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.
6 * Definitions for the AF_INET socket handler.
8 * Version: @(#)sock.h 1.0.4 05/13/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 * David S. Miller : New socket lookup architecture.
26 * Steve Whitehouse: Default routines for sock_ops
27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
28 * protinfo be just a void pointer, as the
29 * protocol specific parts were moved to
30 * respective headers and ipv4/v6, etc now
31 * use private slabcaches for its socks
32 * Pedro Hortas : New flags field for socket options
35 * This program is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU General Public License
37 * as published by the Free Software Foundation; either version
38 * 2 of the License, or (at your option) any later version.
43 #include <linux/hardirq.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/list_nulls.h>
47 #include <linux/timer.h>
48 #include <linux/cache.h>
49 #include <linux/lockdep.h>
50 #include <linux/netdevice.h>
51 #include <linux/skbuff.h> /* struct sk_buff */
53 #include <linux/security.h>
54 #include <linux/slab.h>
55 #include <linux/uaccess.h>
56 #include <linux/memcontrol.h>
57 #include <linux/res_counter.h>
58 #include <linux/jump_label.h>
60 #include <linux/filter.h>
61 #include <linux/rculist_nulls.h>
62 #include <linux/poll.h>
64 #include <linux/atomic.h>
66 #include <net/checksum.h>
71 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss);
72 void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss);
75 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
80 void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss)
85 * This structure really needs to be cleaned up.
86 * Most of it is for TCP, and not used by any of
87 * the other protocols.
90 /* Define this to get the SOCK_DBG debugging facility. */
91 #define SOCK_DEBUGGING
93 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
94 printk(KERN_DEBUG msg); } while (0)
96 /* Validate arguments and do nothing */
97 static inline __printf(2, 3)
98 void SOCK_DEBUG(struct sock *sk, const char *msg, ...)
103 /* This is the per-socket lock. The spinlock provides a synchronization
104 * between user contexts and software interrupt processing, whereas the
105 * mini-semaphore synchronizes multiple users amongst themselves.
110 wait_queue_head_t wq;
112 * We express the mutex-alike socket_lock semantics
113 * to the lock validator by explicitly managing
114 * the slock as a lock variant (in addition to
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 struct lockdep_map dep_map;
127 * struct sock_common - minimal network layer representation of sockets
128 * @skc_daddr: Foreign IPv4 addr
129 * @skc_rcv_saddr: Bound local IPv4 addr
130 * @skc_hash: hash value used with various protocol lookup tables
131 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
132 * @skc_family: network address family
133 * @skc_state: Connection state
134 * @skc_reuse: %SO_REUSEADDR setting
135 * @skc_bound_dev_if: bound device index if != 0
136 * @skc_bind_node: bind hash linkage for various protocol lookup tables
137 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
138 * @skc_prot: protocol handlers inside a network family
139 * @skc_net: reference to the network namespace of this socket
140 * @skc_node: main hash linkage for various protocol lookup tables
141 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
142 * @skc_tx_queue_mapping: tx queue number for this connection
143 * @skc_refcnt: reference count
145 * This is the minimal network layer representation of sockets, the header
146 * for struct sock and struct inet_timewait_sock.
149 /* skc_daddr and skc_rcv_saddr must be grouped :
150 * cf INET_MATCH() and INET_TW_MATCH()
153 __be32 skc_rcv_saddr;
156 unsigned int skc_hash;
157 __u16 skc_u16hashes[2];
159 unsigned short skc_family;
160 volatile unsigned char skc_state;
161 unsigned char skc_reuse;
162 int skc_bound_dev_if;
164 struct hlist_node skc_bind_node;
165 struct hlist_nulls_node skc_portaddr_node;
167 struct proto *skc_prot;
172 * fields between dontcopy_begin/dontcopy_end
173 * are not copied in sock_copy()
176 int skc_dontcopy_begin[0];
179 struct hlist_node skc_node;
180 struct hlist_nulls_node skc_nulls_node;
182 int skc_tx_queue_mapping;
185 int skc_dontcopy_end[0];
191 * struct sock - network layer representation of sockets
192 * @__sk_common: shared layout with inet_timewait_sock
193 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
194 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
195 * @sk_lock: synchronizer
196 * @sk_rcvbuf: size of receive buffer in bytes
197 * @sk_wq: sock wait queue and async head
198 * @sk_dst_cache: destination cache
199 * @sk_dst_lock: destination cache lock
200 * @sk_policy: flow policy
201 * @sk_receive_queue: incoming packets
202 * @sk_wmem_alloc: transmit queue bytes committed
203 * @sk_write_queue: Packet sending queue
204 * @sk_async_wait_queue: DMA copied packets
205 * @sk_omem_alloc: "o" is "option" or "other"
206 * @sk_wmem_queued: persistent queue size
207 * @sk_forward_alloc: space allocated forward
208 * @sk_allocation: allocation mode
209 * @sk_sndbuf: size of send buffer in bytes
210 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
211 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
212 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
213 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
214 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
215 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
216 * @sk_gso_max_size: Maximum GSO segment size to build
217 * @sk_lingertime: %SO_LINGER l_linger setting
218 * @sk_backlog: always used with the per-socket spinlock held
219 * @sk_callback_lock: used with the callbacks in the end of this struct
220 * @sk_error_queue: rarely used
221 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
222 * IPV6_ADDRFORM for instance)
223 * @sk_err: last error
224 * @sk_err_soft: errors that don't cause failure but are the cause of a
225 * persistent failure not just 'timed out'
226 * @sk_drops: raw/udp drops counter
227 * @sk_ack_backlog: current listen backlog
228 * @sk_max_ack_backlog: listen backlog set in listen()
229 * @sk_priority: %SO_PRIORITY setting
230 * @sk_cgrp_prioidx: socket group's priority map index
231 * @sk_type: socket type (%SOCK_STREAM, etc)
232 * @sk_protocol: which protocol this socket belongs in this network family
233 * @sk_peer_pid: &struct pid for this socket's peer
