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/bitops.h>
50 #include <linux/lockdep.h>
51 #include <linux/netdevice.h>
52 #include <linux/skbuff.h> /* struct sk_buff */
54 #include <linux/security.h>
55 #include <linux/slab.h>
56 #include <linux/uaccess.h>
57 #include <linux/memcontrol.h>
58 #include <linux/res_counter.h>
59 #include <linux/static_key.h>
60 #include <linux/aio.h>
61 #include <linux/sched.h>
63 #include <linux/filter.h>
64 #include <linux/rculist_nulls.h>
65 #include <linux/poll.h>
67 #include <linux/atomic.h>
69 #include <net/checksum.h>
74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
88 * This structure really needs to be cleaned up.
89 * Most of it is for TCP, and not used by any of
90 * the other protocols.
93 /* Define this to get the SOCK_DBG debugging facility. */
94 #define SOCK_DEBUGGING
96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
97 printk(KERN_DEBUG msg); } while (0)
99 /* Validate arguments and do nothing */
100 static inline __printf(2, 3)
101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
106 /* This is the per-socket lock. The spinlock provides a synchronization
107 * between user contexts and software interrupt processing, whereas the
108 * mini-semaphore synchronizes multiple users amongst themselves.
113 wait_queue_head_t wq;
115 * We express the mutex-alike socket_lock semantics
116 * to the lock validator by explicitly managing
117 * the slock as a lock variant (in addition to
120 #ifdef CONFIG_DEBUG_LOCK_ALLOC
121 struct lockdep_map dep_map;
130 * struct sock_common - minimal network layer representation of sockets
131 * @skc_daddr: Foreign IPv4 addr
132 * @skc_rcv_saddr: Bound local IPv4 addr
133 * @skc_hash: hash value used with various protocol lookup tables
134 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
135 * @skc_family: network address family
136 * @skc_state: Connection state
137 * @skc_reuse: %SO_REUSEADDR setting
138 * @skc_bound_dev_if: bound device index if != 0
139 * @skc_bind_node: bind hash linkage for various protocol lookup tables
140 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
141 * @skc_prot: protocol handlers inside a network family
142 * @skc_net: reference to the network namespace of this socket
143 * @skc_node: main hash linkage for various protocol lookup tables
144 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
145 * @skc_tx_queue_mapping: tx queue number for this connection
146 * @skc_refcnt: reference count
148 * This is the minimal network layer representation of sockets, the header
149 * for struct sock and struct inet_timewait_sock.
152 /* skc_daddr and skc_rcv_saddr must be grouped :
153 * cf INET_MATCH() and INET_TW_MATCH()
156 __be32 skc_rcv_saddr;
159 unsigned int skc_hash;
160 __u16 skc_u16hashes[2];
162 unsigned short skc_family;
163 volatile unsigned char skc_state;
164 unsigned char skc_reuse;
165 int skc_bound_dev_if;
167 struct hlist_node skc_bind_node;
168 struct hlist_nulls_node skc_portaddr_node;
170 struct proto *skc_prot;
175 * fields between dontcopy_begin/dontcopy_end
176 * are not copied in sock_copy()
179 int skc_dontcopy_begin[0];
182 struct hlist_node skc_node;
183 struct hlist_nulls_node skc_nulls_node;
185 int skc_tx_queue_mapping;
188 int skc_dontcopy_end[0];
194 * struct sock - network layer representation of sockets
195 * @__sk_common: shared layout with inet_timewait_sock
196 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
197 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
198 * @sk_lock: synchronizer
199 * @sk_rcvbuf: size of receive buffer in bytes
200 * @sk_wq: sock wait queue and async head
201 * @sk_rx_dst: receive input route used by early tcp demux
202 * @sk_dst_cache: destination cache
203 * @sk_dst_lock: destination cache lock
204 * @sk_policy: flow policy
205 * @sk_receive_queue: incoming packets
206 * @sk_wmem_alloc: transmit queue bytes committed
207 * @sk_write_queue: Packet sending queue
208 * @sk_async_wait_queue: DMA copied packets
209 * @sk_omem_alloc: "o" is "option" or "other"
210 * @sk_wmem_queued: persistent queue size
211 * @sk_forward_alloc: space allocated forward
212 * @sk_allocation: allocation mode
213 * @sk_sndbuf: size of send buffer in bytes
214 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
215 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
216 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
217 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
218 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
219 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
220 * @sk_gso_max_size: Maximum GSO segment size to build
221 * @sk_lingertime: %SO_LINGER l_linger setting
222 * @sk_backlog: always used with the per-socket spinlock held
223 * @sk_callback_lock: used with the callbacks in the end of this struct
224 * @sk_error_queue: rarely used
225 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
226 * IPV6_ADDRFORM for instance)
227 * @sk_err: last error
228 * @sk_err_soft: errors that don't cause failure but are the cause of a
229 * persistent failure not just 'timed out'
230 * @sk_drops: raw/udp drops counter
231 * @sk_ack_backlog: current listen backlog
232 * @sk_max_ack_backlog: listen backlog set in listen()
233 * @sk_priority: %SO_PRIORITY setting
234 * @sk_cgrp_prioidx: socket group's priority map index
235 * @sk_type: socket type (%SOCK_STREAM, etc)
236 * @sk_protocol: which protocol this socket belongs in this network family
237 * @sk_peer_pid: &struct pid for this socket's peer
238 * @sk_peer_cred: %SO_PEERCRED setting
239 * @sk_rcvlowat: %SO_RCVLOWAT setting
240 * @sk_rcvtimeo: %SO_RCVTIMEO setting
241 * @sk_sndtimeo: %SO_SNDTIMEO setting
242 * @sk_rxhash: flow hash received from netif layer
243 * @sk_filter: socket filtering instructions
244 * @sk_protinfo: private area, net family specific, when not using slab
245 * @sk_timer: sock cleanup timer
246 * @sk_stamp: time stamp of last packet received
247 * @sk_socket: Identd and reporting IO signals
248 * @sk_user_data: RPC layer private data
249 * @sk_sndmsg_page: cached page for sendmsg
250 * @sk_sndmsg_off: cached offset for sendmsg
251 * @sk_peek_off: current peek_offset value
252 * @sk_send_head: front of stuff to transmit
253 * @sk_security: used by security modules
254 * @sk_mark: generic packet mark
255 * @sk_classid: this socket's cgroup classid
256 * @sk_cgrp: this socket's cgroup-specific proto data
257 * @sk_write_pending: a write to stream socket waits to start
258 * @sk_state_change: callback to indicate change in the state of the sock
259 * @sk_data_ready: callback to indicate there is data to be processed
260 * @sk_write_space: callback to indicate there is bf sending space available
261 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
262 * @sk_backlog_rcv: callback to process the backlog
263 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
267 * Now struct inet_timewait_sock also uses sock_common, so please just
268 * don't add nothing before this first member (__sk_common) --acme
270 struct sock_common __sk_common;
271 #define sk_node __sk_common.skc_node
272 #define sk_nulls_node __sk_common.skc_nulls_node
273 #define sk_refcnt __sk_common.skc_refcnt
274 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
276 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
277 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
278 #define sk_hash __sk_common.skc_hash
279 #define sk_family __sk_common.skc_family
280 #define sk_state __sk_common.skc_state
281 #define sk_reuse __sk_common.skc_reuse
282 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
283 #define sk_bind_node __sk_common.skc_bind_node
284 #define sk_prot __sk_common.skc_prot
285 #define sk_net __sk_common.skc_net
286 socket_lock_t sk_lock;
287 struct sk_buff_head sk_receive_queue;
289 * The backlog queue is special, it is always used with
290 * the per-socket spinlock held and requires low latency
291 * access. Therefore we special case it's implementation.
