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

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

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ 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-3.16.85 ] ~ [ 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-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 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
526 {
527         struct dst_entry *dst = __sk_dst_get(sk);
528 
529         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
530                 sk_tx_queue_clear(sk);
531                 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
532                 dst_release(dst);
533                 return NULL;
534         }
535 
536         return dst;
537 }
538 EXPORT_SYMBOL(__sk_dst_check);
539 
540 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
541 {
542         struct dst_entry *dst = sk_dst_get(sk);
543 
544         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
545                 sk_dst_reset(sk);
546                 dst_release(dst);
547                 return NULL;
548         }
549 
550         return dst;
551 }
552 EXPORT_SYMBOL(sk_dst_check);
553 
554 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
555                                 int optlen)
556 {
557         int ret = -ENOPROTOOPT;
558 #ifdef CONFIG_NETDEVICES
559         struct net *net = sock_net(sk);
560         char devname[IFNAMSIZ];
561         int index;
562 
563         /* Sorry... */
564         ret = -EPERM;
565         if (!ns_capable(net->user_ns, CAP_NET_RAW))
566                 goto out;
567 
568         ret = -EINVAL;
569         if (optlen < 0)
570                 goto out;
571 
572         /* Bind this socket to a particular device like "eth0",
573          * as specified in the passed interface name. If the
574          * name is "" or the option length is zero the socket
575          * is not bound.
576          */
577         if (optlen > IFNAMSIZ - 1)
578                 optlen = IFNAMSIZ - 1;
579         memset(devname, 0, sizeof(devname));
580 
581         ret = -EFAULT;
582         if (copy_from_user(devname, optval, optlen))
583                 goto out;
584 
585         index = 0;
586         if (devname[0] != '\0') {
587                 struct net_device *dev;
588 
589                 rcu_read_lock();
590                 dev = dev_get_by_name_rcu(net, devname);
591                 if (dev)
592                         index = dev->ifindex;
593                 rcu_read_unlock();
594                 ret = -ENODEV;
595                 if (!dev)
596                         goto out;
597         }
598 
599         lock_sock(sk);
600         sk->sk_bound_dev_if = index;
601         sk_dst_reset(sk);
602         release_sock(sk);
603 
604         ret = 0;
605 
606 out:
607 #endif
608 
609         return ret;
610 }
611 
612 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
613                                 int __user *optlen, int len)
614 {
615         int ret = -ENOPROTOOPT;
616 #ifdef CONFIG_NETDEVICES
617         struct net *net = sock_net(sk);
618         char devname[IFNAMSIZ];
619 
620         if (sk->sk_bound_dev_if == 0) {
621                 len = 0;
622                 goto zero;
623         }
624 
625         ret = -EINVAL;
626         if (len < IFNAMSIZ)
627                 goto out;
628 
629         ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
630         if (ret)
631                 goto out;
632 
633         len = strlen(devname) + 1;
634 
635         ret = -EFAULT;
636         if (copy_to_user(optval, devname, len))
637                 goto out;
638 
639 zero:
640         ret = -EFAULT;
641         if (put_user(len, optlen))
642                 goto out;
643 
644         ret = 0;
645 
646 out:
647 #endif
648 
649         return ret;
650 }
651 
652 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
653 {
654         if (valbool)
655                 sock_set_flag(sk, bit);
656         else
657                 sock_reset_flag(sk, bit);
658 }
659 
660 bool sk_mc_loop(struct sock *sk)
661 {
662         if (dev_recursion_level())
663                 return false;
664         if (!sk)
665                 return true;
666         switch (sk->sk_family) {
667         case AF_INET:
668                 return inet_sk(sk)->mc_loop;
669 #if IS_ENABLED(CONFIG_IPV6)
670         case AF_INET6:
671                 return inet6_sk(sk)->mc_loop;
672 #endif
673         }
674         WARN_ON(1);
675         return true;
676 }
677 EXPORT_SYMBOL(sk_mc_loop);
678 
679 /*
680  *      This is meant for all protocols to use and covers goings on
681  *      at the socket level. Everything here is generic.
682  */
683 
684 int sock_setsockopt(struct socket *sock, int level, int optname,
685                     char __user *optval, unsigned int optlen)
686 {
687         struct sock *sk = sock->sk;
688         int val;
689         int valbool;
690         struct linger ling;
691         int ret = 0;
692 
693         /*
694          *      Options without arguments
695          */
696 
697         if (optname == SO_BINDTODEVICE)
698                 return sock_setbindtodevice(sk, optval, optlen);
699 
700         if (optlen < sizeof(int))
701                 return -EINVAL;
702 
703         if (get_user(val, (int __user *)optval))
704                 return -EFAULT;
705 
706         valbool = val ? 1 : 0;
707 
708         lock_sock(sk);
709 
710         switch (optname) {
711         case SO_DEBUG:
712                 if (val && !capable(CAP_NET_ADMIN))
713                         ret = -EACCES;
714                 else
715                         sock_valbool_flag(sk, SOCK_DBG, valbool);
716                 break;
717         case SO_REUSEADDR:
718                 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
719                 break;
720         case SO_REUSEPORT:
721                 sk->sk_reuseport = valbool;
722                 break;
723         case SO_TYPE:
724         case SO_PROTOCOL:
725         case SO_DOMAIN:
726         case SO_ERROR:
727                 ret = -ENOPROTOOPT;
728                 break;
729         case SO_DONTROUTE:
730                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
731                 break;
732         case SO_BROADCAST:
733                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
734                 break;
735         case SO_SNDBUF:
736                 /* Don't error on this BSD doesn't and if you think
737                  * about it this is right. Otherwise apps have to
738                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
739                  * are treated in BSD as hints
740                  */
741                 val = min_t(u32, val, sysctl_wmem_max);
742 set_sndbuf:
743                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
744                 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
745                 /* Wake up sending tasks if we upped the value. */
746                 sk->sk_write_space(sk);
747                 break;
748 
749         case SO_SNDBUFFORCE:
750                 if (!capable(CAP_NET_ADMIN)) {
751                         ret = -EPERM;
752                         break;
753                 }
754                 goto set_sndbuf;
755 
756         case SO_RCVBUF:
757                 /* Don't error on this BSD doesn't and if you think
758                  * about it this is right. Otherwise apps have to
759                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
760                  * are treated in BSD as hints
761                  */
762                 val = min_t(u32, val, sysctl_rmem_max);
763 set_rcvbuf:
764                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
765                 /*
766                  * We double it on the way in to account for
767                  * "struct sk_buff" etc. overhead.   Applications
768                  * assume that the SO_RCVBUF setting they make will
769                  * allow that much actual data to be received on that
770                  * socket.
771                  *
772                  * Applications are unaware that "struct sk_buff" and
773                  * other overheads allocate from the receive buffer
774                  * during socket buffer allocation.
775                  *
776                  * And after considering the possible alternatives,
777                  * returning the value we actually used in getsockopt
778                  * is the most desirable behavior.