234 * @sk_peer_cred: %SO_PEERCRED setting
235 * @sk_rcvlowat: %SO_RCVLOWAT setting
236 * @sk_rcvtimeo: %SO_RCVTIMEO setting
237 * @sk_sndtimeo: %SO_SNDTIMEO setting
238 * @sk_rxhash: flow hash received from netif layer
239 * @sk_filter: socket filtering instructions
240 * @sk_protinfo: private area, net family specific, when not using slab
241 * @sk_timer: sock cleanup timer
242 * @sk_stamp: time stamp of last packet received
243 * @sk_socket: Identd and reporting IO signals
244 * @sk_user_data: RPC layer private data
245 * @sk_sndmsg_page: cached page for sendmsg
246 * @sk_sndmsg_off: cached offset for sendmsg
247 * @sk_send_head: front of stuff to transmit
248 * @sk_security: used by security modules
249 * @sk_mark: generic packet mark
250 * @sk_classid: this socket's cgroup classid
251 * @sk_cgrp: this socket's cgroup-specific proto data
252 * @sk_write_pending: a write to stream socket waits to start
253 * @sk_state_change: callback to indicate change in the state of the sock
254 * @sk_data_ready: callback to indicate there is data to be processed
255 * @sk_write_space: callback to indicate there is bf sending space available
256 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
257 * @sk_backlog_rcv: callback to process the backlog
258 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
262 * Now struct inet_timewait_sock also uses sock_common, so please just
263 * don't add nothing before this first member (__sk_common) --acme
265 struct sock_common __sk_common;
266 #define sk_node __sk_common.skc_node
267 #define sk_nulls_node __sk_common.skc_nulls_node
268 #define sk_refcnt __sk_common.skc_refcnt
269 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
271 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
272 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
273 #define sk_hash __sk_common.skc_hash
274 #define sk_family __sk_common.skc_family
275 #define sk_state __sk_common.skc_state
276 #define sk_reuse __sk_common.skc_reuse
277 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
278 #define sk_bind_node __sk_common.skc_bind_node
279 #define sk_prot __sk_common.skc_prot
280 #define sk_net __sk_common.skc_net
281 socket_lock_t sk_lock;
282 struct sk_buff_head sk_receive_queue;
284 * The backlog queue is special, it is always used with
285 * the per-socket spinlock held and requires low latency
286 * access. Therefore we special case it's implementation.
287 * Note : rmem_alloc is in this structure to fill a hole
288 * on 64bit arches, not because its logically part of
294 struct sk_buff *head;
295 struct sk_buff *tail;
297 #define sk_rmem_alloc sk_backlog.rmem_alloc
298 int sk_forward_alloc;
305 struct sk_filter __rcu *sk_filter;
306 struct socket_wq __rcu *sk_wq;
308 #ifdef CONFIG_NET_DMA
309 struct sk_buff_head sk_async_wait_queue;
313 struct xfrm_policy *sk_policy[2];
315 unsigned long sk_flags;
316 struct dst_entry *sk_dst_cache;
317 spinlock_t sk_dst_lock;
318 atomic_t sk_wmem_alloc;
319 atomic_t sk_omem_alloc;
321 struct sk_buff_head sk_write_queue;
322 kmemcheck_bitfield_begin(flags);
323 unsigned int sk_shutdown : 2,
328 kmemcheck_bitfield_end(flags);
331 netdev_features_t sk_route_caps;
332 netdev_features_t sk_route_nocaps;
334 unsigned int sk_gso_max_size;
336 unsigned long sk_lingertime;
337 struct sk_buff_head sk_error_queue;
338 struct proto *sk_prot_creator;
339 rwlock_t sk_callback_lock;
342 unsigned short sk_ack_backlog;
343 unsigned short sk_max_ack_backlog;
345 #ifdef CONFIG_CGROUPS
346 __u32 sk_cgrp_prioidx;
348 struct pid *sk_peer_pid;
349 const struct cred *sk_peer_cred;
353 struct timer_list sk_timer;
355 struct socket *sk_socket;
357 struct page *sk_sndmsg_page;
358 struct sk_buff *sk_send_head;
361 int sk_write_pending;
362 #ifdef CONFIG_SECURITY
367 struct cg_proto *sk_cgrp;
368 void (*sk_state_change)(struct sock *sk);
369 void (*sk_data_ready)(struct sock *sk, int bytes);
370 void (*sk_write_space)(struct sock *sk);
371 void (*sk_error_report)(struct sock *sk);
372 int (*sk_backlog_rcv)(struct sock *sk,
373 struct sk_buff *skb);
374 void (*sk_destruct)(struct sock *sk);
377 static inline int sk_peek_offset(struct sock *sk, int flags)
379 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
380 return sk->sk_peek_off;
385 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
387 if (sk->sk_peek_off >= 0) {
388 if (sk->sk_peek_off >= val)
389 sk->sk_peek_off -= val;
395 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
397 if (sk->sk_peek_off >= 0)
398 sk->sk_peek_off += val;
402 * Hashed lists helper routines
404 static inline struct sock *sk_entry(const struct hlist_node *node)
406 return hlist_entry(node, struct sock, sk_node);
409 static inline struct sock *__sk_head(const struct hlist_head *head)
411 return hlist_entry(head->first, struct sock, sk_node);
414 static inline struct sock *sk_head(const struct hlist_head *head)
416 return hlist_empty(head) ? NULL : __sk_head(head);
419 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
421 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
424 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
426 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
429 static inline struct sock *sk_next(const struct sock *sk)
431 return sk->sk_node.next ?
432 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
435 static inline struct sock *sk_nulls_next(const struct sock *sk)
437 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
438 hlist_nulls_entry(sk->sk_nulls_node.next,
439 struct sock, sk_nulls_node) :
443 static inline int sk_unhashed(const struct sock *sk)
445 return hlist_unhashed(&sk->sk_node);
448 static inline int sk_hashed(const struct sock *sk)
450 return !sk_unhashed(sk);
453 static __inline__ void sk_node_init(struct hlist_node *node)
458 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
463 static __inline__ void __sk_del_node(struct sock *sk)
465 __hlist_del(&sk->sk_node);
468 /* NB: equivalent to hlist_del_init_rcu */
469 static __inline__ int __sk_del_node_init(struct sock *sk)
473 sk_node_init(&sk->sk_node);
479 /* Grab socket reference count. This operation is valid only
480 when sk is ALREADY grabbed f.e. it is found in hash table
481 or a list and the lookup is made under lock preventing hash table
485 static inline void sock_hold(struct sock *sk)
487 atomic_inc(&sk->sk_refcnt);
490 /* Ungrab socket in the context, which assumes that socket refcnt
491 cannot hit zero, f.e. it is true in context of any socketcall.
