Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4

Linux/net/core/sock.c

  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 
138 #include <trace/events/sock.h>
139 
140 #ifdef CONFIG_INET
141 #include <net/tcp.h>
142 #endif
143 
144 #include <net/busy_poll.h>
145 
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
148 
149 /**
150  * sk_ns_capable - General socket capability test
151  * @sk: Socket to use a capability on or through
152  * @user_ns: The user namespace of the capability to use
153  * @cap: The capability to use
154  *
155  * Test to see if the opener of the socket had when the socket was
156  * created and the current process has the capability @cap in the user
157  * namespace @user_ns.
158  */
159 bool sk_ns_capable(const struct sock *sk,
160                    struct user_namespace *user_ns, int cap)
161 {
162         return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163                 ns_capable(user_ns, cap);
164 }
165 EXPORT_SYMBOL(sk_ns_capable);
166 
167 /**
168  * sk_capable - Socket global capability test
169  * @sk: Socket to use a capability on or through
170  * @cap: The global capability to use
171  *
172  * Test to see if the opener of the socket had when the socket was
173  * created and the current process has the capability @cap in all user
174  * namespaces.
175  */
176 bool sk_capable(const struct sock *sk, int cap)
177 {
178         return sk_ns_capable(sk, &init_user_ns, cap);
179 }
180 EXPORT_SYMBOL(sk_capable);
181 
182 /**
183  * sk_net_capable - Network namespace socket capability test
184  * @sk: Socket to use a capability on or through
185  * @cap: The capability to use
186  *
187  * Test to see if the opener of the socket had when the socket was created
188  * and the current process has the capability @cap over the network namespace
189  * the socket is a member of.
190  */
191 bool sk_net_capable(const struct sock *sk, int cap)
192 {
193         return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 }
195 EXPORT_SYMBOL(sk_net_capable);
196 
197 
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
200 {
201         struct proto *proto;
202         int ret = 0;
203 
204         mutex_lock(&proto_list_mutex);
205         list_for_each_entry(proto, &proto_list, node) {
206                 if (proto->init_cgroup) {
207                         ret = proto->init_cgroup(memcg, ss);
208                         if (ret)
209                                 goto out;
210                 }
211         }
212 
213         mutex_unlock(&proto_list_mutex);
214         return ret;
215 out:
216         list_for_each_entry_continue_reverse(proto, &proto_list, node)
217                 if (proto->destroy_cgroup)
218                         proto->destroy_cgroup(memcg);
219         mutex_unlock(&proto_list_mutex);
220         return ret;
221 }
222 
223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
224 {
225         struct proto *proto;
226 
227         mutex_lock(&proto_list_mutex);
228         list_for_each_entry_reverse(proto, &proto_list, node)
229                 if (proto->destroy_cgroup)
230                         proto->destroy_cgroup(memcg);
231         mutex_unlock(&proto_list_mutex);
232 }
233 #endif
234 
235 /*
236  * Each address family might have different locking rules, so we have
237  * one slock key per address family:
238  */
239 static struct lock_class_key af_family_keys[AF_MAX];
240 static struct lock_class_key af_family_slock_keys[AF_MAX];
241 
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled);
245 #endif
246 
247 /*
248  * Make lock validator output more readable. (we pre-construct these
249  * strings build-time, so that runtime initialization of socket
250  * locks is fast):
251  */
252 static const char *const af_family_key_strings[AF_MAX+1] = {
253   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
254   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
255   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
256   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
257   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
258   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
259   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
260   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
261   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
262   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
263   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
264   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
265   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
266   "sk_lock-AF_NFC"   , "sk_lock-AF_VSOCK"    , "sk_lock-AF_MAX"
267 };
268 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
269   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
270   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
271   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
272   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
273   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
274   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
275   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
276   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
277   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
278   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
279   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
280   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
281   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
282   "slock-AF_NFC"   , "slock-AF_VSOCK"    ,"slock-AF_MAX"
283 };
284 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
285   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
286   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
287   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
288   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
289   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
290   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
291   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
292   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
293   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
294   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
295   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
296   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
297   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
298   "clock-AF_NFC"   , "clock-AF_VSOCK"    , "clock-AF_MAX"
299 };
300 
301 /*
302  * sk_callback_lock locking rules are per-address-family,
303  * so split the lock classes by using a per-AF key:
304  */
305 static struct lock_class_key af_callback_keys[AF_MAX];
306 
307 /* Take into consideration the size of the struct sk_buff overhead in the
308  * determination of these values, since that is non-constant across
309  * platforms.  This makes socket queueing behavior and performance
310  * not depend upon such differences.
311  */
312 #define _SK_MEM_PACKETS         256
313 #define _SK_MEM_OVERHEAD        SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
316 
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
319 EXPORT_SYMBOL(sysctl_wmem_max);
320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
321 EXPORT_SYMBOL(sysctl_rmem_max);
322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
324 
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
327 EXPORT_SYMBOL(sysctl_optmem_max);
328 
329 int sysctl_tstamp_allow_data __read_mostly = 1;
330 
331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
332 EXPORT_SYMBOL_GPL(memalloc_socks);
333 
334 /**
335  * sk_set_memalloc - sets %SOCK_MEMALLOC
336  * @sk: socket to set it on
337  *
338  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339  * It's the responsibility of the admin to adjust min_free_kbytes
340  * to meet the requirements
341  */
342 void sk_set_memalloc(struct sock *sk)
343 {
344         sock_set_flag(sk, SOCK_MEMALLOC);
345         sk->sk_allocation |= __GFP_MEMALLOC;
346         static_key_slow_inc(&memalloc_socks);
347 }
348 EXPORT_SYMBOL_GPL(sk_set_memalloc);
349 
350 void sk_clear_memalloc(struct sock *sk)
351 {
352         sock_reset_flag(sk, SOCK_MEMALLOC);
353         sk->sk_allocation &= ~__GFP_MEMALLOC;
354         static_key_slow_dec(&memalloc_socks);
355 
356         /*
357          * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358          * progress of swapping. SOCK_MEMALLOC may be cleared while
359          * it has rmem allocations due to the last swapfile being deactivated
360          * but there is a risk that the socket is unusable due to exceeding
361          * the rmem limits. Reclaim the reserves and obey rmem limits again.
362          */
363         sk_mem_reclaim(sk);
364 }
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
366 
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
368 {
369         int ret;
370         unsigned long pflags = current->flags;
371 
372         /* these should have been dropped before queueing */
373         BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
374 
375         current->flags |= PF_MEMALLOC;
376         ret = sk->sk_backlog_rcv(sk, skb);
377         tsk_restore_flags(current, pflags, PF_MEMALLOC);
378 
379         return ret;
380 }
381 EXPORT_SYMBOL(__sk_backlog_rcv);
382 
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
384 {
385         struct timeval tv;
386 
387         if (optlen < sizeof(tv))
388                 return -EINVAL;
389         if (copy_from_user(&tv, optval, sizeof(tv)))
390                 return -EFAULT;
391         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
392                 return -EDOM;
393 
394         if (tv.tv_sec < 0) {
395                 static int warned __read_mostly;
396 
397                 *timeo_p = 0;
398                 if (warned < 10 && net_ratelimit()) {
399                         warned++;
400                         pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401                                 __func__, current->comm, task_pid_nr(current));
402                 }
403                 return 0;
404         }
405         *timeo_p = MAX_SCHEDULE_TIMEOUT;
406         if (tv.tv_sec == 0 && tv.tv_usec == 0)
407                 return 0;
408         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
410         return 0;
411 }
412 
413 static void sock_warn_obsolete_bsdism(const char *name)
414 {
415         static int warned;
416         static char warncomm[TASK_COMM_LEN];
417         if (strcmp(warncomm, current->comm) && warned < 5) {
418                 strcpy(warncomm,  current->comm);
419                 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
420                         warncomm, name);
421                 warned++;
422         }
423 }
424 
425 static bool sock_needs_netstamp(const struct sock *sk)
426 {
427         switch (sk->sk_family) {
428         case AF_UNSPEC:
429         case AF_UNIX:
430                 return false;
431         default:
432                 return true;
433         }
434 }
435 
436 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
437 {
438         if (sk->sk_flags & flags) {
439                 sk->sk_flags &= ~flags;
440                 if (sock_needs_netstamp(sk) &&
441                     !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
442                         net_disable_timestamp();
443         }
444 }
445 
446 
447 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
448 {
449         int err;
450         unsigned long flags;
451         struct sk_buff_head *list = &sk->sk_receive_queue;
452 
453         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
454                 atomic_inc(&sk->sk_drops);
455                 trace_sock_rcvqueue_full(sk, skb);
456                 return -ENOMEM;
457         }
458 
459         err = sk_filter(sk, skb);
460         if (err)
461                 return err;
462 
463         if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
464                 atomic_inc(&sk->sk_drops);
465                 return -ENOBUFS;
466         }
467 
468         skb->dev = NULL;
469         skb_set_owner_r(skb, sk);
470 
471         /* we escape from rcu protected region, make sure we dont leak
472          * a norefcounted dst
473          */
474         skb_dst_force(skb);
475 
476         spin_lock_irqsave(&list->lock, flags);
477         sock_skb_set_dropcount(sk, skb);
478         __skb_queue_tail(list, skb);
479         spin_unlock_irqrestore(&list->lock, flags);
480 
481         if (!sock_flag(sk, SOCK_DEAD))
482                 sk->sk_data_ready(sk);
483         return 0;
484 }
485 EXPORT_SYMBOL(sock_queue_rcv_skb);
486 
487 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
488 {
489         int rc = NET_RX_SUCCESS;
490 
491         if (sk_filter(sk, skb))
492                 goto discard_and_relse;
493 
494         skb->dev = NULL;
495 
496         if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
497                 atomic_inc(&sk->sk_drops);
498                 goto discard_and_relse;
499         }
500         if (nested)
501                 bh_lock_sock_nested(sk);
502         else
503                 bh_lock_sock(sk);
504         if (!sock_owned_by_user(sk)) {
505                 /*
506                  * trylock + unlock semantics:
507                  */
508                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
509 
510                 rc = sk_backlog_rcv(sk, skb);
511 
512                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
513         } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
514                 bh_unlock_sock(sk);
515                 atomic_inc(&sk->sk_drops);
516                 goto discard_and_relse;
517         }
518 
519         bh_unlock_sock(sk);
520 out:
521         sock_put(sk);
522         return rc;
523 discard_and_relse:
524         kfree_skb(skb);
525         goto out;
526 }
527 EXPORT_SYMBOL(sk_receive_skb);
528 
529 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
530 {
531         struct dst_entry *dst = __sk_dst_get(sk);
532 
533         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
534                 sk_tx_queue_clear(sk);
535                 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
536                 dst_release(dst);
537                 return NULL;
538         }
539 
540         return dst;
541 }
542 EXPORT_SYMBOL(__sk_dst_check);
543 
544 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
545 {
546         struct dst_entry *dst = sk_dst_get(sk);
547 
548         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
549                 sk_dst_reset(sk);
550                 dst_release(dst);
551                 return NULL;
552         }
553 
554         return dst;
555 }
556 EXPORT_SYMBOL(sk_dst_check);
557 
558 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
559                                 int optlen)
560 {
561         int ret = -ENOPROTOOPT;
562 #ifdef CONFIG_NETDEVICES
563         struct net *net = sock_net(sk);
564         char devname[IFNAMSIZ];
565         int index;
566 
567         /* Sorry... */
568         ret = -EPERM;
569         if (!ns_capable(net->user_ns, CAP_NET_RAW))
570                 goto out;
571 
572         ret = -EINVAL;
573         if (optlen < 0)
574                 goto out;
575 
576         /* Bind this socket to a particular device like "eth0",
577          * as specified in the passed interface name. If the
578          * name is "" or the option length is zero the socket
579          * is not bound.