779                  */
780                 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
781                 break;
782 
783         case SO_RCVBUFFORCE:
784                 if (!capable(CAP_NET_ADMIN)) {
785                         ret = -EPERM;
786                         break;
787                 }
788                 goto set_rcvbuf;
789 
790         case SO_KEEPALIVE:
791 #ifdef CONFIG_INET
792                 if (sk->sk_protocol == IPPROTO_TCP &&
793                     sk->sk_type == SOCK_STREAM)
794                         tcp_set_keepalive(sk, valbool);
795 #endif
796                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
797                 break;
798 
799         case SO_OOBINLINE:
800                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
801                 break;
802 
803         case SO_NO_CHECK:
804                 sk->sk_no_check = valbool;
805                 break;
806 
807         case SO_PRIORITY:
808                 if ((val >= 0 && val <= 6) ||
809                     ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
810                         sk->sk_priority = val;
811                 else
812                         ret = -EPERM;
813                 break;
814 
815         case SO_LINGER:
816                 if (optlen < sizeof(ling)) {
817                         ret = -EINVAL;  /* 1003.1g */
818                         break;
819                 }
820                 if (copy_from_user(&ling, optval, sizeof(ling))) {
821                         ret = -EFAULT;
822                         break;
823                 }
824                 if (!ling.l_onoff)
825                         sock_reset_flag(sk, SOCK_LINGER);
826                 else {
827 #if (BITS_PER_LONG == 32)
828                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
829                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
830                         else
831 #endif
832                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
833                         sock_set_flag(sk, SOCK_LINGER);
834                 }
835                 break;
836 
837         case SO_BSDCOMPAT:
838                 sock_warn_obsolete_bsdism("setsockopt");
839                 break;
840 
841         case SO_PASSCRED:
842                 if (valbool)
843                         set_bit(SOCK_PASSCRED, &sock->flags);
844                 else
845                         clear_bit(SOCK_PASSCRED, &sock->flags);
846                 break;
847 
848         case SO_TIMESTAMP:
849         case SO_TIMESTAMPNS:
850                 if (valbool)  {
851                         if (optname == SO_TIMESTAMP)
852                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
853                         else
854                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
855                         sock_set_flag(sk, SOCK_RCVTSTAMP);
856                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
857                 } else {
858                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
859                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
860                 }
861                 break;
862 
863         case SO_TIMESTAMPING:
864                 if (val & ~SOF_TIMESTAMPING_MASK) {
865                         ret = -EINVAL;
866                         break;
867                 }
868                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
869                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
870                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
871                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
872                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
873                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
874                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
875                         sock_enable_timestamp(sk,
876                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
877                 else
878                         sock_disable_timestamp(sk,
879                                                (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
880                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
881                                   val & SOF_TIMESTAMPING_SOFTWARE);
882                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
883                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
884                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
885                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
886                 break;
887 
888         case SO_RCVLOWAT:
889                 if (val < 0)
890                         val = INT_MAX;
891                 sk->sk_rcvlowat = val ? : 1;
892                 break;
893 
894         case SO_RCVTIMEO:
895                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
896                 break;
897 
898         case SO_SNDTIMEO:
899                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
900                 break;
901 
902         case SO_ATTACH_FILTER:
903                 ret = -EINVAL;
904                 if (optlen == sizeof(struct sock_fprog)) {
905                         struct sock_fprog fprog;
906 
907                         ret = -EFAULT;
908                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
909                                 break;
910 
911                         ret = sk_attach_filter(&fprog, sk);
912                 }
913                 break;
914 
915         case SO_DETACH_FILTER:
916                 ret = sk_detach_filter(sk);
917                 break;
918 
919         case SO_LOCK_FILTER:
920                 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
921                         ret = -EPERM;
922                 else
923                         sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
924                 break;
925 
926         case SO_PASSSEC:
927                 if (valbool)
928                         set_bit(SOCK_PASSSEC, &sock->flags);
929                 else
930                         clear_bit(SOCK_PASSSEC, &sock->flags);
931                 break;
932         case SO_MARK:
933                 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
934                         ret = -EPERM;
935                 else
936                         sk->sk_mark = val;
937                 break;
938 
939                 /* We implement the SO_SNDLOWAT etc to
940                    not be settable (1003.1g 5.3) */
941         case SO_RXQ_OVFL:
942                 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
943                 break;
944 
945         case SO_WIFI_STATUS:
946                 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
947                 break;
948 
949         case SO_PEEK_OFF:
950                 if (sock->ops->set_peek_off)
951                         ret = sock->ops->set_peek_off(sk, val);
952                 else
953                         ret = -EOPNOTSUPP;
954                 break;
955 
956         case SO_NOFCS:
957                 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
958                 break;
959 
960         case SO_SELECT_ERR_QUEUE:
961                 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
962                 break;
963 
964 #ifdef CONFIG_NET_RX_BUSY_POLL
965         case SO_BUSY_POLL:
966                 /* allow unprivileged users to decrease the value */
967                 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
968                         ret = -EPERM;
969                 else {
970                         if (val < 0)
971                                 ret = -EINVAL;
972                         else
973                                 sk->sk_ll_usec = val;
974                 }
975                 break;
976 #endif
977 
978         case SO_MAX_PACING_RATE:
979                 sk->sk_max_pacing_rate = val;
980                 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
981                                          sk->sk_max_pacing_rate);
982                 break;
983 
984         default:
985                 ret = -ENOPROTOOPT;
986                 break;
987         }
988         release_sock(sk);
989         return ret;
990 }
991 EXPORT_SYMBOL(sock_setsockopt);
992 
993 
994 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
995                           struct ucred *ucred)
996 {
997         ucred->pid = pid_vnr(pid);
998         ucred->uid = ucred->gid = -1;
999         if (cred) {
1000                 struct user_namespace *current_ns = current_user_ns();
1001 
1002                 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1003                 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1004         }
1005 }
1006 
1007 int sock_getsockopt(struct socket *sock, int level, int optname,
1008                     char __user *optval, int __user *optlen)
1009 {
1010         struct sock *sk = sock->sk;
1011 
1012         union {
1013                 int val;
1014                 struct linger ling;
1015                 struct timeval tm;
1016         } v;
1017 
1018         int lv = sizeof(int);
1019         int len;
1020 
1021         if (get_user(len, optlen))
1022                 return -EFAULT;
1023         if (len < 0)
1024                 return -EINVAL;
1025 
1026         memset(&v, 0, sizeof(v));
1027 
1028         switch (optname) {
1029         case SO_DEBUG:
1030                 v.val = sock_flag(sk, SOCK_DBG);
1031                 break;
1032 
1033         case SO_DONTROUTE:
1034                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1035                 break;
1036 
1037         case SO_BROADCAST:
1038                 v.val = sock_flag(sk, SOCK_BROADCAST);
1039                 break;
1040 
1041         case SO_SNDBUF:
1042                 v.val = sk->sk_sndbuf;
1043                 break;
1044 
1045         case SO_RCVBUF:
1046                 v.val = sk->sk_rcvbuf;
1047                 break;
1048 
1049         case SO_REUSEADDR:
1050                 v.val = sk->sk_reuse;
1051                 break;
1052 
1053         case SO_REUSEPORT:
1054                 v.val = sk->sk_reuseport;
1055                 break;
1056 
1057         case SO_KEEPALIVE:
1058                 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1059                 break;
1060 
1061         case SO_TYPE:
1062                 v.val = sk->sk_type;
1063                 break;
1064 
1065         case SO_PROTOCOL:
1066                 v.val = sk->sk_protocol;
1067                 break;
1068 
1069         case SO_DOMAIN:
1070                 v.val = sk->sk_family;
1071                 break;
1072 
1073         case SO_ERROR:
1074                 v.val = -sock_error(sk);
1075                 if (v.val == 0)
1076                         v.val = xchg(&sk->sk_err_soft, 0);
1077                 break;
1078 
1079         case SO_OOBINLINE:
1080                 v.val = sock_flag(sk, SOCK_URGINLINE);
1081                 break;
1082 
1083         case SO_NO_CHECK:
1084                 v.val = sk->sk_no_check;
1085                 break;
1086 
1087         case SO_PRIORITY:
1088                 v.val = sk->sk_priority;
1089                 break;
1090 
1091         case SO_LINGER:
1092                 lv              = sizeof(v.ling);
1093                 v.ling.l_onoff  = sock_flag(sk, SOCK_LINGER);
1094                 v.ling.l_linger = sk->sk_lingertime / HZ;
1095                 break;
1096 
1097         case SO_BSDCOMPAT:
1098                 sock_warn_obsolete_bsdism("getsockopt");
1099                 break;
1100 
1101         case SO_TIMESTAMP:
1102                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1103                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1104                 break;
1105 
1106         case SO_TIMESTAMPNS:
1107                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1108                 break;
1109 
1110         case SO_TIMESTAMPING:
1111                 v.val = 0;
1112                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1113                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1114                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1115                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1116                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1117                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1118                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1119                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1120                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1121                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
1122                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1123                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1124                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1125                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1126                 break;
1127 
1128         case SO_RCVTIMEO:
1129                 lv = sizeof(struct timeval);
1130                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1131                         v.tm.tv_sec = 0;
1132                         v.tm.tv_usec = 0;
1133                 } else {
1134                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1135                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1136                 }
1137                 break;
1138 
1139         case SO_SNDTIMEO:
1140                 lv = sizeof(struct timeval);
1141                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1142                         v.tm.tv_sec = 0;
1143                         v.tm.tv_usec = 0;
1144                 } else {
1145                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1146                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1147                 }
1148                 break;
1149 
1150         case SO_RCVLOWAT:
1151                 v.val = sk->sk_rcvlowat;
1152                 break;
1153 
1154         case SO_SNDLOWAT:
1155                 v.val = 1;
1156                 break;
1157 
1158         case SO_PASSCRED:
1159                 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1160                 break;
1161 
1162         case SO_PEERCRED:
1163         {
1164                 struct ucred peercred;
1165                 if (len > sizeof(peercred))
1166                         len = sizeof(peercred);
1167                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1168                 if (copy_to_user(optval, &peercred, len))
1169                         return -EFAULT;
1170                 goto lenout;
1171         }
1172 
1173         case SO_PEERNAME:
1174         {
1175                 char address[128];
1176 
1177                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1178                         return -ENOTCONN;
1179                 if (lv < len)
1180                         return -EINVAL;
1181                 if (copy_to_user(optval, address, len))
1182                         return -EFAULT;
1183                 goto lenout;
1184         }
1185 
1186         /* Dubious BSD thing... Probably nobody even uses it, but
1187          * the UNIX standard wants it for whatever reason... -DaveM
1188          */
1189         case SO_ACCEPTCONN:
1190                 v.val = sk->sk_state == TCP_LISTEN;
1191                 break;
1192 
1193         case SO_PASSSEC:
1194                 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1195                 break;
1196 
1197         case SO_PEERSEC:
1198                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1199 
1200         case SO_MARK:
1201                 v.val = sk->sk_mark;
1202                 break;
1203 
1204         case SO_RXQ_OVFL:
1205                 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1206                 break;
1207 
1208         case SO_WIFI_STATUS:
1209                 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1210                 break;
1211 
1212         case SO_PEEK_OFF:
1213                 if (!sock->ops->set_peek_off)
1214                         return -EOPNOTSUPP;
1215 
1216                 v.val = sk->sk_peek_off;
1217                 break;
1218         case SO_NOFCS:
1219                 v.val = sock_flag(sk, SOCK_NOFCS);
1220                 break;
1221 
1222         case SO_BINDTODEVICE:
1223                 return sock_getbindtodevice(sk, optval, optlen, len);
1224 
1225         case SO_GET_FILTER:
1226                 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1227                 if (len < 0)
1228                         return len;
1229 
1230                 goto lenout;
1231 
1232         case SO_LOCK_FILTER:
1233                 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1234                 break;
1235 
1236         case SO_BPF_EXTENSIONS:
1237                 v.val = bpf_tell_extensions();
1238                 break;
1239 
1240         case SO_SELECT_ERR_QUEUE:
1241                 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1242                 break;
1243 
1244 #ifdef CONFIG_NET_RX_BUSY_POLL
1245         case SO_BUSY_POLL:
1246                 v.val = sk->sk_ll_usec;
1247                 break;
1248 #endif
1249 
1250         case SO_MAX_PACING_RATE:
1251                 v.val = sk->sk_max_pacing_rate;
1252                 break;
1253 
1254         default:
1255                 return -ENOPROTOOPT;
1256         }
1257 
1258         if (len > lv)
1259                 len = lv;
1260         if (copy_to_user(optval, &v, len))
1261                 return -EFAULT;
1262 lenout:
1263         if (put_user(len, optlen))
1264                 return -EFAULT;
1265         return 0;
1266 }
1267 
1268 /*
1269  * Initialize an sk_lock.
