNOTSUPP;
1851}
1852EXPORT_SYMBOL(sock_no_sendmsg);
1853
1854int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1855 size_t len, int flags)
1856{
1857 return -EOPNOTSUPP;
1858}
1859EXPORT_SYMBOL(sock_no_recvmsg);
1860
1861int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1862{
1863 /* Mirror missing mmap method error code */
1864 return -ENODEV;
1865}
1866EXPORT_SYMBOL(sock_no_mmap);
1867
1868ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1869{
1870 ssize_t res;
1871 struct msghdr msg = {.msg_flags = flags};
1872 struct kvec iov;
1873 char *kaddr = kmap(page);
1874 iov.iov_base = kaddr + offset;
1875 iov.iov_len = size;
1876 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1877 kunmap(page);
1878 return res;
1879}
1880EXPORT_SYMBOL(sock_no_sendpage);
1881
1882/*
1883 * Default Socket Callbacks
1884 */
1885
1886static void sock_def_wakeup(struct sock *sk)
1887{
1888 struct socket_wq *wq;
1889
1890 rcu_read_lock();
1891 wq = rcu_dereference(sk->sk_wq);
1892 if (wq_has_sleeper(wq))
1893 wake_up_interruptible_all(&wq->wait);
1894 rcu_read_unlock();
1895}
1896
1897static void sock_def_error_report(struct sock *sk)
1898{
1899 struct socket_wq *wq;
1900
1901 rcu_read_lock();
1902 wq = rcu_dereference(sk->sk_wq);
1903 if (wq_has_sleeper(wq))
1904 wake_up_interruptible_poll(&wq->wait, POLLERR);
1905 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1906 rcu_read_unlock();
1907}
1908
1909static void sock_def_readable(struct sock *sk, int len)
1910{
1911 struct socket_wq *wq;
1912
1913 rcu_read_lock();
1914 wq = rcu_dereference(sk->sk_wq);
1915 if (wq_has_sleeper(wq))
1916 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1917 POLLRDNORM | POLLRDBAND);
1918 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1919 rcu_read_unlock();
1920}
1921
1922static void sock_def_write_space(struct sock *sk)
1923{
1924 struct socket_wq *wq;
1925
1926 rcu_read_lock();
1927
1928 /* Do not wake up a writer until he can make "significant"
1929 * progress. --DaveM
1930 */
1931 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1932 wq = rcu_dereference(sk->sk_wq);
1933 if (wq_has_sleeper(wq))
1934 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1935 POLLWRNORM | POLLWRBAND);
1936
1937 /* Should agree with poll, otherwise some programs break */
1938 if (sock_writeable(sk))
1939 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1940 }
1941
1942 rcu_read_unlock();
1943}
1944
1945static void sock_def_destruct(struct sock *sk)
1946{
1947 kfree(sk->sk_protinfo);
1948}
1949
1950void sk_send_sigurg(struct sock *sk)
1951{
1952 if (sk->sk_socket && sk->sk_socket->file)
1953 if (send_sigurg(&sk->sk_socket->file->f_owner))
1954 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1955}
1956EXPORT_SYMBOL(sk_send_sigurg);
1957
1958void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1959 unsigned long expires)
1960{
1961 if (!mod_timer(timer, expires))
1962 sock_hold(sk);
1963}
1964EXPORT_SYMBOL(sk_reset_timer);
1965
1966void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1967{
1968 if (timer_pending(timer) && del_timer(timer))
1969 __sock_put(sk);
1970}
1971EXPORT_SYMBOL(sk_stop_timer);
1972
1973void sock_init_data(struct socket *sock, struct sock *sk)
1974{
1975 skb_queue_head_init(&sk->sk_receive_queue);
1976 skb_queue_head_init(&sk->sk_write_queue);
1977 skb_queue_head_init(&sk->sk_error_queue);
1978#ifdef CONFIG_NET_DMA
1979 skb_queue_head_init(&sk->sk_async_wait_queue);
1980#endif
1981
1982 sk->sk_send_head = NULL;
1983
1984 init_timer(&sk->sk_timer);
1985
1986 sk->sk_allocation = GFP_KERNEL;
1987 sk->sk_rcvbuf = sysctl_rmem_default;
1988 sk->sk_sndbuf = sysctl_wmem_default;
1989 sk->sk_state = TCP_CLOSE;
1990 sk_set_socket(sk, sock);
1991
1992 sock_set_flag(sk, SOCK_ZAPPED);
1993
1994 if (sock) {
1995 sk->sk_type = sock->type;
1996 sk->sk_wq = sock->wq;
1997 sock->sk = sk;
1998 } else
1999 sk->sk_wq = NULL;
2000
2001 spin_lock_init(&sk->sk_dst_lock);
2002 rwlock_init(&sk->sk_callback_lock);
2003 lockdep_set_class_and_name(&sk->sk_callback_lock,
2004 af_callback_keys + sk->sk_family,
