~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/core/sock.c

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
135 #include <linux/filter.h>
136 
137 #include <trace/events/sock.h>
138 
139 #ifdef CONFIG_INET
140 #include <net/tcp.h>
141 #endif
142 
143 #include <net/busy_poll.h>
144 
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
147 
148 /**
149  * sk_ns_capable - General socket capability test
150  * @sk: Socket to use a capability on or through
151  * @user_ns: The user namespace of the capability to use
152  * @cap: The capability to use
153  *
154  * Test to see if the opener of the socket had when the socket was
155  * created and the current process has the capability @cap in the user
156  * namespace @user_ns.
157  */
158 bool sk_ns_capable(const struct sock *sk,
159                    struct user_namespace *user_ns, int cap)
160 {
161         return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
162                 ns_capable(user_ns, cap);
163 }
164 EXPORT_SYMBOL(sk_ns_capable);
165 
166 /**
167  * sk_capable - Socket global capability test
168  * @sk: Socket to use a capability on or through
169  * @cap: The global capbility to use
170  *
171  * Test to see if the opener of the socket had when the socket was
172  * created and the current process has the capability @cap in all user
173  * namespaces.
174  */
175 bool sk_capable(const struct sock *sk, int cap)
176 {
177         return sk_ns_capable(sk, &init_user_ns, cap);
178 }
179 EXPORT_SYMBOL(sk_capable);
180 
181 /**
182  * sk_net_capable - Network namespace socket capability test
183  * @sk: Socket to use a capability on or through
184  * @cap: The capability to use
185  *
186  * Test to see if the opener of the socket had when the socke was created
187  * and the current process has the capability @cap over the network namespace
188  * the socket is a member of.
189  */
190 bool sk_net_capable(const struct sock *sk, int cap)
191 {
192         return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
193 }
194 EXPORT_SYMBOL(sk_net_capable);
195 
196 
197 #ifdef CONFIG_MEMCG_KMEM
198 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
199 {
200         struct proto *proto;
201         int ret = 0;
202 
203         mutex_lock(&proto_list_mutex);
204         list_for_each_entry(proto, &proto_list, node) {
205                 if (proto->init_cgroup) {
206                         ret = proto->init_cgroup(memcg, ss);
207                         if (ret)
208                                 goto out;
209                 }
210         }
211 
212         mutex_unlock(&proto_list_mutex);
213         return ret;
214 out:
215         list_for_each_entry_continue_reverse(proto, &proto_list, node)
216                 if (proto->destroy_cgroup)
217                         proto->destroy_cgroup(memcg);
218         mutex_unlock(&proto_list_mutex);
219         return ret;
220 }
221 
222 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
223 {
224         struct proto *proto;
225 
226         mutex_lock(&proto_list_mutex);
227         list_for_each_entry_reverse(proto, &proto_list, node)
228                 if (proto->destroy_cgroup)
229                         proto->destroy_cgroup(memcg);
230         mutex_unlock(&proto_list_mutex);
231 }
232 #endif
233 
234 /*
235  * Each address family might have different locking rules, so we have
236  * one slock key per address family:
237  */
238 static struct lock_class_key af_family_keys[AF_MAX];
239 static struct lock_class_key af_family_slock_keys[AF_MAX];
240 
241 #if defined(CONFIG_MEMCG_KMEM)
242 struct static_key memcg_socket_limit_enabled;
243 EXPORT_SYMBOL(memcg_socket_limit_enabled);
244 #endif
245 
246 /*
247  * Make lock validator output more readable. (we pre-construct these
248  * strings build-time, so that runtime initialization of socket
249  * locks is fast):
250  */
251 static const char *const af_family_key_strings[AF_MAX+1] = {
252   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
253   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
254   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
255   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
256   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
257   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
258   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
259   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
260   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
261   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
262   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
263   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
264   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
265   "sk_lock-AF_NFC"   , "sk_lock-AF_VSOCK"    , "sk_lock-AF_MAX"
266 };
267 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
268   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
269   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
270   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
271   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
272   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
273   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
274   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
275   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
276   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
277   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
278   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
279   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
280   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
281   "slock-AF_NFC"   , "slock-AF_VSOCK"    ,"slock-AF_MAX"
282 };
283 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
284   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
285   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
286   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
287   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
288   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
289   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
290   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
291   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
292   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
293   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
294   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
295   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
296   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
297   "clock-AF_NFC"   , "clock-AF_VSOCK"    , "clock-AF_MAX"
298 };
299 
300 /*
301  * sk_callback_lock locking rules are per-address-family,
302  * so split the lock classes by using a per-AF key:
303  */
304 static struct lock_class_key af_callback_keys[AF_MAX];
305 
306 /* Take into consideration the size of the struct sk_buff overhead in the
307  * determination of these values, since that is non-constant across
308  * platforms.  This makes socket queueing behavior and performance
309  * not depend upon such differences.
310  */
311 #define _SK_MEM_PACKETS         256
312 #define _SK_MEM_OVERHEAD        SKB_TRUESIZE(256)
313 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
314 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 
316 /* Run time adjustable parameters. */
317 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
318 EXPORT_SYMBOL(sysctl_wmem_max);
319 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
320 EXPORT_SYMBOL(sysctl_rmem_max);
321 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
322 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
323 
324 /* Maximal space eaten by iovec or ancillary data plus some space */
325 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
326 EXPORT_SYMBOL(sysctl_optmem_max);
327 
328 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
329 EXPORT_SYMBOL_GPL(memalloc_socks);
330 
331 /**
332  * sk_set_memalloc - sets %SOCK_MEMALLOC
333  * @sk: socket to set it on
334  *
335  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
336  * It's the responsibility of the admin to adjust min_free_kbytes
337  * to meet the requirements
338  */
339 void sk_set_memalloc(struct sock *sk)
340 {
341         sock_set_flag(sk, SOCK_MEMALLOC);
342         sk->sk_allocation |= __GFP_MEMALLOC;
343         static_key_slow_inc(&memalloc_socks);
344 }
345 EXPORT_SYMBOL_GPL(sk_set_memalloc);
346 
347 void sk_clear_memalloc(struct sock *sk)
348 {
349         sock_reset_flag(sk, SOCK_MEMALLOC);
350         sk->sk_allocation &= ~__GFP_MEMALLOC;
351         static_key_slow_dec(&memalloc_socks);
352 
353         /*
354          * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
355          * progress of swapping. However, if SOCK_MEMALLOC is cleared while
356          * it has rmem allocations there is a risk that the user of the
357          * socket cannot make forward progress due to exceeding the rmem
358          * limits. By rights, sk_clear_memalloc() should only be called
359          * on sockets being torn down but warn and reset the accounting if
360          * that assumption breaks.
361          */
362         if (WARN_ON(sk->sk_forward_alloc))
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 void sock_disable_timestamp(struct sock *sk, unsigned long flags)
426 {
427         if (sk->sk_flags & flags) {
428                 sk->sk_flags &= ~flags;
429                 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
430                         net_disable_timestamp();
431         }
432 }
433 
434 
435 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
436 {
437         int err;
438         int skb_len;
439         unsigned long flags;
440         struct sk_buff_head *list = &sk->sk_receive_queue;
441 
442         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
443                 atomic_inc(&sk->sk_drops);
444                 trace_sock_rcvqueue_full(sk, skb);
445                 return -ENOMEM;
446         }
447 
448         err = sk_filter(sk, skb);
449         if (err)
450                 return err;
451 
452         if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
453                 atomic_inc(&sk->sk_drops);
454                 return -ENOBUFS;
455         }
456 
457         skb->dev = NULL;
458         skb_set_owner_r(skb, sk);
459 
460         /* Cache the SKB length before we tack it onto the receive
461          * queue.  Once it is added it no longer belongs to us and
462          * may be freed by other threads of control pulling packets
463          * from the queue.
464          */
465         skb_len = skb->len;
466 
467         /* we escape from rcu protected region, make sure we dont leak
468          * a norefcounted dst
469          */
470         skb_dst_force(skb);
471 
472         spin_lock_irqsave(&list->lock, flags);
473         skb->dropcount = atomic_read(&sk->sk_drops);
474         __skb_queue_tail(list, skb);
475         spin_unlock_irqrestore(&list->lock, flags);
476 
477         if (!sock_flag(sk, SOCK_DEAD))
478                 sk->sk_data_ready(sk, skb_len);
479         return 0;
480 }
481 EXPORT_SYMBOL(sock_queue_rcv_skb);
482 
483 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
484 {
485         int rc = NET_RX_SUCCESS;
486 
487         if (sk_filter(sk, skb))
488                 goto discard_and_relse;
489 
490         skb->dev = NULL;
491 
492         if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
493                 atomic_inc(&sk->sk_drops);
494                 goto discard_and_relse;
495         }
496         if (nested)
497                 bh_lock_sock_nested(sk);
498         else
499                 bh_lock_sock(sk);
500         if (!sock_owned_by_user(sk)) {
501                 /*
502                  * trylock + unlock semantics:
503                  */
504                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
505 
506                 rc = sk_backlog_rcv(sk, skb);
507 
508                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
509         } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
510                 bh_unlock_sock(sk);
511                 atomic_inc(&sk->sk_drops);
512                 goto discard_and_relse;
513         }
514 
515         bh_unlock_sock(sk);
516 out:
517         sock_put(sk);
518         return rc;
519 discard_and_relse:
520         kfree_skb(skb);
521         goto out;
522 }
523 EXPORT_SYMBOL(sk_receive_skb);
524 
525 void sk_reset_txq(struct sock *sk)
526 {
527         sk_tx_queue_clear(sk);
528 }
529 EXPORT_SYMBOL(sk_reset_txq);
530 
531 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
532 {
533         struct dst_entry *dst = __sk_dst_get(sk);
534 
535         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536                 sk_tx_queue_clear(sk);
537                 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
538                 dst_release(dst);
539                 return NULL;
540         }
541 
542         return dst;
543 }
544 EXPORT_SYMBOL(__sk_dst_check);
545 
546 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
547 {
548         struct dst_entry *dst = sk_dst_get(sk);
549 
550         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
551                 sk_dst_reset(sk);
552                 dst_release(dst);
553                 return NULL;
554         }
555 
556         return dst;
557 }
558 EXPORT_SYMBOL(sk_dst_check);
559 
560 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
561                                 int optlen)
562 {
563         int ret = -ENOPROTOOPT;
564 #ifdef CONFIG_NETDEVICES
565         struct net *net = sock_net(sk);
566         char devname[IFNAMSIZ];
567         int index;
568 
569         /* Sorry... */
570         ret = -EPERM;
571         if (!ns_capable(net->user_ns, CAP_NET_RAW))
572                 goto out;
573 
574         ret = -EINVAL;
575         if (optlen < 0)
576                 goto out;
577 
578         /* Bind this socket to a particular device like "eth0",
579          * as specified in the passed interface name. If the
580          * name is "" or the option length is zero the socket
581          * is not bound.