580          */
581         if (optlen > IFNAMSIZ - 1)
582                 optlen = IFNAMSIZ - 1;
583         memset(devname, 0, sizeof(devname));
584 
585         ret = -EFAULT;
586         if (copy_from_user(devname, optval, optlen))
587                 goto out;
588 
589         index = 0;
590         if (devname[0] != '\0') {
591                 struct net_device *dev;
592 
593                 rcu_read_lock();
594                 dev = dev_get_by_name_rcu(net, devname);
595                 if (dev)
596                         index = dev->ifindex;
597                 rcu_read_unlock();
598                 ret = -ENODEV;
599                 if (!dev)
600                         goto out;
601         }
602 
603         lock_sock(sk);
604         sk->sk_bound_dev_if = index;
605         sk_dst_reset(sk);
606         release_sock(sk);
607 
608         ret = 0;
609 
610 out:
611 #endif
612 
613         return ret;
614 }
615 
616 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
617                                 int __user *optlen, int len)
618 {
619         int ret = -ENOPROTOOPT;
620 #ifdef CONFIG_NETDEVICES
621         struct net *net = sock_net(sk);
622         char devname[IFNAMSIZ];
623 
624         if (sk->sk_bound_dev_if == 0) {
625                 len = 0;
626                 goto zero;
627         }
628 
629         ret = -EINVAL;
630         if (len < IFNAMSIZ)
631                 goto out;
632 
633         ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
634         if (ret)
635                 goto out;
636 
637         len = strlen(devname) + 1;
638 
639         ret = -EFAULT;
640         if (copy_to_user(optval, devname, len))
641                 goto out;
642 
643 zero:
644         ret = -EFAULT;
645         if (put_user(len, optlen))
646                 goto out;
647 
648         ret = 0;
649 
650 out:
651 #endif
652 
653         return ret;
654 }
655 
656 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
657 {
658         if (valbool)
659                 sock_set_flag(sk, bit);
660         else
661                 sock_reset_flag(sk, bit);
662 }
663 
664 bool sk_mc_loop(struct sock *sk)
665 {
666         if (dev_recursion_level())
667                 return false;
668         if (!sk)
669                 return true;
670         switch (sk->sk_family) {
671         case AF_INET:
672                 return inet_sk(sk)->mc_loop;
673 #if IS_ENABLED(CONFIG_IPV6)
674         case AF_INET6:
675                 return inet6_sk(sk)->mc_loop;
676 #endif
677         }
678         WARN_ON(1);
679         return true;
680 }
681 EXPORT_SYMBOL(sk_mc_loop);
682 
683 /*
684  *      This is meant for all protocols to use and covers goings on
685  *      at the socket level. Everything here is generic.
686  */
687 
688 int sock_setsockopt(struct socket *sock, int level, int optname,
689                     char __user *optval, unsigned int optlen)
690 {
691         struct sock *sk = sock->sk;
692         int val;
693         int valbool;
694         struct linger ling;
695         int ret = 0;
696 
697         /*
698          *      Options without arguments
699          */
700 
701         if (optname == SO_BINDTODEVICE)
702                 return sock_setbindtodevice(sk, optval, optlen);
703 
704         if (optlen < sizeof(int))
705                 return -EINVAL;
706 
707         if (get_user(val, (int __user *)optval))
708                 return -EFAULT;
709 
710         valbool = val ? 1 : 0;
711 
712         lock_sock(sk);
713 
714         switch (optname) {
715         case SO_DEBUG:
716                 if (val && !capable(CAP_NET_ADMIN))
717                         ret = -EACCES;
718                 else
719                         sock_valbool_flag(sk, SOCK_DBG, valbool);
720                 break;
721         case SO_REUSEADDR:
722                 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
723                 break;
724         case SO_REUSEPORT:
725                 sk->sk_reuseport = valbool;
726                 break;
727         case SO_TYPE:
728         case SO_PROTOCOL:
729         case SO_DOMAIN:
730         case SO_ERROR:
731                 ret = -ENOPROTOOPT;
732                 break;
733         case SO_DONTROUTE:
734                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
735                 break;
736         case SO_BROADCAST:
737                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
738                 break;
739         case SO_SNDBUF:
740                 /* Don't error on this BSD doesn't and if you think
741                  * about it this is right. Otherwise apps have to
742                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
743                  * are treated in BSD as hints
744                  */
745                 val = min_t(u32, val, sysctl_wmem_max);
746 set_sndbuf:
747                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
748                 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
749                 /* Wake up sending tasks if we upped the value. */
750                 sk->sk_write_space(sk);
751                 break;
752 
753         case SO_SNDBUFFORCE:
754                 if (!capable(CAP_NET_ADMIN)) {
755                         ret = -EPERM;
756                         break;
757                 }
758                 goto set_sndbuf;
759 
760         case SO_RCVBUF:
761                 /* Don't error on this BSD doesn't and if you think
762                  * about it this is right. Otherwise apps have to
763                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
764                  * are treated in BSD as hints
765                  */
766                 val = min_t(u32, val, sysctl_rmem_max);
767 set_rcvbuf:
768                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
769                 /*
770                  * We double it on the way in to account for
771                  * "struct sk_buff" etc. overhead.   Applications
772                  * assume that the SO_RCVBUF setting they make will
773                  * allow that much actual data to be received on that
774                  * socket.
775                  *
776                  * Applications are unaware that "struct sk_buff" and
777                  * other overheads allocate from the receive buffer
778                  * during socket buffer allocation.
779                  *
780                  * And after considering the possible alternatives,
781                  * returning the value we actually used in getsockopt
782                  * is the most desirable behavior.
783                  */
784                 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
785                 break;
786 
787         case SO_RCVBUFFORCE:
788                 if (!capable(CAP_NET_ADMIN)) {
789                         ret = -EPERM;
790                         break;
791                 }
792                 goto set_rcvbuf;
793 
794         case SO_KEEPALIVE:
795 #ifdef CONFIG_INET
796                 if (sk->sk_protocol == IPPROTO_TCP &&
797                     sk->sk_type == SOCK_STREAM)
798                         tcp_set_keepalive(sk, valbool);
799 #endif
800                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
801                 break;
802 
803         case SO_OOBINLINE:
804                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
805                 break;
806 
807         case SO_NO_CHECK:
808                 sk->sk_no_check_tx = valbool;
809                 break;
810 
811         case SO_PRIORITY:
812                 if ((val >= 0 && val <= 6) ||
813                     ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
814                         sk->sk_priority = val;
815                 else
816                         ret = -EPERM;
817                 break;
818 
819         case SO_LINGER:
820                 if (optlen < sizeof(ling)) {
821                         ret = -EINVAL;  /* 1003.1g */
822                         break;
823                 }
824                 if (copy_from_user(&ling, optval, sizeof(ling))) {
825                         ret = -EFAULT;
826                         break;
827                 }
828                 if (!ling.l_onoff)
829                         sock_reset_flag(sk, SOCK_LINGER);
830                 else {
831 #if (BITS_PER_LONG == 32)
832                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
833                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
834                         else
835 #endif
836                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
837                         sock_set_flag(sk, SOCK_LINGER);
838                 }
839                 break;
840 
841         case SO_BSDCOMPAT:
842                 sock_warn_obsolete_bsdism("setsockopt");
843                 break;
844 
845         case SO_PASSCRED:
846                 if (valbool)
847                         set_bit(SOCK_PASSCRED, &sock->flags);
848                 else
849                         clear_bit(SOCK_PASSCRED, &sock->flags);
850                 break;
851 
852         case SO_TIMESTAMP:
853         case SO_TIMESTAMPNS:
854                 if (valbool)  {
855                         if (optname == SO_TIMESTAMP)
856                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
857                         else
858                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
859                         sock_set_flag(sk, SOCK_RCVTSTAMP);
860                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
861                 } else {
862                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
863                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
864                 }
865                 break;
866 
867         case SO_TIMESTAMPING:
868                 if (val & ~SOF_TIMESTAMPING_MASK) {
869                         ret = -EINVAL;
870                         break;
871                 }
872 
873                 if (val & SOF_TIMESTAMPING_OPT_ID &&
874                     !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
875                         if (sk->sk_protocol == IPPROTO_TCP &&
876                             sk->sk_type == SOCK_STREAM) {
877                                 if (sk->sk_state != TCP_ESTABLISHED) {
878                                         ret = -EINVAL;
879                                         break;
880                                 }
881                                 sk->sk_tskey = tcp_sk(sk)->snd_una;
882                         } else {
883                                 sk->sk_tskey = 0;
884                         }
885                 }
886                 sk->sk_tsflags = val;
887                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
888                         sock_enable_timestamp(sk,
889                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
890                 else
891                         sock_disable_timestamp(sk,
892                                                (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
893                 break;
894 
895         case SO_RCVLOWAT:
896                 if (val < 0)
897                         val = INT_MAX;
898                 sk->sk_rcvlowat = val ? : 1;
899                 break;
900 
901         case SO_RCVTIMEO:
902                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
903                 break;
904 
905         case SO_SNDTIMEO:
906                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
907                 break;
908 
909         case SO_ATTACH_FILTER:
910                 ret = -EINVAL;
911                 if (optlen == sizeof(struct sock_fprog)) {
912                         struct sock_fprog fprog;
913 
914                         ret = -EFAULT;
915                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
916                                 break;
917 
918                         ret = sk_attach_filter(&fprog, sk);
919                 }
920                 break;
921 
922         case SO_ATTACH_BPF:
923                 ret = -EINVAL;