2005 af_family_clock_key_strings[sk->sk_family]);
2006
2007 sk->sk_state_change = sock_def_wakeup;
2008 sk->sk_data_ready = sock_def_readable;
2009 sk->sk_write_space = sock_def_write_space;
2010 sk->sk_error_report = sock_def_error_report;
2011 sk->sk_destruct = sock_def_destruct;
2012
2013 sk->sk_sndmsg_page = NULL;
2014 sk->sk_sndmsg_off = 0;
2015
2016 sk->sk_peer_pid = NULL;
2017 sk->sk_peer_cred = NULL;
2018 sk->sk_write_pending = 0;
2019 sk->sk_rcvlowat = 1;
2020 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2021 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2022
2023 sk->sk_stamp = ktime_set(-1L, 0);
2024
2025 /*
2026 * Before updating sk_refcnt, we must commit prior changes to memory
2027 * (Documentation/RCU/rculist_nulls.txt for details)
2028 */
2029 smp_wmb();
2030 atomic_set(&sk->sk_refcnt, 1);
2031 atomic_set(&sk->sk_drops, 0);
2032}
2033EXPORT_SYMBOL(sock_init_data);
2034
2035void lock_sock_nested(struct sock *sk, int subclass)
2036{
2037 might_sleep();
2038 spin_lock_bh(&sk->sk_lock.slock);
2039 if (sk->sk_lock.owned)
2040 __lock_sock(sk);
2041 sk->sk_lock.owned = 1;
2042 spin_unlock(&sk->sk_lock.slock);
2043 /*
2044 * The sk_lock has mutex_lock() semantics here:
2045 */
2046 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2047 local_bh_enable();
2048}
2049EXPORT_SYMBOL(lock_sock_nested);
2050
2051void release_sock(struct sock *sk)
2052{
2053 /*
2054 * The sk_lock has mutex_unlock() semantics:
2055 */
2056 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2057
2058 spin_lock_bh(&sk->sk_lock.slock);
2059 if (sk->sk_backlog.tail)
2060 __release_sock(sk);
2061 sk->sk_lock.owned = 0;
2062 if (waitqueue_active(&sk->sk_lock.wq))
2063 wake_up(&sk->sk_lock.wq);
2064 spin_unlock_bh(&sk->sk_lock.slock);
2065}
2066EXPORT_SYMBOL(release_sock);
2067
2068/**
2069 * lock_sock_fast - fast version of lock_sock
2070 * @sk: socket
2071 *
2072 * This version should be used for very small section, where process wont block
2073 * return false if fast path is taken
2074 * sk_lock.slock locked, owned = 0, BH disabled
2075 * return true if slow path is taken
2076 * sk_lock.slock unlocked, owned = 1, BH enabled
2077 */
2078bool lock_sock_fast(struct sock *sk)
2079{
2080 might_sleep();
2081 spin_lock_bh(&sk->sk_lock.slock);
2082
2083 if (!sk->sk_lock.owned)
2084 /*
2085 * Note : We must disable BH
2086 */
2087 return false;
2088
2089 __lock_sock(sk);
2090 sk->sk_lock.owned = 1;
2091 spin_unlock(&sk->sk_lock.slock);
2092 /*
2093 * The sk_lock has mutex_lock() semantics here:
2094 */
2095 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2096 local_bh_enable();
2097 return true;
2098}
2099EXPORT_SYMBOL(lock_sock_fast);
2100
2101int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2102{
2103 struct timeval tv;
2104 if (!sock_flag(sk, SOCK_TIMESTAMP))
2105 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2106 tv = ktime_to_timeval(sk->sk_stamp);
2107 if (tv.tv_sec == -1)
2108 return -ENOENT;
2109 if (tv.tv_sec == 0) {
2110 sk->sk_stamp = ktime_get_real();
2111 tv = ktime_to_timeval(sk->sk_stamp);
2112 }
2113 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2114}
2115EXPORT_SYMBOL(sock_get_timestamp);
2116
2117int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2118{
2119 struct timespec ts;
2120 if (!sock_flag(sk, SOCK_TIMESTAMP))
2121 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2122 ts = ktime_to_timespec(sk->sk_stamp);
2123 if (ts.tv_sec == -1)
2124 return -ENOENT;
2125 if (ts.tv_sec == 0) {
2126 sk->sk_stamp = ktime_get_real();
2127 ts = ktime_to_timespec(sk->sk_stamp);
2128 }
2129 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2130}
2131EXPORT_SYMBOL(sock_get_timestampns);
2132
2133void sock_enable_timestamp(struct sock *sk, int flag)
2134{
2135 if (!sock_flag(sk, flag)) {
2136 sock_set_flag(sk, flag);
2137 /*
2138 * we just set one of the two flags which require net
2139 * time stamping, but time stamping might have been on
2140 * already because of the other one
2141 */
2142 if (!sock_flag(sk,
2143 flag == SOCK_TIMESTAMP ?