582          */
583         if (optlen > IFNAMSIZ - 1)
584                 optlen = IFNAMSIZ - 1;
585         memset(devname, 0, sizeof(devname));
586 
587         ret = -EFAULT;
588         if (copy_from_user(devname, optval, optlen))
589                 goto out;
590 
591         index = 0;
592         if (devname[0] != '\0') {
593                 struct net_device *dev;
594 
595                 rcu_read_lock();
596                 dev = dev_get_by_name_rcu(net, devname);
597                 if (dev)
598                         index = dev->ifindex;
599                 rcu_read_unlock();
600                 ret = -ENODEV;
601                 if (!dev)
602                         goto out;
603         }
604 
605         lock_sock(sk);
606         sk->sk_bound_dev_if = index;
607         sk_dst_reset(sk);
608         release_sock(sk);
609 
610         ret = 0;
611 
612 out:
613 #endif
614 
615         return ret;
616 }
617 
618 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
619                                 int __user *optlen, int len)
620 {
621         int ret = -ENOPROTOOPT;
622 #ifdef CONFIG_NETDEVICES
623         struct net *net = sock_net(sk);
624         char devname[IFNAMSIZ];
625 
626         if (sk->sk_bound_dev_if == 0) {
627                 len = 0;
628                 goto zero;
629         }
630 
631         ret = -EINVAL;
632         if (len < IFNAMSIZ)
633                 goto out;
634 
635         ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
636         if (ret)
637                 goto out;
638 
639         len = strlen(devname) + 1;
640 
641         ret = -EFAULT;
642         if (copy_to_user(optval, devname, len))
643                 goto out;
644 
645 zero:
646         ret = -EFAULT;
647         if (put_user(len, optlen))
648                 goto out;
649 
650         ret = 0;
651 
652 out:
653 #endif
654 
655         return ret;
656 }
657 
658 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
659 {
660         if (valbool)
661                 sock_set_flag(sk, bit);
662         else
663                 sock_reset_flag(sk, bit);
664 }
665 
666 /*
667  *      This is meant for all protocols to use and covers goings on
668  *      at the socket level. Everything here is generic.
669  */
670 
671 int sock_setsockopt(struct socket *sock, int level, int optname,
672                     char __user *optval, unsigned int optlen)
673 {
674         struct sock *sk = sock->sk;
675         int val;
676         int valbool;
677         struct linger ling;
678         int ret = 0;
679 
680         /*
681          *      Options without arguments
682          */
683 
684         if (optname == SO_BINDTODEVICE)
685                 return sock_setbindtodevice(sk, optval, optlen);
686 
687         if (optlen < sizeof(int))
688                 return -EINVAL;
689 
690         if (get_user(val, (int __user *)optval))
691                 return -EFAULT;
692 
693         valbool = val ? 1 : 0;
694 
695         lock_sock(sk);
696 
697         switch (optname) {
698         case SO_DEBUG:
699                 if (val && !capable(CAP_NET_ADMIN))
700                         ret = -EACCES;
701                 else
702                         sock_valbool_flag(sk, SOCK_DBG, valbool);
703                 break;
704         case SO_REUSEADDR:
705                 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
706                 break;
707         case SO_REUSEPORT:
708                 sk->sk_reuseport = valbool;
709                 break;
710         case SO_TYPE:
711         case SO_PROTOCOL:
712         case SO_DOMAIN:
713         case SO_ERROR:
714                 ret = -ENOPROTOOPT;
715                 break;
716         case SO_DONTROUTE:
717                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
718                 break;
719         case SO_BROADCAST:
720                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
721                 break;
722         case SO_SNDBUF:
723                 /* Don't error on this BSD doesn't and if you think
724                  * about it this is right. Otherwise apps have to
725                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
726                  * are treated in BSD as hints
727                  */
728                 val = min_t(u32, val, sysctl_wmem_max);
729 set_sndbuf:
730                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
731                 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
732                 /* Wake up sending tasks if we upped the value. */
733                 sk->sk_write_space(sk);
734                 break;
735 
736         case SO_SNDBUFFORCE:
737                 if (!capable(CAP_NET_ADMIN)) {
738                         ret = -EPERM;
739                         break;
740                 }
741                 goto set_sndbuf;
742 
743         case SO_RCVBUF:
744                 /* Don't error on this BSD doesn't and if you think
745                  * about it this is right. Otherwise apps have to
746                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
747                  * are treated in BSD as hints
748                  */
749                 val = min_t(u32, val, sysctl_rmem_max);
750 set_rcvbuf:
751                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
752                 /*
753                  * We double it on the way in to account for
754                  * "struct sk_buff" etc. overhead.   Applications
755                  * assume that the SO_RCVBUF setting they make will
756                  * allow that much actual data to be received on that
757                  * socket.
758                  *
759                  * Applications are unaware that "struct sk_buff" and
760                  * other overheads allocate from the receive buffer
761                  * during socket buffer allocation.
762                  *
763                  * And after considering the possible alternatives,
764                  * returning the value we actually used in getsockopt
765                  * is the most desirable behavior.
766                  */
767                 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
768                 break;
769 
770         case SO_RCVBUFFORCE:
771                 if (!capable(CAP_NET_ADMIN)) {
772                         ret = -EPERM;
773                         break;
774                 }
775                 goto set_rcvbuf;
776 
777         case SO_KEEPALIVE:
778 #ifdef CONFIG_INET
779                 if (sk->sk_protocol == IPPROTO_TCP &&
780                     sk->sk_type == SOCK_STREAM)
781                         tcp_set_keepalive(sk, valbool);
782 #endif
783                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
784                 break;
785 
786         case SO_OOBINLINE:
787                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
788                 break;
789 
790         case SO_NO_CHECK:
791                 sk->sk_no_check = valbool;
792                 break;
793 
794         case SO_PRIORITY:
795                 if ((val >= 0 && val <= 6) ||
796                     ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
797                         sk->sk_priority = val;
798                 else
799                         ret = -EPERM;
800                 break;
801 
802         case SO_LINGER:
803                 if (optlen < sizeof(ling)) {
804                         ret = -EINVAL;  /* 1003.1g */
805                         break;
806                 }
807                 if (copy_from_user(&ling, optval, sizeof(ling))) {
808                         ret = -EFAULT;
809                         break;
810                 }
811                 if (!ling.l_onoff)
812                         sock_reset_flag(sk, SOCK_LINGER);
813                 else {
814 #if (BITS_PER_LONG == 32)
815                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
816                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
817                         else
818 #endif
819                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
820                         sock_set_flag(sk, SOCK_LINGER);
821                 }
822                 break;
823 
824         case SO_BSDCOMPAT:
825                 sock_warn_obsolete_bsdism("setsockopt");
826                 break;
827 
828         case SO_PASSCRED:
829                 if (valbool)
830                         set_bit(SOCK_PASSCRED, &sock->flags);
831                 else
832                         clear_bit(SOCK_PASSCRED, &sock->flags);
833                 break;
834 
835         case SO_TIMESTAMP:
836         case SO_TIMESTAMPNS:
837                 if (valbool)  {
838                         if (optname == SO_TIMESTAMP)
839                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
840                         else
841                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
842                         sock_set_flag(sk, SOCK_RCVTSTAMP);
843                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
844                 } else {
845                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
846                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
847                 }
848                 break;
849 
850         case SO_TIMESTAMPING:
851                 if (val & ~SOF_TIMESTAMPING_MASK) {
852                         ret = -EINVAL;
853                         break;
854                 }
855                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
856                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
857                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
858                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
859                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
860                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
861                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
862                         sock_enable_timestamp(sk,
863                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
864                 else
865                         sock_disable_timestamp(sk,
866                                                (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
867                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
868                                   val & SOF_TIMESTAMPING_SOFTWARE);
869                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
870                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
871                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
872                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
873                 break;
874 
875         case SO_RCVLOWAT:
876                 if (val < 0)
877                         val = INT_MAX;
878                 sk->sk_rcvlowat = val ? : 1;
879                 break;
880 
881         case SO_RCVTIMEO:
882                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
883                 break;
884 
885         case SO_SNDTIMEO:
886                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
887                 break;
888 
889         case SO_ATTACH_FILTER:
890                 ret = -EINVAL;
891                 if (optlen == sizeof(struct sock_fprog)) {
892                         struct sock_fprog fprog;
893 
894                         ret = -EFAULT;
895                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
896                                 break;
897 
898                         ret = sk_attach_filter(&fprog, sk);
899                 }
900                 break;
901 
902         case SO_DETACH_FILTER:
903                 ret = sk_detach_filter(sk);
904                 break;
905 
906         case SO_LOCK_FILTER:
907                 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
908                         ret = -EPERM;
909                 else
910                         sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
911                 break;
912 
913         case SO_PASSSEC:
914                 if (valbool)
915                         set_bit(SOCK_PASSSEC, &sock->flags);
916                 else
917                         clear_bit(SOCK_PASSSEC, &sock->flags);
918                 break;
919         case SO_MARK:
920                 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
921                         ret = -EPERM;
922                 else
923                         sk->sk_mark = val;
924                 break;
925 
926                 /* We implement the SO_SNDLOWAT etc to
927                    not be settable (1003.1g 5.3) */
928         case SO_RXQ_OVFL:
929                 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
930                 break;
931 
932         case SO_WIFI_STATUS:
933                 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
934                 break;
935 
936         case SO_PEEK_OFF:
937                 if (sock->ops->set_peek_off)
938                         ret = sock->ops->set_peek_off(sk, val);
939                 else
940                         ret = -EOPNOTSUPP;
941                 break;
942 
943         case SO_NOFCS:
944                 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
945                 break;
946 
947         case SO_SELECT_ERR_QUEUE:
948                 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
949                 break;
950 
951 #ifdef CONFIG_NET_RX_BUSY_POLL
952         case SO_BUSY_POLL:
953                 /* allow unprivileged users to decrease the value */
954                 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
955                         ret = -EPERM;
956                 else {
957                         if (val < 0)
958                                 ret = -EINVAL;
959                         else
960                                 sk->sk_ll_usec = val;
961                 }
962                 break;
963 #endif
964         default:
965                 ret = -ENOPROTOOPT;
966                 break;
967         }
968         release_sock(sk);
969         return ret;
970 }
971 EXPORT_SYMBOL(sock_setsockopt);
972 
973 
974 void cred_to_ucred(struct pid *pid, const struct cred *cred,
975                    struct ucred *ucred)
976 {
977         ucred->pid = pid_vnr(pid);
978         ucred->uid = ucred->gid = -1;
979         if (cred) {
980                 struct user_namespace *current_ns = current_user_ns();
981 
982                 ucred->uid = from_kuid_munged(current_ns, cred->euid);
983                 ucred->gid = from_kgid_munged(current_ns, cred->egid);
984         }
985 }
986 EXPORT_SYMBOL_GPL(cred_to_ucred);
987 
988 int sock_getsockopt(struct socket *sock, int level, int optname,
989                     char __user *optval, int __user *optlen)
990 {
991         struct sock *sk = sock->sk;
992 
993         union {
994                 int val;
995                 struct linger ling;
996                 struct timeval tm;
997         } v;
998 
999         int lv = sizeof(int);
1000         int len;
1001 
1002         if (get_user(len, optlen))
1003                 return -EFAULT;
1004         if (len < 0)
1005                 return -EINVAL;
1006 
1007         memset(&v, 0, sizeof(v));
1008 
1009         switch (optname) {
1010         case SO_DEBUG:
1011                 v.val = sock_flag(sk, SOCK_DBG);
1012                 break;
1013 
1014         case SO_DONTROUTE:
1015                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1016                 break;
1017 
1018         case SO_BROADCAST:
1019                 v.val = sock_flag(sk, SOCK_BROADCAST);
1020                 break;
1021 
1022         case SO_SNDBUF:
1023                 v.val = sk->sk_sndbuf;
1024                 break;
1025 
1026         case SO_RCVBUF:
1027                 v.val = sk->sk_rcvbuf;
1028                 break;
1029 
1030         case SO_REUSEADDR:
1031                 v.val = sk->sk_reuse;
1032                 break;
1033 
1034         case SO_REUSEPORT:
1035                 v.val = sk->sk_reuseport;
1036                 break;
1037 
1038         case SO_KEEPALIVE:
1039                 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1040                 break;
1041 
1042         case SO_TYPE:
1043                 v.val = sk->sk_type;
1044                 break;
1045 
1046         case SO_PROTOCOL:
1047                 v.val = sk->sk_protocol;
1048                 break;
1049 
1050         case SO_DOMAIN:
1051                 v.val = sk->sk_family;
1052                 break;
1053 
1054         case SO_ERROR:
1055                 v.val = -sock_error(sk);
1056                 if (v.val == 0)
1057                         v.val = xchg(&sk->sk_err_soft, 0);
1058                 break;
1059 
1060         case SO_OOBINLINE:
1061                 v.val = sock_flag(sk, SOCK_URGINLINE);
1062                 break;
1063 
1064         case SO_NO_CHECK:
1065                 v.val = sk->sk_no_check;
1066                 break;
1067 
1068         case SO_PRIORITY:
1069                 v.val = sk->sk_priority;
1070                 break;
1071 
1072         case SO_LINGER:
1073                 lv              = sizeof(v.ling);
1074                 v.ling.l_onoff  = sock_flag(sk, SOCK_LINGER);
1075                 v.ling.l_linger = sk->sk_lingertime / HZ;
1076                 break;
1077 
1078         case SO_BSDCOMPAT:
1079                 sock_warn_obsolete_bsdism("getsockopt");
1080                 break;
1081 
1082         case SO_TIMESTAMP:
1083                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1084                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1085                 break;
1086 
1087         case SO_TIMESTAMPNS:
1088                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1089                 break;
1090 
1091         case SO_TIMESTAMPING:
1092                 v.val = 0;
1093                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1094                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1095                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1096                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1097                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1098                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1099                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1100                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1101                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1102                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
1103                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1104                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1105                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1106                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1107                 break;
1108 
1109         case SO_RCVTIMEO:
1110                 lv = sizeof(struct timeval);
1111                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1112                         v.tm.tv_sec = 0;
1113                         v.tm.tv_usec = 0;
1114                 } else {
1115                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1116                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1117                 }
1118                 break;
1119 
1120         case SO_SNDTIMEO:
1121                 lv = sizeof(struct timeval);
1122                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1123                         v.tm.tv_sec = 0;
1124                         v.tm.tv_usec = 0;
1125                 } else {
1126                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1127                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1128                 }
1129                 break;
1130 
1131         case SO_RCVLOWAT:
1132                 v.val = sk->sk_rcvlowat;
1133                 break;
1134 
1135         case SO_SNDLOWAT:
1136                 v.val = 1;
1137                 break;
1138 
1139         case SO_PASSCRED:
1140                 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1141                 break;
1142 
1143         case SO_PEERCRED:
1144         {
1145                 struct ucred peercred;
1146                 if (len > sizeof(peercred))
1147                         len = sizeof(peercred);
1148                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1149                 if (copy_to_user(optval, &peercred, len))
1150                         return -EFAULT;
1151                 goto lenout;
1152         }
1153 
1154         case SO_PEERNAME:
1155         {
1156                 char address[128];
1157 
1158                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1159                         return -ENOTCONN;
1160                 if (lv < len)
1161                         return -EINVAL;
1162                 if (copy_to_user(optval, address, len))
1163                         return -EFAULT;
1164                 goto lenout;
1165         }
1166 
1167         /* Dubious BSD thing... Probably nobody even uses it, but
1168          * the UNIX standard wants it for whatever reason... -DaveM
1169          */
1170         case SO_ACCEPTCONN:
1171                 v.val = sk->sk_state == TCP_LISTEN;
1172                 break;
1173 
1174         case SO_PASSSEC:
1175                 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1176                 break;
1177 
1178         case SO_PEERSEC:
1179                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1180 
1181         case SO_MARK:
1182                 v.val = sk->sk_mark;
1183                 break;
1184 
1185         case SO_RXQ_OVFL:
1186                 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1187                 break;
1188 
1189         case SO_WIFI_STATUS:
1190                 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1191                 break;
1192 
1193         case SO_PEEK_OFF:
1194                 if (!sock->ops->set_peek_off)
1195                         return -EOPNOTSUPP;
1196 
1197                 v.val = sk->sk_peek_off;
1198                 break;
1199         case SO_NOFCS:
1200                 v.val = sock_flag(sk, SOCK_NOFCS);
1201                 break;
1202 
1203         case SO_BINDTODEVICE:
1204                 return sock_getbindtodevice(sk, optval, optlen, len);
1205 
1206         case SO_GET_FILTER:
1207                 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1208                 if (len < 0)
1209                         return len;
1210 
1211                 goto lenout;
1212 
1213         case SO_LOCK_FILTER:
1214                 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1215                 break;
1216 
1217         case SO_SELECT_ERR_QUEUE:
1218                 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1219                 break;
1220 
1221 #ifdef CONFIG_NET_RX_BUSY_POLL
1222         case SO_BUSY_POLL:
1223                 v.val = sk->sk_ll_usec;
1224                 break;
1225 #endif
1226 
1227         default:
1228                 return -ENOPROTOOPT;
1229         }
1230 
1231         if (len > lv)
1232                 len = lv;
1233         if (copy_to_user(optval, &v, len))
1234                 return -EFAULT;
1235 lenout:
1236         if (put_user(len, optlen))
1237                 return -EFAULT;
1238         return 0;
1239 }
1240 
1241 /*
1242  * Initialize an sk_lock.
1243  *
1244  * (We also register the sk_lock with the lock validator.)