924                 if (optlen == sizeof(u32)) {
925                         u32 ufd;
926 
927                         ret = -EFAULT;
928                         if (copy_from_user(&ufd, optval, sizeof(ufd)))
929                                 break;
930 
931                         ret = sk_attach_bpf(ufd, sk);
932                 }
933                 break;
934 
935         case SO_DETACH_FILTER:
936                 ret = sk_detach_filter(sk);
937                 break;
938 
939         case SO_LOCK_FILTER:
940                 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
941                         ret = -EPERM;
942                 else
943                         sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
944                 break;
945 
946         case SO_PASSSEC:
947                 if (valbool)
948                         set_bit(SOCK_PASSSEC, &sock->flags);
949                 else
950                         clear_bit(SOCK_PASSSEC, &sock->flags);
951                 break;
952         case SO_MARK:
953                 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
954                         ret = -EPERM;
955                 else
956                         sk->sk_mark = val;
957                 break;
958 
959         case SO_RXQ_OVFL:
960                 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
961                 break;
962 
963         case SO_WIFI_STATUS:
964                 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
965                 break;
966 
967         case SO_PEEK_OFF:
968                 if (sock->ops->set_peek_off)
969                         ret = sock->ops->set_peek_off(sk, val);
970                 else
971                         ret = -EOPNOTSUPP;
972                 break;
973 
974         case SO_NOFCS:
975                 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
976                 break;
977 
978         case SO_SELECT_ERR_QUEUE:
979                 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
980                 break;
981 
982 #ifdef CONFIG_NET_RX_BUSY_POLL
983         case SO_BUSY_POLL:
984                 /* allow unprivileged users to decrease the value */
985                 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
986                         ret = -EPERM;
987                 else {
988                         if (val < 0)
989                                 ret = -EINVAL;
990                         else
991                                 sk->sk_ll_usec = val;
992                 }
993                 break;
994 #endif
995 
996         case SO_MAX_PACING_RATE:
997                 sk->sk_max_pacing_rate = val;
998                 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
999                                          sk->sk_max_pacing_rate);
1000                 break;
1001 
1002         case SO_INCOMING_CPU:
1003                 sk->sk_incoming_cpu = val;
1004                 break;
1005 
1006         default:
1007                 ret = -ENOPROTOOPT;
1008                 break;
1009         }
1010         release_sock(sk);
1011         return ret;
1012 }
1013 EXPORT_SYMBOL(sock_setsockopt);
1014 
1015 
1016 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1017                           struct ucred *ucred)
1018 {
1019         ucred->pid = pid_vnr(pid);
1020         ucred->uid = ucred->gid = -1;
1021         if (cred) {
1022                 struct user_namespace *current_ns = current_user_ns();
1023 
1024                 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1025                 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1026         }
1027 }
1028 
1029 int sock_getsockopt(struct socket *sock, int level, int optname,
1030                     char __user *optval, int __user *optlen)
1031 {
1032         struct sock *sk = sock->sk;
1033 
1034         union {
1035                 int val;
1036                 struct linger ling;
1037                 struct timeval tm;
1038         } v;
1039 
1040         int lv = sizeof(int);
1041         int len;
1042 
1043         if (get_user(len, optlen))
1044                 return -EFAULT;
1045         if (len < 0)
1046                 return -EINVAL;
1047 
1048         memset(&v, 0, sizeof(v));
1049 
1050         switch (optname) {
1051         case SO_DEBUG:
1052                 v.val = sock_flag(sk, SOCK_DBG);
1053                 break;
1054 
1055         case SO_DONTROUTE:
1056                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1057                 break;
1058 
1059         case SO_BROADCAST:
1060                 v.val = sock_flag(sk, SOCK_BROADCAST);
1061                 break;
1062 
1063         case SO_SNDBUF:
1064                 v.val = sk->sk_sndbuf;
1065                 break;
1066 
1067         case SO_RCVBUF:
1068                 v.val = sk->sk_rcvbuf;
1069                 break;
1070 
1071         case SO_REUSEADDR:
1072                 v.val = sk->sk_reuse;
1073                 break;
1074 
1075         case SO_REUSEPORT:
1076                 v.val = sk->sk_reuseport;
1077                 break;
1078 
1079         case SO_KEEPALIVE:
1080                 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1081                 break;
1082 
1083         case SO_TYPE:
1084                 v.val = sk->sk_type;
1085                 break;
1086 
1087         case SO_PROTOCOL:
1088                 v.val = sk->sk_protocol;
1089                 break;
1090 
1091         case SO_DOMAIN:
1092                 v.val = sk->sk_family;
1093                 break;
1094 
1095         case SO_ERROR:
1096                 v.val = -sock_error(sk);
1097                 if (v.val == 0)
1098                         v.val = xchg(&sk->sk_err_soft, 0);
1099                 break;
1100 
1101         case SO_OOBINLINE:
1102                 v.val = sock_flag(sk, SOCK_URGINLINE);
1103                 break;
1104 
1105         case SO_NO_CHECK:
1106                 v.val = sk->sk_no_check_tx;
1107                 break;
1108 
1109         case SO_PRIORITY:
1110                 v.val = sk->sk_priority;
1111                 break;
1112 
1113         case SO_LINGER:
1114                 lv              = sizeof(v.ling);
1115                 v.ling.l_onoff  = sock_flag(sk, SOCK_LINGER);
1116                 v.ling.l_linger = sk->sk_lingertime / HZ;
1117                 break;
1118 
1119         case SO_BSDCOMPAT:
1120                 sock_warn_obsolete_bsdism("getsockopt");
1121                 break;
1122 
1123         case SO_TIMESTAMP:
1124                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1125                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1126                 break;
1127 
1128         case SO_TIMESTAMPNS:
1129                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1130                 break;
1131 
1132         case SO_TIMESTAMPING:
1133                 v.val = sk->sk_tsflags;
1134                 break;
1135 
1136         case SO_RCVTIMEO:
1137                 lv = sizeof(struct timeval);
1138                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1139                         v.tm.tv_sec = 0;
1140                         v.tm.tv_usec = 0;
1141                 } else {
1142                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1143                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1144                 }
1145                 break;
1146 
1147         case SO_SNDTIMEO:
1148                 lv = sizeof(struct timeval);
1149                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1150                         v.tm.tv_sec = 0;
1151                         v.tm.tv_usec = 0;
1152                 } else {
1153                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1154                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1155                 }
1156                 break;
1157 
1158         case SO_RCVLOWAT:
1159                 v.val = sk->sk_rcvlowat;
1160                 break;
1161 
1162         case SO_SNDLOWAT:
1163                 v.val = 1;
1164                 break;
1165 
1166         case SO_PASSCRED:
1167                 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1168                 break;
1169 
1170         case SO_PEERCRED:
1171         {
1172                 struct ucred peercred;
1173                 if (len > sizeof(peercred))
1174                         len = sizeof(peercred);
1175                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1176                 if (copy_to_user(optval, &peercred, len))
1177                         return -EFAULT;
1178                 goto lenout;
1179         }
1180 
1181         case SO_PEERNAME:
1182         {
1183                 char address[128];
1184 
1185                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1186                         return -ENOTCONN;
1187                 if (lv < len)
1188                         return -EINVAL;
1189                 if (copy_to_user(optval, address, len))
1190                         return -EFAULT;
1191                 goto lenout;
1192         }
1193 
1194         /* Dubious BSD thing... Probably nobody even uses it, but
1195          * the UNIX standard wants it for whatever reason... -DaveM
1196          */
1197         case SO_ACCEPTCONN:
1198                 v.val = sk->sk_state == TCP_LISTEN;
1199                 break;
1200 
1201         case SO_PASSSEC:
1202                 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1203                 break;
1204 
1205         case SO_PEERSEC:
1206                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1207 
1208         case SO_MARK:
1209                 v.val = sk->sk_mark;
1210                 break;
1211 
1212         case SO_RXQ_OVFL:
1213                 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1214                 break;
1215 
1216         case SO_WIFI_STATUS:
1217                 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1218                 break;
1219 
1220         case SO_PEEK_OFF:
1221                 if (!sock->ops->set_peek_off)
1222                         return -EOPNOTSUPP;
1223 
1224                 v.val = sk->sk_peek_off;
1225                 break;
1226         case SO_NOFCS:
1227                 v.val = sock_flag(sk, SOCK_NOFCS);
1228                 break;
1229 
1230         case SO_BINDTODEVICE:
1231                 return sock_getbindtodevice(sk, optval, optlen, len);
1232 
1233         case SO_GET_FILTER:
1234                 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1235                 if (len < 0)
1236                         return len;
1237 
1238                 goto lenout;
1239 
1240         case SO_LOCK_FILTER:
1241                 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1242                 break;
1243 
1244         case SO_BPF_EXTENSIONS:
1245                 v.val = bpf_tell_extensions();
1246                 break;
1247 
1248         case SO_SELECT_ERR_QUEUE:
1249                 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1250                 break;
1251 
1252 #ifdef CONFIG_NET_RX_BUSY_POLL
1253         case SO_BUSY_POLL:
1254                 v.val = sk->sk_ll_usec;
1255                 break;
1256 #endif
1257 
1258         case SO_MAX_PACING_RATE:
1259                 v.val = sk->sk_max_pacing_rate;
1260                 break;
1261 
1262         case SO_INCOMING_CPU:
1263                 v.val = sk->sk_incoming_cpu;
1264                 break;
1265 
1266         default:
1267                 /* We implement the SO_SNDLOWAT etc to not be settable
1268                  * (1003.1g 7).
1269                  */
1270                 return -ENOPROTOOPT;
1271         }
1272 
1273         if (len > lv)
1274                 len = lv;
1275         if (copy_to_user(optval, &v, len))
1276                 return -EFAULT;
1277 lenout:
1278         if (put_user(len, optlen))
1279                 return -EFAULT;
1280         return 0;
1281 }
1282 
1283 /*
1284  * Initialize an sk_lock.
1285  *
1286  * (We also register the sk_lock with the lock validator.)