2144 SOCK_TIMESTAMPING_RX_SOFTWARE :
2145 SOCK_TIMESTAMP))
2146 net_enable_timestamp();
2147 }
2148}
2149
2150/*
2151 * Get a socket option on an socket.
2152 *
2153 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2154 * asynchronous errors should be reported by getsockopt. We assume
2155 * this means if you specify SO_ERROR (otherwise whats the point of it).
2156 */
2157int sock_common_getsockopt(struct socket *sock, int level, int optname,
2158 char __user *optval, int __user *optlen)
2159{
2160 struct sock *sk = sock->sk;
2161
2162 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2163}
2164EXPORT_SYMBOL(sock_common_getsockopt);
2165
2166#ifdef CONFIG_COMPAT
2167int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2168 char __user *optval, int __user *optlen)
2169{
2170 struct sock *sk = sock->sk;
2171
2172 if (sk->sk_prot->compat_getsockopt != NULL)
2173 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2174 optval, optlen);
2175 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2176}
2177EXPORT_SYMBOL(compat_sock_common_getsockopt);
2178#endif
2179
2180int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2181 struct msghdr *msg, size_t size, int flags)
2182{
2183 struct sock *sk = sock->sk;
2184 int addr_len = 0;
2185 int err;
2186
2187 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2188 flags & ~MSG_DONTWAIT, &addr_len);
2189 if (err >= 0)
2190 msg->msg_namelen = addr_len;
2191 return err;
2192}
2193EXPORT_SYMBOL(sock_common_recvmsg);
2194
2195/*
2196 * Set socket options on an inet socket.
2197 */
2198int sock_common_setsockopt(struct socket *sock, int level, int optname,
2199 char __user *optval, unsigned int optlen)
2200{
2201 struct sock *sk = sock->sk;
2202
2203 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2204}
2205EXPORT_SYMBOL(sock_common_setsockopt);
2206
2207#ifdef CONFIG_COMPAT
2208int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2209 char __user *optval, unsigned int optlen)
2210{
2211 struct sock *sk = sock->sk;
2212
2213 if (sk->sk_prot->compat_setsockopt != NULL)
2214 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2215 optval, optlen);
2216 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2217}
2218EXPORT_SYMBOL(compat_sock_common_setsockopt);
2219#endif
2220
2221void sk_common_release(struct sock *sk)
2222{
2223 if (sk->sk_prot->destroy)
2224 sk->sk_prot->destroy(sk);
2225
2226 /*
2227 * Observation: when sock_common_release is called, processes have
2228 * no access to socket. But net still has.
2229 * Step one, detach it from networking:
2230 *
2231 * A. Remove from hash tables.
2232 */
2233
2234 sk->sk_prot->unhash(sk);
2235
2236 /*
2237 * In this point socket cannot receive new packets, but it is possible
2238 * that some packets are in flight because some CPU runs receiver and
2239 * did hash table lookup before we unhashed socket. They will achieve
2240 * receive queue and will be purged by socket destructor.
2241 *
2242 * Also we still have packets pending on receive queue and probably,
2243 * our own packets waiting in device queues. sock_destroy will drain
2244 * receive queue, but transmitted packets will delay socket destruction
2245 * until the last reference will be released.