493 static inline void __sock_put(struct sock *sk)
495 atomic_dec(&sk->sk_refcnt);
498 static __inline__ int sk_del_node_init(struct sock *sk)
500 int rc = __sk_del_node_init(sk);
503 /* paranoid for a while -acme */
504 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
509 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
511 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
514 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
520 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
522 int rc = __sk_nulls_del_node_init_rcu(sk);
525 /* paranoid for a while -acme */
526 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
532 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
534 hlist_add_head(&sk->sk_node, list);
537 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
540 __sk_add_node(sk, list);
543 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
546 hlist_add_head_rcu(&sk->sk_node, list);
549 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
551 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
554 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
557 __sk_nulls_add_node_rcu(sk, list);
560 static __inline__ void __sk_del_bind_node(struct sock *sk)
562 __hlist_del(&sk->sk_bind_node);
565 static __inline__ void sk_add_bind_node(struct sock *sk,
566 struct hlist_head *list)
568 hlist_add_head(&sk->sk_bind_node, list);
571 #define sk_for_each(__sk, node, list) \
572 hlist_for_each_entry(__sk, node, list, sk_node)
573 #define sk_for_each_rcu(__sk, node, list) \
574 hlist_for_each_entry_rcu(__sk, node, list, sk_node)
575 #define sk_nulls_for_each(__sk, node, list) \
576 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
577 #define sk_nulls_for_each_rcu(__sk, node, list) \
578 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
579 #define sk_for_each_from(__sk, node) \
580 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
581 hlist_for_each_entry_from(__sk, node, sk_node)
582 #define sk_nulls_for_each_from(__sk, node) \
583 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
584 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
585 #define sk_for_each_safe(__sk, node, tmp, list) \
586 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
587 #define sk_for_each_bound(__sk, node, list) \
588 hlist_for_each_entry(__sk, node, list, sk_bind_node)
601 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
602 SOCK_DBG, /* %SO_DEBUG setting */
603 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
604 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
605 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
606 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
607 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
608 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
609 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
610 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
611 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
612 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
613 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
614 SOCK_FASYNC, /* fasync() active */
616 SOCK_ZEROCOPY, /* buffers from userspace */
617 SOCK_WIFI_STATUS, /* push wifi status to userspace */
620 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
622 nsk->sk_flags = osk->sk_flags;
625 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
627 __set_bit(flag, &sk->sk_flags);
630 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
632 __clear_bit(flag, &sk->sk_flags);
635 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
637 return test_bit(flag, &sk->sk_flags);
640 static inline void sk_acceptq_removed(struct sock *sk)
642 sk->sk_ack_backlog--;
645 static inline void sk_acceptq_added(struct sock *sk)
647 sk->sk_ack_backlog++;
650 static inline int sk_acceptq_is_full(struct sock *sk)
652 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
656 * Compute minimal free write space needed to queue new packets.
658 static inline int sk_stream_min_wspace(struct sock *sk)
660 return sk->sk_wmem_queued >> 1;
663 static inline int sk_stream_wspace(struct sock *sk)
665 return sk->sk_sndbuf - sk->sk_wmem_queued;
668 extern void sk_stream_write_space(struct sock *sk);
670 static inline int sk_stream_memory_free(struct sock *sk)
672 return sk->sk_wmem_queued < sk->sk_sndbuf;
675 /* OOB backlog add */
676 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
678 /* dont let skb dst not refcounted, we are going to leave rcu lock */
681 if (!sk->sk_backlog.tail)
682 sk->sk_backlog.head = skb;
684 sk->sk_backlog.tail->next = skb;
686 sk->sk_backlog.tail = skb;
691 * Take into account size of receive queue and backlog queue
692 * Do not take into account this skb truesize,
693 * to allow even a single big packet to come.
695 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
697 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
699 return qsize > sk->sk_rcvbuf;
702 /* The per-socket spinlock must be held here. */
703 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
705 if (sk_rcvqueues_full(sk, skb))
708 __sk_add_backlog(sk, skb);
709 sk->sk_backlog.len += skb->truesize;
713 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
715 return sk->sk_backlog_rcv(sk, skb);
718 static inline void sock_rps_record_flow(const struct sock *sk)
721 struct rps_sock_flow_table *sock_flow_table;
724 sock_flow_table = rcu_dereference(rps_sock_flow_table);
725 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
730 static inline void sock_rps_reset_flow(const struct sock *sk)
733 struct rps_sock_flow_table *sock_flow_table;
736 sock_flow_table = rcu_dereference(rps_sock_flow_table);
737 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
742 static inline void sock_rps_save_rxhash(struct sock *sk,
743 const struct sk_buff *skb)
746 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
747 sock_rps_reset_flow(sk);
748 sk->sk_rxhash = skb->rxhash;
753 static inline void sock_rps_reset_rxhash(struct sock *sk)
756 sock_rps_reset_flow(sk);
761 #define sk_wait_event(__sk, __timeo, __condition) \
763 release_sock(__sk); \
764 __rc = __condition; \
766 *(__timeo) = schedule_timeout(*(__timeo)); \
769 __rc = __condition; \
773 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
774 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
775 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
776 extern int sk_stream_error(struct sock *sk, int flags, int err);
777 extern void sk_stream_kill_queues(struct sock *sk);
779 extern int sk_wait_data(struct sock *sk, long *timeo);
781 struct request_sock_ops;
782 struct timewait_sock_ops;
783 struct inet_hashinfo;
787 /* Networking protocol blocks we attach to sockets.