292 * Note : rmem_alloc is in this structure to fill a hole
293 * on 64bit arches, not because its logically part of
299 struct sk_buff *head;
300 struct sk_buff *tail;
302 #define sk_rmem_alloc sk_backlog.rmem_alloc
303 int sk_forward_alloc;
310 struct sk_filter __rcu *sk_filter;
311 struct socket_wq __rcu *sk_wq;
313 #ifdef CONFIG_NET_DMA
314 struct sk_buff_head sk_async_wait_queue;
318 struct xfrm_policy *sk_policy[2];
320 unsigned long sk_flags;
321 struct dst_entry *sk_rx_dst;
322 struct dst_entry *sk_dst_cache;
323 spinlock_t sk_dst_lock;
324 atomic_t sk_wmem_alloc;
325 atomic_t sk_omem_alloc;
327 struct sk_buff_head sk_write_queue;
328 kmemcheck_bitfield_begin(flags);
329 unsigned int sk_shutdown : 2,
334 kmemcheck_bitfield_end(flags);
337 netdev_features_t sk_route_caps;
338 netdev_features_t sk_route_nocaps;
340 unsigned int sk_gso_max_size;
342 unsigned long sk_lingertime;
343 struct sk_buff_head sk_error_queue;
344 struct proto *sk_prot_creator;
345 rwlock_t sk_callback_lock;
348 unsigned short sk_ack_backlog;
349 unsigned short sk_max_ack_backlog;
351 #ifdef CONFIG_CGROUPS
352 __u32 sk_cgrp_prioidx;
354 struct pid *sk_peer_pid;
355 const struct cred *sk_peer_cred;
359 struct timer_list sk_timer;
361 struct socket *sk_socket;
363 struct page *sk_sndmsg_page;
364 struct sk_buff *sk_send_head;
367 int sk_write_pending;
368 #ifdef CONFIG_SECURITY
373 struct cg_proto *sk_cgrp;
374 void (*sk_state_change)(struct sock *sk);
375 void (*sk_data_ready)(struct sock *sk, int bytes);
376 void (*sk_write_space)(struct sock *sk);
377 void (*sk_error_report)(struct sock *sk);
378 int (*sk_backlog_rcv)(struct sock *sk,
379 struct sk_buff *skb);
380 void (*sk_destruct)(struct sock *sk);
384 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
385 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
386 * on a socket means that the socket will reuse everybody else's port
387 * without looking at the other's sk_reuse value.
390 #define SK_NO_REUSE 0
391 #define SK_CAN_REUSE 1
392 #define SK_FORCE_REUSE 2
394 static inline int sk_peek_offset(struct sock *sk, int flags)
396 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
397 return sk->sk_peek_off;
402 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
404 if (sk->sk_peek_off >= 0) {
405 if (sk->sk_peek_off >= val)
406 sk->sk_peek_off -= val;
412 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
414 if (sk->sk_peek_off >= 0)
415 sk->sk_peek_off += val;
419 * Hashed lists helper routines
421 static inline struct sock *sk_entry(const struct hlist_node *node)
423 return hlist_entry(node, struct sock, sk_node);
426 static inline struct sock *__sk_head(const struct hlist_head *head)
428 return hlist_entry(head->first, struct sock, sk_node);
431 static inline struct sock *sk_head(const struct hlist_head *head)
433 return hlist_empty(head) ? NULL : __sk_head(head);
436 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
438 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
441 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
443 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
446 static inline struct sock *sk_next(const struct sock *sk)
448 return sk->sk_node.next ?
449 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
452 static inline struct sock *sk_nulls_next(const struct sock *sk)
454 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
455 hlist_nulls_entry(sk->sk_nulls_node.next,
456 struct sock, sk_nulls_node) :
460 static inline bool sk_unhashed(const struct sock *sk)
462 return hlist_unhashed(&sk->sk_node);
465 static inline bool sk_hashed(const struct sock *sk)
467 return !sk_unhashed(sk);
470 static inline void sk_node_init(struct hlist_node *node)
475 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
480 static inline void __sk_del_node(struct sock *sk)
482 __hlist_del(&sk->sk_node);
485 /* NB: equivalent to hlist_del_init_rcu */
486 static inline bool __sk_del_node_init(struct sock *sk)
490 sk_node_init(&sk->sk_node);
496 /* Grab socket reference count. This operation is valid only
497 when sk is ALREADY grabbed f.e. it is found in hash table
498 or a list and the lookup is made under lock preventing hash table
502 static inline void sock_hold(struct sock *sk)
504 atomic_inc(&sk->sk_refcnt);
507 /* Ungrab socket in the context, which assumes that socket refcnt
508 cannot hit zero, f.e. it is true in context of any socketcall.