1245  */
1246 static inline void sock_lock_init(struct sock *sk)
1247 {
1248         sock_lock_init_class_and_name(sk,
1249                         af_family_slock_key_strings[sk->sk_family],
1250                         af_family_slock_keys + sk->sk_family,
1251                         af_family_key_strings[sk->sk_family],
1252                         af_family_keys + sk->sk_family);
1253 }
1254 
1255 /*
1256  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1257  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1258  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1259  */
1260 static void sock_copy(struct sock *nsk, const struct sock *osk)
1261 {
1262 #ifdef CONFIG_SECURITY_NETWORK
1263         void *sptr = nsk->sk_security;
1264 #endif
1265         memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1266 
1267         memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1268                osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1269 
1270 #ifdef CONFIG_SECURITY_NETWORK
1271         nsk->sk_security = sptr;
1272         security_sk_clone(osk, nsk);
1273 #endif
1274 }
1275 
1276 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1277 {
1278         unsigned long nulls1, nulls2;
1279 
1280         nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1281         nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1282         if (nulls1 > nulls2)
1283                 swap(nulls1, nulls2);
1284 
1285         if (nulls1 != 0)
1286                 memset((char *)sk, 0, nulls1);
1287         memset((char *)sk + nulls1 + sizeof(void *), 0,
1288                nulls2 - nulls1 - sizeof(void *));
1289         memset((char *)sk + nulls2 + sizeof(void *), 0,
1290                size - nulls2 - sizeof(void *));
1291 }
1292 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1293 
1294 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1295                 int family)
1296 {
1297         struct sock *sk;
1298         struct kmem_cache *slab;
1299 
1300         slab = prot->slab;
1301         if (slab != NULL) {
1302                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1303                 if (!sk)
1304                         return sk;
1305                 if (priority & __GFP_ZERO) {
1306                         if (prot->clear_sk)
1307                                 prot->clear_sk(sk, prot->obj_size);
1308                         else
1309                                 sk_prot_clear_nulls(sk, prot->obj_size);
1310                 }
1311         } else
1312                 sk = kmalloc(prot->obj_size, priority);
1313 
1314         if (sk != NULL) {
1315                 kmemcheck_annotate_bitfield(sk, flags);
1316 
1317                 if (security_sk_alloc(sk, family, priority))
1318                         goto out_free;
1319 
1320                 if (!try_module_get(prot->owner))
1321                         goto out_free_sec;
1322                 sk_tx_queue_clear(sk);
1323         }
1324 
1325         return sk;
1326 
1327 out_free_sec:
1328         security_sk_free(sk);
1329 out_free:
1330         if (slab != NULL)
1331                 kmem_cache_free(slab, sk);
1332         else
1333                 kfree(sk);
1334         return NULL;
1335 }
1336 
1337 static void sk_prot_free(struct proto *prot, struct sock *sk)
1338 {
1339         struct kmem_cache *slab;
1340         struct module *owner;
1341 
1342         owner = prot->owner;
1343         slab = prot->slab;
1344 
1345         security_sk_free(sk);
1346         if (slab != NULL)
1347                 kmem_cache_free(slab, sk);
1348         else
1349                 kfree(sk);
1350         module_put(owner);
1351 }
1352 
1353 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1354 void sock_update_classid(struct sock *sk)
1355 {
1356         u32 classid;
1357 
1358         classid = task_cls_classid(current);
1359         if (classid != sk->sk_classid)
1360                 sk->sk_classid = classid;
1361 }
1362 EXPORT_SYMBOL(sock_update_classid);
1363 #endif
1364 
1365 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1366 void sock_update_netprioidx(struct sock *sk)
1367 {
1368         if (in_interrupt())
1369                 return;
1370 
1371         sk->sk_cgrp_prioidx = task_netprioidx(current);
1372 }
1373 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1374 #endif
1375 
1376 /**
1377  *      sk_alloc - All socket objects are allocated here
1378  *      @net: the applicable net namespace
1379  *      @family: protocol family
1380  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1381  *      @prot: struct proto associated with this new sock instance
1382  */
1383 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1384                       struct proto *prot)
1385 {
1386         struct sock *sk;
1387 
1388         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1389         if (sk) {
1390                 sk->sk_family = family;
1391                 /*
1392                  * See comment in struct sock definition to understand
1393                  * why we need sk_prot_creator -acme
1394                  */
1395                 sk->sk_prot = sk->sk_prot_creator = prot;
1396                 sock_lock_init(sk);
1397                 sock_net_set(sk, get_net(net));
1398                 atomic_set(&sk->sk_wmem_alloc, 1);
1399 
1400                 sock_update_classid(sk);
1401                 sock_update_netprioidx(sk);
1402         }
1403 
1404         return sk;
1405 }
1406 EXPORT_SYMBOL(sk_alloc);
1407 
1408 static void __sk_free(struct sock *sk)
1409 {
1410         struct sk_filter *filter;
1411 
1412         if (sk->sk_destruct)
1413                 sk->sk_destruct(sk);
1414 
1415         filter = rcu_dereference_check(sk->sk_filter,
1416                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1417         if (filter) {
1418                 sk_filter_uncharge(sk, filter);
1419                 RCU_INIT_POINTER(sk->sk_filter, NULL);
1420         }
1421 
1422         sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1423 
1424         if (atomic_read(&sk->sk_omem_alloc))
1425                 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1426                          __func__, atomic_read(&sk->sk_omem_alloc));
1427 
1428         if (sk->sk_frag.page) {
1429                 put_page(sk->sk_frag.page);
1430                 sk->sk_frag.page = NULL;
1431         }
1432 
1433         if (sk->sk_peer_cred)
1434                 put_cred(sk->sk_peer_cred);
1435         put_pid(sk->sk_peer_pid);
1436         put_net(sock_net(sk));
1437         sk_prot_free(sk->sk_prot_creator, sk);
1438 }
1439 
1440 void sk_free(struct sock *sk)
1441 {
1442         /*
1443          * We subtract one from sk_wmem_alloc and can know if
1444          * some packets are still in some tx queue.
1445          * If not null, sock_wfree() will call __sk_free(sk) later
1446          */
1447         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1448                 __sk_free(sk);
1449 }
1450 EXPORT_SYMBOL(sk_free);
1451 
1452 /*
1453  * Last sock_put should drop reference to sk->sk_net. It has already
1454  * been dropped in sk_change_net. Taking reference to stopping namespace
1455  * is not an option.
1456  * Take reference to a socket to remove it from hash _alive_ and after that
1457  * destroy it in the context of init_net.
1458  */
1459 void sk_release_kernel(struct sock *sk)
1460 {
1461         if (sk == NULL || sk->sk_socket == NULL)
1462                 return;
1463 
1464         sock_hold(sk);
1465         sock_release(sk->sk_socket);
1466         release_net(sock_net(sk));
1467         sock_net_set(sk, get_net(&init_net));
1468         sock_put(sk);
1469 }
1470 EXPORT_SYMBOL(sk_release_kernel);
1471 
1472 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1473 {
1474         if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1475                 sock_update_memcg(newsk);
1476 }
1477 
1478 /**
1479  *      sk_clone_lock - clone a socket, and lock its clone
1480  *      @sk: the socket to clone
1481  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1482  *
1483  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1484  */
1485 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1486 {
1487         struct sock *newsk;
1488 
1489         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1490         if (newsk != NULL) {
1491                 struct sk_filter *filter;
1492 
1493                 sock_copy(newsk, sk);
1494 
1495                 /* SANITY */
1496                 get_net(sock_net(newsk));
1497                 sk_node_init(&newsk->sk_node);
1498                 sock_lock_init(newsk);
1499                 bh_lock_sock(newsk);
1500                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1501                 newsk->sk_backlog.len = 0;
1502 
1503                 atomic_set(&newsk->sk_rmem_alloc, 0);
1504                 /*
1505                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1506                  */
1507                 atomic_set(&newsk->sk_wmem_alloc, 1);
1508                 atomic_set(&newsk->sk_omem_alloc, 0);
1509                 skb_queue_head_init(&newsk->sk_receive_queue);
1510                 skb_queue_head_init(&newsk->sk_write_queue);
1511 #ifdef CONFIG_NET_DMA
1512                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1513 #endif
1514 
1515                 spin_lock_init(&newsk->sk_dst_lock);
1516                 rwlock_init(&newsk->sk_callback_lock);
1517                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1518                                 af_callback_keys + newsk->sk_family,
1519                                 af_family_clock_key_strings[newsk->sk_family]);
1520 
1521                 newsk->sk_dst_cache     = NULL;
1522                 newsk->sk_wmem_queued   = 0;
1523                 newsk->sk_forward_alloc = 0;
1524                 newsk->sk_send_head     = NULL;
1525                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1526 
1527                 sock_reset_flag(newsk, SOCK_DONE);
1528                 skb_queue_head_init(&newsk->sk_error_queue);
1529 
1530                 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1531                 if (filter != NULL)
1532                         sk_filter_charge(newsk, filter);
1533 
1534                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1535                         /* It is still raw copy of parent, so invalidate
1536                          * destructor and make plain sk_free() */
1537                         newsk->sk_destruct = NULL;
1538                         bh_unlock_sock(newsk);
1539                         sk_free(newsk);
1540                         newsk = NULL;
1541                         goto out;
1542                 }
1543 
1544                 newsk->sk_err      = 0;
1545                 newsk->sk_err_soft = 0;
1546                 newsk->sk_priority = 0;
1547                 /*
1548                  * Before updating sk_refcnt, we must commit prior changes to memory
1549                  * (Documentation/RCU/rculist_nulls.txt for details)
1550                  */
1551                 smp_wmb();
1552                 atomic_set(&newsk->sk_refcnt, 2);
1553 
1554                 /*
1555                  * Increment the counter in the same struct proto as the master
1556                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1557                  * is the same as sk->sk_prot->socks, as this field was copied
1558                  * with memcpy).
1559                  *
1560                  * This _changes_ the previous behaviour, where
1561                  * tcp_create_openreq_child always was incrementing the
1562                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1563                  * to be taken into account in all callers. -acme
1564                  */
1565                 sk_refcnt_debug_inc(newsk);
1566                 sk_set_socket(newsk, NULL);
1567                 newsk->sk_wq = NULL;
1568 
1569                 sk_update_clone(sk, newsk);
1570 
1571                 if (newsk->sk_prot->sockets_allocated)
1572                         sk_sockets_allocated_inc(newsk);
1573 
1574                 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1575                         net_enable_timestamp();
1576         }
1577 out:
1578         return newsk;
1579 }
1580 EXPORT_SYMBOL_GPL(sk_clone_lock);
1581 
1582 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1583 {
1584         __sk_dst_set(sk, dst);
1585         sk->sk_route_caps = dst->dev->features;
1586         if (sk->sk_route_caps & NETIF_F_GSO)
1587                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1588         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1589         if (sk_can_gso(sk)) {
1590                 if (dst->header_len) {
1591                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1592                 } else {
1593                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1594                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1595                         sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1596                 }
1597         }
1598 }
1599 EXPORT_SYMBOL_GPL(sk_setup_caps);
1600 
1601 /*
1602  *      Simple resource managers for sockets.
1603  */
1604 
1605 
1606 /*
1607  * Write buffer destructor automatically called from kfree_skb.
1608  */
1609 void sock_wfree(struct sk_buff *skb)
1610 {
1611         struct sock *sk = skb->sk;
1612         unsigned int len = skb->truesize;
1613 
1614         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1615                 /*
1616                  * Keep a reference on sk_wmem_alloc, this will be released
1617                  * after sk_write_space() call
1618                  */
1619                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1620                 sk->sk_write_space(sk);
1621                 len = 1;
1622         }
1623         /*
1624          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1625          * could not do because of in-flight packets
1626          */
1627         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1628                 __sk_free(sk);
1629 }
1630 EXPORT_SYMBOL(sock_wfree);
1631 
1632 void skb_orphan_partial(struct sk_buff *skb)
1633 {
1634         /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1635          * so we do not completely orphan skb, but transfert all
1636          * accounted bytes but one, to avoid unexpected reorders.
1637          */
1638         if (skb->destructor == sock_wfree
1639 #ifdef CONFIG_INET
1640             || skb->destructor == tcp_wfree
1641 #endif
1642                 ) {
1643                 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1644                 skb->truesize = 1;
1645         } else {
1646                 skb_orphan(skb);
1647         }
1648 }
1649 EXPORT_SYMBOL(skb_orphan_partial);
1650 
1651 /*
1652  * Read buffer destructor automatically called from kfree_skb.