788 * socket layer -> transport layer interface
789 * transport -> network interface is defined by struct inet_proto
792 void (*close)(struct sock *sk,
794 int (*connect)(struct sock *sk,
795 struct sockaddr *uaddr,
797 int (*disconnect)(struct sock *sk, int flags);
799 struct sock * (*accept) (struct sock *sk, int flags, int *err);
801 int (*ioctl)(struct sock *sk, int cmd,
803 int (*init)(struct sock *sk);
804 void (*destroy)(struct sock *sk);
805 void (*shutdown)(struct sock *sk, int how);
806 int (*setsockopt)(struct sock *sk, int level,
807 int optname, char __user *optval,
808 unsigned int optlen);
809 int (*getsockopt)(struct sock *sk, int level,
810 int optname, char __user *optval,
813 int (*compat_setsockopt)(struct sock *sk,
815 int optname, char __user *optval,
816 unsigned int optlen);
817 int (*compat_getsockopt)(struct sock *sk,
819 int optname, char __user *optval,
821 int (*compat_ioctl)(struct sock *sk,
822 unsigned int cmd, unsigned long arg);
824 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
825 struct msghdr *msg, size_t len);
826 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
828 size_t len, int noblock, int flags,
830 int (*sendpage)(struct sock *sk, struct page *page,
831 int offset, size_t size, int flags);
832 int (*bind)(struct sock *sk,
833 struct sockaddr *uaddr, int addr_len);
835 int (*backlog_rcv) (struct sock *sk,
836 struct sk_buff *skb);
838 /* Keeping track of sk's, looking them up, and port selection methods. */
839 void (*hash)(struct sock *sk);
840 void (*unhash)(struct sock *sk);
841 void (*rehash)(struct sock *sk);
842 int (*get_port)(struct sock *sk, unsigned short snum);
843 void (*clear_sk)(struct sock *sk, int size);
845 /* Keeping track of sockets in use */
846 #ifdef CONFIG_PROC_FS
847 unsigned int inuse_idx;
850 /* Memory pressure */
851 void (*enter_memory_pressure)(struct sock *sk);
852 atomic_long_t *memory_allocated; /* Current allocated memory. */
853 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
855 * Pressure flag: try to collapse.
856 * Technical note: it is used by multiple contexts non atomically.
857 * All the __sk_mem_schedule() is of this nature: accounting
858 * is strict, actions are advisory and have some latency.
860 int *memory_pressure;
867 struct kmem_cache *slab;
868 unsigned int obj_size;
871 struct percpu_counter *orphan_count;
873 struct request_sock_ops *rsk_prot;
874 struct timewait_sock_ops *twsk_prot;
877 struct inet_hashinfo *hashinfo;
878 struct udp_table *udp_table;
879 struct raw_hashinfo *raw_hash;
882 struct module *owner;
886 struct list_head node;
887 #ifdef SOCK_REFCNT_DEBUG
890 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
892 * cgroup specific init/deinit functions. Called once for all
893 * protocols that implement it, from cgroups populate function.
894 * This function has to setup any files the protocol want to
895 * appear in the kmem cgroup filesystem.
897 int (*init_cgroup)(struct cgroup *cgrp,
898 struct cgroup_subsys *ss);
899 void (*destroy_cgroup)(struct cgroup *cgrp,
900 struct cgroup_subsys *ss);
901 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
906 void (*enter_memory_pressure)(struct sock *sk);
907 struct res_counter *memory_allocated; /* Current allocated memory. */
908 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
909 int *memory_pressure;
912 * memcg field is used to find which memcg we belong directly
913 * Each memcg struct can hold more than one cg_proto, so container_of
916 * The elegant solution would be having an inverse function to
917 * proto_cgroup in struct proto, but that means polluting the structure
918 * for everybody, instead of just for memcg users.
920 struct mem_cgroup *memcg;
923 extern int proto_register(struct proto *prot, int alloc_slab);
924 extern void proto_unregister(struct proto *prot);
926 #ifdef SOCK_REFCNT_DEBUG
927 static inline void sk_refcnt_debug_inc(struct sock *sk)
929 atomic_inc(&sk->sk_prot->socks);
932 static inline void sk_refcnt_debug_dec(struct sock *sk)
934 atomic_dec(&sk->sk_prot->socks);
935 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
936 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
939 inline void sk_refcnt_debug_release(const struct sock *sk)
941 if (atomic_read(&sk->sk_refcnt) != 1)
942 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
943 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
945 #else /* SOCK_REFCNT_DEBUG */
946 #define sk_refcnt_debug_inc(sk) do { } while (0)
947 #define sk_refcnt_debug_dec(sk) do { } while (0)
948 #define sk_refcnt_debug_release(sk) do { } while (0)
949 #endif /* SOCK_REFCNT_DEBUG */
951 #if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM) && defined(CONFIG_NET)
952 extern struct jump_label_key memcg_socket_limit_enabled;
953 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
954 struct cg_proto *cg_proto)
956 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
958 #define mem_cgroup_sockets_enabled static_branch(&memcg_socket_limit_enabled)
960 #define mem_cgroup_sockets_enabled 0
961 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
962 struct cg_proto *cg_proto)
969 static inline bool sk_has_memory_pressure(const struct sock *sk)
971 return sk->sk_prot->memory_pressure != NULL;
974 static inline bool sk_under_memory_pressure(const struct sock *sk)
976 if (!sk->sk_prot->memory_pressure)
979 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
980 return !!*sk->sk_cgrp->memory_pressure;
982 return !!*sk->sk_prot->memory_pressure;
985 static inline void sk_leave_memory_pressure(struct sock *sk)
987 int *memory_pressure = sk->sk_prot->memory_pressure;
989 if (!memory_pressure)
992 if (*memory_pressure)
993 *memory_pressure = 0;
995 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
996 struct cg_proto *cg_proto = sk->sk_cgrp;
997 struct proto *prot = sk->sk_prot;
999 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1000 if (*cg_proto->memory_pressure)
1001 *cg_proto->memory_pressure = 0;
1006 static inline void sk_enter_memory_pressure(struct sock *sk)
1008 if (!sk->sk_prot->enter_memory_pressure)
1011 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1012 struct cg_proto *cg_proto = sk->sk_cgrp;
1013 struct proto *prot = sk->sk_prot;
1015 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1016 cg_proto->enter_memory_pressure(sk);
1019 sk->sk_prot->enter_memory_pressure(sk);
1022 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1024 long *prot = sk->sk_prot->sysctl_mem;
1025 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1026 prot = sk->sk_cgrp->sysctl_mem;