510 static inline void __sock_put(struct sock *sk)
512 atomic_dec(&sk->sk_refcnt);
515 static inline bool sk_del_node_init(struct sock *sk)
517 bool rc = __sk_del_node_init(sk);
520 /* paranoid for a while -acme */
521 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
526 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
528 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
531 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
537 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
539 bool rc = __sk_nulls_del_node_init_rcu(sk);
542 /* paranoid for a while -acme */
543 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
549 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
551 hlist_add_head(&sk->sk_node, list);
554 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
557 __sk_add_node(sk, list);
560 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
563 hlist_add_head_rcu(&sk->sk_node, list);
566 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
568 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
571 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
574 __sk_nulls_add_node_rcu(sk, list);
577 static inline void __sk_del_bind_node(struct sock *sk)
579 __hlist_del(&sk->sk_bind_node);
582 static inline void sk_add_bind_node(struct sock *sk,
583 struct hlist_head *list)
585 hlist_add_head(&sk->sk_bind_node, list);
588 #define sk_for_each(__sk, node, list) \
589 hlist_for_each_entry(__sk, node, list, sk_node)
590 #define sk_for_each_rcu(__sk, node, list) \
591 hlist_for_each_entry_rcu(__sk, node, list, sk_node)
592 #define sk_nulls_for_each(__sk, node, list) \
593 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
594 #define sk_nulls_for_each_rcu(__sk, node, list) \
595 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
596 #define sk_for_each_from(__sk, node) \
597 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
598 hlist_for_each_entry_from(__sk, node, sk_node)
599 #define sk_nulls_for_each_from(__sk, node) \
600 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
601 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
602 #define sk_for_each_safe(__sk, node, tmp, list) \
603 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
604 #define sk_for_each_bound(__sk, node, list) \
605 hlist_for_each_entry(__sk, node, list, sk_bind_node)
618 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
619 SOCK_DBG, /* %SO_DEBUG setting */
620 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
621 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
622 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
623 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
624 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
625 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
626 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
627 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
628 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
629 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
630 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
631 SOCK_FASYNC, /* fasync() active */
633 SOCK_ZEROCOPY, /* buffers from userspace */
634 SOCK_WIFI_STATUS, /* push wifi status to userspace */
635 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
636 * Will use last 4 bytes of packet sent from
637 * user-space instead.
641 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
643 nsk->sk_flags = osk->sk_flags;
646 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
648 __set_bit(flag, &sk->sk_flags);
651 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
653 __clear_bit(flag, &sk->sk_flags);
656 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
658 return test_bit(flag, &sk->sk_flags);
661 static inline void sk_acceptq_removed(struct sock *sk)
663 sk->sk_ack_backlog--;
666 static inline void sk_acceptq_added(struct sock *sk)
668 sk->sk_ack_backlog++;
671 static inline bool sk_acceptq_is_full(const struct sock *sk)
673 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
677 * Compute minimal free write space needed to queue new packets.
679 static inline int sk_stream_min_wspace(const struct sock *sk)
681 return sk->sk_wmem_queued >> 1;
684 static inline int sk_stream_wspace(const struct sock *sk)
686 return sk->sk_sndbuf - sk->sk_wmem_queued;
689 extern void sk_stream_write_space(struct sock *sk);
691 static inline bool sk_stream_memory_free(const struct sock *sk)
693 return sk->sk_wmem_queued < sk->sk_sndbuf;
696 /* OOB backlog add */
697 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
699 /* dont let skb dst not refcounted, we are going to leave rcu lock */
702 if (!sk->sk_backlog.tail)
703 sk->sk_backlog.head = skb;
705 sk->sk_backlog.tail->next = skb;
707 sk->sk_backlog.tail = skb;
712 * Take into account size of receive queue and backlog queue
713 * Do not take into account this skb truesize,
714 * to allow even a single big packet to come.
716 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
719 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
721 return qsize > limit;
724 /* The per-socket spinlock must be held here. */
725 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
728 if (sk_rcvqueues_full(sk, skb, limit))
731 __sk_add_backlog(sk, skb);
732 sk->sk_backlog.len += skb->truesize;
736 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
738 return sk->sk_backlog_rcv(sk, skb);
741 static inline void sock_rps_record_flow(const struct sock *sk)
744 struct rps_sock_flow_table *sock_flow_table;
747 sock_flow_table = rcu_dereference(rps_sock_flow_table);
748 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
753 static inline void sock_rps_reset_flow(const struct sock *sk)
756 struct rps_sock_flow_table *sock_flow_table;
759 sock_flow_table = rcu_dereference(rps_sock_flow_table);
760 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
765 static inline void sock_rps_save_rxhash(struct sock *sk,
766 const struct sk_buff *skb)
769 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
770 sock_rps_reset_flow(sk);
771 sk->sk_rxhash = skb->rxhash;
776 static inline void sock_rps_reset_rxhash(struct sock *sk)
779 sock_rps_reset_flow(sk);
784 #define sk_wait_event(__sk, __timeo, __condition) \
786 release_sock(__sk); \
787 __rc = __condition; \
789 *(__timeo) = schedule_timeout(*(__timeo)); \
792 __rc = __condition; \
796 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
797 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
798 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
799 extern int sk_stream_error(struct sock *sk, int flags, int err);
800 extern void sk_stream_kill_queues(struct sock *sk);
802 extern int sk_wait_data(struct sock *sk, long *timeo);
804 struct request_sock_ops;
805 struct timewait_sock_ops;
806 struct inet_hashinfo;
810 /* Networking protocol blocks we attach to sockets.