1653  */
1654 void sock_rfree(struct sk_buff *skb)
1655 {
1656         struct sock *sk = skb->sk;
1657         unsigned int len = skb->truesize;
1658 
1659         atomic_sub(len, &sk->sk_rmem_alloc);
1660         sk_mem_uncharge(sk, len);
1661 }
1662 EXPORT_SYMBOL(sock_rfree);
1663 
1664 void sock_efree(struct sk_buff *skb)
1665 {
1666         sock_put(skb->sk);
1667 }
1668 EXPORT_SYMBOL(sock_efree);
1669 
1670 void sock_edemux(struct sk_buff *skb)
1671 {
1672         struct sock *sk = skb->sk;
1673 
1674 #ifdef CONFIG_INET
1675         if (sk->sk_state == TCP_TIME_WAIT)
1676                 inet_twsk_put(inet_twsk(sk));
1677         else
1678 #endif
1679                 sock_put(sk);
1680 }
1681 EXPORT_SYMBOL(sock_edemux);
1682 
1683 kuid_t sock_i_uid(struct sock *sk)
1684 {
1685         kuid_t uid;
1686 
1687         read_lock_bh(&sk->sk_callback_lock);
1688         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1689         read_unlock_bh(&sk->sk_callback_lock);
1690         return uid;
1691 }
1692 EXPORT_SYMBOL(sock_i_uid);
1693 
1694 unsigned long sock_i_ino(struct sock *sk)
1695 {
1696         unsigned long ino;
1697 
1698         read_lock_bh(&sk->sk_callback_lock);
1699         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1700         read_unlock_bh(&sk->sk_callback_lock);
1701         return ino;
1702 }
1703 EXPORT_SYMBOL(sock_i_ino);
1704 
1705 /*
1706  * Allocate a skb from the socket's send buffer.
1707  */
1708 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1709                              gfp_t priority)
1710 {
1711         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1712                 struct sk_buff *skb = alloc_skb(size, priority);
1713                 if (skb) {
1714                         skb_set_owner_w(skb, sk);
1715                         return skb;
1716                 }
1717         }
1718         return NULL;
1719 }
1720 EXPORT_SYMBOL(sock_wmalloc);
1721 
1722 /*
1723  * Allocate a skb from the socket's receive buffer.
1724  */
1725 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1726                              gfp_t priority)
1727 {
1728         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1729                 struct sk_buff *skb = alloc_skb(size, priority);
1730                 if (skb) {
1731                         skb_set_owner_r(skb, sk);
1732                         return skb;
1733                 }
1734         }
1735         return NULL;
1736 }
1737 
1738 /*
1739  * Allocate a memory block from the socket's option memory buffer.
1740  */
1741 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1742 {
1743         if ((unsigned int)size <= sysctl_optmem_max &&
1744             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1745                 void *mem;
1746                 /* First do the add, to avoid the race if kmalloc
1747                  * might sleep.
1748                  */
1749                 atomic_add(size, &sk->sk_omem_alloc);
1750                 mem = kmalloc(size, priority);
1751                 if (mem)
1752                         return mem;
1753                 atomic_sub(size, &sk->sk_omem_alloc);
1754         }
1755         return NULL;
1756 }
1757 EXPORT_SYMBOL(sock_kmalloc);
1758 
1759 /*
1760  * Free an option memory block.
1761  */
1762 void sock_kfree_s(struct sock *sk, void *mem, int size)
1763 {
1764         kfree(mem);
1765         atomic_sub(size, &sk->sk_omem_alloc);
1766 }
1767 EXPORT_SYMBOL(sock_kfree_s);
1768 
1769 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1770    I think, these locks should be removed for datagram sockets.
1771  */
1772 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1773 {
1774         DEFINE_WAIT(wait);
1775 
1776         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1777         for (;;) {
1778                 if (!timeo)
1779                         break;
1780                 if (signal_pending(current))
1781                         break;
1782                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1783                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1784                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1785                         break;
1786                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1787                         break;
1788                 if (sk->sk_err)
1789                         break;
1790                 timeo = schedule_timeout(timeo);
1791         }
1792         finish_wait(sk_sleep(sk), &wait);
1793         return timeo;
1794 }
1795 
1796 
1797 /*
1798  *      Generic send/receive buffer handlers
1799  */
1800 
1801 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1802                                      unsigned long data_len, int noblock,
1803                                      int *errcode, int max_page_order)
1804 {
1805         struct sk_buff *skb = NULL;
1806         unsigned long chunk;
1807         gfp_t gfp_mask;
1808         long timeo;
1809         int err;
1810         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1811         struct page *page;
1812         int i;
1813 
1814         err = -EMSGSIZE;
1815         if (npages > MAX_SKB_FRAGS)
1816                 goto failure;
1817 
1818         timeo = sock_sndtimeo(sk, noblock);
1819         while (!skb) {
1820                 err = sock_error(sk);
1821                 if (err != 0)
1822                         goto failure;
1823 
1824                 err = -EPIPE;
1825                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1826                         goto failure;
1827 
1828                 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1829                         set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1830                         set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1831                         err = -EAGAIN;
1832                         if (!timeo)
1833                                 goto failure;
1834                         if (signal_pending(current))
1835                                 goto interrupted;
1836                         timeo = sock_wait_for_wmem(sk, timeo);
1837                         continue;
1838                 }
1839 
1840                 err = -ENOBUFS;
1841                 gfp_mask = sk->sk_allocation;
1842                 if (gfp_mask & __GFP_WAIT)
1843                         gfp_mask |= __GFP_REPEAT;
1844 
1845                 skb = alloc_skb(header_len, gfp_mask);
1846                 if (!skb)
1847                         goto failure;
1848 
1849                 skb->truesize += data_len;
1850 
1851                 for (i = 0; npages > 0; i++) {
1852                         int order = max_page_order;
1853 
1854                         while (order) {
1855                                 if (npages >= 1 << order) {
1856                                         page = alloc_pages(sk->sk_allocation |
1857                                                            __GFP_COMP |
1858                                                            __GFP_NOWARN |
1859                                                            __GFP_NORETRY,
1860                                                            order);
1861                                         if (page)
1862                                                 goto fill_page;
1863                                 }
1864                                 order--;
1865                         }
1866                         page = alloc_page(sk->sk_allocation);
1867                         if (!page)
1868                                 goto failure;
1869 fill_page:
1870                         chunk = min_t(unsigned long, data_len,
1871                                       PAGE_SIZE << order);
1872                         skb_fill_page_desc(skb, i, page, 0, chunk);
1873                         data_len -= chunk;
1874                         npages -= 1 << order;
1875                 }
1876         }
1877 
1878         skb_set_owner_w(skb, sk);
1879         return skb;
1880 
1881 interrupted:
1882         err = sock_intr_errno(timeo);
1883 failure:
1884         kfree_skb(skb);
1885         *errcode = err;
1886         return NULL;
1887 }
1888 EXPORT_SYMBOL(sock_alloc_send_pskb);
1889 
1890 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1891                                     int noblock, int *errcode)
1892 {
1893         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1894 }
1895 EXPORT_SYMBOL(sock_alloc_send_skb);
1896 
1897 /* On 32bit arches, an skb frag is limited to 2^15 */
1898 #define SKB_FRAG_PAGE_ORDER     get_order(32768)
1899 
1900 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1901 {
1902         int order;
1903 
1904         if (pfrag->page) {
1905                 if (atomic_read(&pfrag->page->_count) == 1) {
1906                         pfrag->offset = 0;
1907                         return true;
1908                 }
1909                 if (pfrag->offset < pfrag->size)
1910                         return true;
1911                 put_page(pfrag->page);
1912         }
1913 
1914         /* We restrict high order allocations to users that can afford to wait */
1915         order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1916 
1917         do {
1918                 gfp_t gfp = sk->sk_allocation;
1919 
1920                 if (order) {
1921                         gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
1922                         gfp &= ~__GFP_WAIT;
1923                 }
1924                 pfrag->page = alloc_pages(gfp, order);
1925                 if (likely(pfrag->page)) {
1926                         pfrag->offset = 0;
1927                         pfrag->size = PAGE_SIZE << order;
1928                         return true;
1929                 }
1930         } while (--order >= 0);
1931 
1932         sk_enter_memory_pressure(sk);
1933         sk_stream_moderate_sndbuf(sk);
1934         return false;
1935 }
1936 EXPORT_SYMBOL(sk_page_frag_refill);
1937 
1938 static void __lock_sock(struct sock *sk)
1939         __releases(&sk->sk_lock.slock)
1940         __acquires(&sk->sk_lock.slock)
1941 {
1942         DEFINE_WAIT(wait);
1943 
1944         for (;;) {
1945                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1946                                         TASK_UNINTERRUPTIBLE);
1947                 spin_unlock_bh(&sk->sk_lock.slock);
1948                 schedule();
1949                 spin_lock_bh(&sk->sk_lock.slock);
1950                 if (!sock_owned_by_user(sk))
1951                         break;
1952         }
1953         finish_wait(&sk->sk_lock.wq, &wait);
1954 }
1955 
1956 static void __release_sock(struct sock *sk)
1957         __releases(&sk->sk_lock.slock)
1958         __acquires(&sk->sk_lock.slock)
1959 {
1960         struct sk_buff *skb = sk->sk_backlog.head;
1961 
1962         do {
1963                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1964                 bh_unlock_sock(sk);
1965 
1966                 do {
1967                         struct sk_buff *next = skb->next;
1968 
1969                         prefetch(next);
1970                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1971                         skb->next = NULL;
1972                         sk_backlog_rcv(sk, skb);
1973 
1974                         /*
1975                          * We are in process context here with softirqs
1976                          * disabled, use cond_resched_softirq() to preempt.
1977                          * This is safe to do because we've taken the backlog
1978                          * queue private:
1979                          */
1980                         cond_resched_softirq();
1981 
1982                         skb = next;
1983                 } while (skb != NULL);
1984 
1985                 bh_lock_sock(sk);
1986         } while ((skb = sk->sk_backlog.head) != NULL);
1987 
1988         /*
1989          * Doing the zeroing here guarantee we can not loop forever
1990          * while a wild producer attempts to flood us.
1991          */
1992         sk->sk_backlog.len = 0;
1993 }
1994 
1995 /**
1996  * sk_wait_data - wait for data to arrive at sk_receive_queue
1997  * @sk:    sock to wait on
1998  * @timeo: for how long
1999  *
2000  * Now socket state including sk->sk_err is changed only under lock,
2001  * hence we may omit checks after joining wait queue.
2002  * We check receive queue before schedule() only as optimization;
2003  * it is very likely that release_sock() added new data.