1030 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1034 struct res_counter *fail;
1037 ret = res_counter_charge_nofail(prot->memory_allocated,
1038 amt << PAGE_SHIFT, &fail);
1040 *parent_status = OVER_LIMIT;
1043 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1046 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1049 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1052 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1053 return ret >> PAGE_SHIFT;
1057 sk_memory_allocated(const struct sock *sk)
1059 struct proto *prot = sk->sk_prot;
1060 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1061 return memcg_memory_allocated_read(sk->sk_cgrp);
1063 return atomic_long_read(prot->memory_allocated);
1067 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1069 struct proto *prot = sk->sk_prot;
1071 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1072 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1073 /* update the root cgroup regardless */
1074 atomic_long_add_return(amt, prot->memory_allocated);
1075 return memcg_memory_allocated_read(sk->sk_cgrp);
1078 return atomic_long_add_return(amt, prot->memory_allocated);
1082 sk_memory_allocated_sub(struct sock *sk, int amt)
1084 struct proto *prot = sk->sk_prot;
1086 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1087 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1089 atomic_long_sub(amt, prot->memory_allocated);
1092 static inline void sk_sockets_allocated_dec(struct sock *sk)
1094 struct proto *prot = sk->sk_prot;
1096 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1097 struct cg_proto *cg_proto = sk->sk_cgrp;
1099 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1100 percpu_counter_dec(cg_proto->sockets_allocated);
1103 percpu_counter_dec(prot->sockets_allocated);
1106 static inline void sk_sockets_allocated_inc(struct sock *sk)
1108 struct proto *prot = sk->sk_prot;
1110 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1111 struct cg_proto *cg_proto = sk->sk_cgrp;
1113 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1114 percpu_counter_inc(cg_proto->sockets_allocated);
1117 percpu_counter_inc(prot->sockets_allocated);
1121 sk_sockets_allocated_read_positive(struct sock *sk)
1123 struct proto *prot = sk->sk_prot;
1125 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1126 return percpu_counter_sum_positive(sk->sk_cgrp->sockets_allocated);
1128 return percpu_counter_sum_positive(prot->sockets_allocated);
1132 proto_sockets_allocated_sum_positive(struct proto *prot)
1134 return percpu_counter_sum_positive(prot->sockets_allocated);
1138 proto_memory_allocated(struct proto *prot)
1140 return atomic_long_read(prot->memory_allocated);
1144 proto_memory_pressure(struct proto *prot)
1146 if (!prot->memory_pressure)
1148 return !!*prot->memory_pressure;
1152 #ifdef CONFIG_PROC_FS
1153 /* Called with local bh disabled */
1154 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1155 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1157 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
1164 /* With per-bucket locks this operation is not-atomic, so that
1165 * this version is not worse.
1167 static inline void __sk_prot_rehash(struct sock *sk)
1169 sk->sk_prot->unhash(sk);
1170 sk->sk_prot->hash(sk);
1173 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1175 /* About 10 seconds */
1176 #define SOCK_DESTROY_TIME (10*HZ)
1178 /* Sockets 0-1023 can't be bound to unless you are superuser */
1179 #define PROT_SOCK 1024
1181 #define SHUTDOWN_MASK 3
1182 #define RCV_SHUTDOWN 1
1183 #define SEND_SHUTDOWN 2
1185 #define SOCK_SNDBUF_LOCK 1
1186 #define SOCK_RCVBUF_LOCK 2
1187 #define SOCK_BINDADDR_LOCK 4
1188 #define SOCK_BINDPORT_LOCK 8
1190 /* sock_iocb: used to kick off async processing of socket ios */
1192 struct list_head list;
1196 struct socket *sock;
1198 struct scm_cookie *scm;
1199 struct msghdr *msg, async_msg;
1200 struct kiocb *kiocb;
1203 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1205 return (struct sock_iocb *)iocb->private;
1208 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1213 struct socket_alloc {
1214 struct socket socket;
1215 struct inode vfs_inode;
1218 static inline struct socket *SOCKET_I(struct inode *inode)
1220 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1223 static inline struct inode *SOCK_INODE(struct socket *socket)
1225 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1229 * Functions for memory accounting
1231 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1232 extern void __sk_mem_reclaim(struct sock *sk);
1234 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1235 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1236 #define SK_MEM_SEND 0
1237 #define SK_MEM_RECV 1
1239 static inline int sk_mem_pages(int amt)
1241 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1244 static inline int sk_has_account(struct sock *sk)
1246 /* return true if protocol supports memory accounting */
1247 return !!sk->sk_prot->memory_allocated;
1250 static inline int sk_wmem_schedule(struct sock *sk, int size)
1252 if (!sk_has_account(sk))
1254 return size <= sk->sk_forward_alloc ||
1255 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1258 static inline int sk_rmem_schedule(struct sock *sk, int size)
1260 if (!sk_has_account(sk))
1262 return size <= sk->sk_forward_alloc ||
1263 __sk_mem_schedule(sk, size, SK_MEM_RECV);
1266 static inline void sk_mem_reclaim(struct sock *sk)
1268 if (!sk_has_account(sk))
1270 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1271 __sk_mem_reclaim(sk);
1274 static inline void sk_mem_reclaim_partial(struct sock *sk)
1276 if (!sk_has_account(sk))
1278 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1279 __sk_mem_reclaim(sk);
1282 static inline void sk_mem_charge(struct sock *sk, int size)
1284 if (!sk_has_account(sk))
1286 sk->sk_forward_alloc -= size;
1289 static inline void sk_mem_uncharge(struct sock *sk, int size)
1291 if (!sk_has_account(sk))
1293 sk->sk_forward_alloc += size;
1296 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1298 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1299 sk->sk_wmem_queued -= skb->truesize;
1300 sk_mem_uncharge(sk, skb->truesize);
1304 /* Used by processes to "lock" a socket state, so that
1305 * interrupts and bottom half handlers won't change it
1306 * from under us. It essentially blocks any incoming
1307 * packets, so that we won't get any new data or any
1308 * packets that change the state of the socket.