1287  */
1288 static inline void sock_lock_init(struct sock *sk)
1289 {
1290         sock_lock_init_class_and_name(sk,
1291                         af_family_slock_key_strings[sk->sk_family],
1292                         af_family_slock_keys + sk->sk_family,
1293                         af_family_key_strings[sk->sk_family],
1294                         af_family_keys + sk->sk_family);
1295 }
1296 
1297 /*
1298  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1299  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1300  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1301  */
1302 static void sock_copy(struct sock *nsk, const struct sock *osk)
1303 {
1304 #ifdef CONFIG_SECURITY_NETWORK
1305         void *sptr = nsk->sk_security;
1306 #endif
1307         memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1308 
1309         memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1310                osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1311 
1312 #ifdef CONFIG_SECURITY_NETWORK
1313         nsk->sk_security = sptr;
1314         security_sk_clone(osk, nsk);
1315 #endif
1316 }
1317 
1318 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1319 {
1320         unsigned long nulls1, nulls2;
1321 
1322         nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1323         nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1324         if (nulls1 > nulls2)
1325                 swap(nulls1, nulls2);
1326 
1327         if (nulls1 != 0)
1328                 memset((char *)sk, 0, nulls1);
1329         memset((char *)sk + nulls1 + sizeof(void *), 0,
1330                nulls2 - nulls1 - sizeof(void *));
1331         memset((char *)sk + nulls2 + sizeof(void *), 0,
1332                size - nulls2 - sizeof(void *));
1333 }
1334 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1335 
1336 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1337                 int family)
1338 {
1339         struct sock *sk;
1340         struct kmem_cache *slab;
1341 
1342         slab = prot->slab;
1343         if (slab != NULL) {
1344                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1345                 if (!sk)
1346                         return sk;
1347                 if (priority & __GFP_ZERO) {
1348                         if (prot->clear_sk)
1349                                 prot->clear_sk(sk, prot->obj_size);
1350                         else
1351                                 sk_prot_clear_nulls(sk, prot->obj_size);
1352                 }
1353         } else
1354                 sk = kmalloc(prot->obj_size, priority);
1355 
1356         if (sk != NULL) {
1357                 kmemcheck_annotate_bitfield(sk, flags);
1358 
1359                 if (security_sk_alloc(sk, family, priority))
1360                         goto out_free;
1361 
1362                 if (!try_module_get(prot->owner))
1363                         goto out_free_sec;
1364                 sk_tx_queue_clear(sk);
1365         }
1366 
1367         return sk;
1368 
1369 out_free_sec:
1370         security_sk_free(sk);
1371 out_free:
1372         if (slab != NULL)
1373                 kmem_cache_free(slab, sk);
1374         else
1375                 kfree(sk);
1376         return NULL;
1377 }
1378 
1379 static void sk_prot_free(struct proto *prot, struct sock *sk)
1380 {
1381         struct kmem_cache *slab;
1382         struct module *owner;
1383 
1384         owner = prot->owner;
1385         slab = prot->slab;
1386 
1387         security_sk_free(sk);
1388         if (slab != NULL)
1389                 kmem_cache_free(slab, sk);
1390         else
1391                 kfree(sk);
1392         module_put(owner);
1393 }
1394 
1395 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1396 void sock_update_netprioidx(struct sock *sk)
1397 {
1398         if (in_interrupt())
1399                 return;
1400 
1401         sk->sk_cgrp_prioidx = task_netprioidx(current);
1402 }
1403 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1404 #endif
1405 
1406 /**
1407  *      sk_alloc - All socket objects are allocated here
1408  *      @net: the applicable net namespace
1409  *      @family: protocol family
1410  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1411  *      @prot: struct proto associated with this new sock instance
1412  *      @kern: is this to be a kernel socket?
1413  */
1414 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1415                       struct proto *prot, int kern)
1416 {
1417         struct sock *sk;
1418 
1419         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1420         if (sk) {
1421                 sk->sk_family = family;
1422                 /*
1423                  * See comment in struct sock definition to understand
1424                  * why we need sk_prot_creator -acme
1425                  */
1426                 sk->sk_prot = sk->sk_prot_creator = prot;
1427                 sock_lock_init(sk);
1428                 sk->sk_net_refcnt = kern ? 0 : 1;
1429                 if (likely(sk->sk_net_refcnt))
1430                         get_net(net);
1431                 sock_net_set(sk, net);
1432                 atomic_set(&sk->sk_wmem_alloc, 1);
1433 
1434                 sock_update_classid(sk);
1435                 sock_update_netprioidx(sk);
1436         }
1437 
1438         return sk;
1439 }
1440 EXPORT_SYMBOL(sk_alloc);
1441 
1442 void sk_destruct(struct sock *sk)
1443 {
1444         struct sk_filter *filter;
1445 
1446         if (sk->sk_destruct)
1447                 sk->sk_destruct(sk);
1448 
1449         filter = rcu_dereference_check(sk->sk_filter,
1450                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1451         if (filter) {
1452                 sk_filter_uncharge(sk, filter);
1453                 RCU_INIT_POINTER(sk->sk_filter, NULL);
1454         }
1455 
1456         sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1457 
1458         if (atomic_read(&sk->sk_omem_alloc))
1459                 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1460                          __func__, atomic_read(&sk->sk_omem_alloc));
1461 
1462         if (sk->sk_peer_cred)
1463                 put_cred(sk->sk_peer_cred);
1464         put_pid(sk->sk_peer_pid);
1465         if (likely(sk->sk_net_refcnt))
1466                 put_net(sock_net(sk));
1467         sk_prot_free(sk->sk_prot_creator, sk);
1468 }
1469 
1470 static void __sk_free(struct sock *sk)
1471 {
1472         if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1473                 sock_diag_broadcast_destroy(sk);
1474         else
1475                 sk_destruct(sk);
1476 }
1477 
1478 void sk_free(struct sock *sk)
1479 {
1480         /*
1481          * We subtract one from sk_wmem_alloc and can know if
1482          * some packets are still in some tx queue.
1483          * If not null, sock_wfree() will call __sk_free(sk) later
1484          */
1485         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1486                 __sk_free(sk);
1487 }
1488 EXPORT_SYMBOL(sk_free);
1489 
1490 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1491 {
1492         if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1493                 sock_update_memcg(newsk);
1494 }
1495 
1496 /**
1497  *      sk_clone_lock - clone a socket, and lock its clone
1498  *      @sk: the socket to clone
1499  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1500  *
1501  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1502  */
1503 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1504 {
1505         struct sock *newsk;
1506         bool is_charged = true;
1507 
1508         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1509         if (newsk != NULL) {
1510                 struct sk_filter *filter;
1511 
1512                 sock_copy(newsk, sk);
1513 
1514                 /* SANITY */
1515                 if (likely(newsk->sk_net_refcnt))
1516                         get_net(sock_net(newsk));
1517                 sk_node_init(&newsk->sk_node);
1518                 sock_lock_init(newsk);
1519                 bh_lock_sock(newsk);
1520                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1521                 newsk->sk_backlog.len = 0;
1522 
1523                 atomic_set(&newsk->sk_rmem_alloc, 0);
1524                 /*
1525                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1526                  */
1527                 atomic_set(&newsk->sk_wmem_alloc, 1);
1528                 atomic_set(&newsk->sk_omem_alloc, 0);
1529                 skb_queue_head_init(&newsk->sk_receive_queue);
1530                 skb_queue_head_init(&newsk->sk_write_queue);
1531 
1532                 rwlock_init(&newsk->sk_callback_lock);
1533                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1534                                 af_callback_keys + newsk->sk_family,
1535                                 af_family_clock_key_strings[newsk->sk_family]);
1536 
1537                 newsk->sk_dst_cache     = NULL;
1538                 newsk->sk_wmem_queued   = 0;
1539                 newsk->sk_forward_alloc = 0;
1540                 newsk->sk_send_head     = NULL;
1541                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1542 
1543                 sock_reset_flag(newsk, SOCK_DONE);
1544                 skb_queue_head_init(&newsk->sk_error_queue);
1545 
1546                 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1547                 if (filter != NULL)
1548                         /* though it's an empty new sock, the charging may fail
1549                          * if sysctl_optmem_max was changed between creation of
1550                          * original socket and cloning
1551                          */
1552                         is_charged = sk_filter_charge(newsk, filter);
1553 
1554                 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1555                         /* It is still raw copy of parent, so invalidate
1556                          * destructor and make plain sk_free() */
1557                         newsk->sk_destruct = NULL;
1558                         bh_unlock_sock(newsk);
1559                         sk_free(newsk);
1560                         newsk = NULL;
1561                         goto out;
1562                 }
1563 
1564                 newsk->sk_err      = 0;
1565                 newsk->sk_priority = 0;
1566                 newsk->sk_incoming_cpu = raw_smp_processor_id();
1567                 atomic64_set(&newsk->sk_cookie, 0);
1568                 /*
1569                  * Before updating sk_refcnt, we must commit prior changes to memory
1570                  * (Documentation/RCU/rculist_nulls.txt for details)
1571                  */
1572                 smp_wmb();
1573                 atomic_set(&newsk->sk_refcnt, 2);
1574 
1575                 /*
1576                  * Increment the counter in the same struct proto as the master
1577                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1578                  * is the same as sk->sk_prot->socks, as this field was copied
1579                  * with memcpy).
1580                  *
1581                  * This _changes_ the previous behaviour, where
1582                  * tcp_create_openreq_child always was incrementing the
1583                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1584                  * to be taken into account in all callers. -acme
1585                  */
1586                 sk_refcnt_debug_inc(newsk);
1587                 sk_set_socket(newsk, NULL);
1588                 newsk->sk_wq = NULL;
1589 
1590                 sk_update_clone(sk, newsk);
1591 
1592                 if (newsk->sk_prot->sockets_allocated)
1593                         sk_sockets_allocated_inc(newsk);
1594 
1595                 if (sock_needs_netstamp(sk) &&
1596                     newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1597                         net_enable_timestamp();
1598         }
1599 out:
1600         return newsk;
1601 }
1602 EXPORT_SYMBOL_GPL(sk_clone_lock);
1603 
1604 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1605 {
1606         u32 max_segs = 1;
1607 
1608         sk_dst_set(sk, dst);
1609         sk->sk_route_caps = dst->dev->features;
1610         if (sk->sk_route_caps & NETIF_F_GSO)
1611                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1612         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1613         if (sk_can_gso(sk)) {
1614                 if (dst->header_len) {
1615                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1616                 } else {
1617                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1618                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1619                         max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1620                 }
1621         }
1622         sk->sk_gso_max_segs = max_segs;
1623 }
1624 EXPORT_SYMBOL_GPL(sk_setup_caps);
1625 
1626 /*
1627  *      Simple resource managers for sockets.
1628  */
1629 
1630 
1631 /*
1632  * Write buffer destructor automatically called from kfree_skb.