2246 */
2247
2248 sock_orphan(sk);
2249
2250 xfrm_sk_free_policy(sk);
2251
2252 sk_refcnt_debug_release(sk);
2253 sock_put(sk);
2254}
2255EXPORT_SYMBOL(sk_common_release);
2256
2257static DEFINE_RWLOCK(proto_list_lock);
2258static LIST_HEAD(proto_list);
2259
2260#ifdef CONFIG_PROC_FS
2261#define PROTO_INUSE_NR 64 /* should be enough for the first time */
2262struct prot_inuse {
2263 int val[PROTO_INUSE_NR];
2264};
2265
2266static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2267
2268#ifdef CONFIG_NET_NS
2269void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2270{
2271 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2272}
2273EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2274
2275int sock_prot_inuse_get(struct net *net, struct proto *prot)
2276{
2277 int cpu, idx = prot->inuse_idx;
2278 int res = 0;
2279
2280 for_each_possible_cpu(cpu)
2281 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2282
2283 return res >= 0 ? res : 0;
2284}
2285EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2286
2287static int __net_init sock_inuse_init_net(struct net *net)
2288{
2289 net->core.inuse = alloc_percpu(struct prot_inuse);
2290 return net->core.inuse ? 0 : -ENOMEM;
2291}
2292
2293static void __net_exit sock_inuse_exit_net(struct net *net)
2294{
2295 free_percpu(net->core.inuse);
2296}
2297
2298static struct pernet_operations net_inuse_ops = {
2299 .init = sock_inuse_init_net,
2300 .exit = sock_inuse_exit_net,
2301};
2302
2303static __init int net_inuse_init(void)
2304{
2305 if (register_pernet_subsys(&net_inuse_ops))
2306 panic("Cannot initialize net inuse counters");
2307
2308 return 0;
2309}
2310
2311core_initcall(net_inuse_init);
2312#else
2313static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2314
2315void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2316{
2317 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2318}
2319EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2320
2321int sock_prot_inuse_get(struct net *net, struct proto *prot)
2322{
2323 int cpu, idx = prot->inuse_idx;
2324 int res = 0;
2325
2326 for_each_possible_cpu(cpu)
2327 res += per_cpu(prot_inuse, cpu).val[idx];
2328
2329 return res >= 0 ? res : 0;
2330}
2331EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2332#endif
2333
2334static void assign_proto_idx(struct proto *prot)
2335{
2336 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2337
2338 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2339 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2340 return;
2341 }
2342
2343 set_bit(prot->inuse_idx, proto_inuse_idx);
2344}
2345
2346static void release_proto_idx(struct proto *prot)
2347{
2348 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2349 clear_bit(prot->inuse_idx, proto_inuse_idx);
2350}
2351#else
2352static inline void assign_proto_idx(struct proto *prot)
2353{
2354}
2355
2356static inline void release_proto_idx(struct proto *prot)
2357{
2358}
2359#endif
2360
2361int proto_register(struct proto *prot, int alloc_slab)
2362{
2363 if (alloc_slab) {
2364 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2365 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2366 NULL);
2367
2368 if (prot->slab == NULL) {
2369 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2370 prot->name);
2371 goto out;
2372 }
2373
2374 if (prot->rsk_prot != NULL) {
2375 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2376 if (prot->rsk_prot->slab_name == NULL)
2377 goto out_free_sock_slab;
2378
2379 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2380 prot->rsk_prot->obj_size, 0,
2381 SLAB_HWCACHE_ALIGN, NULL);
2382
2383 if (prot->rsk_prot->slab == NULL) {
2384 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2385 prot->name);
2386 goto out_free_request_sock_slab_name;
2387 }
2388 }
2389
2390 if (prot->twsk_prot != NULL) {
2391 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2392
2393 if (prot->twsk_prot->twsk_slab_name == NULL)
2394 goto out_free_request_sock_slab;
2395
2396 prot->twsk_prot->twsk_slab =
2397 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2398 prot->twsk_prot->twsk_obj_size,
2399 0,
2400 SLAB_HWCACHE_ALIGN |
2401 prot->slab_flags,
2402 NULL);
2403 if (prot->twsk_prot->twsk_slab == NULL)
2404 goto out_free_timewait_sock_slab_name;
2405 }
2406 }
2407
2408 write_lock(&proto_list_lock);
2409 list_add(&prot->node, &proto_list);
2410 assign_proto_idx(prot);
2411 write_unlock(&proto_list_lock);
2412 return 0;
2413
2414out_free_timewait_sock_slab_name:
2415 kfree(prot->twsk_prot->twsk_slab_name);
2416out_free_request_sock_slab:
2417 if (prot->rsk_prot && prot->rsk_prot->slab) {
2418 kmem_cache_destroy(prot->rsk_prot->slab);
2419 prot->rsk_prot->slab = NULL;