811 * socket layer -> transport layer interface
812 * transport -> network interface is defined by struct inet_proto
815 void (*close)(struct sock *sk,
817 int (*connect)(struct sock *sk,
818 struct sockaddr *uaddr,
820 int (*disconnect)(struct sock *sk, int flags);
822 struct sock * (*accept)(struct sock *sk, int flags, int *err);
824 int (*ioctl)(struct sock *sk, int cmd,
826 int (*init)(struct sock *sk);
827 void (*destroy)(struct sock *sk);
828 void (*shutdown)(struct sock *sk, int how);
829 int (*setsockopt)(struct sock *sk, int level,
830 int optname, char __user *optval,
831 unsigned int optlen);
832 int (*getsockopt)(struct sock *sk, int level,
833 int optname, char __user *optval,
836 int (*compat_setsockopt)(struct sock *sk,
838 int optname, char __user *optval,
839 unsigned int optlen);
840 int (*compat_getsockopt)(struct sock *sk,
842 int optname, char __user *optval,
844 int (*compat_ioctl)(struct sock *sk,
845 unsigned int cmd, unsigned long arg);
847 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
848 struct msghdr *msg, size_t len);
849 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
851 size_t len, int noblock, int flags,
853 int (*sendpage)(struct sock *sk, struct page *page,
854 int offset, size_t size, int flags);
855 int (*bind)(struct sock *sk,
856 struct sockaddr *uaddr, int addr_len);
858 int (*backlog_rcv) (struct sock *sk,
859 struct sk_buff *skb);
861 /* Keeping track of sk's, looking them up, and port selection methods. */
862 void (*hash)(struct sock *sk);
863 void (*unhash)(struct sock *sk);
864 void (*rehash)(struct sock *sk);
865 int (*get_port)(struct sock *sk, unsigned short snum);
866 void (*clear_sk)(struct sock *sk, int size);
868 /* Keeping track of sockets in use */
869 #ifdef CONFIG_PROC_FS
870 unsigned int inuse_idx;
873 /* Memory pressure */
874 void (*enter_memory_pressure)(struct sock *sk);
875 atomic_long_t *memory_allocated; /* Current allocated memory. */
876 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
878 * Pressure flag: try to collapse.
879 * Technical note: it is used by multiple contexts non atomically.
880 * All the __sk_mem_schedule() is of this nature: accounting
881 * is strict, actions are advisory and have some latency.
883 int *memory_pressure;
890 struct kmem_cache *slab;
891 unsigned int obj_size;
894 struct percpu_counter *orphan_count;
896 struct request_sock_ops *rsk_prot;
897 struct timewait_sock_ops *twsk_prot;
900 struct inet_hashinfo *hashinfo;
901 struct udp_table *udp_table;
902 struct raw_hashinfo *raw_hash;
905 struct module *owner;
909 struct list_head node;
910 #ifdef SOCK_REFCNT_DEBUG
913 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
915 * cgroup specific init/deinit functions. Called once for all
916 * protocols that implement it, from cgroups populate function.
917 * This function has to setup any files the protocol want to
918 * appear in the kmem cgroup filesystem.
920 int (*init_cgroup)(struct mem_cgroup *memcg,
921 struct cgroup_subsys *ss);
922 void (*destroy_cgroup)(struct mem_cgroup *memcg);
923 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
928 * Bits in struct cg_proto.flags
930 enum cg_proto_flags {
931 /* Currently active and new sockets should be assigned to cgroups */
933 /* It was ever activated; we must disarm static keys on destruction */
934 MEMCG_SOCK_ACTIVATED,
938 void (*enter_memory_pressure)(struct sock *sk);
939 struct res_counter *memory_allocated; /* Current allocated memory. */
940 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
941 int *memory_pressure;
945 * memcg field is used to find which memcg we belong directly
946 * Each memcg struct can hold more than one cg_proto, so container_of
949 * The elegant solution would be having an inverse function to
950 * proto_cgroup in struct proto, but that means polluting the structure
951 * for everybody, instead of just for memcg users.
953 struct mem_cgroup *memcg;
956 extern int proto_register(struct proto *prot, int alloc_slab);
957 extern void proto_unregister(struct proto *prot);
959 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
961 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
964 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
966 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
969 #ifdef SOCK_REFCNT_DEBUG
970 static inline void sk_refcnt_debug_inc(struct sock *sk)
972 atomic_inc(&sk->sk_prot->socks);
975 static inline void sk_refcnt_debug_dec(struct sock *sk)
977 atomic_dec(&sk->sk_prot->socks);
978 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
979 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
982 inline void sk_refcnt_debug_release(const struct sock *sk)
984 if (atomic_read(&sk->sk_refcnt) != 1)
985 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
986 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
988 #else /* SOCK_REFCNT_DEBUG */
989 #define sk_refcnt_debug_inc(sk) do { } while (0)
990 #define sk_refcnt_debug_dec(sk) do { } while (0)
991 #define sk_refcnt_debug_release(sk) do { } while (0)
992 #endif /* SOCK_REFCNT_DEBUG */
994 #if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM) && defined(CONFIG_NET)
995 extern struct static_key memcg_socket_limit_enabled;
996 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
997 struct cg_proto *cg_proto)
999 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1001 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1003 #define mem_cgroup_sockets_enabled 0
1004 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1005 struct cg_proto *cg_proto)
1012 static inline bool sk_has_memory_pressure(const struct sock *sk)
1014 return sk->sk_prot->memory_pressure != NULL;
1017 static inline bool sk_under_memory_pressure(const struct sock *sk)
1019 if (!sk->sk_prot->memory_pressure)
1022 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1023 return !!*sk->sk_cgrp->memory_pressure;
1025 return !!*sk->sk_prot->memory_pressure;
1028 static inline void sk_leave_memory_pressure(struct sock *sk)
1030 int *memory_pressure = sk->sk_prot->memory_pressure;
1032 if (!memory_pressure)
1035 if (*memory_pressure)
1036 *memory_pressure = 0;
1038 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1039 struct cg_proto *cg_proto = sk->sk_cgrp;
1040 struct proto *prot = sk->sk_prot;
1042 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1043 if (*cg_proto->memory_pressure)
1044 *cg_proto->memory_pressure = 0;
1049 static inline void sk_enter_memory_pressure(struct sock *sk)
1051 if (!sk->sk_prot->enter_memory_pressure)
1054 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1055 struct cg_proto *cg_proto = sk->sk_cgrp;
1056 struct proto *prot = sk->sk_prot;
1058 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1059 cg_proto->enter_memory_pressure(sk);