1633  */
1634 void sock_wfree(struct sk_buff *skb)
1635 {
1636         struct sock *sk = skb->sk;
1637         unsigned int len = skb->truesize;
1638 
1639         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1640                 /*
1641                  * Keep a reference on sk_wmem_alloc, this will be released
1642                  * after sk_write_space() call
1643                  */
1644                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1645                 sk->sk_write_space(sk);
1646                 len = 1;
1647         }
1648         /*
1649          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1650          * could not do because of in-flight packets
1651          */
1652         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1653                 __sk_free(sk);
1654 }
1655 EXPORT_SYMBOL(sock_wfree);
1656 
1657 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1658 {
1659         skb_orphan(skb);
1660         skb->sk = sk;
1661 #ifdef CONFIG_INET
1662         if (unlikely(!sk_fullsock(sk))) {
1663                 skb->destructor = sock_edemux;
1664                 sock_hold(sk);
1665                 return;
1666         }
1667 #endif
1668         skb->destructor = sock_wfree;
1669         skb_set_hash_from_sk(skb, sk);
1670         /*
1671          * We used to take a refcount on sk, but following operation
1672          * is enough to guarantee sk_free() wont free this sock until
1673          * all in-flight packets are completed
1674          */
1675         atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1676 }
1677 EXPORT_SYMBOL(skb_set_owner_w);
1678 
1679 void skb_orphan_partial(struct sk_buff *skb)
1680 {
1681         /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1682          * so we do not completely orphan skb, but transfert all
1683          * accounted bytes but one, to avoid unexpected reorders.
1684          */
1685         if (skb->destructor == sock_wfree
1686 #ifdef CONFIG_INET
1687             || skb->destructor == tcp_wfree
1688 #endif
1689                 ) {
1690                 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1691                 skb->truesize = 1;
1692         } else {
1693                 skb_orphan(skb);
1694         }
1695 }
1696 EXPORT_SYMBOL(skb_orphan_partial);
1697 
1698 /*
1699  * Read buffer destructor automatically called from kfree_skb.
1700  */
1701 void sock_rfree(struct sk_buff *skb)
1702 {
1703         struct sock *sk = skb->sk;
1704         unsigned int len = skb->truesize;
1705 
1706         atomic_sub(len, &sk->sk_rmem_alloc);
1707         sk_mem_uncharge(sk, len);
1708 }
1709 EXPORT_SYMBOL(sock_rfree);
1710 
1711 /*
1712  * Buffer destructor for skbs that are not used directly in read or write
1713  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1714  */
1715 void sock_efree(struct sk_buff *skb)
1716 {
1717         sock_put(skb->sk);
1718 }
1719 EXPORT_SYMBOL(sock_efree);
1720 
1721 kuid_t sock_i_uid(struct sock *sk)
1722 {
1723         kuid_t uid;
1724 
1725         read_lock_bh(&sk->sk_callback_lock);
1726         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1727         read_unlock_bh(&sk->sk_callback_lock);
1728         return uid;
1729 }
1730 EXPORT_SYMBOL(sock_i_uid);
1731 
1732 unsigned long sock_i_ino(struct sock *sk)
1733 {
1734         unsigned long ino;
1735 
1736         read_lock_bh(&sk->sk_callback_lock);
1737         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1738         read_unlock_bh(&sk->sk_callback_lock);
1739         return ino;
1740 }
1741 EXPORT_SYMBOL(sock_i_ino);
1742 
1743 /*
1744  * Allocate a skb from the socket's send buffer.
1745  */
1746 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1747                              gfp_t priority)
1748 {
1749         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1750                 struct sk_buff *skb = alloc_skb(size, priority);
1751                 if (skb) {
1752                         skb_set_owner_w(skb, sk);
1753                         return skb;
1754                 }
1755         }
1756         return NULL;
1757 }
1758 EXPORT_SYMBOL(sock_wmalloc);
1759 
1760 /*
1761  * Allocate a memory block from the socket's option memory buffer.
1762  */
1763 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1764 {
1765         if ((unsigned int)size <= sysctl_optmem_max &&
1766             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1767                 void *mem;
1768                 /* First do the add, to avoid the race if kmalloc
1769                  * might sleep.
1770                  */
1771                 atomic_add(size, &sk->sk_omem_alloc);
1772                 mem = kmalloc(size, priority);
1773                 if (mem)
1774                         return mem;
1775                 atomic_sub(size, &sk->sk_omem_alloc);
1776         }
1777         return NULL;
1778 }
1779 EXPORT_SYMBOL(sock_kmalloc);
1780 
1781 /* Free an option memory block. Note, we actually want the inline
1782  * here as this allows gcc to detect the nullify and fold away the
1783  * condition entirely.
1784  */
1785 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1786                                   const bool nullify)
1787 {
1788         if (WARN_ON_ONCE(!mem))
1789                 return;
1790         if (nullify)
1791                 kzfree(mem);
1792         else
1793                 kfree(mem);
1794         atomic_sub(size, &sk->sk_omem_alloc);
1795 }
1796 
1797 void sock_kfree_s(struct sock *sk, void *mem, int size)
1798 {
1799         __sock_kfree_s(sk, mem, size, false);
1800 }
1801 EXPORT_SYMBOL(sock_kfree_s);
1802 
1803 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1804 {
1805         __sock_kfree_s(sk, mem, size, true);
1806 }
1807 EXPORT_SYMBOL(sock_kzfree_s);
1808 
1809 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1810    I think, these locks should be removed for datagram sockets.
1811  */
1812 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1813 {
1814         DEFINE_WAIT(wait);
1815 
1816         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1817         for (;;) {
1818                 if (!timeo)
1819                         break;
1820                 if (signal_pending(current))
1821                         break;
1822                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1823                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1824                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1825                         break;
1826                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1827                         break;
1828                 if (sk->sk_err)
1829                         break;
1830                 timeo = schedule_timeout(timeo);
1831         }
1832         finish_wait(sk_sleep(sk), &wait);
1833         return timeo;
1834 }
1835 
1836 
1837 /*
1838  *      Generic send/receive buffer handlers
1839  */
1840 
1841 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1842                                      unsigned long data_len, int noblock,
1843                                      int *errcode, int max_page_order)
1844 {
1845         struct sk_buff *skb;
1846         long timeo;
1847         int err;
1848 
1849         timeo = sock_sndtimeo(sk, noblock);
1850         for (;;) {
1851                 err = sock_error(sk);
1852                 if (err != 0)
1853                         goto failure;
1854 
1855                 err = -EPIPE;
1856                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1857                         goto failure;
1858 
1859                 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1860                         break;
1861 
1862                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1863                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1864                 err = -EAGAIN;
1865                 if (!timeo)
1866                         goto failure;
1867                 if (signal_pending(current))
1868                         goto interrupted;
1869                 timeo = sock_wait_for_wmem(sk, timeo);
1870         }
1871         skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1872                                    errcode, sk->sk_allocation);
1873         if (skb)
1874                 skb_set_owner_w(skb, sk);
1875         return skb;
1876 
1877 interrupted:
1878         err = sock_intr_errno(timeo);
1879 failure:
1880         *errcode = err;
1881         return NULL;
1882 }
1883 EXPORT_SYMBOL(sock_alloc_send_pskb);
1884 
1885 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1886                                     int noblock, int *errcode)
1887 {
1888         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1889 }
1890 EXPORT_SYMBOL(sock_alloc_send_skb);
1891 
1892 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1893                    struct sockcm_cookie *sockc)
1894 {
1895         struct cmsghdr *cmsg;
1896 
1897         for_each_cmsghdr(cmsg, msg) {
1898                 if (!CMSG_OK(msg, cmsg))
1899                         return -EINVAL;
1900                 if (cmsg->cmsg_level != SOL_SOCKET)
1901                         continue;
1902                 switch (cmsg->cmsg_type) {
1903                 case SO_MARK:
1904                         if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1905                                 return -EPERM;
1906                         if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1907                                 return -EINVAL;
1908                         sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1909                         break;
1910                 default:
1911                         return -EINVAL;
1912                 }
1913         }
1914         return 0;
1915 }
1916 EXPORT_SYMBOL(sock_cmsg_send);
1917 
1918 /* On 32bit arches, an skb frag is limited to 2^15 */
1919 #define SKB_FRAG_PAGE_ORDER     get_order(32768)
1920 
1921 /**
1922  * skb_page_frag_refill - check that a page_frag contains enough room
1923  * @sz: minimum size of the fragment we want to get
1924  * @pfrag: pointer to page_frag
1925  * @gfp: priority for memory allocation
1926  *
1927  * Note: While this allocator tries to use high order pages, there is
1928  * no guarantee that allocations succeed. Therefore, @sz MUST be
1929  * less or equal than PAGE_SIZE.
1930  */
1931 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1932 {
1933         if (pfrag->page) {
1934                 if (atomic_read(&pfrag->page->_count) == 1) {
1935                         pfrag->offset = 0;
1936                         return true;
1937                 }
1938                 if (pfrag->offset + sz <= pfrag->size)
1939                         return true;
1940                 put_page(pfrag->page);
1941         }
1942 
1943         pfrag->offset = 0;
1944         if (SKB_FRAG_PAGE_ORDER) {
1945                 /* Avoid direct reclaim but allow kswapd to wake */
1946                 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1947                                           __GFP_COMP | __GFP_NOWARN |
1948                                           __GFP_NORETRY,
1949                                           SKB_FRAG_PAGE_ORDER);
1950                 if (likely(pfrag->page)) {
1951                         pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1952                         return true;
1953                 }
1954         }
1955         pfrag->page = alloc_page(gfp);
1956         if (likely(pfrag->page)) {
1957                 pfrag->size = PAGE_SIZE;
1958                 return true;
1959         }
1960         return false;
1961 }
1962 EXPORT_SYMBOL(skb_page_frag_refill);
1963 
1964 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1965 {
1966         if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1967                 return true;
1968 
1969         sk_enter_memory_pressure(sk);
1970         sk_stream_moderate_sndbuf(sk);
1971         return false;
1972 }
1973 EXPORT_SYMBOL(sk_page_frag_refill);
1974 
1975 static void __lock_sock(struct sock *sk)
1976         __releases(&sk->sk_lock.slock)
1977         __acquires(&sk->sk_lock.slock)
1978 {
1979         DEFINE_WAIT(wait);
1980 
1981         for (;;) {
1982                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1983                                         TASK_UNINTERRUPTIBLE);
1984                 spin_unlock_bh(&sk->sk_lock.slock);
1985                 schedule();
1986                 spin_lock_bh(&sk->sk_lock.slock);
1987                 if (!sock_owned_by_user(sk))
1988                         break;
1989         }
1990         finish_wait(&sk->sk_lock.wq, &wait);
1991 }
1992 
1993 static void __release_sock(struct sock *sk)
1994         __releases(&sk->sk_lock.slock)
1995         __acquires(&sk->sk_lock.slock)
1996 {
1997         struct sk_buff *skb = sk->sk_backlog.head;
1998 
1999         do {
2000                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2001                 bh_unlock_sock(sk);
2002 
2003                 do {
2004                         struct sk_buff *next = skb->next;
2005 
2006                         prefetch(next);
2007                         WARN_ON_ONCE(skb_dst_is_noref(skb));
2008                         skb->next = NULL;
2009                         sk_backlog_rcv(sk, skb);
2010 
2011                         /*
2012                          * We are in process context here with softirqs
2013                          * disabled, use cond_resched_softirq() to preempt.