1270  *
1271  * (We also register the sk_lock with the lock validator.)
1272  */
1273 static inline void sock_lock_init(struct sock *sk)
1274 {
1275         sock_lock_init_class_and_name(sk,
1276                         af_family_slock_key_strings[sk->sk_family],
1277                         af_family_slock_keys + sk->sk_family,
1278                         af_family_key_strings[sk->sk_family],
1279                         af_family_keys + sk->sk_family);
1280 }
1281 
1282 /*
1283  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1284  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1285  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1286  */
1287 static void sock_copy(struct sock *nsk, const struct sock *osk)
1288 {
1289 #ifdef CONFIG_SECURITY_NETWORK
1290         void *sptr = nsk->sk_security;
1291 #endif
1292         memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1293 
1294         memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1295                osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1296 
1297 #ifdef CONFIG_SECURITY_NETWORK
1298         nsk->sk_security = sptr;
1299         security_sk_clone(osk, nsk);
1300 #endif
1301 }
1302 
1303 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1304 {
1305         unsigned long nulls1, nulls2;
1306 
1307         nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1308         nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1309         if (nulls1 > nulls2)
1310                 swap(nulls1, nulls2);
1311 
1312         if (nulls1 != 0)
1313                 memset((char *)sk, 0, nulls1);
1314         memset((char *)sk + nulls1 + sizeof(void *), 0,
1315                nulls2 - nulls1 - sizeof(void *));
1316         memset((char *)sk + nulls2 + sizeof(void *), 0,
1317                size - nulls2 - sizeof(void *));
1318 }
1319 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1320 
1321 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1322                 int family)
1323 {
1324         struct sock *sk;
1325         struct kmem_cache *slab;
1326 
1327         slab = prot->slab;
1328         if (slab != NULL) {
1329                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1330                 if (!sk)
1331                         return sk;
1332                 if (priority & __GFP_ZERO) {
1333                         if (prot->clear_sk)
1334                                 prot->clear_sk(sk, prot->obj_size);
1335                         else
1336                                 sk_prot_clear_nulls(sk, prot->obj_size);
1337                 }
1338         } else
1339                 sk = kmalloc(prot->obj_size, priority);
1340 
1341         if (sk != NULL) {
1342                 kmemcheck_annotate_bitfield(sk, flags);
1343 
1344                 if (security_sk_alloc(sk, family, priority))
1345                         goto out_free;
1346 
1347                 if (!try_module_get(prot->owner))
1348                         goto out_free_sec;
1349                 sk_tx_queue_clear(sk);
1350         }
1351 
1352         return sk;
1353 
1354 out_free_sec:
1355         security_sk_free(sk);
1356 out_free:
1357         if (slab != NULL)
1358                 kmem_cache_free(slab, sk);
1359         else
1360                 kfree(sk);
1361         return NULL;
1362 }
1363 
1364 static void sk_prot_free(struct proto *prot, struct sock *sk)
1365 {
1366         struct kmem_cache *slab;
1367         struct module *owner;
1368 
1369         owner = prot->owner;
1370         slab = prot->slab;
1371 
1372         security_sk_free(sk);
1373         if (slab != NULL)
1374                 kmem_cache_free(slab, sk);
1375         else
1376                 kfree(sk);
1377         module_put(owner);
1378 }
1379 
1380 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1381 void sock_update_netprioidx(struct sock *sk)
1382 {
1383         if (in_interrupt())
1384                 return;
1385 
1386         sk->sk_cgrp_prioidx = task_netprioidx(current);
1387 }
1388 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1389 #endif
1390 
1391 /**
1392  *      sk_alloc - All socket objects are allocated here
1393  *      @net: the applicable net namespace
1394  *      @family: protocol family
1395  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1396  *      @prot: struct proto associated with this new sock instance
1397  */
1398 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1399                       struct proto *prot)
1400 {
1401         struct sock *sk;
1402 
1403         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1404         if (sk) {
1405                 sk->sk_family = family;
1406                 /*
1407                  * See comment in struct sock definition to understand
1408                  * why we need sk_prot_creator -acme
1409                  */
1410                 sk->sk_prot = sk->sk_prot_creator = prot;
1411                 sock_lock_init(sk);
1412                 sock_net_set(sk, get_net(net));
1413                 atomic_set(&sk->sk_wmem_alloc, 1);
1414 
1415                 sock_update_classid(sk);
1416                 sock_update_netprioidx(sk);
1417         }
1418 
1419         return sk;
1420 }
1421 EXPORT_SYMBOL(sk_alloc);
1422 
1423 static void __sk_free(struct sock *sk)
1424 {
1425         struct sk_filter *filter;
1426 
1427         if (sk->sk_destruct)
1428                 sk->sk_destruct(sk);
1429 
1430         filter = rcu_dereference_check(sk->sk_filter,
1431                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1432         if (filter) {
1433                 sk_filter_uncharge(sk, filter);
1434                 RCU_INIT_POINTER(sk->sk_filter, NULL);
1435         }
1436 
1437         sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1438 
1439         if (atomic_read(&sk->sk_omem_alloc))
1440                 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1441                          __func__, atomic_read(&sk->sk_omem_alloc));
1442 
1443         if (sk->sk_peer_cred)
1444                 put_cred(sk->sk_peer_cred);
1445         put_pid(sk->sk_peer_pid);
1446         put_net(sock_net(sk));
1447         sk_prot_free(sk->sk_prot_creator, sk);
1448 }
1449 
1450 void sk_free(struct sock *sk)
1451 {
1452         /*
1453          * We subtract one from sk_wmem_alloc and can know if
1454          * some packets are still in some tx queue.
1455          * If not null, sock_wfree() will call __sk_free(sk) later
1456          */
1457         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1458                 __sk_free(sk);
1459 }
1460 EXPORT_SYMBOL(sk_free);
1461 
1462 /*
1463  * Last sock_put should drop reference to sk->sk_net. It has already
1464  * been dropped in sk_change_net. Taking reference to stopping namespace
1465  * is not an option.
1466  * Take reference to a socket to remove it from hash _alive_ and after that
1467  * destroy it in the context of init_net.
1468  */
1469 void sk_release_kernel(struct sock *sk)
1470 {
1471         if (sk == NULL || sk->sk_socket == NULL)
1472                 return;
1473 
1474         sock_hold(sk);
1475         sock_release(sk->sk_socket);
1476         release_net(sock_net(sk));
1477         sock_net_set(sk, get_net(&init_net));
1478         sock_put(sk);
1479 }
1480 EXPORT_SYMBOL(sk_release_kernel);
1481 
1482 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1483 {
1484         if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1485                 sock_update_memcg(newsk);
1486 }
1487 
1488 /**
1489  *      sk_clone_lock - clone a socket, and lock its clone
1490  *      @sk: the socket to clone
1491  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1492  *
1493  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1494  */
1495 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1496 {
1497         struct sock *newsk;
1498 
1499         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1500         if (newsk != NULL) {
1501                 struct sk_filter *filter;
1502 
1503                 sock_copy(newsk, sk);
1504 
1505                 /* SANITY */
1506                 get_net(sock_net(newsk));
1507                 sk_node_init(&newsk->sk_node);
1508                 sock_lock_init(newsk);
1509                 bh_lock_sock(newsk);
1510                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1511                 newsk->sk_backlog.len = 0;
1512 
1513                 atomic_set(&newsk->sk_rmem_alloc, 0);
1514                 /*
1515                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1516                  */
1517                 atomic_set(&newsk->sk_wmem_alloc, 1);
1518                 atomic_set(&newsk->sk_omem_alloc, 0);
1519                 skb_queue_head_init(&newsk->sk_receive_queue);
1520                 skb_queue_head_init(&newsk->sk_write_queue);
1521 #ifdef CONFIG_NET_DMA
1522                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1523 #endif
1524 
1525                 spin_lock_init(&newsk->sk_dst_lock);
1526                 rwlock_init(&newsk->sk_callback_lock);
1527                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1528                                 af_callback_keys + newsk->sk_family,
1529                                 af_family_clock_key_strings[newsk->sk_family]);
1530 
1531                 newsk->sk_dst_cache     = NULL;
1532                 newsk->sk_wmem_queued   = 0;
1533                 newsk->sk_forward_alloc = 0;
1534                 newsk->sk_send_head     = NULL;
1535                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1536 
1537                 sock_reset_flag(newsk, SOCK_DONE);
1538                 skb_queue_head_init(&newsk->sk_error_queue);
1539 
1540                 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1541                 if (filter != NULL)
1542                         sk_filter_charge(newsk, filter);
1543 
1544                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1545                         /* It is still raw copy of parent, so invalidate
1546                          * destructor and make plain sk_free() */
1547                         newsk->sk_destruct = NULL;
1548                         bh_unlock_sock(newsk);
1549                         sk_free(newsk);
1550                         newsk = NULL;
1551                         goto out;
1552                 }
1553 
1554                 newsk->sk_err      = 0;
1555                 newsk->sk_priority = 0;
1556                 /*
1557                  * Before updating sk_refcnt, we must commit prior changes to memory
1558                  * (Documentation/RCU/rculist_nulls.txt for details)
1559                  */
1560                 smp_wmb();
1561                 atomic_set(&newsk->sk_refcnt, 2);
1562 
1563                 /*
1564                  * Increment the counter in the same struct proto as the master
1565                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1566                  * is the same as sk->sk_prot->socks, as this field was copied
1567                  * with memcpy).