1062 sk->sk_prot->enter_memory_pressure(sk);
1065 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1067 long *prot = sk->sk_prot->sysctl_mem;
1068 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1069 prot = sk->sk_cgrp->sysctl_mem;
1073 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1077 struct res_counter *fail;
1080 ret = res_counter_charge_nofail(prot->memory_allocated,
1081 amt << PAGE_SHIFT, &fail);
1083 *parent_status = OVER_LIMIT;
1086 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1089 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1092 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1095 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1096 return ret >> PAGE_SHIFT;
1100 sk_memory_allocated(const struct sock *sk)
1102 struct proto *prot = sk->sk_prot;
1103 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1104 return memcg_memory_allocated_read(sk->sk_cgrp);
1106 return atomic_long_read(prot->memory_allocated);
1110 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1112 struct proto *prot = sk->sk_prot;
1114 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1115 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1116 /* update the root cgroup regardless */
1117 atomic_long_add_return(amt, prot->memory_allocated);
1118 return memcg_memory_allocated_read(sk->sk_cgrp);
1121 return atomic_long_add_return(amt, prot->memory_allocated);
1125 sk_memory_allocated_sub(struct sock *sk, int amt)
1127 struct proto *prot = sk->sk_prot;
1129 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1130 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1132 atomic_long_sub(amt, prot->memory_allocated);
1135 static inline void sk_sockets_allocated_dec(struct sock *sk)
1137 struct proto *prot = sk->sk_prot;
1139 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1140 struct cg_proto *cg_proto = sk->sk_cgrp;
1142 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1143 percpu_counter_dec(cg_proto->sockets_allocated);
1146 percpu_counter_dec(prot->sockets_allocated);
1149 static inline void sk_sockets_allocated_inc(struct sock *sk)
1151 struct proto *prot = sk->sk_prot;
1153 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1154 struct cg_proto *cg_proto = sk->sk_cgrp;
1156 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1157 percpu_counter_inc(cg_proto->sockets_allocated);
1160 percpu_counter_inc(prot->sockets_allocated);
1164 sk_sockets_allocated_read_positive(struct sock *sk)
1166 struct proto *prot = sk->sk_prot;
1168 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1169 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1171 return percpu_counter_read_positive(prot->sockets_allocated);
1175 proto_sockets_allocated_sum_positive(struct proto *prot)
1177 return percpu_counter_sum_positive(prot->sockets_allocated);
1181 proto_memory_allocated(struct proto *prot)
1183 return atomic_long_read(prot->memory_allocated);
1187 proto_memory_pressure(struct proto *prot)
1189 if (!prot->memory_pressure)
1191 return !!*prot->memory_pressure;
1195 #ifdef CONFIG_PROC_FS
1196 /* Called with local bh disabled */
1197 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1198 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1200 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1207 /* With per-bucket locks this operation is not-atomic, so that
1208 * this version is not worse.
1210 static inline void __sk_prot_rehash(struct sock *sk)
1212 sk->sk_prot->unhash(sk);
1213 sk->sk_prot->hash(sk);
1216 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1218 /* About 10 seconds */
1219 #define SOCK_DESTROY_TIME (10*HZ)
1221 /* Sockets 0-1023 can't be bound to unless you are superuser */
1222 #define PROT_SOCK 1024
1224 #define SHUTDOWN_MASK 3
1225 #define RCV_SHUTDOWN 1
1226 #define SEND_SHUTDOWN 2
1228 #define SOCK_SNDBUF_LOCK 1
1229 #define SOCK_RCVBUF_LOCK 2
1230 #define SOCK_BINDADDR_LOCK 4
1231 #define SOCK_BINDPORT_LOCK 8
1233 /* sock_iocb: used to kick off async processing of socket ios */
1235 struct list_head list;
1239 struct socket *sock;
1241 struct scm_cookie *scm;
1242 struct msghdr *msg, async_msg;
1243 struct kiocb *kiocb;
1246 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1248 return (struct sock_iocb *)iocb->private;
1251 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1256 struct socket_alloc {
1257 struct socket socket;
1258 struct inode vfs_inode;
1261 static inline struct socket *SOCKET_I(struct inode *inode)
1263 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1266 static inline struct inode *SOCK_INODE(struct socket *socket)
1268 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1272 * Functions for memory accounting
1274 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1275 extern void __sk_mem_reclaim(struct sock *sk);
1277 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1278 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1279 #define SK_MEM_SEND 0
1280 #define SK_MEM_RECV 1
1282 static inline int sk_mem_pages(int amt)
1284 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1287 static inline bool sk_has_account(struct sock *sk)
1289 /* return true if protocol supports memory accounting */
1290 return !!sk->sk_prot->memory_allocated;
1293 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1295 if (!sk_has_account(sk))
1297 return size <= sk->sk_forward_alloc ||
1298 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1301 static inline bool sk_rmem_schedule(struct sock *sk, int size)
1303 if (!sk_has_account(sk))
1305 return size <= sk->sk_forward_alloc ||
1306 __sk_mem_schedule(sk, size, SK_MEM_RECV);
1309 static inline void sk_mem_reclaim(struct sock *sk)
1311 if (!sk_has_account(sk))
1313 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1314 __sk_mem_reclaim(sk);
1317 static inline void sk_mem_reclaim_partial(struct sock *sk)
1319 if (!sk_has_account(sk))
1321 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1322 __sk_mem_reclaim(sk);
1325 static inline void sk_mem_charge(struct sock *sk, int size)
1327 if (!sk_has_account(sk))
1329 sk->sk_forward_alloc -= size;
1332 static inline void sk_mem_uncharge(struct sock *sk, int size)
1334 if (!sk_has_account(sk))
1336 sk->sk_forward_alloc += size;
1339 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1341 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1342 sk->sk_wmem_queued -= skb->truesize;
1343 sk_mem_uncharge(sk, skb->truesize);
1347 /* Used by processes to "lock" a socket state, so that
1348 * interrupts and bottom half handlers won't change it
1349 * from under us. It essentially blocks any incoming
1350 * packets, so that we won't get any new data or any
1351 * packets that change the state of the socket.