2014                          * This is safe to do because we've taken the backlog
2015                          * queue private:
2016                          */
2017                         cond_resched_softirq();
2018 
2019                         skb = next;
2020                 } while (skb != NULL);
2021 
2022                 bh_lock_sock(sk);
2023         } while ((skb = sk->sk_backlog.head) != NULL);
2024 
2025         /*
2026          * Doing the zeroing here guarantee we can not loop forever
2027          * while a wild producer attempts to flood us.
2028          */
2029         sk->sk_backlog.len = 0;
2030 }
2031 
2032 /**
2033  * sk_wait_data - wait for data to arrive at sk_receive_queue
2034  * @sk:    sock to wait on
2035  * @timeo: for how long
2036  * @skb:   last skb seen on sk_receive_queue
2037  *
2038  * Now socket state including sk->sk_err is changed only under lock,
2039  * hence we may omit checks after joining wait queue.
2040  * We check receive queue before schedule() only as optimization;
2041  * it is very likely that release_sock() added new data.
2042  */
2043 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2044 {
2045         int rc;
2046         DEFINE_WAIT(wait);
2047 
2048         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2049         sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2050         rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2051         sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2052         finish_wait(sk_sleep(sk), &wait);
2053         return rc;
2054 }
2055 EXPORT_SYMBOL(sk_wait_data);
2056 
2057 /**
2058  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2059  *      @sk: socket
2060  *      @size: memory size to allocate
2061  *      @kind: allocation type
2062  *
2063  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2064  *      rmem allocation. This function assumes that protocols which have
2065  *      memory_pressure use sk_wmem_queued as write buffer accounting.
2066  */
2067 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2068 {
2069         struct proto *prot = sk->sk_prot;
2070         int amt = sk_mem_pages(size);
2071         long allocated;
2072         int parent_status = UNDER_LIMIT;
2073 
2074         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2075 
2076         allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2077 
2078         /* Under limit. */
2079         if (parent_status == UNDER_LIMIT &&
2080                         allocated <= sk_prot_mem_limits(sk, 0)) {
2081                 sk_leave_memory_pressure(sk);
2082                 return 1;
2083         }
2084 
2085         /* Under pressure. (we or our parents) */
2086         if ((parent_status > SOFT_LIMIT) ||
2087                         allocated > sk_prot_mem_limits(sk, 1))
2088                 sk_enter_memory_pressure(sk);
2089 
2090         /* Over hard limit (we or our parents) */
2091         if ((parent_status == OVER_LIMIT) ||
2092                         (allocated > sk_prot_mem_limits(sk, 2)))
2093                 goto suppress_allocation;
2094 
2095         /* guarantee minimum buffer size under pressure */
2096         if (kind == SK_MEM_RECV) {
2097                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2098                         return 1;
2099 
2100         } else { /* SK_MEM_SEND */
2101                 if (sk->sk_type == SOCK_STREAM) {
2102                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2103                                 return 1;
2104                 } else if (atomic_read(&sk->sk_wmem_alloc) <
2105                            prot->sysctl_wmem[0])
2106                                 return 1;
2107         }
2108 
2109         if (sk_has_memory_pressure(sk)) {
2110                 int alloc;
2111 
2112                 if (!sk_under_memory_pressure(sk))
2113                         return 1;
2114                 alloc = sk_sockets_allocated_read_positive(sk);
2115                 if (sk_prot_mem_limits(sk, 2) > alloc *
2116                     sk_mem_pages(sk->sk_wmem_queued +
2117                                  atomic_read(&sk->sk_rmem_alloc) +
2118                                  sk->sk_forward_alloc))
2119                         return 1;
2120         }
2121 
2122 suppress_allocation:
2123 
2124         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2125                 sk_stream_moderate_sndbuf(sk);
2126 
2127                 /* Fail only if socket is _under_ its sndbuf.
2128                  * In this case we cannot block, so that we have to fail.
2129                  */
2130                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2131                         return 1;
2132         }
2133 
2134         trace_sock_exceed_buf_limit(sk, prot, allocated);
2135 
2136         /* Alas. Undo changes. */
2137         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2138 
2139         sk_memory_allocated_sub(sk, amt);
2140 
2141         return 0;
2142 }
2143 EXPORT_SYMBOL(__sk_mem_schedule);
2144 
2145 /**
2146  *      __sk_mem_reclaim - reclaim memory_allocated
2147  *      @sk: socket
2148  *      @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2149  */
2150 void __sk_mem_reclaim(struct sock *sk, int amount)
2151 {
2152         amount >>= SK_MEM_QUANTUM_SHIFT;
2153         sk_memory_allocated_sub(sk, amount);
2154         sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2155 
2156         if (sk_under_memory_pressure(sk) &&
2157             (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2158                 sk_leave_memory_pressure(sk);
2159 }
2160 EXPORT_SYMBOL(__sk_mem_reclaim);
2161 
2162 
2163 /*
2164  * Set of default routines for initialising struct proto_ops when
2165  * the protocol does not support a particular function. In certain
2166  * cases where it makes no sense for a protocol to have a "do nothing"
2167  * function, some default processing is provided.
2168  */
2169 
2170 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2171 {
2172         return -EOPNOTSUPP;
2173 }
2174 EXPORT_SYMBOL(sock_no_bind);
2175 
2176 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2177                     int len, int flags)
2178 {
2179         return -EOPNOTSUPP;
2180 }
2181 EXPORT_SYMBOL(sock_no_connect);
2182 
2183 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2184 {
2185         return -EOPNOTSUPP;
2186 }
2187 EXPORT_SYMBOL(sock_no_socketpair);
2188 
2189 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2190 {
2191         return -EOPNOTSUPP;
2192 }
2193 EXPORT_SYMBOL(sock_no_accept);
2194 
2195 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2196                     int *len, int peer)
2197 {
2198         return -EOPNOTSUPP;
2199 }
2200 EXPORT_SYMBOL(sock_no_getname);
2201 
2202 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2203 {
2204         return 0;
2205 }
2206 EXPORT_SYMBOL(sock_no_poll);
2207 
2208 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2209 {
2210         return -EOPNOTSUPP;
2211 }
2212 EXPORT_SYMBOL(sock_no_ioctl);
2213 
2214 int sock_no_listen(struct socket *sock, int backlog)
2215 {
2216         return -EOPNOTSUPP;
2217 }
2218 EXPORT_SYMBOL(sock_no_listen);
2219 
2220 int sock_no_shutdown(struct socket *sock, int how)
2221 {
2222         return -EOPNOTSUPP;
2223 }
2224 EXPORT_SYMBOL(sock_no_shutdown);
2225 
2226 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2227                     char __user *optval, unsigned int optlen)
2228 {
2229         return -EOPNOTSUPP;
2230 }
2231 EXPORT_SYMBOL(sock_no_setsockopt);
2232 
2233 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2234                     char __user *optval, int __user *optlen)
2235 {
2236         return -EOPNOTSUPP;
2237 }
2238 EXPORT_SYMBOL(sock_no_getsockopt);
2239 
2240 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2241 {
2242         return -EOPNOTSUPP;
2243 }
2244 EXPORT_SYMBOL(sock_no_sendmsg);
2245 
2246 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2247                     int flags)
2248 {
2249         return -EOPNOTSUPP;
2250 }
2251 EXPORT_SYMBOL(sock_no_recvmsg);
2252 
2253 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2254 {
2255         /* Mirror missing mmap method error code */
2256         return -ENODEV;
2257 }
2258 EXPORT_SYMBOL(sock_no_mmap);
2259 
2260 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2261 {
2262         ssize_t res;
2263         struct msghdr msg = {.msg_flags = flags};
2264         struct kvec iov;
2265         char *kaddr = kmap(page);
2266         iov.iov_base = kaddr + offset;
2267         iov.iov_len = size;
2268         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2269         kunmap(page);
2270         return res;
2271 }
2272 EXPORT_SYMBOL(sock_no_sendpage);
2273 
2274 /*
2275  *      Default Socket Callbacks
2276  */
2277 
2278 static void sock_def_wakeup(struct sock *sk)
2279 {
2280         struct socket_wq *wq;
2281 
2282         rcu_read_lock();
2283         wq = rcu_dereference(sk->sk_wq);
2284         if (wq_has_sleeper(wq))
2285                 wake_up_interruptible_all(&wq->wait);
2286         rcu_read_unlock();
2287 }
2288 
2289 static void sock_def_error_report(struct sock *sk)
2290 {
2291         struct socket_wq *wq;
2292 
2293         rcu_read_lock();
2294         wq = rcu_dereference(sk->sk_wq);
2295         if (wq_has_sleeper(wq))
2296                 wake_up_interruptible_poll(&wq->wait, POLLERR);
2297         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2298         rcu_read_unlock();
2299 }
2300 
2301 static void sock_def_readable(struct sock *sk)
2302 {
2303         struct socket_wq *wq;
2304 
2305         rcu_read_lock();
2306         wq = rcu_dereference(sk->sk_wq);
2307         if (wq_has_sleeper(wq))
2308                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2309                                                 POLLRDNORM | POLLRDBAND);
2310         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2311         rcu_read_unlock();
2312 }
2313 
2314 static void sock_def_write_space(struct sock *sk)
2315 {
2316         struct socket_wq *wq;
2317 
2318         rcu_read_lock();
2319 
2320         /* Do not wake up a writer until he can make "significant"
2321          * progress.  --DaveM
2322          */
2323         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2324                 wq = rcu_dereference(sk->sk_wq);
2325                 if (wq_has_sleeper(wq))
2326                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2327                                                 POLLWRNORM | POLLWRBAND);
2328 
2329                 /* Should agree with poll, otherwise some programs break */
2330                 if (sock_writeable(sk))
2331                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2332         }
2333 
2334         rcu_read_unlock();
2335 }
2336 
2337 static void sock_def_destruct(struct sock *sk)
2338 {
2339 }
2340 
2341 void sk_send_sigurg(struct sock *sk)
2342 {
2343         if (sk->sk_socket && sk->sk_socket->file)
2344                 if (send_sigurg(&sk->sk_socket->file->f_owner))
2345                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2346 }
2347 EXPORT_SYMBOL(sk_send_sigurg);
2348 
2349 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2350                     unsigned long expires)
2351 {
2352         if (!mod_timer(timer, expires))
2353                 sock_hold(sk);
2354 }
2355 EXPORT_SYMBOL(sk_reset_timer);
2356 
2357 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2358 {
2359         if (del_timer(timer))