1568                  *
1569                  * This _changes_ the previous behaviour, where
1570                  * tcp_create_openreq_child always was incrementing the
1571                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1572                  * to be taken into account in all callers. -acme
1573                  */
1574                 sk_refcnt_debug_inc(newsk);
1575                 sk_set_socket(newsk, NULL);
1576                 newsk->sk_wq = NULL;
1577 
1578                 sk_update_clone(sk, newsk);
1579 
1580                 if (newsk->sk_prot->sockets_allocated)
1581                         sk_sockets_allocated_inc(newsk);
1582 
1583                 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1584                         net_enable_timestamp();
1585         }
1586 out:
1587         return newsk;
1588 }
1589 EXPORT_SYMBOL_GPL(sk_clone_lock);
1590 
1591 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1592 {
1593         __sk_dst_set(sk, dst);
1594         sk->sk_route_caps = dst->dev->features;
1595         if (sk->sk_route_caps & NETIF_F_GSO)
1596                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1597         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1598         if (sk_can_gso(sk)) {
1599                 if (dst->header_len) {
1600                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1601                 } else {
1602                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1603                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1604                         sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1605                 }
1606         }
1607 }
1608 EXPORT_SYMBOL_GPL(sk_setup_caps);
1609 
1610 /*
1611  *      Simple resource managers for sockets.
1612  */
1613 
1614 
1615 /*
1616  * Write buffer destructor automatically called from kfree_skb.
1617  */
1618 void sock_wfree(struct sk_buff *skb)
1619 {
1620         struct sock *sk = skb->sk;
1621         unsigned int len = skb->truesize;
1622 
1623         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1624                 /*
1625                  * Keep a reference on sk_wmem_alloc, this will be released
1626                  * after sk_write_space() call
1627                  */
1628                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1629                 sk->sk_write_space(sk);
1630                 len = 1;
1631         }
1632         /*
1633          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1634          * could not do because of in-flight packets
1635          */
1636         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1637                 __sk_free(sk);
1638 }
1639 EXPORT_SYMBOL(sock_wfree);
1640 
1641 void skb_orphan_partial(struct sk_buff *skb)
1642 {
1643         /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1644          * so we do not completely orphan skb, but transfert all
1645          * accounted bytes but one, to avoid unexpected reorders.
1646          */
1647         if (skb->destructor == sock_wfree
1648 #ifdef CONFIG_INET
1649             || skb->destructor == tcp_wfree
1650 #endif
1651                 ) {
1652                 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1653                 skb->truesize = 1;
1654         } else {
1655                 skb_orphan(skb);
1656         }
1657 }
1658 EXPORT_SYMBOL(skb_orphan_partial);
1659 
1660 /*
1661  * Read buffer destructor automatically called from kfree_skb.
1662  */
1663 void sock_rfree(struct sk_buff *skb)
1664 {
1665         struct sock *sk = skb->sk;
1666         unsigned int len = skb->truesize;
1667 
1668         atomic_sub(len, &sk->sk_rmem_alloc);
1669         sk_mem_uncharge(sk, len);
1670 }
1671 EXPORT_SYMBOL(sock_rfree);
1672 
1673 void sock_edemux(struct sk_buff *skb)
1674 {
1675         struct sock *sk = skb->sk;
1676 
1677 #ifdef CONFIG_INET
1678         if (sk->sk_state == TCP_TIME_WAIT)
1679                 inet_twsk_put(inet_twsk(sk));
1680         else
1681 #endif
1682                 sock_put(sk);
1683 }
1684 EXPORT_SYMBOL(sock_edemux);
1685 
1686 kuid_t sock_i_uid(struct sock *sk)
1687 {
1688         kuid_t uid;
1689 
1690         read_lock_bh(&sk->sk_callback_lock);
1691         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1692         read_unlock_bh(&sk->sk_callback_lock);
1693         return uid;
1694 }
1695 EXPORT_SYMBOL(sock_i_uid);
1696 
1697 unsigned long sock_i_ino(struct sock *sk)
1698 {
1699         unsigned long ino;
1700 
1701         read_lock_bh(&sk->sk_callback_lock);
1702         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1703         read_unlock_bh(&sk->sk_callback_lock);
1704         return ino;
1705 }
1706 EXPORT_SYMBOL(sock_i_ino);
1707 
1708 /*
1709  * Allocate a skb from the socket's send buffer.
1710  */
1711 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1712                              gfp_t priority)
1713 {
1714         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1715                 struct sk_buff *skb = alloc_skb(size, priority);
1716                 if (skb) {
1717                         skb_set_owner_w(skb, sk);
1718                         return skb;
1719                 }
1720         }
1721         return NULL;
1722 }
1723 EXPORT_SYMBOL(sock_wmalloc);
1724 
1725 /*
1726  * Allocate a memory block from the socket's option memory buffer.
1727  */
1728 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1729 {
1730         if ((unsigned int)size <= sysctl_optmem_max &&
1731             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1732                 void *mem;
1733                 /* First do the add, to avoid the race if kmalloc
1734                  * might sleep.
1735                  */
1736                 atomic_add(size, &sk->sk_omem_alloc);
1737                 mem = kmalloc(size, priority);
1738                 if (mem)
1739                         return mem;
1740                 atomic_sub(size, &sk->sk_omem_alloc);
1741         }
1742         return NULL;
1743 }
1744 EXPORT_SYMBOL(sock_kmalloc);
1745 
1746 /*
1747  * Free an option memory block.
1748  */
1749 void sock_kfree_s(struct sock *sk, void *mem, int size)
1750 {
1751         kfree(mem);
1752         atomic_sub(size, &sk->sk_omem_alloc);
1753 }
1754 EXPORT_SYMBOL(sock_kfree_s);
1755 
1756 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1757    I think, these locks should be removed for datagram sockets.
1758  */
1759 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1760 {
1761         DEFINE_WAIT(wait);
1762 
1763         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1764         for (;;) {
1765                 if (!timeo)
1766                         break;
1767                 if (signal_pending(current))
1768                         break;
1769                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1770                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1771                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1772                         break;
1773                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1774                         break;
1775                 if (sk->sk_err)
1776                         break;
1777                 timeo = schedule_timeout(timeo);
1778         }
1779         finish_wait(sk_sleep(sk), &wait);
1780         return timeo;
1781 }
1782 
1783 
1784 /*
1785  *      Generic send/receive buffer handlers
1786  */
1787 
1788 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1789                                      unsigned long data_len, int noblock,
1790                                      int *errcode, int max_page_order)
1791 {
1792         struct sk_buff *skb = NULL;
1793         unsigned long chunk;
1794         gfp_t gfp_mask;
1795         long timeo;
1796         int err;
1797         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1798         struct page *page;
1799         int i;
1800 
1801         err = -EMSGSIZE;
1802         if (npages > MAX_SKB_FRAGS)
1803                 goto failure;
1804 
1805         timeo = sock_sndtimeo(sk, noblock);
1806         while (!skb) {
1807                 err = sock_error(sk);
1808                 if (err != 0)
1809                         goto failure;
1810 
1811                 err = -EPIPE;
1812                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1813                         goto failure;
1814 
1815                 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1816                         set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1817                         set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1818                         err = -EAGAIN;
1819                         if (!timeo)
1820                                 goto failure;
1821                         if (signal_pending(current))
1822                                 goto interrupted;
1823                         timeo = sock_wait_for_wmem(sk, timeo);
1824                         continue;
1825                 }
1826 
1827                 err = -ENOBUFS;
1828                 gfp_mask = sk->sk_allocation;
1829                 if (gfp_mask & __GFP_WAIT)
1830                         gfp_mask |= __GFP_REPEAT;
1831 
1832                 skb = alloc_skb(header_len, gfp_mask);
1833                 if (!skb)
1834                         goto failure;
1835 
1836                 skb->truesize += data_len;
1837 
1838                 for (i = 0; npages > 0; i++) {
1839                         int order = max_page_order;
1840 
1841                         while (order) {
1842                                 if (npages >= 1 << order) {
1843                                         page = alloc_pages(sk->sk_allocation |
1844                                                            __GFP_COMP |
1845                                                            __GFP_NOWARN |
1846                                                            __GFP_NORETRY,
1847                                                            order);
1848                                         if (page)
1849                                                 goto fill_page;
1850                                 }
1851                                 order--;
1852                         }
1853                         page = alloc_page(sk->sk_allocation);
1854                         if (!page)
1855                                 goto failure;
1856 fill_page:
1857                         chunk = min_t(unsigned long, data_len,
1858                                       PAGE_SIZE << order);
1859                         skb_fill_page_desc(skb, i, page, 0, chunk);
1860                         data_len -= chunk;
1861                         npages -= 1 << order;
1862                 }
1863         }
1864 
1865         skb_set_owner_w(skb, sk);
1866         return skb;
1867 
1868 interrupted:
1869         err = sock_intr_errno(timeo);
1870 failure:
1871         kfree_skb(skb);
1872         *errcode = err;
1873         return NULL;
1874 }
1875 EXPORT_SYMBOL(sock_alloc_send_pskb);
1876 
1877 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1878                                     int noblock, int *errcode)
1879 {
1880         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1881 }
1882 EXPORT_SYMBOL(sock_alloc_send_skb);
1883 
1884 /* On 32bit arches, an skb frag is limited to 2^15 */
1885 #define SKB_FRAG_PAGE_ORDER     get_order(32768)
1886 
1887 /**
1888  * skb_page_frag_refill - check that a page_frag contains enough room
1889  * @sz: minimum size of the fragment we want to get
1890  * @pfrag: pointer to page_frag
1891  * @prio: priority for memory allocation
1892  *
1893  * Note: While this allocator tries to use high order pages, there is
1894  * no guarantee that allocations succeed. Therefore, @sz MUST be
1895  * less or equal than PAGE_SIZE.