1353 * While locked, BH processing will add new packets to
1354 * the backlog queue. This queue is processed by the
1355 * owner of the socket lock right before it is released.
1357 * Since ~2.3.5 it is also exclusive sleep lock serializing
1358 * accesses from user process context.
1360 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1363 * Macro so as to not evaluate some arguments when
1364 * lockdep is not enabled.
1366 * Mark both the sk_lock and the sk_lock.slock as a
1367 * per-address-family lock class.
1369 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1371 sk->sk_lock.owned = 0; \
1372 init_waitqueue_head(&sk->sk_lock.wq); \
1373 spin_lock_init(&(sk)->sk_lock.slock); \
1374 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1375 sizeof((sk)->sk_lock)); \
1376 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1378 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1381 extern void lock_sock_nested(struct sock *sk, int subclass);
1383 static inline void lock_sock(struct sock *sk)
1385 lock_sock_nested(sk, 0);
1388 extern void release_sock(struct sock *sk);
1390 /* BH context may only use the following locking interface. */
1391 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1392 #define bh_lock_sock_nested(__sk) \
1393 spin_lock_nested(&((__sk)->sk_lock.slock), \
1394 SINGLE_DEPTH_NESTING)
1395 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1397 extern bool lock_sock_fast(struct sock *sk);
1399 * unlock_sock_fast - complement of lock_sock_fast
1403 * fast unlock socket for user context.
1404 * If slow mode is on, we call regular release_sock()
1406 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1411 spin_unlock_bh(&sk->sk_lock.slock);
1415 extern struct sock *sk_alloc(struct net *net, int family,
1417 struct proto *prot);
1418 extern void sk_free(struct sock *sk);
1419 extern void sk_release_kernel(struct sock *sk);
1420 extern struct sock *sk_clone_lock(const struct sock *sk,
1421 const gfp_t priority);
1423 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1424 unsigned long size, int force,
1426 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1427 unsigned long size, int force,
1429 extern void sock_wfree(struct sk_buff *skb);
1430 extern void sock_rfree(struct sk_buff *skb);
1431 extern void sock_edemux(struct sk_buff *skb);
1433 extern int sock_setsockopt(struct socket *sock, int level,
1434 int op, char __user *optval,
1435 unsigned int optlen);
1437 extern int sock_getsockopt(struct socket *sock, int level,
1438 int op, char __user *optval,
1439 int __user *optlen);
1440 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1444 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1445 unsigned long header_len,
1446 unsigned long data_len,
1449 extern void *sock_kmalloc(struct sock *sk, int size,
1451 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1452 extern void sk_send_sigurg(struct sock *sk);
1454 #ifdef CONFIG_CGROUPS
1455 extern void sock_update_classid(struct sock *sk);
1457 static inline void sock_update_classid(struct sock *sk)
1463 * Functions to fill in entries in struct proto_ops when a protocol
1464 * does not implement a particular function.
1466 extern int sock_no_bind(struct socket *,
1467 struct sockaddr *, int);
1468 extern int sock_no_connect(struct socket *,
1469 struct sockaddr *, int, int);
1470 extern int sock_no_socketpair(struct socket *,
1472 extern int sock_no_accept(struct socket *,
1473 struct socket *, int);
1474 extern int sock_no_getname(struct socket *,
1475 struct sockaddr *, int *, int);
1476 extern unsigned int sock_no_poll(struct file *, struct socket *,
1477 struct poll_table_struct *);
1478 extern int sock_no_ioctl(struct socket *, unsigned int,
1480 extern int sock_no_listen(struct socket *, int);
1481 extern int sock_no_shutdown(struct socket *, int);
1482 extern int sock_no_getsockopt(struct socket *, int , int,
1483 char __user *, int __user *);
1484 extern int sock_no_setsockopt(struct socket *, int, int,
1485 char __user *, unsigned int);
1486 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1487 struct msghdr *, size_t);
1488 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1489 struct msghdr *, size_t, int);
1490 extern int sock_no_mmap(struct file *file,
1491 struct socket *sock,
1492 struct vm_area_struct *vma);
1493 extern ssize_t sock_no_sendpage(struct socket *sock,
1495 int offset, size_t size,
1499 * Functions to fill in entries in struct proto_ops when a protocol
1500 * uses the inet style.
1502 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1503 char __user *optval, int __user *optlen);
1504 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1505 struct msghdr *msg, size_t size, int flags);
1506 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1507 char __user *optval, unsigned int optlen);
1508 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1509 int optname, char __user *optval, int __user *optlen);
1510 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1511 int optname, char __user *optval, unsigned int optlen);
1513 extern void sk_common_release(struct sock *sk);
1516 * Default socket callbacks and setup code
1519 /* Initialise core socket variables */
1520 extern void sock_init_data(struct socket *sock, struct sock *sk);
1522 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1525 * sk_filter_release - release a socket filter
1526 * @fp: filter to remove
1528 * Remove a filter from a socket and release its resources.
1531 static inline void sk_filter_release(struct sk_filter *fp)
1533 if (atomic_dec_and_test(&fp->refcnt))
1534 call_rcu(&fp->rcu, sk_filter_release_rcu);
1537 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1539 unsigned int size = sk_filter_len(fp);
1541 atomic_sub(size, &sk->sk_omem_alloc);
1542 sk_filter_release(fp);
1545 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1547 atomic_inc(&fp->refcnt);
1548 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1552 * Socket reference counting postulates.
1554 * * Each user of socket SHOULD hold a reference count.
1555 * * Each access point to socket (an hash table bucket, reference from a list,
1556 * running timer, skb in flight MUST hold a reference count.
1557 * * When reference count hits 0, it means it will never increase back.
1558 * * When reference count hits 0, it means that no references from
1559 * outside exist to this socket and current process on current CPU
1560 * is last user and may/should destroy this socket.
1561 * * sk_free is called from any context: process, BH, IRQ. When
1562 * it is called, socket has no references from outside -> sk_free
1563 * may release descendant resources allocated by the socket, but
1564 * to the time when it is called, socket is NOT referenced by any
1565 * hash tables, lists etc.