2360                 __sock_put(sk);
2361 }
2362 EXPORT_SYMBOL(sk_stop_timer);
2363 
2364 void sock_init_data(struct socket *sock, struct sock *sk)
2365 {
2366         skb_queue_head_init(&sk->sk_receive_queue);
2367         skb_queue_head_init(&sk->sk_write_queue);
2368         skb_queue_head_init(&sk->sk_error_queue);
2369 
2370         sk->sk_send_head        =       NULL;
2371 
2372         init_timer(&sk->sk_timer);
2373 
2374         sk->sk_allocation       =       GFP_KERNEL;
2375         sk->sk_rcvbuf           =       sysctl_rmem_default;
2376         sk->sk_sndbuf           =       sysctl_wmem_default;
2377         sk->sk_state            =       TCP_CLOSE;
2378         sk_set_socket(sk, sock);
2379 
2380         sock_set_flag(sk, SOCK_ZAPPED);
2381 
2382         if (sock) {
2383                 sk->sk_type     =       sock->type;
2384                 sk->sk_wq       =       sock->wq;
2385                 sock->sk        =       sk;
2386         } else
2387                 sk->sk_wq       =       NULL;
2388 
2389         rwlock_init(&sk->sk_callback_lock);
2390         lockdep_set_class_and_name(&sk->sk_callback_lock,
2391                         af_callback_keys + sk->sk_family,
2392                         af_family_clock_key_strings[sk->sk_family]);
2393 
2394         sk->sk_state_change     =       sock_def_wakeup;
2395         sk->sk_data_ready       =       sock_def_readable;
2396         sk->sk_write_space      =       sock_def_write_space;
2397         sk->sk_error_report     =       sock_def_error_report;
2398         sk->sk_destruct         =       sock_def_destruct;
2399 
2400         sk->sk_frag.page        =       NULL;
2401         sk->sk_frag.offset      =       0;
2402         sk->sk_peek_off         =       -1;
2403 
2404         sk->sk_peer_pid         =       NULL;
2405         sk->sk_peer_cred        =       NULL;
2406         sk->sk_write_pending    =       0;
2407         sk->sk_rcvlowat         =       1;
2408         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2409         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2410 
2411         sk->sk_stamp = ktime_set(-1L, 0);
2412 
2413 #ifdef CONFIG_NET_RX_BUSY_POLL
2414         sk->sk_napi_id          =       0;
2415         sk->sk_ll_usec          =       sysctl_net_busy_read;
2416 #endif
2417 
2418         sk->sk_max_pacing_rate = ~0U;
2419         sk->sk_pacing_rate = ~0U;
2420         sk->sk_incoming_cpu = -1;
2421         /*
2422          * Before updating sk_refcnt, we must commit prior changes to memory
2423          * (Documentation/RCU/rculist_nulls.txt for details)
2424          */
2425         smp_wmb();
2426         atomic_set(&sk->sk_refcnt, 1);
2427         atomic_set(&sk->sk_drops, 0);
2428 }
2429 EXPORT_SYMBOL(sock_init_data);
2430 
2431 void lock_sock_nested(struct sock *sk, int subclass)
2432 {
2433         might_sleep();
2434         spin_lock_bh(&sk->sk_lock.slock);
2435         if (sk->sk_lock.owned)
2436                 __lock_sock(sk);
2437         sk->sk_lock.owned = 1;
2438         spin_unlock(&sk->sk_lock.slock);
2439         /*
2440          * The sk_lock has mutex_lock() semantics here:
2441          */
2442         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2443         local_bh_enable();
2444 }
2445 EXPORT_SYMBOL(lock_sock_nested);
2446 
2447 void release_sock(struct sock *sk)
2448 {
2449         /*
2450          * The sk_lock has mutex_unlock() semantics:
2451          */
2452         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2453 
2454         spin_lock_bh(&sk->sk_lock.slock);
2455         if (sk->sk_backlog.tail)
2456                 __release_sock(sk);
2457 
2458         /* Warning : release_cb() might need to release sk ownership,
2459          * ie call sock_release_ownership(sk) before us.
2460          */
2461         if (sk->sk_prot->release_cb)
2462                 sk->sk_prot->release_cb(sk);
2463 
2464         sock_release_ownership(sk);
2465         if (waitqueue_active(&sk->sk_lock.wq))
2466                 wake_up(&sk->sk_lock.wq);
2467         spin_unlock_bh(&sk->sk_lock.slock);
2468 }
2469 EXPORT_SYMBOL(release_sock);
2470 
2471 /**
2472  * lock_sock_fast - fast version of lock_sock
2473  * @sk: socket
2474  *
2475  * This version should be used for very small section, where process wont block
2476  * return false if fast path is taken
2477  *   sk_lock.slock locked, owned = 0, BH disabled
2478  * return true if slow path is taken
2479  *   sk_lock.slock unlocked, owned = 1, BH enabled
2480  */
2481 bool lock_sock_fast(struct sock *sk)
2482 {
2483         might_sleep();
2484         spin_lock_bh(&sk->sk_lock.slock);
2485 
2486         if (!sk->sk_lock.owned)
2487                 /*
2488                  * Note : We must disable BH
2489                  */
2490                 return false;
2491 
2492         __lock_sock(sk);
2493         sk->sk_lock.owned = 1;
2494         spin_unlock(&sk->sk_lock.slock);
2495         /*
2496          * The sk_lock has mutex_lock() semantics here:
2497          */
2498         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2499         local_bh_enable();
2500         return true;
2501 }
2502 EXPORT_SYMBOL(lock_sock_fast);
2503 
2504 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2505 {
2506         struct timeval tv;
2507         if (!sock_flag(sk, SOCK_TIMESTAMP))
2508                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2509         tv = ktime_to_timeval(sk->sk_stamp);
2510         if (tv.tv_sec == -1)
2511                 return -ENOENT;
2512         if (tv.tv_sec == 0) {
2513                 sk->sk_stamp = ktime_get_real();
2514                 tv = ktime_to_timeval(sk->sk_stamp);
2515         }
2516         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2517 }
2518 EXPORT_SYMBOL(sock_get_timestamp);
2519 
2520 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2521 {
2522         struct timespec ts;
2523         if (!sock_flag(sk, SOCK_TIMESTAMP))
2524                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2525         ts = ktime_to_timespec(sk->sk_stamp);
2526         if (ts.tv_sec == -1)
2527                 return -ENOENT;
2528         if (ts.tv_sec == 0) {
2529                 sk->sk_stamp = ktime_get_real();
2530                 ts = ktime_to_timespec(sk->sk_stamp);
2531         }
2532         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2533 }
2534 EXPORT_SYMBOL(sock_get_timestampns);
2535 
2536 void sock_enable_timestamp(struct sock *sk, int flag)
2537 {
2538         if (!sock_flag(sk, flag)) {
2539                 unsigned long previous_flags = sk->sk_flags;
2540 
2541                 sock_set_flag(sk, flag);
2542                 /*
2543                  * we just set one of the two flags which require net
2544                  * time stamping, but time stamping might have been on
2545                  * already because of the other one
2546                  */
2547                 if (sock_needs_netstamp(sk) &&
2548                     !(previous_flags & SK_FLAGS_TIMESTAMP))
2549                         net_enable_timestamp();
2550         }
2551 }
2552 
2553 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2554                        int level, int type)
2555 {
2556         struct sock_exterr_skb *serr;
2557         struct sk_buff *skb;
2558         int copied, err;
2559 
2560         err = -EAGAIN;
2561         skb = sock_dequeue_err_skb(sk);
2562         if (skb == NULL)
2563                 goto out;
2564 
2565         copied = skb->len;
2566         if (copied > len) {
2567                 msg->msg_flags |= MSG_TRUNC;
2568                 copied = len;
2569         }
2570         err = skb_copy_datagram_msg(skb, 0, msg, copied);
2571         if (err)
2572                 goto out_free_skb;
2573 
2574         sock_recv_timestamp(msg, sk, skb);
2575 
2576         serr = SKB_EXT_ERR(skb);
2577         put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2578 
2579         msg->msg_flags |= MSG_ERRQUEUE;
2580         err = copied;
2581 
2582 out_free_skb:
2583         kfree_skb(skb);
2584 out:
2585         return err;
2586 }
2587 EXPORT_SYMBOL(sock_recv_errqueue);
2588 
2589 /*
2590  *      Get a socket option on an socket.
2591  *
2592  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2593  *      asynchronous errors should be reported by getsockopt. We assume
2594  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2595  */
2596 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2597                            char __user *optval, int __user *optlen)
2598 {
2599         struct sock *sk = sock->sk;
2600 
2601         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2602 }
2603 EXPORT_SYMBOL(sock_common_getsockopt);
2604 
2605 #ifdef CONFIG_COMPAT
2606 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2607                                   char __user *optval, int __user *optlen)
2608 {
2609         struct sock *sk = sock->sk;
2610 
2611         if (sk->sk_prot->compat_getsockopt != NULL)
2612                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2613                                                       optval, optlen);
2614         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2615 }
2616 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2617 #endif
2618 
2619 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2620                         int flags)
2621 {
2622         struct sock *sk = sock->sk;
2623         int addr_len = 0;
2624         int err;
2625 
2626         err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2627                                    flags & ~MSG_DONTWAIT, &addr_len);
2628         if (err >= 0)
2629                 msg->msg_namelen = addr_len;
2630         return err;
2631 }
2632 EXPORT_SYMBOL(sock_common_recvmsg);
2633 
2634 /*
2635  *      Set socket options on an inet socket.
2636  */
2637 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2638                            char __user *optval, unsigned int optlen)
2639 {
2640         struct sock *sk = sock->sk;
2641 
2642         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2643 }
2644 EXPORT_SYMBOL(sock_common_setsockopt);
2645 
2646 #ifdef CONFIG_COMPAT
2647 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2648                                   char __user *optval, unsigned int optlen)
2649 {
2650         struct sock *sk = sock->sk;
2651 
2652         if (sk->sk_prot->compat_setsockopt != NULL)
2653                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2654                                                       optval, optlen);
2655         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2656 }
2657 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2658 #endif
2659 
2660 void sk_common_release(struct sock *sk)
2661 {
2662         if (sk->sk_prot->destroy)
2663                 sk->sk_prot->destroy(sk);
2664 
2665         /*
2666          * Observation: when sock_common_release is called, processes have
2667          * no access to socket. But net still has.
2668          * Step one, detach it from networking:
2669          *
2670          * A. Remove from hash tables.
2671          */
2672 
2673         sk->sk_prot->unhash(sk);
2674 
2675         /*
2676          * In this point socket cannot receive new packets, but it is possible
2677          * that some packets are in flight because some CPU runs receiver and
2678          * did hash table lookup before we unhashed socket. They will achieve
2679          * receive queue and will be purged by socket destructor.