1896  */
1897 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio)
1898 {
1899         int order;
1900 
1901         if (pfrag->page) {
1902                 if (atomic_read(&pfrag->page->_count) == 1) {
1903                         pfrag->offset = 0;
1904                         return true;
1905                 }
1906                 if (pfrag->offset + sz <= pfrag->size)
1907                         return true;
1908                 put_page(pfrag->page);
1909         }
1910 
1911         order = SKB_FRAG_PAGE_ORDER;
1912         do {
1913                 gfp_t gfp = prio;
1914 
1915                 if (order) {
1916                         gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
1917                         gfp &= ~__GFP_WAIT;
1918                 }
1919                 pfrag->page = alloc_pages(gfp, order);
1920                 if (likely(pfrag->page)) {
1921                         pfrag->offset = 0;
1922                         pfrag->size = PAGE_SIZE << order;
1923                         return true;
1924                 }
1925         } while (--order >= 0);
1926 
1927         return false;
1928 }
1929 EXPORT_SYMBOL(skb_page_frag_refill);
1930 
1931 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1932 {
1933         if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1934                 return true;
1935 
1936         sk_enter_memory_pressure(sk);
1937         sk_stream_moderate_sndbuf(sk);
1938         return false;
1939 }
1940 EXPORT_SYMBOL(sk_page_frag_refill);
1941 
1942 static void __lock_sock(struct sock *sk)
1943         __releases(&sk->sk_lock.slock)
1944         __acquires(&sk->sk_lock.slock)
1945 {
1946         DEFINE_WAIT(wait);
1947 
1948         for (;;) {
1949                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1950                                         TASK_UNINTERRUPTIBLE);
1951                 spin_unlock_bh(&sk->sk_lock.slock);
1952                 schedule();
1953                 spin_lock_bh(&sk->sk_lock.slock);
1954                 if (!sock_owned_by_user(sk))
1955                         break;
1956         }
1957         finish_wait(&sk->sk_lock.wq, &wait);
1958 }
1959 
1960 static void __release_sock(struct sock *sk)
1961         __releases(&sk->sk_lock.slock)
1962         __acquires(&sk->sk_lock.slock)
1963 {
1964         struct sk_buff *skb = sk->sk_backlog.head;
1965 
1966         do {
1967                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1968                 bh_unlock_sock(sk);
1969 
1970                 do {
1971                         struct sk_buff *next = skb->next;
1972 
1973                         prefetch(next);
1974                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1975                         skb->next = NULL;
1976                         sk_backlog_rcv(sk, skb);
1977 
1978                         /*
1979                          * We are in process context here with softirqs
1980                          * disabled, use cond_resched_softirq() to preempt.
1981                          * This is safe to do because we've taken the backlog
1982                          * queue private:
1983                          */
1984                         cond_resched_softirq();
1985 
1986                         skb = next;
1987                 } while (skb != NULL);
1988 
1989                 bh_lock_sock(sk);
1990         } while ((skb = sk->sk_backlog.head) != NULL);
1991 
1992         /*
1993          * Doing the zeroing here guarantee we can not loop forever
1994          * while a wild producer attempts to flood us.
1995          */
1996         sk->sk_backlog.len = 0;
1997 }
1998 
1999 /**
2000  * sk_wait_data - wait for data to arrive at sk_receive_queue
2001  * @sk:    sock to wait on
2002  * @timeo: for how long
2003  *
2004  * Now socket state including sk->sk_err is changed only under lock,
2005  * hence we may omit checks after joining wait queue.
2006  * We check receive queue before schedule() only as optimization;
2007  * it is very likely that release_sock() added new data.
2008  */
2009 int sk_wait_data(struct sock *sk, long *timeo)
2010 {
2011         int rc;
2012         DEFINE_WAIT(wait);
2013 
2014         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2015         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2016         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
2017         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2018         finish_wait(sk_sleep(sk), &wait);
2019         return rc;
2020 }
2021 EXPORT_SYMBOL(sk_wait_data);
2022 
2023 /**
2024  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2025  *      @sk: socket
2026  *      @size: memory size to allocate
2027  *      @kind: allocation type
2028  *
2029  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2030  *      rmem allocation. This function assumes that protocols which have
2031  *      memory_pressure use sk_wmem_queued as write buffer accounting.
2032  */
2033 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2034 {
2035         struct proto *prot = sk->sk_prot;
2036         int amt = sk_mem_pages(size);
2037         long allocated;
2038         int parent_status = UNDER_LIMIT;
2039 
2040         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2041 
2042         allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2043 
2044         /* Under limit. */
2045         if (parent_status == UNDER_LIMIT &&
2046                         allocated <= sk_prot_mem_limits(sk, 0)) {
2047                 sk_leave_memory_pressure(sk);
2048                 return 1;
2049         }
2050 
2051         /* Under pressure. (we or our parents) */
2052         if ((parent_status > SOFT_LIMIT) ||
2053                         allocated > sk_prot_mem_limits(sk, 1))
2054                 sk_enter_memory_pressure(sk);
2055 
2056         /* Over hard limit (we or our parents) */
2057         if ((parent_status == OVER_LIMIT) ||
2058                         (allocated > sk_prot_mem_limits(sk, 2)))
2059                 goto suppress_allocation;
2060 
2061         /* guarantee minimum buffer size under pressure */
2062         if (kind == SK_MEM_RECV) {
2063                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2064                         return 1;
2065 
2066         } else { /* SK_MEM_SEND */
2067                 if (sk->sk_type == SOCK_STREAM) {
2068                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2069                                 return 1;
2070                 } else if (atomic_read(&sk->sk_wmem_alloc) <
2071                            prot->sysctl_wmem[0])
2072                                 return 1;
2073         }
2074 
2075         if (sk_has_memory_pressure(sk)) {
2076                 int alloc;
2077 
2078                 if (!sk_under_memory_pressure(sk))
2079                         return 1;
2080                 alloc = sk_sockets_allocated_read_positive(sk);
2081                 if (sk_prot_mem_limits(sk, 2) > alloc *
2082                     sk_mem_pages(sk->sk_wmem_queued +
2083                                  atomic_read(&sk->sk_rmem_alloc) +
2084                                  sk->sk_forward_alloc))
2085                         return 1;
2086         }
2087 
2088 suppress_allocation:
2089 
2090         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2091                 sk_stream_moderate_sndbuf(sk);
2092 
2093                 /* Fail only if socket is _under_ its sndbuf.
2094                  * In this case we cannot block, so that we have to fail.
2095                  */
2096                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2097                         return 1;
2098         }
2099 
2100         trace_sock_exceed_buf_limit(sk, prot, allocated);
2101 
2102         /* Alas. Undo changes. */
2103         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2104 
2105         sk_memory_allocated_sub(sk, amt);
2106 
2107         return 0;
2108 }
2109 EXPORT_SYMBOL(__sk_mem_schedule);
2110 
2111 /**
2112  *      __sk_reclaim - reclaim memory_allocated
2113  *      @sk: socket
2114  */
2115 void __sk_mem_reclaim(struct sock *sk)
2116 {
2117         sk_memory_allocated_sub(sk,
2118                                 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2119         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2120 
2121         if (sk_under_memory_pressure(sk) &&
2122             (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2123                 sk_leave_memory_pressure(sk);
2124 }
2125 EXPORT_SYMBOL(__sk_mem_reclaim);
2126 
2127 
2128 /*
2129  * Set of default routines for initialising struct proto_ops when
2130  * the protocol does not support a particular function. In certain
2131  * cases where it makes no sense for a protocol to have a "do nothing"
2132  * function, some default processing is provided.