1566 * * Packets, delivered from outside (from network or from another process)
1567 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1568 * when they sit in queue. Otherwise, packets will leak to hole, when
1569 * socket is looked up by one cpu and unhasing is made by another CPU.
1570 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1571 * (leak to backlog). Packet socket does all the processing inside
1572 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1573 * use separate SMP lock, so that they are prone too.
1576 /* Ungrab socket and destroy it, if it was the last reference. */
1577 static inline void sock_put(struct sock *sk)
1579 if (atomic_dec_and_test(&sk->sk_refcnt))
1583 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1586 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1588 sk->sk_tx_queue_mapping = tx_queue;
1591 static inline void sk_tx_queue_clear(struct sock *sk)
1593 sk->sk_tx_queue_mapping = -1;
1596 static inline int sk_tx_queue_get(const struct sock *sk)
1598 return sk ? sk->sk_tx_queue_mapping : -1;
1601 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1603 sk_tx_queue_clear(sk);
1604 sk->sk_socket = sock;
1607 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1609 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1610 return &rcu_dereference_raw(sk->sk_wq)->wait;
1612 /* Detach socket from process context.
1613 * Announce socket dead, detach it from wait queue and inode.
1614 * Note that parent inode held reference count on this struct sock,
1615 * we do not release it in this function, because protocol
1616 * probably wants some additional cleanups or even continuing
1617 * to work with this socket (TCP).
1619 static inline void sock_orphan(struct sock *sk)
1621 write_lock_bh(&sk->sk_callback_lock);
1622 sock_set_flag(sk, SOCK_DEAD);
1623 sk_set_socket(sk, NULL);
1625 write_unlock_bh(&sk->sk_callback_lock);
1628 static inline void sock_graft(struct sock *sk, struct socket *parent)
1630 write_lock_bh(&sk->sk_callback_lock);
1631 sk->sk_wq = parent->wq;
1633 sk_set_socket(sk, parent);
1634 security_sock_graft(sk, parent);
1635 write_unlock_bh(&sk->sk_callback_lock);
1638 extern int sock_i_uid(struct sock *sk);
1639 extern unsigned long sock_i_ino(struct sock *sk);
1641 static inline struct dst_entry *
1642 __sk_dst_get(struct sock *sk)
1644 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1645 lockdep_is_held(&sk->sk_lock.slock));
1648 static inline struct dst_entry *
1649 sk_dst_get(struct sock *sk)
1651 struct dst_entry *dst;
1654 dst = rcu_dereference(sk->sk_dst_cache);
1661 extern void sk_reset_txq(struct sock *sk);
1663 static inline void dst_negative_advice(struct sock *sk)
1665 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1667 if (dst && dst->ops->negative_advice) {
1668 ndst = dst->ops->negative_advice(dst);
1671 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1678 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1680 struct dst_entry *old_dst;
1682 sk_tx_queue_clear(sk);
1684 * This can be called while sk is owned by the caller only,
1685 * with no state that can be checked in a rcu_dereference_check() cond
1687 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1688 rcu_assign_pointer(sk->sk_dst_cache, dst);
1689 dst_release(old_dst);
1693 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1695 spin_lock(&sk->sk_dst_lock);
1696 __sk_dst_set(sk, dst);
1697 spin_unlock(&sk->sk_dst_lock);
1701 __sk_dst_reset(struct sock *sk)
1703 __sk_dst_set(sk, NULL);
1707 sk_dst_reset(struct sock *sk)
1709 spin_lock(&sk->sk_dst_lock);
1711 spin_unlock(&sk->sk_dst_lock);
1714 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1716 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1718 static inline bool sk_can_gso(const struct sock *sk)
1720 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1723 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1725 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1727 sk->sk_route_nocaps |= flags;
1728 sk->sk_route_caps &= ~flags;
1731 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1732 char __user *from, char *to,
1733 int copy, int offset)
1735 if (skb->ip_summed == CHECKSUM_NONE) {
1737 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1740 skb->csum = csum_block_add(skb->csum, csum, offset);
1741 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1742 if (!access_ok(VERIFY_READ, from, copy) ||
1743 __copy_from_user_nocache(to, from, copy))
1745 } else if (copy_from_user(to, from, copy))
1751 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1752 char __user *from, int copy)
1754 int err, offset = skb->len;
1756 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1759 __skb_trim(skb, offset);
1764 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1765 struct sk_buff *skb,
1771 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1777 skb->data_len += copy;
1778 skb->truesize += copy;
1779 sk->sk_wmem_queued += copy;
1780 sk_mem_charge(sk, copy);
1784 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1785 struct sk_buff *skb, struct page *page,
1788 if (skb->ip_summed == CHECKSUM_NONE) {
1790 __wsum csum = csum_and_copy_from_user(from,
1791 page_address(page) + off,
1795 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1796 } else if (copy_from_user(page_address(page) + off, from, copy))
1800 skb->data_len += copy;
1801 skb->truesize += copy;
1802 sk->sk_wmem_queued += copy;
1803 sk_mem_charge(sk, copy);
1808 * sk_wmem_alloc_get - returns write allocations
1811 * Returns sk_wmem_alloc minus initial offset of one
1813 static inline int sk_wmem_alloc_get(const struct sock *sk)
1815 return atomic_read(&sk->sk_wmem_alloc) - 1;
1819 * sk_rmem_alloc_get - returns read allocations
1822 * Returns sk_rmem_alloc
1824 static inline int sk_rmem_alloc_get(const struct sock *sk)
1826 return atomic_read(&sk->sk_rmem_alloc);
1830 * sk_has_allocations - check if allocations are outstanding
1833 * Returns true if socket has write or read allocations
1835 static inline bool sk_has_allocations(const struct sock *sk)
1837 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1841 * wq_has_sleeper - check if there are any waiting processes
1842 * @wq: struct socket_wq
1844 * Returns true if socket_wq has waiting processes
1846 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1847 * barrier call. They were added due to the race found within the tcp code.
1849 * Consider following tcp code paths:
1853 * sys_select receive packet
1855 * __add_wait_queue update tp->rcv_nxt
1857 * tp->rcv_nxt check sock_def_readable
1859 * schedule rcu_read_lock();
1860 * wq = rcu_dereference(sk->sk_wq);
1861 * if (wq && waitqueue_active(&wq->wait))
1862 * wake_up_interruptible(&wq->wait)
1866 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1867 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1868 * could then endup calling schedule and sleep forever if there are no more
1869 * data on the socket.