2004  */
2005 int sk_wait_data(struct sock *sk, long *timeo)
2006 {
2007         int rc;
2008         DEFINE_WAIT(wait);
2009 
2010         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2011         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2012         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
2013         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2014         finish_wait(sk_sleep(sk), &wait);
2015         return rc;
2016 }
2017 EXPORT_SYMBOL(sk_wait_data);
2018 
2019 /**
2020  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2021  *      @sk: socket
2022  *      @size: memory size to allocate
2023  *      @kind: allocation type
2024  *
2025  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2026  *      rmem allocation. This function assumes that protocols which have
2027  *      memory_pressure use sk_wmem_queued as write buffer accounting.
2028  */
2029 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2030 {
2031         struct proto *prot = sk->sk_prot;
2032         int amt = sk_mem_pages(size);
2033         long allocated;
2034         int parent_status = UNDER_LIMIT;
2035 
2036         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2037 
2038         allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2039 
2040         /* Under limit. */
2041         if (parent_status == UNDER_LIMIT &&
2042                         allocated <= sk_prot_mem_limits(sk, 0)) {
2043                 sk_leave_memory_pressure(sk);
2044                 return 1;
2045         }
2046 
2047         /* Under pressure. (we or our parents) */
2048         if ((parent_status > SOFT_LIMIT) ||
2049                         allocated > sk_prot_mem_limits(sk, 1))
2050                 sk_enter_memory_pressure(sk);
2051 
2052         /* Over hard limit (we or our parents) */
2053         if ((parent_status == OVER_LIMIT) ||
2054                         (allocated > sk_prot_mem_limits(sk, 2)))
2055                 goto suppress_allocation;
2056 
2057         /* guarantee minimum buffer size under pressure */
2058         if (kind == SK_MEM_RECV) {
2059                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2060                         return 1;
2061 
2062         } else { /* SK_MEM_SEND */
2063                 if (sk->sk_type == SOCK_STREAM) {
2064                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2065                                 return 1;
2066                 } else if (atomic_read(&sk->sk_wmem_alloc) <
2067                            prot->sysctl_wmem[0])
2068                                 return 1;
2069         }
2070 
2071         if (sk_has_memory_pressure(sk)) {
2072                 int alloc;
2073 
2074                 if (!sk_under_memory_pressure(sk))
2075                         return 1;
2076                 alloc = sk_sockets_allocated_read_positive(sk);
2077                 if (sk_prot_mem_limits(sk, 2) > alloc *
2078                     sk_mem_pages(sk->sk_wmem_queued +
2079                                  atomic_read(&sk->sk_rmem_alloc) +
2080                                  sk->sk_forward_alloc))
2081                         return 1;
2082         }
2083 
2084 suppress_allocation:
2085 
2086         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2087                 sk_stream_moderate_sndbuf(sk);
2088 
2089                 /* Fail only if socket is _under_ its sndbuf.
2090                  * In this case we cannot block, so that we have to fail.
2091                  */
2092                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2093                         return 1;
2094         }
2095 
2096         trace_sock_exceed_buf_limit(sk, prot, allocated);
2097 
2098         /* Alas. Undo changes. */
2099         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2100 
2101         sk_memory_allocated_sub(sk, amt);
2102 
2103         return 0;
2104 }
2105 EXPORT_SYMBOL(__sk_mem_schedule);
2106 
2107 /**
2108  *      __sk_reclaim - reclaim memory_allocated
2109  *      @sk: socket
2110  *      @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2111  */
2112 void __sk_mem_reclaim(struct sock *sk, int amount)
2113 {
2114         amount >>= SK_MEM_QUANTUM_SHIFT;
2115         sk_memory_allocated_sub(sk, amount);
2116         sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2117 
2118         if (sk_under_memory_pressure(sk) &&
2119             (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2120                 sk_leave_memory_pressure(sk);
2121 }
2122 EXPORT_SYMBOL(__sk_mem_reclaim);
2123 
2124 
2125 /*
2126  * Set of default routines for initialising struct proto_ops when
2127  * the protocol does not support a particular function. In certain
2128  * cases where it makes no sense for a protocol to have a "do nothing"
2129  * function, some default processing is provided.
2130  */
2131 
2132 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2133 {
2134         return -EOPNOTSUPP;
2135 }
2136 EXPORT_SYMBOL(sock_no_bind);
2137 
2138 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2139                     int len, int flags)
2140 {
2141         return -EOPNOTSUPP;
2142 }
2143 EXPORT_SYMBOL(sock_no_connect);
2144 
2145 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2146 {
2147         return -EOPNOTSUPP;
2148 }
2149 EXPORT_SYMBOL(sock_no_socketpair);
2150 
2151 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2152 {
2153         return -EOPNOTSUPP;
2154 }
2155 EXPORT_SYMBOL(sock_no_accept);
2156 
2157 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2158                     int *len, int peer)
2159 {
2160         return -EOPNOTSUPP;
2161 }
2162 EXPORT_SYMBOL(sock_no_getname);
2163 
2164 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2165 {
2166         return 0;
2167 }
2168 EXPORT_SYMBOL(sock_no_poll);
2169 
2170 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2171 {
2172         return -EOPNOTSUPP;
2173 }
2174 EXPORT_SYMBOL(sock_no_ioctl);
2175 
2176 int sock_no_listen(struct socket *sock, int backlog)
2177 {
2178         return -EOPNOTSUPP;
2179 }
2180 EXPORT_SYMBOL(sock_no_listen);
2181 
2182 int sock_no_shutdown(struct socket *sock, int how)
2183 {
2184         return -EOPNOTSUPP;
2185 }
2186 EXPORT_SYMBOL(sock_no_shutdown);
2187 
2188 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2189                     char __user *optval, unsigned int optlen)
2190 {
2191         return -EOPNOTSUPP;
2192 }
2193 EXPORT_SYMBOL(sock_no_setsockopt);
2194 
2195 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2196                     char __user *optval, int __user *optlen)
2197 {
2198         return -EOPNOTSUPP;
2199 }
2200 EXPORT_SYMBOL(sock_no_getsockopt);
2201 
2202 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2203                     size_t len)
2204 {
2205         return -EOPNOTSUPP;
2206 }
2207 EXPORT_SYMBOL(sock_no_sendmsg);
2208 
2209 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2210                     size_t len, int flags)
2211 {
2212         return -EOPNOTSUPP;
2213 }
2214 EXPORT_SYMBOL(sock_no_recvmsg);
2215 
2216 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2217 {
2218         /* Mirror missing mmap method error code */
2219         return -ENODEV;
2220 }
2221 EXPORT_SYMBOL(sock_no_mmap);
2222 
2223 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2224 {
2225         ssize_t res;
2226         struct msghdr msg = {.msg_flags = flags};
2227         struct kvec iov;
2228         char *kaddr = kmap(page);
2229         iov.iov_base = kaddr + offset;
2230         iov.iov_len = size;
2231         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2232         kunmap(page);
2233         return res;
2234 }
2235 EXPORT_SYMBOL(sock_no_sendpage);
2236 
2237 /*
2238  *      Default Socket Callbacks
2239  */
2240 
2241 static void sock_def_wakeup(struct sock *sk)
2242 {
2243         struct socket_wq *wq;
2244 
2245         rcu_read_lock();
2246         wq = rcu_dereference(sk->sk_wq);
2247         if (wq_has_sleeper(wq))
2248                 wake_up_interruptible_all(&wq->wait);
2249         rcu_read_unlock();
2250 }
2251 
2252 static void sock_def_error_report(struct sock *sk)
2253 {
2254         struct socket_wq *wq;
2255 
2256         rcu_read_lock();
2257         wq = rcu_dereference(sk->sk_wq);
2258         if (wq_has_sleeper(wq))
2259                 wake_up_interruptible_poll(&wq->wait, POLLERR);
2260         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2261         rcu_read_unlock();
2262 }
2263 
2264 static void sock_def_readable(struct sock *sk, int len)
2265 {
2266         struct socket_wq *wq;
2267 
2268         rcu_read_lock();
2269         wq = rcu_dereference(sk->sk_wq);
2270         if (wq_has_sleeper(wq))
2271                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2272                                                 POLLRDNORM | POLLRDBAND);
2273         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2274         rcu_read_unlock();
2275 }
2276 
2277 static void sock_def_write_space(struct sock *sk)
2278 {
2279         struct socket_wq *wq;
2280 
2281         rcu_read_lock();
2282 
2283         /* Do not wake up a writer until he can make "significant"
2284          * progress.  --DaveM
2285          */
2286         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2287                 wq = rcu_dereference(sk->sk_wq);
2288                 if (wq_has_sleeper(wq))
2289                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2290                                                 POLLWRNORM | POLLWRBAND);
2291 
2292                 /* Should agree with poll, otherwise some programs break */
2293                 if (sock_writeable(sk))
2294                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2295         }
2296 
2297         rcu_read_unlock();
2298 }
2299 
2300 static void sock_def_destruct(struct sock *sk)
2301 {
2302         kfree(sk->sk_protinfo);
2303 }
2304 
2305 void sk_send_sigurg(struct sock *sk)
2306 {
2307         if (sk->sk_socket && sk->sk_socket->file)
2308                 if (send_sigurg(&sk->sk_socket->file->f_owner))
2309                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2310 }
2311 EXPORT_SYMBOL(sk_send_sigurg);
2312 
2313 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2314                     unsigned long expires)
2315 {
2316         if (!mod_timer(timer, expires))
2317                 sock_hold(sk);
2318 }
2319 EXPORT_SYMBOL(sk_reset_timer);
2320 
2321 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2322 {
2323         if (del_timer(timer))
2324                 __sock_put(sk);
2325 }
2326 EXPORT_SYMBOL(sk_stop_timer);
2327 
2328 void sock_init_data(struct socket *sock, struct sock *sk)
2329 {
2330         skb_queue_head_init(&sk->sk_receive_queue);
2331         skb_queue_head_init(&sk->sk_write_queue);
2332         skb_queue_head_init(&sk->sk_error_queue);
2333 #ifdef CONFIG_NET_DMA
2334         skb_queue_head_init(&sk->sk_async_wait_queue);
2335 #endif
2336 
2337         sk->sk_send_head        =       NULL;
2338 
2339         init_timer(&sk->sk_timer);
2340 
2341         sk->sk_allocation       =       GFP_KERNEL;
2342         sk->sk_rcvbuf           =       sysctl_rmem_default;
2343         sk->sk_sndbuf           =       sysctl_wmem_default;
2344         sk->sk_state            =       TCP_CLOSE;
2345         sk_set_socket(sk, sock);
2346 
2347         sock_set_flag(sk, SOCK_ZAPPED);
2348 
2349         if (sock) {
2350                 sk->sk_type     =       sock->type;
2351                 sk->sk_wq       =       sock->wq;
2352                 sock->sk        =       sk;
2353         } else
2354                 sk->sk_wq       =       NULL;
2355 
2356         spin_lock_init(&sk->sk_dst_lock);
2357         rwlock_init(&sk->sk_callback_lock);
2358         lockdep_set_class_and_name(&sk->sk_callback_lock,
2359                         af_callback_keys + sk->sk_family,
2360                         af_family_clock_key_strings[sk->sk_family]);
2361 
2362         sk->sk_state_change     =       sock_def_wakeup;
2363         sk->sk_data_ready       =       sock_def_readable;
2364         sk->sk_write_space      =       sock_def_write_space;
2365         sk->sk_error_report     =       sock_def_error_report;
2366         sk->sk_destruct         =       sock_def_destruct;
2367 
2368         sk->sk_frag.page        =       NULL;
2369         sk->sk_frag.offset      =       0;
2370         sk->sk_peek_off         =       -1;
2371 
2372         sk->sk_peer_pid         =       NULL;
2373         sk->sk_peer_cred        =       NULL;
2374         sk->sk_write_pending    =       0;
2375         sk->sk_rcvlowat         =       1;
2376         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2377         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2378 
2379         sk->sk_stamp = ktime_set(-1L, 0);
2380 
2381 #ifdef CONFIG_NET_RX_BUSY_POLL
2382         sk->sk_napi_id          =       0;
2383         sk->sk_ll_usec          =       sysctl_net_busy_read;
2384 #endif
2385 
2386         sk->sk_pacing_rate = ~0U;
2387         /*
2388          * Before updating sk_refcnt, we must commit prior changes to memory
2389          * (Documentation/RCU/rculist_nulls.txt for details)
2390          */
2391         smp_wmb();
2392         atomic_set(&sk->sk_refcnt, 1);
2393         atomic_set(&sk->sk_drops, 0);
2394 }
2395 EXPORT_SYMBOL(sock_init_data);
2396 
2397 void lock_sock_nested(struct sock *sk, int subclass)
2398 {
2399         might_sleep();
2400         spin_lock_bh(&sk->sk_lock.slock);
2401         if (sk->sk_lock.owned)
2402                 __lock_sock(sk);
2403         sk->sk_lock.owned = 1;
2404         spin_unlock(&sk->sk_lock.slock);
2405         /*
2406          * The sk_lock has mutex_lock() semantics here:
2407          */
2408         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2409         local_bh_enable();
2410 }
2411 EXPORT_SYMBOL(lock_sock_nested);
2412 
2413 void release_sock(struct sock *sk)
2414 {
2415         /*
2416          * The sk_lock has mutex_unlock() semantics:
2417          */
2418         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2419 
2420         spin_lock_bh(&sk->sk_lock.slock);
2421         if (sk->sk_backlog.tail)
2422                 __release_sock(sk);
2423 
2424         /* Warning : release_cb() might need to release sk ownership,
2425          * ie call sock_release_ownership(sk) before us.