2680          *
2681          * Also we still have packets pending on receive queue and probably,
2682          * our own packets waiting in device queues. sock_destroy will drain
2683          * receive queue, but transmitted packets will delay socket destruction
2684          * until the last reference will be released.
2685          */
2686 
2687         sock_orphan(sk);
2688 
2689         xfrm_sk_free_policy(sk);
2690 
2691         sk_refcnt_debug_release(sk);
2692 
2693         if (sk->sk_frag.page) {
2694                 put_page(sk->sk_frag.page);
2695                 sk->sk_frag.page = NULL;
2696         }
2697 
2698         sock_put(sk);
2699 }
2700 EXPORT_SYMBOL(sk_common_release);
2701 
2702 #ifdef CONFIG_PROC_FS
2703 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2704 struct prot_inuse {
2705         int val[PROTO_INUSE_NR];
2706 };
2707 
2708 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2709 
2710 #ifdef CONFIG_NET_NS
2711 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2712 {
2713         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2714 }
2715 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2716 
2717 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2718 {
2719         int cpu, idx = prot->inuse_idx;
2720         int res = 0;
2721 
2722         for_each_possible_cpu(cpu)
2723                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2724 
2725         return res >= 0 ? res : 0;
2726 }
2727 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2728 
2729 static int __net_init sock_inuse_init_net(struct net *net)
2730 {
2731         net->core.inuse = alloc_percpu(struct prot_inuse);
2732         return net->core.inuse ? 0 : -ENOMEM;
2733 }
2734 
2735 static void __net_exit sock_inuse_exit_net(struct net *net)
2736 {
2737         free_percpu(net->core.inuse);
2738 }
2739 
2740 static struct pernet_operations net_inuse_ops = {
2741         .init = sock_inuse_init_net,
2742         .exit = sock_inuse_exit_net,
2743 };
2744 
2745 static __init int net_inuse_init(void)
2746 {
2747         if (register_pernet_subsys(&net_inuse_ops))
2748                 panic("Cannot initialize net inuse counters");
2749 
2750         return 0;
2751 }
2752 
2753 core_initcall(net_inuse_init);
2754 #else
2755 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2756 
2757 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2758 {
2759         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2760 }
2761 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2762 
2763 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2764 {
2765         int cpu, idx = prot->inuse_idx;
2766         int res = 0;
2767 
2768         for_each_possible_cpu(cpu)
2769                 res += per_cpu(prot_inuse, cpu).val[idx];
2770 
2771         return res >= 0 ? res : 0;
2772 }
2773 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2774 #endif
2775 
2776 static void assign_proto_idx(struct proto *prot)
2777 {
2778         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2779 
2780         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2781                 pr_err("PROTO_INUSE_NR exhausted\n");
2782                 return;
2783         }
2784 
2785         set_bit(prot->inuse_idx, proto_inuse_idx);
2786 }
2787 
2788 static void release_proto_idx(struct proto *prot)
2789 {
2790         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2791                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2792 }
2793 #else
2794 static inline void assign_proto_idx(struct proto *prot)
2795 {
2796 }
2797 
2798 static inline void release_proto_idx(struct proto *prot)
2799 {
2800 }
2801 #endif
2802 
2803 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2804 {
2805         if (!rsk_prot)
2806                 return;
2807         kfree(rsk_prot->slab_name);
2808         rsk_prot->slab_name = NULL;
2809         kmem_cache_destroy(rsk_prot->slab);
2810         rsk_prot->slab = NULL;
2811 }
2812 
2813 static int req_prot_init(const struct proto *prot)
2814 {
2815         struct request_sock_ops *rsk_prot = prot->rsk_prot;
2816 
2817         if (!rsk_prot)
2818                 return 0;
2819 
2820         rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2821                                         prot->name);
2822         if (!rsk_prot->slab_name)
2823                 return -ENOMEM;
2824 
2825         rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2826                                            rsk_prot->obj_size, 0,
2827                                            prot->slab_flags, NULL);
2828 
2829         if (!rsk_prot->slab) {
2830                 pr_crit("%s: Can't create request sock SLAB cache!\n",
2831                         prot->name);
2832                 return -ENOMEM;
2833         }
2834         return 0;
2835 }
2836 
2837 int proto_register(struct proto *prot, int alloc_slab)
2838 {
2839         if (alloc_slab) {
2840                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2841                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2842                                         NULL);
2843 
2844                 if (prot->slab == NULL) {
2845                         pr_crit("%s: Can't create sock SLAB cache!\n",
2846                                 prot->name);
2847                         goto out;
2848                 }
2849 
2850                 if (req_prot_init(prot))
2851                         goto out_free_request_sock_slab;
2852 
2853                 if (prot->twsk_prot != NULL) {
2854                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2855 
2856                         if (prot->twsk_prot->twsk_slab_name == NULL)
2857                                 goto out_free_request_sock_slab;
2858 
2859                         prot->twsk_prot->twsk_slab =
2860                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2861                                                   prot->twsk_prot->twsk_obj_size,
2862                                                   0,
2863                                                   prot->slab_flags,
2864                                                   NULL);
2865                         if (prot->twsk_prot->twsk_slab == NULL)
2866                                 goto out_free_timewait_sock_slab_name;
2867                 }
2868         }
2869 
2870         mutex_lock(&proto_list_mutex);
2871         list_add(&prot->node, &proto_list);
2872         assign_proto_idx(prot);
2873         mutex_unlock(&proto_list_mutex);
2874         return 0;
2875 
2876 out_free_timewait_sock_slab_name:
2877         kfree(prot->twsk_prot->twsk_slab_name);
2878 out_free_request_sock_slab:
2879         req_prot_cleanup(prot->rsk_prot);
2880 
2881         kmem_cache_destroy(prot->slab);
2882         prot->slab = NULL;
2883 out:
2884         return -ENOBUFS;
2885 }
2886 EXPORT_SYMBOL(proto_register);
2887 
2888 void proto_unregister(struct proto *prot)
2889 {
2890         mutex_lock(&proto_list_mutex);
2891         release_proto_idx(prot);
2892         list_del(&prot->node);
2893         mutex_unlock(&proto_list_mutex);
2894 
2895         kmem_cache_destroy(prot->slab);
2896         prot->slab = NULL;
2897 
2898         req_prot_cleanup(prot->rsk_prot);
2899 
2900         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2901                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2902                 kfree(prot->twsk_prot->twsk_slab_name);
2903                 prot->twsk_prot->twsk_slab = NULL;
2904         }
2905 }
2906 EXPORT_SYMBOL(proto_unregister);
2907 
2908 #ifdef CONFIG_PROC_FS
2909 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2910         __acquires(proto_list_mutex)
2911 {
2912         mutex_lock(&proto_list_mutex);
2913         return seq_list_start_head(&proto_list, *pos);
2914 }
2915 
2916 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2917 {
2918         return seq_list_next(v, &proto_list, pos);
2919 }
2920 
2921 static void proto_seq_stop(struct seq_file *seq, void *v)
2922         __releases(proto_list_mutex)
2923 {
2924         mutex_unlock(&proto_list_mutex);
2925 }
2926 
2927 static char proto_method_implemented(const void *method)
2928 {
2929         return method == NULL ? 'n' : 'y';
2930 }
2931 static long sock_prot_memory_allocated(struct proto *proto)
2932 {
2933         return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2934 }
2935 
2936 static char *sock_prot_memory_pressure(struct proto *proto)
2937 {
2938         return proto->memory_pressure != NULL ?
2939         proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2940 }
2941 
2942 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2943 {
2944 
2945         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2946                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2947                    proto->name,
2948                    proto->obj_size,
2949                    sock_prot_inuse_get(seq_file_net(seq), proto),
2950                    sock_prot_memory_allocated(proto),
2951                    sock_prot_memory_pressure(proto),
2952                    proto->max_header,
2953                    proto->slab == NULL ? "no" : "yes",
2954                    module_name(proto->owner),
2955                    proto_method_implemented(proto->close),
2956                    proto_method_implemented(proto->connect),
2957                    proto_method_implemented(proto->disconnect),
2958                    proto_method_implemented(proto->accept),
2959                    proto_method_implemented(proto->ioctl),
2960                    proto_method_implemented(proto->init),
2961                    proto_method_implemented(proto->destroy),
2962                    proto_method_implemented(proto->shutdown),
2963                    proto_method_implemented(proto->setsockopt),
2964                    proto_method_implemented(proto->getsockopt),
2965                    proto_method_implemented(proto->sendmsg),
2966                    proto_method_implemented(proto->recvmsg),
2967                    proto_method_implemented(proto->sendpage),
2968                    proto_method_implemented(proto->bind),
2969                    proto_method_implemented(proto->backlog_rcv),
2970                    proto_method_implemented(proto->hash),
2971                    proto_method_implemented(proto->unhash),
2972                    proto_method_implemented(proto->get_port),
2973                    proto_method_implemented(proto->enter_memory_pressure));
2974 }
2975 
2976 static int proto_seq_show(struct seq_file *seq, void *v)
2977 {
2978         if (v == &proto_list)
2979                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2980                            "protocol",
2981                            "size",
2982                            "sockets",
2983                            "memory",
2984                            "press",
2985                            "maxhdr",
2986                            "slab",
2987                            "module",
2988                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2989         else
2990                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2991         return 0;
2992 }
2993 
2994 static const struct seq_operations proto_seq_ops = {
2995         .start  = proto_seq_start,
2996         .next   = proto_seq_next,
2997         .stop   = proto_seq_stop,
2998         .show   = proto_seq_show,
2999 };
3000 
3001 static int proto_seq_open(struct inode *inode, struct file *file)
3002 {
3003         return seq_open_net(inode, file, &proto_seq_ops,
3004                             sizeof(struct seq_net_private));
3005 }
3006 
3007 static const struct file_operations proto_seq_fops = {
3008         .owner          = THIS_MODULE,
3009         .open           = proto_seq_open,
3010         .read           = seq_read,
3011         .llseek         = seq_lseek,
3012         .release        = seq_release_net,
3013 };
3014 
3015 static __net_init int proto_init_net(struct net *net)
3016 {
3017         if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3018                 return -ENOMEM;
3019 
3020         return 0;
3021 }
3022 
3023 static __net_exit void proto_exit_net(struct net *net)
3024 {
3025         remove_proc_entry("protocols", net->proc_net);
3026 }
3027 
3028 
3029 static __net_initdata struct pernet_operations proto_net_ops = {
3030         .init = proto_init_net,
3031         .exit = proto_exit_net,
3032 };
3033 
3034 static int __init proto_init(void)
3035 {
3036         return register_pernet_subsys(&proto_net_ops);
3037 }
3038 
3039 subsys_initcall(proto_init);
3040 
3041 #endif /* PROC_FS */
3042 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us