1872 static inline bool wq_has_sleeper(struct socket_wq *wq)
1874 /* We need to be sure we are in sync with the
1875 * add_wait_queue modifications to the wait queue.
1877 * This memory barrier is paired in the sock_poll_wait.
1880 return wq && waitqueue_active(&wq->wait);
1884 * sock_poll_wait - place memory barrier behind the poll_wait call.
1886 * @wait_address: socket wait queue
1889 * See the comments in the wq_has_sleeper function.
1891 static inline void sock_poll_wait(struct file *filp,
1892 wait_queue_head_t *wait_address, poll_table *p)
1894 if (!poll_does_not_wait(p) && wait_address) {
1895 poll_wait(filp, wait_address, p);
1896 /* We need to be sure we are in sync with the
1897 * socket flags modification.
1899 * This memory barrier is paired in the wq_has_sleeper.
1906 * Queue a received datagram if it will fit. Stream and sequenced
1907 * protocols can't normally use this as they need to fit buffers in
1908 * and play with them.
1910 * Inlined as it's very short and called for pretty much every
1911 * packet ever received.
1914 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1918 skb->destructor = sock_wfree;
1920 * We used to take a refcount on sk, but following operation
1921 * is enough to guarantee sk_free() wont free this sock until
1922 * all in-flight packets are completed
1924 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1927 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1931 skb->destructor = sock_rfree;
1932 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1933 sk_mem_charge(sk, skb->truesize);
1936 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1937 unsigned long expires);
1939 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1941 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1943 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1946 * Recover an error report and clear atomically
1949 static inline int sock_error(struct sock *sk)
1952 if (likely(!sk->sk_err))
1954 err = xchg(&sk->sk_err, 0);
1958 static inline unsigned long sock_wspace(struct sock *sk)
1962 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1963 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1970 static inline void sk_wake_async(struct sock *sk, int how, int band)
1972 if (sock_flag(sk, SOCK_FASYNC))
1973 sock_wake_async(sk->sk_socket, how, band);
1976 #define SOCK_MIN_SNDBUF 2048
1978 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1979 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1981 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1983 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1985 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1986 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1987 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1991 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1993 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1995 struct page *page = NULL;
1997 page = alloc_pages(sk->sk_allocation, 0);
1999 sk_enter_memory_pressure(sk);
2000 sk_stream_moderate_sndbuf(sk);
2006 * Default write policy as shown to user space via poll/select/SIGIO
2008 static inline bool sock_writeable(const struct sock *sk)
2010 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2013 static inline gfp_t gfp_any(void)
2015 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2018 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2020 return noblock ? 0 : sk->sk_rcvtimeo;
2023 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2025 return noblock ? 0 : sk->sk_sndtimeo;
2028 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2030 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2033 /* Alas, with timeout socket operations are not restartable.
2034 * Compare this to poll().
2036 static inline int sock_intr_errno(long timeo)
2038 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2041 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2042 struct sk_buff *skb);
2043 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2044 struct sk_buff *skb);
2047 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2049 ktime_t kt = skb->tstamp;
2050 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2053 * generate control messages if
2054 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2055 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2056 * - software time stamp available and wanted
2057 * (SOCK_TIMESTAMPING_SOFTWARE)
2058 * - hardware time stamps available and wanted
2059 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2060 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2062 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2063 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2064 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2065 (hwtstamps->hwtstamp.tv64 &&
2066 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2067 (hwtstamps->syststamp.tv64 &&
2068 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2069 __sock_recv_timestamp(msg, sk, skb);
2073 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2074 __sock_recv_wifi_status(msg, sk, skb);
2077 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2078 struct sk_buff *skb);
2080 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2081 struct sk_buff *skb)
2083 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2084 (1UL << SOCK_RCVTSTAMP) | \
2085 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2086 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2087 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2088 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2090 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2091 __sock_recv_ts_and_drops(msg, sk, skb);
2093 sk->sk_stamp = skb->tstamp;
2097 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2098 * @sk: socket sending this packet
2099 * @tx_flags: filled with instructions for time stamping
2101 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2102 * parameters are invalid.
2104 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2107 * sk_eat_skb - Release a skb if it is no longer needed
2108 * @sk: socket to eat this skb from
2109 * @skb: socket buffer to eat
2110 * @copied_early: flag indicating whether DMA operations copied this data early
2112 * This routine must be called with interrupts disabled or with the socket
2113 * locked so that the sk_buff queue operation is ok.
2115 #ifdef CONFIG_NET_DMA
2116 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2118 __skb_unlink(skb, &sk->sk_receive_queue);
2122 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2125 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2127 __skb_unlink(skb, &sk->sk_receive_queue);
2133 struct net *sock_net(const struct sock *sk)
2135 return read_pnet(&sk->sk_net);
2139 void sock_net_set(struct sock *sk, struct net *net)
2141 write_pnet(&sk->sk_net, net);
2145 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2146 * They should not hold a reference to a namespace in order to allow
2148 * Sockets after sk_change_net should be released using sk_release_kernel
2150 static inline void sk_change_net(struct sock *sk, struct net *net)
2152 put_net(sock_net(sk));
2153 sock_net_set(sk, hold_net(net));
2156 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2159 struct sock *sk = skb->sk;
2161 skb->destructor = NULL;
2168 extern void sock_enable_timestamp(struct sock *sk, int flag);
2169 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2170 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2173 * Enable debug/info messages
2175 extern int net_msg_warn;
2176 #define NETDEBUG(fmt, args...) \
2177 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2179 #define LIMIT_NETDEBUG(fmt, args...) \
2180 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2182 extern __u32 sysctl_wmem_max;
2183 extern __u32 sysctl_rmem_max;
2185 extern void sk_init(void);
2187 extern int sysctl_optmem_max;
2189 extern __u32 sysctl_wmem_default;
2190 extern __u32 sysctl_rmem_default;
2192 #endif /* _SOCK_H */