1310 * While locked, BH processing will add new packets to
1311 * the backlog queue. This queue is processed by the
1312 * owner of the socket lock right before it is released.
1314 * Since ~2.3.5 it is also exclusive sleep lock serializing
1315 * accesses from user process context.
1317 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1320 * Macro so as to not evaluate some arguments when
1321 * lockdep is not enabled.
1323 * Mark both the sk_lock and the sk_lock.slock as a
1324 * per-address-family lock class.
1326 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1328 sk->sk_lock.owned = 0; \
1329 init_waitqueue_head(&sk->sk_lock.wq); \
1330 spin_lock_init(&(sk)->sk_lock.slock); \
1331 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1332 sizeof((sk)->sk_lock)); \
1333 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1335 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1338 extern void lock_sock_nested(struct sock *sk, int subclass);
1340 static inline void lock_sock(struct sock *sk)
1342 lock_sock_nested(sk, 0);
1345 extern void release_sock(struct sock *sk);
1347 /* BH context may only use the following locking interface. */
1348 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1349 #define bh_lock_sock_nested(__sk) \
1350 spin_lock_nested(&((__sk)->sk_lock.slock), \
1351 SINGLE_DEPTH_NESTING)
1352 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1354 extern bool lock_sock_fast(struct sock *sk);
1356 * unlock_sock_fast - complement of lock_sock_fast
1360 * fast unlock socket for user context.
1361 * If slow mode is on, we call regular release_sock()
1363 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1368 spin_unlock_bh(&sk->sk_lock.slock);
1372 extern struct sock *sk_alloc(struct net *net, int family,
1374 struct proto *prot);
1375 extern void sk_free(struct sock *sk);
1376 extern void sk_release_kernel(struct sock *sk);
1377 extern struct sock *sk_clone_lock(const struct sock *sk,
1378 const gfp_t priority);
1380 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1381 unsigned long size, int force,
1383 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1384 unsigned long size, int force,
1386 extern void sock_wfree(struct sk_buff *skb);
1387 extern void sock_rfree(struct sk_buff *skb);
1389 extern int sock_setsockopt(struct socket *sock, int level,
1390 int op, char __user *optval,
1391 unsigned int optlen);
1393 extern int sock_getsockopt(struct socket *sock, int level,
1394 int op, char __user *optval,
1395 int __user *optlen);
1396 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1400 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1401 unsigned long header_len,
1402 unsigned long data_len,
1405 extern void *sock_kmalloc(struct sock *sk, int size,
1407 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1408 extern void sk_send_sigurg(struct sock *sk);
1410 #ifdef CONFIG_CGROUPS
1411 extern void sock_update_classid(struct sock *sk);
1413 static inline void sock_update_classid(struct sock *sk)
1419 * Functions to fill in entries in struct proto_ops when a protocol
1420 * does not implement a particular function.
1422 extern int sock_no_bind(struct socket *,
1423 struct sockaddr *, int);
1424 extern int sock_no_connect(struct socket *,
1425 struct sockaddr *, int, int);
1426 extern int sock_no_socketpair(struct socket *,
1428 extern int sock_no_accept(struct socket *,
1429 struct socket *, int);
1430 extern int sock_no_getname(struct socket *,
1431 struct sockaddr *, int *, int);
1432 extern unsigned int sock_no_poll(struct file *, struct socket *,
1433 struct poll_table_struct *);
1434 extern int sock_no_ioctl(struct socket *, unsigned int,
1436 extern int sock_no_listen(struct socket *, int);
1437 extern int sock_no_shutdown(struct socket *, int);
1438 extern int sock_no_getsockopt(struct socket *, int , int,
1439 char __user *, int __user *);
1440 extern int sock_no_setsockopt(struct socket *, int, int,
1441 char __user *, unsigned int);
1442 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1443 struct msghdr *, size_t);
1444 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1445 struct msghdr *, size_t, int);
1446 extern int sock_no_mmap(struct file *file,
1447 struct socket *sock,
1448 struct vm_area_struct *vma);
1449 extern ssize_t sock_no_sendpage(struct socket *sock,
1451 int offset, size_t size,
1455 * Functions to fill in entries in struct proto_ops when a protocol
1456 * uses the inet style.
1458 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1459 char __user *optval, int __user *optlen);
1460 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1461 struct msghdr *msg, size_t size, int flags);
1462 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1463 char __user *optval, unsigned int optlen);
1464 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1465 int optname, char __user *optval, int __user *optlen);
1466 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1467 int optname, char __user *optval, unsigned int optlen);
1469 extern void sk_common_release(struct sock *sk);
1472 * Default socket callbacks and setup code
1475 /* Initialise core socket variables */
1476 extern void sock_init_data(struct socket *sock, struct sock *sk);
1478 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1481 * sk_filter_release - release a socket filter
1482 * @fp: filter to remove
1484 * Remove a filter from a socket and release its resources.
1487 static inline void sk_filter_release(struct sk_filter *fp)
1489 if (atomic_dec_and_test(&fp->refcnt))
1490 call_rcu(&fp->rcu, sk_filter_release_rcu);
1493 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1495 unsigned int size = sk_filter_len(fp);
1497 atomic_sub(size, &sk->sk_omem_alloc);
1498 sk_filter_release(fp);
1501 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1503 atomic_inc(&fp->refcnt);
1504 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1508 * Socket reference counting postulates.
1510 * * Each user of socket SHOULD hold a reference count.
1511 * * Each access point to socket (an hash table bucket, reference from a list,
1512 * running timer, skb in flight MUST hold a reference count.
1513 * * When reference count hits 0, it means it will never increase back.
1514 * * When reference count hits 0, it means that no references from
1515 * outside exist to this socket and current process on current CPU
1516 * is last user and may/should destroy this socket.
1517 * * sk_free is called from any context: process, BH, IRQ. When
1518 * it is called, socket has no references from outside -> sk_free
1519 * may release descendant resources allocated by the socket, but
1520 * to the time when it is called, socket is NOT referenced by any
1521 * hash tables, lists etc.