2133  */
2134 
2135 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2136 {
2137         return -EOPNOTSUPP;
2138 }
2139 EXPORT_SYMBOL(sock_no_bind);
2140 
2141 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2142                     int len, int flags)
2143 {
2144         return -EOPNOTSUPP;
2145 }
2146 EXPORT_SYMBOL(sock_no_connect);
2147 
2148 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2149 {
2150         return -EOPNOTSUPP;
2151 }
2152 EXPORT_SYMBOL(sock_no_socketpair);
2153 
2154 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2155 {
2156         return -EOPNOTSUPP;
2157 }
2158 EXPORT_SYMBOL(sock_no_accept);
2159 
2160 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2161                     int *len, int peer)
2162 {
2163         return -EOPNOTSUPP;
2164 }
2165 EXPORT_SYMBOL(sock_no_getname);
2166 
2167 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2168 {
2169         return 0;
2170 }
2171 EXPORT_SYMBOL(sock_no_poll);
2172 
2173 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2174 {
2175         return -EOPNOTSUPP;
2176 }
2177 EXPORT_SYMBOL(sock_no_ioctl);
2178 
2179 int sock_no_listen(struct socket *sock, int backlog)
2180 {
2181         return -EOPNOTSUPP;
2182 }
2183 EXPORT_SYMBOL(sock_no_listen);
2184 
2185 int sock_no_shutdown(struct socket *sock, int how)
2186 {
2187         return -EOPNOTSUPP;
2188 }
2189 EXPORT_SYMBOL(sock_no_shutdown);
2190 
2191 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2192                     char __user *optval, unsigned int optlen)
2193 {
2194         return -EOPNOTSUPP;
2195 }
2196 EXPORT_SYMBOL(sock_no_setsockopt);
2197 
2198 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2199                     char __user *optval, int __user *optlen)
2200 {
2201         return -EOPNOTSUPP;
2202 }
2203 EXPORT_SYMBOL(sock_no_getsockopt);
2204 
2205 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2206                     size_t len)
2207 {
2208         return -EOPNOTSUPP;
2209 }
2210 EXPORT_SYMBOL(sock_no_sendmsg);
2211 
2212 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2213                     size_t len, int flags)
2214 {
2215         return -EOPNOTSUPP;
2216 }
2217 EXPORT_SYMBOL(sock_no_recvmsg);
2218 
2219 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2220 {
2221         /* Mirror missing mmap method error code */
2222         return -ENODEV;
2223 }
2224 EXPORT_SYMBOL(sock_no_mmap);
2225 
2226 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2227 {
2228         ssize_t res;
2229         struct msghdr msg = {.msg_flags = flags};
2230         struct kvec iov;
2231         char *kaddr = kmap(page);
2232         iov.iov_base = kaddr + offset;
2233         iov.iov_len = size;
2234         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2235         kunmap(page);
2236         return res;
2237 }
2238 EXPORT_SYMBOL(sock_no_sendpage);
2239 
2240 /*
2241  *      Default Socket Callbacks
2242  */
2243 
2244 static void sock_def_wakeup(struct sock *sk)
2245 {
2246         struct socket_wq *wq;
2247 
2248         rcu_read_lock();
2249         wq = rcu_dereference(sk->sk_wq);
2250         if (wq_has_sleeper(wq))
2251                 wake_up_interruptible_all(&wq->wait);
2252         rcu_read_unlock();
2253 }
2254 
2255 static void sock_def_error_report(struct sock *sk)
2256 {
2257         struct socket_wq *wq;
2258 
2259         rcu_read_lock();
2260         wq = rcu_dereference(sk->sk_wq);
2261         if (wq_has_sleeper(wq))
2262                 wake_up_interruptible_poll(&wq->wait, POLLERR);
2263         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2264         rcu_read_unlock();
2265 }
2266 
2267 static void sock_def_readable(struct sock *sk, int len)
2268 {
2269         struct socket_wq *wq;
2270 
2271         rcu_read_lock();
2272         wq = rcu_dereference(sk->sk_wq);
2273         if (wq_has_sleeper(wq))
2274                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2275                                                 POLLRDNORM | POLLRDBAND);
2276         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2277         rcu_read_unlock();
2278 }
2279 
2280 static void sock_def_write_space(struct sock *sk)
2281 {
2282         struct socket_wq *wq;
2283 
2284         rcu_read_lock();
2285 
2286         /* Do not wake up a writer until he can make "significant"
2287          * progress.  --DaveM
2288          */
2289         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2290                 wq = rcu_dereference(sk->sk_wq);
2291                 if (wq_has_sleeper(wq))
2292                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2293                                                 POLLWRNORM | POLLWRBAND);
2294 
2295                 /* Should agree with poll, otherwise some programs break */
2296                 if (sock_writeable(sk))
2297                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2298         }
2299 
2300         rcu_read_unlock();
2301 }
2302 
2303 static void sock_def_destruct(struct sock *sk)
2304 {
2305         kfree(sk->sk_protinfo);
2306 }
2307 
2308 void sk_send_sigurg(struct sock *sk)
2309 {
2310         if (sk->sk_socket && sk->sk_socket->file)
2311                 if (send_sigurg(&sk->sk_socket->file->f_owner))
2312                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2313 }
2314 EXPORT_SYMBOL(sk_send_sigurg);
2315 
2316 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2317                     unsigned long expires)
2318 {
2319         if (!mod_timer(timer, expires))
2320                 sock_hold(sk);
2321 }
2322 EXPORT_SYMBOL(sk_reset_timer);
2323 
2324 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2325 {
2326         if (del_timer(timer))
2327                 __sock_put(sk);
2328 }
2329 EXPORT_SYMBOL(sk_stop_timer);
2330 
2331 void sock_init_data(struct socket *sock, struct sock *sk)
2332 {
2333         skb_queue_head_init(&sk->sk_receive_queue);
2334         skb_queue_head_init(&sk->sk_write_queue);
2335         skb_queue_head_init(&sk->sk_error_queue);
2336 #ifdef CONFIG_NET_DMA
2337         skb_queue_head_init(&sk->sk_async_wait_queue);
2338 #endif
2339 
2340         sk->sk_send_head        =       NULL;
2341 
2342         init_timer(&sk->sk_timer);
2343 
2344         sk->sk_allocation       =       GFP_KERNEL;
2345         sk->sk_rcvbuf           =       sysctl_rmem_default;
2346         sk->sk_sndbuf           =       sysctl_wmem_default;
2347         sk->sk_state            =       TCP_CLOSE;
2348         sk_set_socket(sk, sock);
2349 
2350         sock_set_flag(sk, SOCK_ZAPPED);
2351 
2352         if (sock) {
2353                 sk->sk_type     =       sock->type;
2354                 sk->sk_wq       =       sock->wq;
2355                 sock->sk        =       sk;
2356         } else
2357                 sk->sk_wq       =       NULL;
2358 
2359         spin_lock_init(&sk->sk_dst_lock);
2360         rwlock_init(&sk->sk_callback_lock);
2361         lockdep_set_class_and_name(&sk->sk_callback_lock,
2362                         af_callback_keys + sk->sk_family,
2363                         af_family_clock_key_strings[sk->sk_family]);
2364 
2365         sk->sk_state_change     =       sock_def_wakeup;
2366         sk->sk_data_ready       =       sock_def_readable;
2367         sk->sk_write_space      =       sock_def_write_space;
2368         sk->sk_error_report     =       sock_def_error_report;
2369         sk->sk_destruct         =       sock_def_destruct;
2370 
2371         sk->sk_frag.page        =       NULL;
2372         sk->sk_frag.offset      =       0;
2373         sk->sk_peek_off         =       -1;
2374 
2375         sk->sk_peer_pid         =       NULL;
2376         sk->sk_peer_cred        =       NULL;
2377         sk->sk_write_pending    =       0;
2378         sk->sk_rcvlowat         =       1;
2379         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2380         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2381 
2382         sk->sk_stamp = ktime_set(-1L, 0);
2383 
2384 #ifdef CONFIG_NET_RX_BUSY_POLL
2385         sk->sk_napi_id          =       0;
2386         sk->sk_ll_usec          =       sysctl_net_busy_read;
2387 #endif
2388 
2389         sk->sk_max_pacing_rate = ~0U;
2390         sk->sk_pacing_rate = ~0U;
2391         /*
2392          * Before updating sk_refcnt, we must commit prior changes to memory
2393          * (Documentation/RCU/rculist_nulls.txt for details)
2394          */
2395         smp_wmb();
2396         atomic_set(&sk->sk_refcnt, 1);
2397         atomic_set(&sk->sk_drops, 0);
2398 }
2399 EXPORT_SYMBOL(sock_init_data);
2400 
2401 void lock_sock_nested(struct sock *sk, int subclass)
2402 {
2403         might_sleep();
2404         spin_lock_bh(&sk->sk_lock.slock);
2405         if (sk->sk_lock.owned)
2406                 __lock_sock(sk);
2407         sk->sk_lock.owned = 1;
2408         spin_unlock(&sk->sk_lock.slock);
2409         /*
2410          * The sk_lock has mutex_lock() semantics here:
2411          */
2412         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2413         local_bh_enable();
2414 }
2415 EXPORT_SYMBOL(lock_sock_nested);
2416 
2417 void release_sock(struct sock *sk)
2418 {
2419         /*
2420          * The sk_lock has mutex_unlock() semantics:
2421          */
2422         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2423 
2424         spin_lock_bh(&sk->sk_lock.slock);
2425         if (sk->sk_backlog.tail)
2426                 __release_sock(sk);
2427 
2428         /* Warning : release_cb() might need to release sk ownership,
2429          * ie call sock_release_ownership(sk) before us.