2426          */
2427         if (sk->sk_prot->release_cb)
2428                 sk->sk_prot->release_cb(sk);
2429 
2430         sock_release_ownership(sk);
2431         if (waitqueue_active(&sk->sk_lock.wq))
2432                 wake_up(&sk->sk_lock.wq);
2433         spin_unlock_bh(&sk->sk_lock.slock);
2434 }
2435 EXPORT_SYMBOL(release_sock);
2436 
2437 /**
2438  * lock_sock_fast - fast version of lock_sock
2439  * @sk: socket
2440  *
2441  * This version should be used for very small section, where process wont block
2442  * return false if fast path is taken
2443  *   sk_lock.slock locked, owned = 0, BH disabled
2444  * return true if slow path is taken
2445  *   sk_lock.slock unlocked, owned = 1, BH enabled
2446  */
2447 bool lock_sock_fast(struct sock *sk)
2448 {
2449         might_sleep();
2450         spin_lock_bh(&sk->sk_lock.slock);
2451 
2452         if (!sk->sk_lock.owned)
2453                 /*
2454                  * Note : We must disable BH
2455                  */
2456                 return false;
2457 
2458         __lock_sock(sk);
2459         sk->sk_lock.owned = 1;
2460         spin_unlock(&sk->sk_lock.slock);
2461         /*
2462          * The sk_lock has mutex_lock() semantics here:
2463          */
2464         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2465         local_bh_enable();
2466         return true;
2467 }
2468 EXPORT_SYMBOL(lock_sock_fast);
2469 
2470 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2471 {
2472         struct timeval tv;
2473         if (!sock_flag(sk, SOCK_TIMESTAMP))
2474                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2475         tv = ktime_to_timeval(sk->sk_stamp);
2476         if (tv.tv_sec == -1)
2477                 return -ENOENT;
2478         if (tv.tv_sec == 0) {
2479                 sk->sk_stamp = ktime_get_real();
2480                 tv = ktime_to_timeval(sk->sk_stamp);
2481         }
2482         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2483 }
2484 EXPORT_SYMBOL(sock_get_timestamp);
2485 
2486 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2487 {
2488         struct timespec ts;
2489         if (!sock_flag(sk, SOCK_TIMESTAMP))
2490                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2491         ts = ktime_to_timespec(sk->sk_stamp);
2492         if (ts.tv_sec == -1)
2493                 return -ENOENT;
2494         if (ts.tv_sec == 0) {
2495                 sk->sk_stamp = ktime_get_real();
2496                 ts = ktime_to_timespec(sk->sk_stamp);
2497         }
2498         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2499 }
2500 EXPORT_SYMBOL(sock_get_timestampns);
2501 
2502 void sock_enable_timestamp(struct sock *sk, int flag)
2503 {
2504         if (!sock_flag(sk, flag)) {
2505                 unsigned long previous_flags = sk->sk_flags;
2506 
2507                 sock_set_flag(sk, flag);
2508                 /*
2509                  * we just set one of the two flags which require net
2510                  * time stamping, but time stamping might have been on
2511                  * already because of the other one
2512                  */
2513                 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2514                         net_enable_timestamp();
2515         }
2516 }
2517 
2518 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2519                        int level, int type)
2520 {
2521         struct sock_exterr_skb *serr;
2522         struct sk_buff *skb, *skb2;
2523         int copied, err;
2524 
2525         err = -EAGAIN;
2526         skb = skb_dequeue(&sk->sk_error_queue);
2527         if (skb == NULL)
2528                 goto out;
2529 
2530         copied = skb->len;
2531         if (copied > len) {
2532                 msg->msg_flags |= MSG_TRUNC;
2533                 copied = len;
2534         }
2535         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2536         if (err)
2537                 goto out_free_skb;
2538 
2539         sock_recv_timestamp(msg, sk, skb);
2540 
2541         serr = SKB_EXT_ERR(skb);
2542         put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2543 
2544         msg->msg_flags |= MSG_ERRQUEUE;
2545         err = copied;
2546 
2547         /* Reset and regenerate socket error */
2548         spin_lock_bh(&sk->sk_error_queue.lock);
2549         sk->sk_err = 0;
2550         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2551                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2552                 spin_unlock_bh(&sk->sk_error_queue.lock);
2553                 sk->sk_error_report(sk);
2554         } else
2555                 spin_unlock_bh(&sk->sk_error_queue.lock);
2556 
2557 out_free_skb:
2558         kfree_skb(skb);
2559 out:
2560         return err;
2561 }
2562 EXPORT_SYMBOL(sock_recv_errqueue);
2563 
2564 /*
2565  *      Get a socket option on an socket.
2566  *
2567  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2568  *      asynchronous errors should be reported by getsockopt. We assume
2569  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2570  */
2571 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2572                            char __user *optval, int __user *optlen)
2573 {
2574         struct sock *sk = sock->sk;
2575 
2576         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2577 }
2578 EXPORT_SYMBOL(sock_common_getsockopt);
2579 
2580 #ifdef CONFIG_COMPAT
2581 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2582                                   char __user *optval, int __user *optlen)
2583 {
2584         struct sock *sk = sock->sk;
2585 
2586         if (sk->sk_prot->compat_getsockopt != NULL)
2587                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2588                                                       optval, optlen);
2589         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2590 }
2591 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2592 #endif
2593 
2594 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2595                         struct msghdr *msg, size_t size, int flags)
2596 {
2597         struct sock *sk = sock->sk;
2598         int addr_len = 0;
2599         int err;
2600 
2601         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2602                                    flags & ~MSG_DONTWAIT, &addr_len);
2603         if (err >= 0)
2604                 msg->msg_namelen = addr_len;
2605         return err;
2606 }
2607 EXPORT_SYMBOL(sock_common_recvmsg);
2608 
2609 /*
2610  *      Set socket options on an inet socket.
2611  */
2612 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2613                            char __user *optval, unsigned int optlen)
2614 {
2615         struct sock *sk = sock->sk;
2616 
2617         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2618 }
2619 EXPORT_SYMBOL(sock_common_setsockopt);
2620 
2621 #ifdef CONFIG_COMPAT
2622 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2623                                   char __user *optval, unsigned int optlen)
2624 {
2625         struct sock *sk = sock->sk;
2626 
2627         if (sk->sk_prot->compat_setsockopt != NULL)
2628                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2629                                                       optval, optlen);
2630         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2631 }
2632 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2633 #endif
2634 
2635 void sk_common_release(struct sock *sk)
2636 {
2637         if (sk->sk_prot->destroy)
2638                 sk->sk_prot->destroy(sk);
2639 
2640         /*
2641          * Observation: when sock_common_release is called, processes have
2642          * no access to socket. But net still has.
2643          * Step one, detach it from networking:
2644          *
2645          * A. Remove from hash tables.
2646          */
2647 
2648         sk->sk_prot->unhash(sk);
2649 
2650         /*
2651          * In this point socket cannot receive new packets, but it is possible
2652          * that some packets are in flight because some CPU runs receiver and
2653          * did hash table lookup before we unhashed socket. They will achieve
2654          * receive queue and will be purged by socket destructor.
2655          *
2656          * Also we still have packets pending on receive queue and probably,
2657          * our own packets waiting in device queues. sock_destroy will drain
2658          * receive queue, but transmitted packets will delay socket destruction
2659          * until the last reference will be released.