1522 * * Packets, delivered from outside (from network or from another process)
1523 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1524 * when they sit in queue. Otherwise, packets will leak to hole, when
1525 * socket is looked up by one cpu and unhasing is made by another CPU.
1526 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1527 * (leak to backlog). Packet socket does all the processing inside
1528 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1529 * use separate SMP lock, so that they are prone too.
1532 /* Ungrab socket and destroy it, if it was the last reference. */
1533 static inline void sock_put(struct sock *sk)
1535 if (atomic_dec_and_test(&sk->sk_refcnt))
1539 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1542 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1544 sk->sk_tx_queue_mapping = tx_queue;
1547 static inline void sk_tx_queue_clear(struct sock *sk)
1549 sk->sk_tx_queue_mapping = -1;
1552 static inline int sk_tx_queue_get(const struct sock *sk)
1554 return sk ? sk->sk_tx_queue_mapping : -1;
1557 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1559 sk_tx_queue_clear(sk);
1560 sk->sk_socket = sock;
1563 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1565 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1566 return &rcu_dereference_raw(sk->sk_wq)->wait;
1568 /* Detach socket from process context.
1569 * Announce socket dead, detach it from wait queue and inode.
1570 * Note that parent inode held reference count on this struct sock,
1571 * we do not release it in this function, because protocol
1572 * probably wants some additional cleanups or even continuing
1573 * to work with this socket (TCP).
1575 static inline void sock_orphan(struct sock *sk)
1577 write_lock_bh(&sk->sk_callback_lock);
1578 sock_set_flag(sk, SOCK_DEAD);
1579 sk_set_socket(sk, NULL);
1581 write_unlock_bh(&sk->sk_callback_lock);
1584 static inline void sock_graft(struct sock *sk, struct socket *parent)
1586 write_lock_bh(&sk->sk_callback_lock);
1587 sk->sk_wq = parent->wq;
1589 sk_set_socket(sk, parent);
1590 security_sock_graft(sk, parent);
1591 write_unlock_bh(&sk->sk_callback_lock);
1594 extern int sock_i_uid(struct sock *sk);
1595 extern unsigned long sock_i_ino(struct sock *sk);
1597 static inline struct dst_entry *
1598 __sk_dst_get(struct sock *sk)
1600 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1601 lockdep_is_held(&sk->sk_lock.slock));
1604 static inline struct dst_entry *
1605 sk_dst_get(struct sock *sk)
1607 struct dst_entry *dst;
1610 dst = rcu_dereference(sk->sk_dst_cache);
1617 extern void sk_reset_txq(struct sock *sk);
1619 static inline void dst_negative_advice(struct sock *sk)
1621 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1623 if (dst && dst->ops->negative_advice) {
1624 ndst = dst->ops->negative_advice(dst);
1627 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1634 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1636 struct dst_entry *old_dst;
1638 sk_tx_queue_clear(sk);
1640 * This can be called while sk is owned by the caller only,
1641 * with no state that can be checked in a rcu_dereference_check() cond
1643 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1644 rcu_assign_pointer(sk->sk_dst_cache, dst);
1645 dst_release(old_dst);
1649 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1651 spin_lock(&sk->sk_dst_lock);
1652 __sk_dst_set(sk, dst);
1653 spin_unlock(&sk->sk_dst_lock);
1657 __sk_dst_reset(struct sock *sk)
1659 __sk_dst_set(sk, NULL);
1663 sk_dst_reset(struct sock *sk)
1665 spin_lock(&sk->sk_dst_lock);
1667 spin_unlock(&sk->sk_dst_lock);
1670 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1672 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1674 static inline int sk_can_gso(const struct sock *sk)
1676 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1679 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1681 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1683 sk->sk_route_nocaps |= flags;
1684 sk->sk_route_caps &= ~flags;
1687 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1688 char __user *from, char *to,
1689 int copy, int offset)
1691 if (skb->ip_summed == CHECKSUM_NONE) {
1693 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1696 skb->csum = csum_block_add(skb->csum, csum, offset);
1697 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1698 if (!access_ok(VERIFY_READ, from, copy) ||
1699 __copy_from_user_nocache(to, from, copy))
1701 } else if (copy_from_user(to, from, copy))
1707 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1708 char __user *from, int copy)
1710 int err, offset = skb->len;
1712 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1715 __skb_trim(skb, offset);
1720 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1721 struct sk_buff *skb,
1727 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1733 skb->data_len += copy;
1734 skb->truesize += copy;
1735 sk->sk_wmem_queued += copy;
1736 sk_mem_charge(sk, copy);
1740 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1741 struct sk_buff *skb, struct page *page,
1744 if (skb->ip_summed == CHECKSUM_NONE) {
1746 __wsum csum = csum_and_copy_from_user(from,
1747 page_address(page) + off,
1751 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1752 } else if (copy_from_user(page_address(page) + off, from, copy))
1756 skb->data_len += copy;
1757 skb->truesize += copy;
1758 sk->sk_wmem_queued += copy;
1759 sk_mem_charge(sk, copy);
1764 * sk_wmem_alloc_get - returns write allocations
1767 * Returns sk_wmem_alloc minus initial offset of one
1769 static inline int sk_wmem_alloc_get(const struct sock *sk)
1771 return atomic_read(&sk->sk_wmem_alloc) - 1;
1775 * sk_rmem_alloc_get - returns read allocations
1778 * Returns sk_rmem_alloc
1780 static inline int sk_rmem_alloc_get(const struct sock *sk)
1782 return atomic_read(&sk->sk_rmem_alloc);
1786 * sk_has_allocations - check if allocations are outstanding
1789 * Returns true if socket has write or read allocations
1791 static inline int sk_has_allocations(const struct sock *sk)
1793 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1797 * wq_has_sleeper - check if there are any waiting processes
1798 * @wq: struct socket_wq
1800 * Returns true if socket_wq has waiting processes
1802 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1803 * barrier call. They were added due to the race found within the tcp code.
1805 * Consider following tcp code paths:
1809 * sys_select receive packet
1811 * __add_wait_queue update tp->rcv_nxt
1813 * tp->rcv_nxt check sock_def_readable
1815 * schedule rcu_read_lock();
1816 * wq = rcu_dereference(sk->sk_wq);
1817 * if (wq && waitqueue_active(&wq->wait))
1818 * wake_up_interruptible(&wq->wait)
1822 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1823 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1824 * could then endup calling schedule and sleep forever if there are no more
1825 * data on the socket.