2430          */
2431         if (sk->sk_prot->release_cb)
2432                 sk->sk_prot->release_cb(sk);
2433 
2434         sock_release_ownership(sk);
2435         if (waitqueue_active(&sk->sk_lock.wq))
2436                 wake_up(&sk->sk_lock.wq);
2437         spin_unlock_bh(&sk->sk_lock.slock);
2438 }
2439 EXPORT_SYMBOL(release_sock);
2440 
2441 /**
2442  * lock_sock_fast - fast version of lock_sock
2443  * @sk: socket
2444  *
2445  * This version should be used for very small section, where process wont block
2446  * return false if fast path is taken
2447  *   sk_lock.slock locked, owned = 0, BH disabled
2448  * return true if slow path is taken
2449  *   sk_lock.slock unlocked, owned = 1, BH enabled
2450  */
2451 bool lock_sock_fast(struct sock *sk)
2452 {
2453         might_sleep();
2454         spin_lock_bh(&sk->sk_lock.slock);
2455 
2456         if (!sk->sk_lock.owned)
2457                 /*
2458                  * Note : We must disable BH
2459                  */
2460                 return false;
2461 
2462         __lock_sock(sk);
2463         sk->sk_lock.owned = 1;
2464         spin_unlock(&sk->sk_lock.slock);
2465         /*
2466          * The sk_lock has mutex_lock() semantics here:
2467          */
2468         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2469         local_bh_enable();
2470         return true;
2471 }
2472 EXPORT_SYMBOL(lock_sock_fast);
2473 
2474 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2475 {
2476         struct timeval tv;
2477         if (!sock_flag(sk, SOCK_TIMESTAMP))
2478                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2479         tv = ktime_to_timeval(sk->sk_stamp);
2480         if (tv.tv_sec == -1)
2481                 return -ENOENT;
2482         if (tv.tv_sec == 0) {
2483                 sk->sk_stamp = ktime_get_real();
2484                 tv = ktime_to_timeval(sk->sk_stamp);
2485         }
2486         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2487 }
2488 EXPORT_SYMBOL(sock_get_timestamp);
2489 
2490 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2491 {
2492         struct timespec ts;
2493         if (!sock_flag(sk, SOCK_TIMESTAMP))
2494                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2495         ts = ktime_to_timespec(sk->sk_stamp);
2496         if (ts.tv_sec == -1)
2497                 return -ENOENT;
2498         if (ts.tv_sec == 0) {
2499                 sk->sk_stamp = ktime_get_real();
2500                 ts = ktime_to_timespec(sk->sk_stamp);
2501         }
2502         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2503 }
2504 EXPORT_SYMBOL(sock_get_timestampns);
2505 
2506 void sock_enable_timestamp(struct sock *sk, int flag)
2507 {
2508         if (!sock_flag(sk, flag)) {
2509                 unsigned long previous_flags = sk->sk_flags;
2510 
2511                 sock_set_flag(sk, flag);
2512                 /*
2513                  * we just set one of the two flags which require net
2514                  * time stamping, but time stamping might have been on
2515                  * already because of the other one
2516                  */
2517                 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2518                         net_enable_timestamp();
2519         }
2520 }
2521 
2522 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2523                        int level, int type)
2524 {
2525         struct sock_exterr_skb *serr;
2526         struct sk_buff *skb, *skb2;
2527         int copied, err;
2528 
2529         err = -EAGAIN;
2530         skb = skb_dequeue(&sk->sk_error_queue);
2531         if (skb == NULL)
2532                 goto out;
2533 
2534         copied = skb->len;
2535         if (copied > len) {
2536                 msg->msg_flags |= MSG_TRUNC;
2537                 copied = len;
2538         }
2539         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2540         if (err)
2541                 goto out_free_skb;
2542 
2543         sock_recv_timestamp(msg, sk, skb);
2544 
2545         serr = SKB_EXT_ERR(skb);
2546         put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2547 
2548         msg->msg_flags |= MSG_ERRQUEUE;
2549         err = copied;
2550 
2551         /* Reset and regenerate socket error */
2552         spin_lock_bh(&sk->sk_error_queue.lock);
2553         sk->sk_err = 0;
2554         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2555                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2556                 spin_unlock_bh(&sk->sk_error_queue.lock);
2557                 sk->sk_error_report(sk);
2558         } else
2559                 spin_unlock_bh(&sk->sk_error_queue.lock);
2560 
2561 out_free_skb:
2562         kfree_skb(skb);
2563 out:
2564         return err;
2565 }
2566 EXPORT_SYMBOL(sock_recv_errqueue);
2567 
2568 /*
2569  *      Get a socket option on an socket.
2570  *
2571  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2572  *      asynchronous errors should be reported by getsockopt. We assume
2573  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2574  */
2575 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2576                            char __user *optval, int __user *optlen)
2577 {
2578         struct sock *sk = sock->sk;
2579 
2580         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2581 }
2582 EXPORT_SYMBOL(sock_common_getsockopt);
2583 
2584 #ifdef CONFIG_COMPAT
2585 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2586                                   char __user *optval, int __user *optlen)
2587 {
2588         struct sock *sk = sock->sk;
2589 
2590         if (sk->sk_prot->compat_getsockopt != NULL)
2591                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2592                                                       optval, optlen);
2593         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2594 }
2595 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2596 #endif
2597 
2598 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2599                         struct msghdr *msg, size_t size, int flags)
2600 {
2601         struct sock *sk = sock->sk;
2602         int addr_len = 0;
2603         int err;
2604 
2605         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2606                                    flags & ~MSG_DONTWAIT, &addr_len);
2607         if (err >= 0)
2608                 msg->msg_namelen = addr_len;
2609         return err;
2610 }
2611 EXPORT_SYMBOL(sock_common_recvmsg);
2612 
2613 /*
2614  *      Set socket options on an inet socket.
2615  */
2616 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2617                            char __user *optval, unsigned int optlen)
2618 {
2619         struct sock *sk = sock->sk;
2620 
2621         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2622 }
2623 EXPORT_SYMBOL(sock_common_setsockopt);
2624 
2625 #ifdef CONFIG_COMPAT
2626 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2627                                   char __user *optval, unsigned int optlen)
2628 {
2629         struct sock *sk = sock->sk;
2630 
2631         if (sk->sk_prot->compat_setsockopt != NULL)
2632                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2633                                                       optval, optlen);
2634         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2635 }
2636 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2637 #endif
2638 
2639 void sk_common_release(struct sock *sk)
2640 {
2641         if (sk->sk_prot->destroy)
2642                 sk->sk_prot->destroy(sk);
2643 
2644         /*
2645          * Observation: when sock_common_release is called, processes have
2646          * no access to socket. But net still has.
2647          * Step one, detach it from networking:
2648          *
2649          * A. Remove from hash tables.
2650          */
2651 
2652         sk->sk_prot->unhash(sk);
2653 
2654         /*
2655          * In this point socket cannot receive new packets, but it is possible
2656          * that some packets are in flight because some CPU runs receiver and
2657          * did hash table lookup before we unhashed socket. They will achieve
2658          * receive queue and will be purged by socket destructor.
2659          *
2660          * Also we still have packets pending on receive queue and probably,
2661          * our own packets waiting in device queues. sock_destroy will drain
2662          * receive queue, but transmitted packets will delay socket destruction
2663          * until the last reference will be released.
2664          */
2665 
2666         sock_orphan(sk);
2667 
2668         xfrm_sk_free_policy(sk);
2669 
2670         sk_refcnt_debug_release(sk);
2671 
2672         if (sk->sk_frag.page) {
2673                 put_page(sk->sk_frag.page);
2674                 sk->sk_frag.page = NULL;
2675         }
2676 
2677         sock_put(sk);
2678 }
2679 EXPORT_SYMBOL(sk_common_release);
2680 
2681 #ifdef CONFIG_PROC_FS
2682 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2683 struct prot_inuse {
2684         int val[PROTO_INUSE_NR];
2685 };
2686 
2687 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2688 
2689 #ifdef CONFIG_NET_NS
2690 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2691 {
2692         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2693 }
2694 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2695 
2696 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2697 {
2698         int cpu, idx = prot->inuse_idx;
2699         int res = 0;
2700 
2701         for_each_possible_cpu(cpu)
2702                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2703 
2704         return res >= 0 ? res : 0;
2705 }
2706 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2707 
2708 static int __net_init sock_inuse_init_net(struct net *net)
2709 {
2710         net->core.inuse = alloc_percpu(struct prot_inuse);
2711         return net->core.inuse ? 0 : -ENOMEM;
2712 }
2713 
2714 static void __net_exit sock_inuse_exit_net(struct net *net)
2715 {
2716         free_percpu(net->core.inuse);
2717 }
2718 
2719 static struct pernet_operations net_inuse_ops = {
2720         .init = sock_inuse_init_net,
2721         .exit = sock_inuse_exit_net,
2722 };
2723 
2724 static __init int net_inuse_init(void)
2725 {
2726         if (register_pernet_subsys(&net_inuse_ops))
2727                 panic("Cannot initialize net inuse counters");
2728 
2729         return 0;
2730 }
2731 
2732 core_initcall(net_inuse_init);
2733 #else
2734 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2735 
2736 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2737 {
2738         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2739 }
2740 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2741 
2742 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2743 {
2744         int cpu, idx = prot->inuse_idx;
2745         int res = 0;
2746 
2747         for_each_possible_cpu(cpu)
2748                 res += per_cpu(prot_inuse, cpu).val[idx];
2749 
2750         return res >= 0 ? res : 0;
2751 }
2752 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2753 #endif
2754 
2755 static void assign_proto_idx(struct proto *prot)
2756 {
2757         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2758 
2759         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2760                 pr_err("PROTO_INUSE_NR exhausted\n");
2761                 return;
2762         }
2763 
2764         set_bit(prot->inuse_idx, proto_inuse_idx);
2765 }
2766 
2767 static void release_proto_idx(struct proto *prot)
2768 {
2769         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2770                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2771 }
2772 #else
2773 static inline void assign_proto_idx(struct proto *prot)
2774 {
2775 }
2776 
2777 static inline void release_proto_idx(struct proto *prot)
2778 {
2779 }
2780 #endif
2781 
2782 int proto_register(struct proto *prot, int alloc_slab)
2783 {
2784         if (alloc_slab) {
2785                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2786                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2787                                         NULL);
2788 
2789                 if (prot->slab == NULL) {
2790                         pr_crit("%s: Can't create sock SLAB cache!\n",
2791                                 prot->name);
2792                         goto out;
2793                 }
2794 
2795                 if (prot->rsk_prot != NULL) {
2796                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2797                         if (prot->rsk_prot->slab_name == NULL)
2798                                 goto out_free_sock_slab;
2799 
2800                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2801                                                                  prot->rsk_prot->obj_size, 0,
2802                                                                  SLAB_HWCACHE_ALIGN, NULL);
2803 
2804                         if (prot->rsk_prot->slab == NULL) {
2805                                 pr_crit("%s: Can't create request sock SLAB cache!\n",
2806                                         prot->name);
2807                                 goto out_free_request_sock_slab_name;
2808                         }
2809                 }
2810 
2811                 if (prot->twsk_prot != NULL) {
2812                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2813 
2814                         if (prot->twsk_prot->twsk_slab_name == NULL)
2815                                 goto out_free_request_sock_slab;
2816 
2817                         prot->twsk_prot->twsk_slab =
2818                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2819                                                   prot->twsk_prot->twsk_obj_size,
2820                                                   0,
2821                                                   SLAB_HWCACHE_ALIGN |
2822                                                         prot->slab_flags,
2823                                                   NULL);
2824                         if (prot->twsk_prot->twsk_slab == NULL)
2825                                 goto out_free_timewait_sock_slab_name;
2826                 }
2827         }
2828 
2829         mutex_lock(&proto_list_mutex);
2830         list_add(&prot->node, &proto_list);
2831         assign_proto_idx(prot);
2832         mutex_unlock(&proto_list_mutex);
2833         return 0;
2834 
2835 out_free_timewait_sock_slab_name:
2836         kfree(prot->twsk_prot->twsk_slab_name);
2837 out_free_request_sock_slab:
2838         if (prot->rsk_prot && prot->rsk_prot->slab) {
2839                 kmem_cache_destroy(prot->rsk_prot->slab);
2840                 prot->rsk_prot->slab = NULL;
2841         }
2842 out_free_request_sock_slab_name:
2843         if (prot->rsk_prot)
2844                 kfree(prot->rsk_prot->slab_name);
2845 out_free_sock_slab:
2846         kmem_cache_destroy(prot->slab);
2847         prot->slab = NULL;
2848 out:
2849         return -ENOBUFS;
2850 }
2851 EXPORT_SYMBOL(proto_register);
2852 
2853 void proto_unregister(struct proto *prot)
2854 {
2855         mutex_lock(&proto_list_mutex);
2856         release_proto_idx(prot);
2857         list_del(&prot->node);
2858         mutex_unlock(&proto_list_mutex);
2859 
2860         if (prot->slab != NULL) {
2861                 kmem_cache_destroy(prot->slab);
2862                 prot->slab = NULL;
2863         }
2864 
2865         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2866                 kmem_cache_destroy(prot->rsk_prot->slab);
2867                 kfree(prot->rsk_prot->slab_name);
2868                 prot->rsk_prot->slab = NULL;
2869         }
2870 
2871         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2872                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2873                 kfree(prot->twsk_prot->twsk_slab_name);
2874                 prot->twsk_prot->twsk_slab = NULL;
2875         }
2876 }
2877 EXPORT_SYMBOL(proto_unregister);
2878 
2879 #ifdef CONFIG_PROC_FS
2880 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2881         __acquires(proto_list_mutex)
2882 {
2883         mutex_lock(&proto_list_mutex);
2884         return seq_list_start_head(&proto_list, *pos);
2885 }
2886 
2887 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2888 {
2889         return seq_list_next(v, &proto_list, pos);
2890 }
2891 
2892 static void proto_seq_stop(struct seq_file *seq, void *v)
2893         __releases(proto_list_mutex)
2894 {
2895         mutex_unlock(&proto_list_mutex);
2896 }
2897 
2898 static char proto_method_implemented(const void *method)
2899 {
2900         return method == NULL ? 'n' : 'y';
2901 }
2902 static long sock_prot_memory_allocated(struct proto *proto)
2903 {
2904         return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2905 }
2906 
2907 static char *sock_prot_memory_pressure(struct proto *proto)
2908 {
2909         return proto->memory_pressure != NULL ?
2910         proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2911 }
2912 
2913 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2914 {
2915 
2916         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2917                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2918                    proto->name,
2919                    proto->obj_size,
2920                    sock_prot_inuse_get(seq_file_net(seq), proto),
2921                    sock_prot_memory_allocated(proto),
2922                    sock_prot_memory_pressure(proto),
2923                    proto->max_header,
2924                    proto->slab == NULL ? "no" : "yes",
2925                    module_name(proto->owner),
2926                    proto_method_implemented(proto->close),
2927                    proto_method_implemented(proto->connect),
2928                    proto_method_implemented(proto->disconnect),
2929                    proto_method_implemented(proto->accept),
2930                    proto_method_implemented(proto->ioctl),
2931                    proto_method_implemented(proto->init),
2932                    proto_method_implemented(proto->destroy),
2933                    proto_method_implemented(proto->shutdown),
2934                    proto_method_implemented(proto->setsockopt),
2935                    proto_method_implemented(proto->getsockopt),
2936                    proto_method_implemented(proto->sendmsg),
2937                    proto_method_implemented(proto->recvmsg),
2938                    proto_method_implemented(proto->sendpage),
2939                    proto_method_implemented(proto->bind),
2940                    proto_method_implemented(proto->backlog_rcv),
2941                    proto_method_implemented(proto->hash),
2942                    proto_method_implemented(proto->unhash),
2943                    proto_method_implemented(proto->get_port),
2944                    proto_method_implemented(proto->enter_memory_pressure));
2945 }
2946 
2947 static int proto_seq_show(struct seq_file *seq, void *v)
2948 {
2949         if (v == &proto_list)
2950                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2951                            "protocol",
2952                            "size",
2953                            "sockets",
2954                            "memory",
2955                            "press",
2956                            "maxhdr",
2957                            "slab",
2958                            "module",
2959                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2960         else
2961                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2962         return 0;
2963 }
2964 
2965 static const struct seq_operations proto_seq_ops = {
2966         .start  = proto_seq_start,
2967         .next   = proto_seq_next,
2968         .stop   = proto_seq_stop,
2969         .show   = proto_seq_show,
2970 };
2971 
2972 static int proto_seq_open(struct inode *inode, struct file *file)
2973 {
2974         return seq_open_net(inode, file, &proto_seq_ops,
2975                             sizeof(struct seq_net_private));
2976 }
2977 
2978 static const struct file_operations proto_seq_fops = {
2979         .owner          = THIS_MODULE,
2980         .open           = proto_seq_open,
2981         .read           = seq_read,
2982         .llseek         = seq_lseek,
2983         .release        = seq_release_net,
2984 };
2985 
2986 static __net_init int proto_init_net(struct net *net)
2987 {
2988         if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2989                 return -ENOMEM;
2990 
2991         return 0;
2992 }
2993 
2994 static __net_exit void proto_exit_net(struct net *net)
2995 {
2996         remove_proc_entry("protocols", net->proc_net);
2997 }
2998 
2999 
3000 static __net_initdata struct pernet_operations proto_net_ops = {
3001         .init = proto_init_net,
3002         .exit = proto_exit_net,
3003 };
3004 
3005 static int __init proto_init(void)
3006 {
3007         return register_pernet_subsys(&proto_net_ops);
3008 }
3009 
3010 subsys_initcall(proto_init);
3011 
3012 #endif /* PROC_FS */
3013 

~ [ 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