2420 }
2421out_free_request_sock_slab_name:
2422 if (prot->rsk_prot)
2423 kfree(prot->rsk_prot->slab_name);
2424out_free_sock_slab:
2425 kmem_cache_destroy(prot->slab);
2426 prot->slab = NULL;
2427out:
2428 return -ENOBUFS;
2429}
2430EXPORT_SYMBOL(proto_register);
2431
2432void proto_unregister(struct proto *prot)
2433{
2434 write_lock(&proto_list_lock);
2435 release_proto_idx(prot);
2436 list_del(&prot->node);
2437 write_unlock(&proto_list_lock);
2438
2439 if (prot->slab != NULL) {
2440 kmem_cache_destroy(prot->slab);
2441 prot->slab = NULL;
2442 }
2443
2444 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2445 kmem_cache_destroy(prot->rsk_prot->slab);
2446 kfree(prot->rsk_prot->slab_name);
2447 prot->rsk_prot->slab = NULL;
2448 }
2449
2450 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2451 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2452 kfree(prot->twsk_prot->twsk_slab_name);
2453 prot->twsk_prot->twsk_slab = NULL;
2454 }
2455}
2456EXPORT_SYMBOL(proto_unregister);
2457
2458#ifdef CONFIG_PROC_FS
2459static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2460 __acquires(proto_list_lock)
2461{
2462 read_lock(&proto_list_lock);
2463 return seq_list_start_head(&proto_list, *pos);
2464}
2465
2466static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2467{
2468 return seq_list_next(v, &proto_list, pos);
2469}
2470
2471static void proto_seq_stop(struct seq_file *seq, void *v)
2472 __releases(proto_list_lock)
2473{
2474 read_unlock(&proto_list_lock);
2475}
2476
2477static char proto_method_implemented(const void *method)
2478{
2479 return method == NULL ? 'n' : 'y';
2480}
2481
2482static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2483{
2484 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2485 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2486 proto->name,
2487 proto->obj_size,
2488 sock_prot_inuse_get(seq_file_net(seq), proto),
2489 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2490 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2491 proto->max_header,
2492 proto->slab == NULL ? "no" : "yes",
2493 module_name(proto->owner),
2494 proto_method_implemented(proto->close),
2495 proto_method_implemented(proto->connect),
2496 proto_method_implemented(proto->disconnect),
2497 proto_method_implemented(proto->accept),
2498 proto_method_implemented(proto->ioctl),
2499 proto_method_implemented(proto->init),
2500 proto_method_implemented(proto->destroy),
2501 proto_method_implemented(proto->shutdown),
2502 proto_method_implemented(proto->setsockopt),
2503 proto_method_implemented(proto->getsockopt),
2504 proto_method_implemented(proto->sendmsg),
2505 proto_method_implemented(proto->recvmsg),
2506 proto_method_implemented(proto->sendpage),
2507 proto_method_implemented(proto->bind),
2508 proto_method_implemented(proto->backlog_rcv),
2509 proto_method_implemented(proto->hash),
2510 proto_method_implemented(proto->unhash),
2511 proto_method_implemented(proto->get_port),
2512 proto_method_implemented(proto->enter_memory_pressure));
2513}
2514
2515static int proto_seq_show(struct seq_file *seq, void *v)
2516{
2517 if (v == &proto_list)
2518 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2519 "protocol",
2520 "size",
2521 "sockets",
2522 "memory",
2523 "press",
2524 "maxhdr",
2525 "slab",
2526 "module",
2527 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2528 else
2529 proto_seq_printf(seq, list_entry(v, struct proto, node));
2530 return 0;
2531}
2532
2533static const struct seq_operations proto_seq_ops = {
2534 .start = proto_seq_start,
2535 .next = proto_seq_next,
2536 .stop = proto_seq_stop,
2537 .show = proto_seq_show,
2538};
2539
2540static int proto_seq_open(struct inode *inode, struct file *file)
2541{
2542 return seq_open_net(inode, file, &proto_seq_ops,
2543 sizeof(struct seq_net_private));
2544}
2545
2546static const struct file_operations proto_seq_fops = {
2547 .owner = THIS_MODULE,
2548 .open = proto_seq_open,
2549 .read = seq_read,
2550 .llseek = seq_lseek,
2551 .release = seq_release_net,
2552};
2553
2554static __net_init int proto_init_net(struct net *net)
2555{
2556 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2557 return -ENOMEM;
2558
2559 return 0;
2560}
2561
2562static __net_exit void proto_exit_net(struct net *net)
2563{
2564 proc_net_remove(net, "protocols");
2565}
2566
2567
2568static __net_initdata struct pernet_operations proto_net_ops = {
2569 .init = proto_init_net,
2570 .exit = proto_exit_net,
2571};
2572
2573static int __init proto_init(void)
2574{
2575 return register_pernet_subsys(&proto_net_ops);
2576}
2577
2578subsys_initcall(proto_init);
2579
2580#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 */