1828 static inline bool wq_has_sleeper(struct socket_wq *wq)
1832 * We need to be sure we are in sync with the
1833 * add_wait_queue modifications to the wait queue.
1835 * This memory barrier is paired in the sock_poll_wait.
1838 return wq && waitqueue_active(&wq->wait);
1842 * sock_poll_wait - place memory barrier behind the poll_wait call.
1844 * @wait_address: socket wait queue
1847 * See the comments in the wq_has_sleeper function.
1849 static inline void sock_poll_wait(struct file *filp,
1850 wait_queue_head_t *wait_address, poll_table *p)
1852 if (p && wait_address) {
1853 poll_wait(filp, wait_address, p);
1855 * We need to be sure we are in sync with the
1856 * socket flags modification.
1858 * This memory barrier is paired in the wq_has_sleeper.
1865 * Queue a received datagram if it will fit. Stream and sequenced
1866 * protocols can't normally use this as they need to fit buffers in
1867 * and play with them.
1869 * Inlined as it's very short and called for pretty much every
1870 * packet ever received.
1873 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1877 skb->destructor = sock_wfree;
1879 * We used to take a refcount on sk, but following operation
1880 * is enough to guarantee sk_free() wont free this sock until
1881 * all in-flight packets are completed
1883 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1886 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1890 skb->destructor = sock_rfree;
1891 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1892 sk_mem_charge(sk, skb->truesize);
1895 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1896 unsigned long expires);
1898 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1900 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1902 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1905 * Recover an error report and clear atomically
1908 static inline int sock_error(struct sock *sk)
1911 if (likely(!sk->sk_err))
1913 err = xchg(&sk->sk_err, 0);
1917 static inline unsigned long sock_wspace(struct sock *sk)
1921 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1922 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1929 static inline void sk_wake_async(struct sock *sk, int how, int band)
1931 if (sock_flag(sk, SOCK_FASYNC))
1932 sock_wake_async(sk->sk_socket, how, band);
1935 #define SOCK_MIN_SNDBUF 2048
1937 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1938 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1940 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1942 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1944 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1945 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1946 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1950 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1952 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1954 struct page *page = NULL;
1956 page = alloc_pages(sk->sk_allocation, 0);
1958 sk_enter_memory_pressure(sk);
1959 sk_stream_moderate_sndbuf(sk);
1965 * Default write policy as shown to user space via poll/select/SIGIO
1967 static inline int sock_writeable(const struct sock *sk)
1969 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1972 static inline gfp_t gfp_any(void)
1974 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1977 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1979 return noblock ? 0 : sk->sk_rcvtimeo;
1982 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1984 return noblock ? 0 : sk->sk_sndtimeo;
1987 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1989 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1992 /* Alas, with timeout socket operations are not restartable.
1993 * Compare this to poll().
1995 static inline int sock_intr_errno(long timeo)
1997 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2000 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2001 struct sk_buff *skb);
2002 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2003 struct sk_buff *skb);
2005 static __inline__ void
2006 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2008 ktime_t kt = skb->tstamp;
2009 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2012 * generate control messages if
2013 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2014 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2015 * - software time stamp available and wanted
2016 * (SOCK_TIMESTAMPING_SOFTWARE)
2017 * - hardware time stamps available and wanted
2018 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2019 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2021 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2022 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2023 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2024 (hwtstamps->hwtstamp.tv64 &&
2025 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2026 (hwtstamps->syststamp.tv64 &&
2027 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2028 __sock_recv_timestamp(msg, sk, skb);
2032 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2033 __sock_recv_wifi_status(msg, sk, skb);
2036 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2037 struct sk_buff *skb);
2039 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2040 struct sk_buff *skb)
2042 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2043 (1UL << SOCK_RCVTSTAMP) | \
2044 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2045 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2046 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2047 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2049 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2050 __sock_recv_ts_and_drops(msg, sk, skb);
2052 sk->sk_stamp = skb->tstamp;
2056 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2057 * @sk: socket sending this packet
2058 * @tx_flags: filled with instructions for time stamping
2060 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2061 * parameters are invalid.
2063 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2066 * sk_eat_skb - Release a skb if it is no longer needed
2067 * @sk: socket to eat this skb from
2068 * @skb: socket buffer to eat
2069 * @copied_early: flag indicating whether DMA operations copied this data early
2071 * This routine must be called with interrupts disabled or with the socket
2072 * locked so that the sk_buff queue operation is ok.
2074 #ifdef CONFIG_NET_DMA
2075 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
2077 __skb_unlink(skb, &sk->sk_receive_queue);
2081 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2084 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
2086 __skb_unlink(skb, &sk->sk_receive_queue);
2092 struct net *sock_net(const struct sock *sk)
2094 return read_pnet(&sk->sk_net);
2098 void sock_net_set(struct sock *sk, struct net *net)
2100 write_pnet(&sk->sk_net, net);
2104 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2105 * They should not hold a reference to a namespace in order to allow
2107 * Sockets after sk_change_net should be released using sk_release_kernel
2109 static inline void sk_change_net(struct sock *sk, struct net *net)
2111 put_net(sock_net(sk));
2112 sock_net_set(sk, hold_net(net));
2115 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2117 if (unlikely(skb->sk)) {
2118 struct sock *sk = skb->sk;
2120 skb->destructor = NULL;
2127 extern void sock_enable_timestamp(struct sock *sk, int flag);
2128 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2129 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2132 * Enable debug/info messages
2134 extern int net_msg_warn;
2135 #define NETDEBUG(fmt, args...) \
2136 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2138 #define LIMIT_NETDEBUG(fmt, args...) \
2139 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2141 extern __u32 sysctl_wmem_max;
2142 extern __u32 sysctl_rmem_max;
2144 extern void sk_init(void);
2146 extern int sysctl_optmem_max;
2148 extern __u32 sysctl_wmem_default;
2149 extern __u32 sysctl_rmem_default;
2151 #endif /* _SOCK_H */