2660          */
2661 
2662         sock_orphan(sk);
2663 
2664         xfrm_sk_free_policy(sk);
2665 
2666         sk_refcnt_debug_release(sk);
2667 
2668         sock_put(sk);
2669 }
2670 EXPORT_SYMBOL(sk_common_release);
2671 
2672 #ifdef CONFIG_PROC_FS
2673 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2674 struct prot_inuse {
2675         int val[PROTO_INUSE_NR];
2676 };
2677 
2678 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2679 
2680 #ifdef CONFIG_NET_NS
2681 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2682 {
2683         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2684 }
2685 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2686 
2687 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2688 {
2689         int cpu, idx = prot->inuse_idx;
2690         int res = 0;
2691 
2692         for_each_possible_cpu(cpu)
2693                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2694 
2695         return res >= 0 ? res : 0;
2696 }
2697 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2698 
2699 static int __net_init sock_inuse_init_net(struct net *net)
2700 {
2701         net->core.inuse = alloc_percpu(struct prot_inuse);
2702         return net->core.inuse ? 0 : -ENOMEM;
2703 }
2704 
2705 static void __net_exit sock_inuse_exit_net(struct net *net)
2706 {
2707         free_percpu(net->core.inuse);
2708 }
2709 
2710 static struct pernet_operations net_inuse_ops = {
2711         .init = sock_inuse_init_net,
2712         .exit = sock_inuse_exit_net,
2713 };
2714 
2715 static __init int net_inuse_init(void)
2716 {
2717         if (register_pernet_subsys(&net_inuse_ops))
2718                 panic("Cannot initialize net inuse counters");
2719 
2720         return 0;
2721 }
2722 
2723 core_initcall(net_inuse_init);
2724 #else
2725 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2726 
2727 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2728 {
2729         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2730 }
2731 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2732 
2733 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2734 {
2735         int cpu, idx = prot->inuse_idx;
2736         int res = 0;
2737 
2738         for_each_possible_cpu(cpu)
2739                 res += per_cpu(prot_inuse, cpu).val[idx];
2740 
2741         return res >= 0 ? res : 0;
2742 }
2743 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2744 #endif
2745 
2746 static void assign_proto_idx(struct proto *prot)
2747 {
2748         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2749 
2750         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2751                 pr_err("PROTO_INUSE_NR exhausted\n");
2752                 return;
2753         }
2754 
2755         set_bit(prot->inuse_idx, proto_inuse_idx);
2756 }
2757 
2758 static void release_proto_idx(struct proto *prot)
2759 {
2760         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2761                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2762 }
2763 #else
2764 static inline void assign_proto_idx(struct proto *prot)
2765 {
2766 }
2767 
2768 static inline void release_proto_idx(struct proto *prot)
2769 {
2770 }
2771 #endif
2772 
2773 int proto_register(struct proto *prot, int alloc_slab)
2774 {
2775         if (alloc_slab) {
2776                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2777                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2778                                         NULL);
2779 
2780                 if (prot->slab == NULL) {
2781                         pr_crit("%s: Can't create sock SLAB cache!\n",
2782                                 prot->name);
2783                         goto out;
2784                 }
2785 
2786                 if (prot->rsk_prot != NULL) {
2787                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2788                         if (prot->rsk_prot->slab_name == NULL)
2789                                 goto out_free_sock_slab;
2790 
2791                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2792                                                                  prot->rsk_prot->obj_size, 0,
2793                                                                  SLAB_HWCACHE_ALIGN, NULL);
2794 
2795                         if (prot->rsk_prot->slab == NULL) {
2796                                 pr_crit("%s: Can't create request sock SLAB cache!\n",
2797                                         prot->name);
2798                                 goto out_free_request_sock_slab_name;
2799                         }
2800                 }
2801 
2802                 if (prot->twsk_prot != NULL) {
2803                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2804 
2805                         if (prot->twsk_prot->twsk_slab_name == NULL)
2806                                 goto out_free_request_sock_slab;
2807 
2808                         prot->twsk_prot->twsk_slab =
2809                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2810                                                   prot->twsk_prot->twsk_obj_size,
2811                                                   0,
2812                                                   SLAB_HWCACHE_ALIGN |
2813                                                         prot->slab_flags,
2814                                                   NULL);
2815                         if (prot->twsk_prot->twsk_slab == NULL)
2816                                 goto out_free_timewait_sock_slab_name;
2817                 }
2818         }
2819 
2820         mutex_lock(&proto_list_mutex);
2821         list_add(&prot->node, &proto_list);
2822         assign_proto_idx(prot);
2823         mutex_unlock(&proto_list_mutex);
2824         return 0;
2825 
2826 out_free_timewait_sock_slab_name:
2827         kfree(prot->twsk_prot->twsk_slab_name);
2828 out_free_request_sock_slab:
2829         if (prot->rsk_prot && prot->rsk_prot->slab) {
2830                 kmem_cache_destroy(prot->rsk_prot->slab);
2831                 prot->rsk_prot->slab = NULL;
2832         }
2833 out_free_request_sock_slab_name:
2834         if (prot->rsk_prot)
2835                 kfree(prot->rsk_prot->slab_name);
2836 out_free_sock_slab:
2837         kmem_cache_destroy(prot->slab);
2838         prot->slab = NULL;
2839 out:
2840         return -ENOBUFS;
2841 }
2842 EXPORT_SYMBOL(proto_register);
2843 
2844 void proto_unregister(struct proto *prot)
2845 {
2846         mutex_lock(&proto_list_mutex);
2847         release_proto_idx(prot);
2848         list_del(&prot->node);
2849         mutex_unlock(&proto_list_mutex);
2850 
2851         if (prot->slab != NULL) {
2852                 kmem_cache_destroy(prot->slab);
2853                 prot->slab = NULL;
2854         }
2855 
2856         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2857                 kmem_cache_destroy(prot->rsk_prot->slab);
2858                 kfree(prot->rsk_prot->slab_name);
2859                 prot->rsk_prot->slab = NULL;
2860         }
2861 
2862         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2863                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2864                 kfree(prot->twsk_prot->twsk_slab_name);
2865                 prot->twsk_prot->twsk_slab = NULL;
2866         }
2867 }
2868 EXPORT_SYMBOL(proto_unregister);
2869 
2870 #ifdef CONFIG_PROC_FS
2871 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2872         __acquires(proto_list_mutex)
2873 {
2874         mutex_lock(&proto_list_mutex);
2875         return seq_list_start_head(&proto_list, *pos);
2876 }
2877 
2878 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2879 {
2880         return seq_list_next(v, &proto_list, pos);
2881 }
2882 
2883 static void proto_seq_stop(struct seq_file *seq, void *v)
2884         __releases(proto_list_mutex)
2885 {
2886         mutex_unlock(&proto_list_mutex);
2887 }
2888 
2889 static char proto_method_implemented(const void *method)
2890 {
2891         return method == NULL ? 'n' : 'y';
2892 }
2893 static long sock_prot_memory_allocated(struct proto *proto)
2894 {
2895         return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2896 }
2897 
2898 static char *sock_prot_memory_pressure(struct proto *proto)
2899 {
2900         return proto->memory_pressure != NULL ?
2901         proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2902 }
2903 
2904 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2905 {
2906 
2907         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2908                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2909                    proto->name,
2910                    proto->obj_size,
2911                    sock_prot_inuse_get(seq_file_net(seq), proto),
2912                    sock_prot_memory_allocated(proto),
2913                    sock_prot_memory_pressure(proto),
2914                    proto->max_header,
2915                    proto->slab == NULL ? "no" : "yes",
2916                    module_name(proto->owner),
2917                    proto_method_implemented(proto->close),
2918                    proto_method_implemented(proto->connect),
2919                    proto_method_implemented(proto->disconnect),
2920                    proto_method_implemented(proto->accept),
2921                    proto_method_implemented(proto->ioctl),
2922                    proto_method_implemented(proto->init),
2923                    proto_method_implemented(proto->destroy),
2924                    proto_method_implemented(proto->shutdown),
2925                    proto_method_implemented(proto->setsockopt),
2926                    proto_method_implemented(proto->getsockopt),
2927                    proto_method_implemented(proto->sendmsg),
2928                    proto_method_implemented(proto->recvmsg),
2929                    proto_method_implemented(proto->sendpage),
2930                    proto_method_implemented(proto->bind),
2931                    proto_method_implemented(proto->backlog_rcv),
2932                    proto_method_implemented(proto->hash),
2933                    proto_method_implemented(proto->unhash),
2934                    proto_method_implemented(proto->get_port),
2935                    proto_method_implemented(proto->enter_memory_pressure));
2936 }
2937 
2938 static int proto_seq_show(struct seq_file *seq, void *v)
2939 {
2940         if (v == &proto_list)
2941                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2942                            "protocol",
2943                            "size",
2944                            "sockets",
2945                            "memory",
2946                            "press",
2947                            "maxhdr",
2948                            "slab",
2949                            "module",
2950                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2951         else
2952                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2953         return 0;
2954 }
2955 
2956 static const struct seq_operations proto_seq_ops = {
2957         .start  = proto_seq_start,
2958         .next   = proto_seq_next,
2959         .stop   = proto_seq_stop,
2960         .show   = proto_seq_show,
2961 };
2962 
2963 static int proto_seq_open(struct inode *inode, struct file *file)
2964 {
2965         return seq_open_net(inode, file, &proto_seq_ops,
2966                             sizeof(struct seq_net_private));
2967 }
2968 
2969 static const struct file_operations proto_seq_fops = {
2970         .owner          = THIS_MODULE,
2971         .open           = proto_seq_open,
2972         .read           = seq_read,
2973         .llseek         = seq_lseek,
2974         .release        = seq_release_net,
2975 };
2976 
2977 static __net_init int proto_init_net(struct net *net)
2978 {
2979         if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2980                 return -ENOMEM;
2981 
2982         return 0;
2983 }
2984 
2985 static __net_exit void proto_exit_net(struct net *net)
2986 {
2987         remove_proc_entry("protocols", net->proc_net);
2988 }
2989 
2990 
2991 static __net_initdata struct pernet_operations proto_net_ops = {
2992         .init = proto_init_net,
2993         .exit = proto_exit_net,
2994 };
2995 
2996 static int __init proto_init(void)
2997 {
2998         return register_pernet_subsys(&proto_net_ops);
2999 }
3000 
3001 subsys_initcall(proto_init);
3002 
3003 #endif /* PROC_FS */
3004 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp