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;
129 typedef __u32 __bitwise __portpair;
130 typedef __u64 __bitwise __addrpair;
133 * struct sock_common - minimal network layer representation of sockets
134 * @skc_daddr: Foreign IPv4 addr
135 * @skc_rcv_saddr: Bound local IPv4 addr
136 * @skc_hash: hash value used with various protocol lookup tables
137 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
138 * @skc_dport: placeholder for inet_dport/tw_dport
139 * @skc_num: placeholder for inet_num/tw_num
140 * @skc_family: network address family
141 * @skc_state: Connection state
142 * @skc_reuse: %SO_REUSEADDR setting
143 * @skc_reuseport: %SO_REUSEPORT setting
144 * @skc_bound_dev_if: bound device index if != 0
145 * @skc_bind_node: bind hash linkage for various protocol lookup tables
146 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
147 * @skc_prot: protocol handlers inside a network family
148 * @skc_net: reference to the network namespace of this socket
149 * @skc_node: main hash linkage for various protocol lookup tables
150 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
151 * @skc_tx_queue_mapping: tx queue number for this connection
152 * @skc_refcnt: reference count
154 * This is the minimal network layer representation of sockets, the header
155 * for struct sock and struct inet_timewait_sock.
158 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
159 * address on 64bit arches : cf INET_MATCH()
162 __addrpair skc_addrpair;
165 __be32 skc_rcv_saddr;
169 unsigned int skc_hash;
170 __u16 skc_u16hashes[2];
172 /* skc_dport && skc_num must be grouped as well */
174 __portpair skc_portpair;
181 unsigned short skc_family;
182 volatile unsigned char skc_state;
183 unsigned char skc_reuse:4;
184 unsigned char skc_reuseport:4;
185 int skc_bound_dev_if;
187 struct hlist_node skc_bind_node;
188 struct hlist_nulls_node skc_portaddr_node;
190 struct proto *skc_prot;
195 * fields between dontcopy_begin/dontcopy_end
196 * are not copied in sock_copy()
199 int skc_dontcopy_begin[0];
202 struct hlist_node skc_node;
203 struct hlist_nulls_node skc_nulls_node;
205 int skc_tx_queue_mapping;
208 int skc_dontcopy_end[0];
214 * struct sock - network layer representation of sockets
215 * @__sk_common: shared layout with inet_timewait_sock
216 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
217 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
218 * @sk_lock: synchronizer
219 * @sk_rcvbuf: size of receive buffer in bytes
220 * @sk_wq: sock wait queue and async head
221 * @sk_rx_dst: receive input route used by early demux
222 * @sk_dst_cache: destination cache
223 * @sk_dst_lock: destination cache lock
224 * @sk_policy: flow policy
225 * @sk_receive_queue: incoming packets
226 * @sk_wmem_alloc: transmit queue bytes committed
227 * @sk_write_queue: Packet sending queue
228 * @sk_async_wait_queue: DMA copied packets
229 * @sk_omem_alloc: "o" is "option" or "other"
230 * @sk_wmem_queued: persistent queue size
231 * @sk_forward_alloc: space allocated forward
232 * @sk_napi_id: id of the last napi context to receive data for sk
233 * @sk_ll_usec: usecs to busypoll when there is no data
234 * @sk_allocation: allocation mode
235 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
236 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
237 * @sk_sndbuf: size of send buffer in bytes
238 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
239 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
240 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets
241 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
242 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
243 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
244 * @sk_gso_max_size: Maximum GSO segment size to build
245 * @sk_gso_max_segs: Maximum number of GSO segments
246 * @sk_lingertime: %SO_LINGER l_linger setting
247 * @sk_backlog: always used with the per-socket spinlock held
248 * @sk_callback_lock: used with the callbacks in the end of this struct
249 * @sk_error_queue: rarely used
250 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
251 * IPV6_ADDRFORM for instance)
252 * @sk_err: last error
253 * @sk_err_soft: errors that don't cause failure but are the cause of a
254 * persistent failure not just 'timed out'
255 * @sk_drops: raw/udp drops counter
256 * @sk_ack_backlog: current listen backlog
257 * @sk_max_ack_backlog: listen backlog set in listen()
258 * @sk_priority: %SO_PRIORITY setting
259 * @sk_cgrp_prioidx: socket group's priority map index
260 * @sk_type: socket type (%SOCK_STREAM, etc)
261 * @sk_protocol: which protocol this socket belongs in this network family
262 * @sk_peer_pid: &struct pid for this socket's peer
263 * @sk_peer_cred: %SO_PEERCRED setting
264 * @sk_rcvlowat: %SO_RCVLOWAT setting
265 * @sk_rcvtimeo: %SO_RCVTIMEO setting
266 * @sk_sndtimeo: %SO_SNDTIMEO setting
267 * @sk_rxhash: flow hash received from netif layer
268 * @sk_filter: socket filtering instructions
269 * @sk_protinfo: private area, net family specific, when not using slab
270 * @sk_timer: sock cleanup timer
271 * @sk_stamp: time stamp of last packet received
272 * @sk_socket: Identd and reporting IO signals
273 * @sk_user_data: RPC layer private data
274 * @sk_frag: cached page frag
275 * @sk_peek_off: current peek_offset value
276 * @sk_send_head: front of stuff to transmit
277 * @sk_security: used by security modules
278 * @sk_mark: generic packet mark
279 * @sk_classid: this socket's cgroup classid
280 * @sk_cgrp: this socket's cgroup-specific proto data
281 * @sk_write_pending: a write to stream socket waits to start
282 * @sk_state_change: callback to indicate change in the state of the sock
283 * @sk_data_ready: callback to indicate there is data to be processed
284 * @sk_write_space: callback to indicate there is bf sending space available
285 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
286 * @sk_backlog_rcv: callback to process the backlog
287 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
291 * Now struct inet_timewait_sock also uses sock_common, so please just
292 * don't add nothing before this first member (__sk_common) --acme
294 struct sock_common __sk_common;
295 #define sk_node __sk_common.skc_node
296 #define sk_nulls_node __sk_common.skc_nulls_node
297 #define sk_refcnt __sk_common.skc_refcnt
298 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
300 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
301 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
302 #define sk_hash __sk_common.skc_hash
303 #define sk_portpair __sk_common.skc_portpair
304 #define sk_num __sk_common.skc_num
305 #define sk_dport __sk_common.skc_dport
306 #define sk_addrpair __sk_common.skc_addrpair
307 #define sk_daddr __sk_common.skc_daddr
308 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
309 #define sk_family __sk_common.skc_family
310 #define sk_state __sk_common.skc_state
311 #define sk_reuse __sk_common.skc_reuse
312 #define sk_reuseport __sk_common.skc_reuseport
313 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
314 #define sk_bind_node __sk_common.skc_bind_node
315 #define sk_prot __sk_common.skc_prot
316 #define sk_net __sk_common.skc_net
317 socket_lock_t sk_lock;
318 struct sk_buff_head sk_receive_queue;
320 * The backlog queue is special, it is always used with
321 * the per-socket spinlock held and requires low latency
322 * access. Therefore we special case it's implementation.
323 * Note : rmem_alloc is in this structure to fill a hole
324 * on 64bit arches, not because its logically part of
330 struct sk_buff *head;
331 struct sk_buff *tail;
333 #define sk_rmem_alloc sk_backlog.rmem_alloc
334 int sk_forward_alloc;
338 #ifdef CONFIG_NET_RX_BUSY_POLL
339 unsigned int sk_napi_id;
340 unsigned int sk_ll_usec;
345 struct sk_filter __rcu *sk_filter;
346 struct socket_wq __rcu *sk_wq;
348 #ifdef CONFIG_NET_DMA
349 struct sk_buff_head sk_async_wait_queue;
353 struct xfrm_policy *sk_policy[2];
355 unsigned long sk_flags;
356 struct dst_entry *sk_rx_dst;
357 struct dst_entry __rcu *sk_dst_cache;
358 spinlock_t sk_dst_lock;
359 atomic_t sk_wmem_alloc;
360 atomic_t sk_omem_alloc;
362 struct sk_buff_head sk_write_queue;
363 kmemcheck_bitfield_begin(flags);
364 unsigned int sk_shutdown : 2,
369 kmemcheck_bitfield_end(flags);
372 u32 sk_pacing_rate; /* bytes per second */
373 u32 sk_max_pacing_rate;
374 netdev_features_t sk_route_caps;
375 netdev_features_t sk_route_nocaps;
377 unsigned int sk_gso_max_size;
380 unsigned long sk_lingertime;
381 struct sk_buff_head sk_error_queue;
382 struct proto *sk_prot_creator;
383 rwlock_t sk_callback_lock;
386 unsigned short sk_ack_backlog;
387 unsigned short sk_max_ack_backlog;
389 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
390 __u32 sk_cgrp_prioidx;
392 struct pid *sk_peer_pid;
393 const struct cred *sk_peer_cred;
397 struct timer_list sk_timer;
399 struct socket *sk_socket;
401 struct page_frag sk_frag;
402 struct sk_buff *sk_send_head;
404 int sk_write_pending;
405 #ifdef CONFIG_SECURITY
410 struct cg_proto *sk_cgrp;
411 void (*sk_state_change)(struct sock *sk);
412 void (*sk_data_ready)(struct sock *sk, int bytes);
413 void (*sk_write_space)(struct sock *sk);
414 void (*sk_error_report)(struct sock *sk);
415 int (*sk_backlog_rcv)(struct sock *sk,
416 struct sk_buff *skb);
417 void (*sk_destruct)(struct sock *sk);
420 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
422 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
423 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
426 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
427 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
428 * on a socket means that the socket will reuse everybody else's port
429 * without looking at the other's sk_reuse value.
432 #define SK_NO_REUSE 0
433 #define SK_CAN_REUSE 1
434 #define SK_FORCE_REUSE 2
436 static inline int sk_peek_offset(struct sock *sk, int flags)
438 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
439 return sk->sk_peek_off;
444 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
446 if (sk->sk_peek_off >= 0) {
447 if (sk->sk_peek_off >= val)
448 sk->sk_peek_off -= val;
454 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
456 if (sk->sk_peek_off >= 0)
457 sk->sk_peek_off += val;
461 * Hashed lists helper routines
463 static inline struct sock *sk_entry(const struct hlist_node *node)
465 return hlist_entry(node, struct sock, sk_node);
468 static inline struct sock *__sk_head(const struct hlist_head *head)
470 return hlist_entry(head->first, struct sock, sk_node);
473 static inline struct sock *sk_head(const struct hlist_head *head)
475 return hlist_empty(head) ? NULL : __sk_head(head);
478 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
480 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
483 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
485 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
488 static inline struct sock *sk_next(const struct sock *sk)
490 return sk->sk_node.next ?
491 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
494 static inline struct sock *sk_nulls_next(const struct sock *sk)
496 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
497 hlist_nulls_entry(sk->sk_nulls_node.next,
498 struct sock, sk_nulls_node) :
502 static inline bool sk_unhashed(const struct sock *sk)
504 return hlist_unhashed(&sk->sk_node);
507 static inline bool sk_hashed(const struct sock *sk)
509 return !sk_unhashed(sk);
512 static inline void sk_node_init(struct hlist_node *node)
517 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
522 static inline void __sk_del_node(struct sock *sk)
524 __hlist_del(&sk->sk_node);
527 /* NB: equivalent to hlist_del_init_rcu */
528 static inline bool __sk_del_node_init(struct sock *sk)
532 sk_node_init(&sk->sk_node);
538 /* Grab socket reference count. This operation is valid only
539 when sk is ALREADY grabbed f.e. it is found in hash table
540 or a list and the lookup is made under lock preventing hash table
544 static inline void sock_hold(struct sock *sk)
546 atomic_inc(&sk->sk_refcnt);
549 /* Ungrab socket in the context, which assumes that socket refcnt
550 cannot hit zero, f.e. it is true in context of any socketcall.
552 static inline void __sock_put(struct sock *sk)
554 atomic_dec(&sk->sk_refcnt);
557 static inline bool sk_del_node_init(struct sock *sk)
559 bool rc = __sk_del_node_init(sk);
562 /* paranoid for a while -acme */
563 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
568 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
570 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
573 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
579 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
581 bool rc = __sk_nulls_del_node_init_rcu(sk);
584 /* paranoid for a while -acme */
585 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
591 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
593 hlist_add_head(&sk->sk_node, list);
596 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
599 __sk_add_node(sk, list);
602 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
605 hlist_add_head_rcu(&sk->sk_node, list);
608 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
610 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
613 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
616 __sk_nulls_add_node_rcu(sk, list);
619 static inline void __sk_del_bind_node(struct sock *sk)
621 __hlist_del(&sk->sk_bind_node);
624 static inline void sk_add_bind_node(struct sock *sk,
625 struct hlist_head *list)
627 hlist_add_head(&sk->sk_bind_node, list);
630 #define sk_for_each(__sk, list) \
631 hlist_for_each_entry(__sk, list, sk_node)
632 #define sk_for_each_rcu(__sk, list) \
633 hlist_for_each_entry_rcu(__sk, list, sk_node)
634 #define sk_nulls_for_each(__sk, node, list) \
635 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
636 #define sk_nulls_for_each_rcu(__sk, node, list) \
637 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
638 #define sk_for_each_from(__sk) \
639 hlist_for_each_entry_from(__sk, sk_node)
640 #define sk_nulls_for_each_from(__sk, node) \
641 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
642 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
643 #define sk_for_each_safe(__sk, tmp, list) \
644 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
645 #define sk_for_each_bound(__sk, list) \
646 hlist_for_each_entry(__sk, list, sk_bind_node)
648 static inline struct user_namespace *sk_user_ns(struct sock *sk)
650 /* Careful only use this in a context where these parameters
651 * can not change and must all be valid, such as recvmsg from
654 return sk->sk_socket->file->f_cred->user_ns;
668 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
669 SOCK_DBG, /* %SO_DEBUG setting */
670 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
671 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
672 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
673 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
674 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
675 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
676 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
677 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
678 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
679 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
680 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
681 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
682 SOCK_FASYNC, /* fasync() active */
684 SOCK_ZEROCOPY, /* buffers from userspace */
685 SOCK_WIFI_STATUS, /* push wifi status to userspace */
686 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
687 * Will use last 4 bytes of packet sent from
688 * user-space instead.
690 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
691 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
694 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
696 nsk->sk_flags = osk->sk_flags;
699 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
701 __set_bit(flag, &sk->sk_flags);
704 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
706 __clear_bit(flag, &sk->sk_flags);
709 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
711 return test_bit(flag, &sk->sk_flags);
715 extern struct static_key memalloc_socks;
716 static inline int sk_memalloc_socks(void)
718 return static_key_false(&memalloc_socks);
722 static inline int sk_memalloc_socks(void)
729 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
731 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
734 static inline void sk_acceptq_removed(struct sock *sk)
736 sk->sk_ack_backlog--;
739 static inline void sk_acceptq_added(struct sock *sk)
741 sk->sk_ack_backlog++;
744 static inline bool sk_acceptq_is_full(const struct sock *sk)
746 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
750 * Compute minimal free write space needed to queue new packets.
752 static inline int sk_stream_min_wspace(const struct sock *sk)
754 return sk->sk_wmem_queued >> 1;
757 static inline int sk_stream_wspace(const struct sock *sk)
759 return sk->sk_sndbuf - sk->sk_wmem_queued;
762 void sk_stream_write_space(struct sock *sk);
764 /* OOB backlog add */
765 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
767 /* dont let skb dst not refcounted, we are going to leave rcu lock */
770 if (!sk->sk_backlog.tail)
771 sk->sk_backlog.head = skb;
773 sk->sk_backlog.tail->next = skb;
775 sk->sk_backlog.tail = skb;
780 * Take into account size of receive queue and backlog queue
781 * Do not take into account this skb truesize,
782 * to allow even a single big packet to come.
784 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
787 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
789 return qsize > limit;
792 /* The per-socket spinlock must be held here. */
793 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
796 if (sk_rcvqueues_full(sk, skb, limit))
799 __sk_add_backlog(sk, skb);
800 sk->sk_backlog.len += skb->truesize;
804 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
806 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
808 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
809 return __sk_backlog_rcv(sk, skb);
811 return sk->sk_backlog_rcv(sk, skb);
814 static inline void sock_rps_record_flow(const struct sock *sk)
817 struct rps_sock_flow_table *sock_flow_table;
820 sock_flow_table = rcu_dereference(rps_sock_flow_table);
821 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
826 static inline void sock_rps_reset_flow(const struct sock *sk)
829 struct rps_sock_flow_table *sock_flow_table;
832 sock_flow_table = rcu_dereference(rps_sock_flow_table);
833 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
838 static inline void sock_rps_save_rxhash(struct sock *sk,
839 const struct sk_buff *skb)
842 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
843 sock_rps_reset_flow(sk);
844 sk->sk_rxhash = skb->rxhash;
849 static inline void sock_rps_reset_rxhash(struct sock *sk)
852 sock_rps_reset_flow(sk);
857 #define sk_wait_event(__sk, __timeo, __condition) \
859 release_sock(__sk); \
860 __rc = __condition; \
862 *(__timeo) = schedule_timeout(*(__timeo)); \
865 __rc = __condition; \
869 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
870 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
871 void sk_stream_wait_close(struct sock *sk, long timeo_p);
872 int sk_stream_error(struct sock *sk, int flags, int err);
873 void sk_stream_kill_queues(struct sock *sk);
874 void sk_set_memalloc(struct sock *sk);
875 void sk_clear_memalloc(struct sock *sk);
877 int sk_wait_data(struct sock *sk, long *timeo);
879 struct request_sock_ops;
880 struct timewait_sock_ops;
881 struct inet_hashinfo;
886 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
887 * un-modified. Special care is taken when initializing object to zero.
889 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
891 if (offsetof(struct sock, sk_node.next) != 0)
892 memset(sk, 0, offsetof(struct sock, sk_node.next));
893 memset(&sk->sk_node.pprev, 0,
894 size - offsetof(struct sock, sk_node.pprev));
897 /* Networking protocol blocks we attach to sockets.
898 * socket layer -> transport layer interface
899 * transport -> network interface is defined by struct inet_proto
902 void (*close)(struct sock *sk,
904 int (*connect)(struct sock *sk,
905 struct sockaddr *uaddr,
907 int (*disconnect)(struct sock *sk, int flags);
909 struct sock * (*accept)(struct sock *sk, int flags, int *err);
911 int (*ioctl)(struct sock *sk, int cmd,
913 int (*init)(struct sock *sk);
914 void (*destroy)(struct sock *sk);
915 void (*shutdown)(struct sock *sk, int how);
916 int (*setsockopt)(struct sock *sk, int level,
917 int optname, char __user *optval,
918 unsigned int optlen);
919 int (*getsockopt)(struct sock *sk, int level,
920 int optname, char __user *optval,
923 int (*compat_setsockopt)(struct sock *sk,
925 int optname, char __user *optval,
926 unsigned int optlen);
927 int (*compat_getsockopt)(struct sock *sk,
929 int optname, char __user *optval,
931 int (*compat_ioctl)(struct sock *sk,
932 unsigned int cmd, unsigned long arg);
934 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
935 struct msghdr *msg, size_t len);
936 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
938 size_t len, int noblock, int flags,
940 int (*sendpage)(struct sock *sk, struct page *page,
941 int offset, size_t size, int flags);
942 int (*bind)(struct sock *sk,
943 struct sockaddr *uaddr, int addr_len);
945 int (*backlog_rcv) (struct sock *sk,
946 struct sk_buff *skb);
948 void (*release_cb)(struct sock *sk);
949 void (*mtu_reduced)(struct sock *sk);
951 /* Keeping track of sk's, looking them up, and port selection methods. */
952 void (*hash)(struct sock *sk);
953 void (*unhash)(struct sock *sk);
954 void (*rehash)(struct sock *sk);
955 int (*get_port)(struct sock *sk, unsigned short snum);
956 void (*clear_sk)(struct sock *sk, int size);
958 /* Keeping track of sockets in use */
959 #ifdef CONFIG_PROC_FS
960 unsigned int inuse_idx;
963 bool (*stream_memory_free)(const struct sock *sk);
964 /* Memory pressure */
965 void (*enter_memory_pressure)(struct sock *sk);
966 atomic_long_t *memory_allocated; /* Current allocated memory. */
967 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
969 * Pressure flag: try to collapse.
970 * Technical note: it is used by multiple contexts non atomically.
971 * All the __sk_mem_schedule() is of this nature: accounting
972 * is strict, actions are advisory and have some latency.
974 int *memory_pressure;
981 struct kmem_cache *slab;
982 unsigned int obj_size;
985 struct percpu_counter *orphan_count;
987 struct request_sock_ops *rsk_prot;
988 struct timewait_sock_ops *twsk_prot;
991 struct inet_hashinfo *hashinfo;
992 struct udp_table *udp_table;
993 struct raw_hashinfo *raw_hash;
996 struct module *owner;
1000 struct list_head node;
1001 #ifdef SOCK_REFCNT_DEBUG
1004 #ifdef CONFIG_MEMCG_KMEM
1006 * cgroup specific init/deinit functions. Called once for all
1007 * protocols that implement it, from cgroups populate function.
1008 * This function has to setup any files the protocol want to
1009 * appear in the kmem cgroup filesystem.
1011 int (*init_cgroup)(struct mem_cgroup *memcg,
1012 struct cgroup_subsys *ss);
1013 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1014 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1019 * Bits in struct cg_proto.flags
1021 enum cg_proto_flags {
1022 /* Currently active and new sockets should be assigned to cgroups */
1024 /* It was ever activated; we must disarm static keys on destruction */
1025 MEMCG_SOCK_ACTIVATED,
1029 void (*enter_memory_pressure)(struct sock *sk);
1030 struct res_counter *memory_allocated; /* Current allocated memory. */
1031 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1032 int *memory_pressure;
1034 unsigned long flags;
1036 * memcg field is used to find which memcg we belong directly
1037 * Each memcg struct can hold more than one cg_proto, so container_of
1040 * The elegant solution would be having an inverse function to
1041 * proto_cgroup in struct proto, but that means polluting the structure
1042 * for everybody, instead of just for memcg users.
1044 struct mem_cgroup *memcg;
1047 int proto_register(struct proto *prot, int alloc_slab);
1048 void proto_unregister(struct proto *prot);
1050 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1052 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1055 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1057 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1060 #ifdef SOCK_REFCNT_DEBUG
1061 static inline void sk_refcnt_debug_inc(struct sock *sk)
1063 atomic_inc(&sk->sk_prot->socks);
1066 static inline void sk_refcnt_debug_dec(struct sock *sk)
1068 atomic_dec(&sk->sk_prot->socks);
1069 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1070 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1073 static inline void sk_refcnt_debug_release(const struct sock *sk)
1075 if (atomic_read(&sk->sk_refcnt) != 1)
1076 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1077 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1079 #else /* SOCK_REFCNT_DEBUG */
1080 #define sk_refcnt_debug_inc(sk) do { } while (0)
1081 #define sk_refcnt_debug_dec(sk) do { } while (0)
1082 #define sk_refcnt_debug_release(sk) do { } while (0)
1083 #endif /* SOCK_REFCNT_DEBUG */
1085 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1086 extern struct static_key memcg_socket_limit_enabled;
1087 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1088 struct cg_proto *cg_proto)
1090 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1092 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1094 #define mem_cgroup_sockets_enabled 0
1095 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1096 struct cg_proto *cg_proto)
1102 static inline bool sk_stream_memory_free(const struct sock *sk)
1104 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1107 return sk->sk_prot->stream_memory_free ?
1108 sk->sk_prot->stream_memory_free(sk) : true;
1111 static inline bool sk_stream_is_writeable(const struct sock *sk)
1113 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1114 sk_stream_memory_free(sk);
1118 static inline bool sk_has_memory_pressure(const struct sock *sk)
1120 return sk->sk_prot->memory_pressure != NULL;
1123 static inline bool sk_under_memory_pressure(const struct sock *sk)
1125 if (!sk->sk_prot->memory_pressure)
1128 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1129 return !!*sk->sk_cgrp->memory_pressure;
1131 return !!*sk->sk_prot->memory_pressure;
1134 static inline void sk_leave_memory_pressure(struct sock *sk)
1136 int *memory_pressure = sk->sk_prot->memory_pressure;
1138 if (!memory_pressure)
1141 if (*memory_pressure)
1142 *memory_pressure = 0;
1144 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1145 struct cg_proto *cg_proto = sk->sk_cgrp;
1146 struct proto *prot = sk->sk_prot;
1148 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1149 if (*cg_proto->memory_pressure)
1150 *cg_proto->memory_pressure = 0;
1155 static inline void sk_enter_memory_pressure(struct sock *sk)
1157 if (!sk->sk_prot->enter_memory_pressure)
1160 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1161 struct cg_proto *cg_proto = sk->sk_cgrp;
1162 struct proto *prot = sk->sk_prot;
1164 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1165 cg_proto->enter_memory_pressure(sk);
1168 sk->sk_prot->enter_memory_pressure(sk);
1171 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1173 long *prot = sk->sk_prot->sysctl_mem;
1174 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1175 prot = sk->sk_cgrp->sysctl_mem;
1179 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1183 struct res_counter *fail;
1186 ret = res_counter_charge_nofail(prot->memory_allocated,
1187 amt << PAGE_SHIFT, &fail);
1189 *parent_status = OVER_LIMIT;
1192 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1195 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1198 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1201 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1202 return ret >> PAGE_SHIFT;
1206 sk_memory_allocated(const struct sock *sk)
1208 struct proto *prot = sk->sk_prot;
1209 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1210 return memcg_memory_allocated_read(sk->sk_cgrp);
1212 return atomic_long_read(prot->memory_allocated);
1216 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1218 struct proto *prot = sk->sk_prot;
1220 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1221 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1222 /* update the root cgroup regardless */
1223 atomic_long_add_return(amt, prot->memory_allocated);
1224 return memcg_memory_allocated_read(sk->sk_cgrp);
1227 return atomic_long_add_return(amt, prot->memory_allocated);
1231 sk_memory_allocated_sub(struct sock *sk, int amt)
1233 struct proto *prot = sk->sk_prot;
1235 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1236 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1238 atomic_long_sub(amt, prot->memory_allocated);
1241 static inline void sk_sockets_allocated_dec(struct sock *sk)
1243 struct proto *prot = sk->sk_prot;
1245 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1246 struct cg_proto *cg_proto = sk->sk_cgrp;
1248 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1249 percpu_counter_dec(cg_proto->sockets_allocated);
1252 percpu_counter_dec(prot->sockets_allocated);
1255 static inline void sk_sockets_allocated_inc(struct sock *sk)
1257 struct proto *prot = sk->sk_prot;
1259 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1260 struct cg_proto *cg_proto = sk->sk_cgrp;
1262 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1263 percpu_counter_inc(cg_proto->sockets_allocated);
1266 percpu_counter_inc(prot->sockets_allocated);
1270 sk_sockets_allocated_read_positive(struct sock *sk)
1272 struct proto *prot = sk->sk_prot;
1274 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1275 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1277 return percpu_counter_read_positive(prot->sockets_allocated);
1281 proto_sockets_allocated_sum_positive(struct proto *prot)
1283 return percpu_counter_sum_positive(prot->sockets_allocated);
1287 proto_memory_allocated(struct proto *prot)
1289 return atomic_long_read(prot->memory_allocated);
1293 proto_memory_pressure(struct proto *prot)
1295 if (!prot->memory_pressure)
1297 return !!*prot->memory_pressure;
1301 #ifdef CONFIG_PROC_FS
1302 /* Called with local bh disabled */
1303 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1304 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1306 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1313 /* With per-bucket locks this operation is not-atomic, so that
1314 * this version is not worse.
1316 static inline void __sk_prot_rehash(struct sock *sk)
1318 sk->sk_prot->unhash(sk);
1319 sk->sk_prot->hash(sk);
1322 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1324 /* About 10 seconds */
1325 #define SOCK_DESTROY_TIME (10*HZ)
1327 /* Sockets 0-1023 can't be bound to unless you are superuser */
1328 #define PROT_SOCK 1024
1330 #define SHUTDOWN_MASK 3
1331 #define RCV_SHUTDOWN 1
1332 #define SEND_SHUTDOWN 2
1334 #define SOCK_SNDBUF_LOCK 1
1335 #define SOCK_RCVBUF_LOCK 2
1336 #define SOCK_BINDADDR_LOCK 4
1337 #define SOCK_BINDPORT_LOCK 8
1339 /* sock_iocb: used to kick off async processing of socket ios */
1341 struct list_head list;
1345 struct socket *sock;
1347 struct scm_cookie *scm;
1348 struct msghdr *msg, async_msg;
1349 struct kiocb *kiocb;
1352 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1354 return (struct sock_iocb *)iocb->private;
1357 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1362 struct socket_alloc {
1363 struct socket socket;
1364 struct inode vfs_inode;
1367 static inline struct socket *SOCKET_I(struct inode *inode)
1369 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1372 static inline struct inode *SOCK_INODE(struct socket *socket)
1374 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1378 * Functions for memory accounting
1380 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1381 void __sk_mem_reclaim(struct sock *sk);
1383 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1384 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1385 #define SK_MEM_SEND 0
1386 #define SK_MEM_RECV 1
1388 static inline int sk_mem_pages(int amt)
1390 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1393 static inline bool sk_has_account(struct sock *sk)
1395 /* return true if protocol supports memory accounting */
1396 return !!sk->sk_prot->memory_allocated;
1399 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1401 if (!sk_has_account(sk))
1403 return size <= sk->sk_forward_alloc ||
1404 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1408 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1410 if (!sk_has_account(sk))
1412 return size<= sk->sk_forward_alloc ||
1413 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1414 skb_pfmemalloc(skb);
1417 static inline void sk_mem_reclaim(struct sock *sk)
1419 if (!sk_has_account(sk))
1421 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1422 __sk_mem_reclaim(sk);
1425 static inline void sk_mem_reclaim_partial(struct sock *sk)
1427 if (!sk_has_account(sk))
1429 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1430 __sk_mem_reclaim(sk);
1433 static inline void sk_mem_charge(struct sock *sk, int size)
1435 if (!sk_has_account(sk))
1437 sk->sk_forward_alloc -= size;
1440 static inline void sk_mem_uncharge(struct sock *sk, int size)
1442 if (!sk_has_account(sk))
1444 sk->sk_forward_alloc += size;
1447 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1449 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1450 sk->sk_wmem_queued -= skb->truesize;
1451 sk_mem_uncharge(sk, skb->truesize);
1455 /* Used by processes to "lock" a socket state, so that
1456 * interrupts and bottom half handlers won't change it
1457 * from under us. It essentially blocks any incoming
1458 * packets, so that we won't get any new data or any
1459 * packets that change the state of the socket.
1461 * While locked, BH processing will add new packets to
1462 * the backlog queue. This queue is processed by the
1463 * owner of the socket lock right before it is released.
1465 * Since ~2.3.5 it is also exclusive sleep lock serializing
1466 * accesses from user process context.
1468 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1471 * Macro so as to not evaluate some arguments when
1472 * lockdep is not enabled.
1474 * Mark both the sk_lock and the sk_lock.slock as a
1475 * per-address-family lock class.
1477 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1479 sk->sk_lock.owned = 0; \
1480 init_waitqueue_head(&sk->sk_lock.wq); \
1481 spin_lock_init(&(sk)->sk_lock.slock); \
1482 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1483 sizeof((sk)->sk_lock)); \
1484 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1486 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1489 void lock_sock_nested(struct sock *sk, int subclass);
1491 static inline void lock_sock(struct sock *sk)
1493 lock_sock_nested(sk, 0);
1496 void release_sock(struct sock *sk);
1498 /* BH context may only use the following locking interface. */
1499 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1500 #define bh_lock_sock_nested(__sk) \
1501 spin_lock_nested(&((__sk)->sk_lock.slock), \
1502 SINGLE_DEPTH_NESTING)
1503 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1505 bool lock_sock_fast(struct sock *sk);
1507 * unlock_sock_fast - complement of lock_sock_fast
1511 * fast unlock socket for user context.
1512 * If slow mode is on, we call regular release_sock()
1514 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1519 spin_unlock_bh(&sk->sk_lock.slock);
1523 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1524 struct proto *prot);
1525 void sk_free(struct sock *sk);
1526 void sk_release_kernel(struct sock *sk);
1527 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1529 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1531 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1533 void sock_wfree(struct sk_buff *skb);
1534 void skb_orphan_partial(struct sk_buff *skb);
1535 void sock_rfree(struct sk_buff *skb);
1536 void sock_edemux(struct sk_buff *skb);
1538 int sock_setsockopt(struct socket *sock, int level, int op,
1539 char __user *optval, unsigned int optlen);
1541 int sock_getsockopt(struct socket *sock, int level, int op,
1542 char __user *optval, int __user *optlen);
1543 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1544 int noblock, int *errcode);
1545 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1546 unsigned long data_len, int noblock,
1547 int *errcode, int max_page_order);
1548 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1549 void sock_kfree_s(struct sock *sk, void *mem, int size);
1550 void sk_send_sigurg(struct sock *sk);
1553 * Functions to fill in entries in struct proto_ops when a protocol
1554 * does not implement a particular function.
1556 int sock_no_bind(struct socket *, struct sockaddr *, int);
1557 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1558 int sock_no_socketpair(struct socket *, struct socket *);
1559 int sock_no_accept(struct socket *, struct socket *, int);
1560 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1561 unsigned int sock_no_poll(struct file *, struct socket *,
1562 struct poll_table_struct *);
1563 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1564 int sock_no_listen(struct socket *, int);
1565 int sock_no_shutdown(struct socket *, int);
1566 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1567 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1568 int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t);
1569 int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t,
1571 int sock_no_mmap(struct file *file, struct socket *sock,
1572 struct vm_area_struct *vma);
1573 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1574 size_t size, int flags);
1577 * Functions to fill in entries in struct proto_ops when a protocol
1578 * uses the inet style.
1580 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1581 char __user *optval, int __user *optlen);
1582 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1583 struct msghdr *msg, size_t size, int flags);
1584 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1585 char __user *optval, unsigned int optlen);
1586 int compat_sock_common_getsockopt(struct socket *sock, int level,
1587 int optname, char __user *optval, int __user *optlen);
1588 int compat_sock_common_setsockopt(struct socket *sock, int level,
1589 int optname, char __user *optval, unsigned int optlen);
1591 void sk_common_release(struct sock *sk);
1594 * Default socket callbacks and setup code
1597 /* Initialise core socket variables */
1598 void sock_init_data(struct socket *sock, struct sock *sk);
1600 void sk_filter_release_rcu(struct rcu_head *rcu);
1603 * sk_filter_release - release a socket filter
1604 * @fp: filter to remove
1606 * Remove a filter from a socket and release its resources.
1609 static inline void sk_filter_release(struct sk_filter *fp)
1611 if (atomic_dec_and_test(&fp->refcnt))
1612 call_rcu(&fp->rcu, sk_filter_release_rcu);
1615 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1617 atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1618 sk_filter_release(fp);
1621 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1623 atomic_inc(&fp->refcnt);
1624 atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1628 * Socket reference counting postulates.
1630 * * Each user of socket SHOULD hold a reference count.
1631 * * Each access point to socket (an hash table bucket, reference from a list,
1632 * running timer, skb in flight MUST hold a reference count.
1633 * * When reference count hits 0, it means it will never increase back.
1634 * * When reference count hits 0, it means that no references from
1635 * outside exist to this socket and current process on current CPU
1636 * is last user and may/should destroy this socket.
1637 * * sk_free is called from any context: process, BH, IRQ. When
1638 * it is called, socket has no references from outside -> sk_free
1639 * may release descendant resources allocated by the socket, but
1640 * to the time when it is called, socket is NOT referenced by any
1641 * hash tables, lists etc.
1642 * * Packets, delivered from outside (from network or from another process)
1643 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1644 * when they sit in queue. Otherwise, packets will leak to hole, when
1645 * socket is looked up by one cpu and unhasing is made by another CPU.
1646 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1647 * (leak to backlog). Packet socket does all the processing inside
1648 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1649 * use separate SMP lock, so that they are prone too.
1652 /* Ungrab socket and destroy it, if it was the last reference. */
1653 static inline void sock_put(struct sock *sk)
1655 if (atomic_dec_and_test(&sk->sk_refcnt))
1658 /* Generic version of sock_put(), dealing with all sockets
1659 * (TCP_TIMEWAIT, ESTABLISHED...)
1661 void sock_gen_put(struct sock *sk);
1663 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1665 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1667 sk->sk_tx_queue_mapping = tx_queue;
1670 static inline void sk_tx_queue_clear(struct sock *sk)
1672 sk->sk_tx_queue_mapping = -1;
1675 static inline int sk_tx_queue_get(const struct sock *sk)
1677 return sk ? sk->sk_tx_queue_mapping : -1;
1680 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1682 sk_tx_queue_clear(sk);
1683 sk->sk_socket = sock;
1686 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1688 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1689 return &rcu_dereference_raw(sk->sk_wq)->wait;
1691 /* Detach socket from process context.
1692 * Announce socket dead, detach it from wait queue and inode.
1693 * Note that parent inode held reference count on this struct sock,
1694 * we do not release it in this function, because protocol
1695 * probably wants some additional cleanups or even continuing
1696 * to work with this socket (TCP).
1698 static inline void sock_orphan(struct sock *sk)
1700 write_lock_bh(&sk->sk_callback_lock);
1701 sock_set_flag(sk, SOCK_DEAD);
1702 sk_set_socket(sk, NULL);
1704 write_unlock_bh(&sk->sk_callback_lock);
1707 static inline void sock_graft(struct sock *sk, struct socket *parent)
1709 write_lock_bh(&sk->sk_callback_lock);
1710 sk->sk_wq = parent->wq;
1712 sk_set_socket(sk, parent);
1713 security_sock_graft(sk, parent);
1714 write_unlock_bh(&sk->sk_callback_lock);
1717 kuid_t sock_i_uid(struct sock *sk);
1718 unsigned long sock_i_ino(struct sock *sk);
1720 static inline struct dst_entry *
1721 __sk_dst_get(struct sock *sk)
1723 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1724 lockdep_is_held(&sk->sk_lock.slock));
1727 static inline struct dst_entry *
1728 sk_dst_get(struct sock *sk)
1730 struct dst_entry *dst;
1733 dst = rcu_dereference(sk->sk_dst_cache);
1740 void sk_reset_txq(struct sock *sk);
1742 static inline void dst_negative_advice(struct sock *sk)
1744 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1746 if (dst && dst->ops->negative_advice) {
1747 ndst = dst->ops->negative_advice(dst);
1750 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1757 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1759 struct dst_entry *old_dst;
1761 sk_tx_queue_clear(sk);
1763 * This can be called while sk is owned by the caller only,
1764 * with no state that can be checked in a rcu_dereference_check() cond
1766 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1767 rcu_assign_pointer(sk->sk_dst_cache, dst);
1768 dst_release(old_dst);
1772 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1774 spin_lock(&sk->sk_dst_lock);
1775 __sk_dst_set(sk, dst);
1776 spin_unlock(&sk->sk_dst_lock);
1780 __sk_dst_reset(struct sock *sk)
1782 __sk_dst_set(sk, NULL);
1786 sk_dst_reset(struct sock *sk)
1788 spin_lock(&sk->sk_dst_lock);
1790 spin_unlock(&sk->sk_dst_lock);
1793 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1795 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1797 static inline bool sk_can_gso(const struct sock *sk)
1799 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1802 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1804 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1806 sk->sk_route_nocaps |= flags;
1807 sk->sk_route_caps &= ~flags;
1810 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1811 char __user *from, char *to,
1812 int copy, int offset)
1814 if (skb->ip_summed == CHECKSUM_NONE) {
1816 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1819 skb->csum = csum_block_add(skb->csum, csum, offset);
1820 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1821 if (!access_ok(VERIFY_READ, from, copy) ||
1822 __copy_from_user_nocache(to, from, copy))
1824 } else if (copy_from_user(to, from, copy))
1830 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1831 char __user *from, int copy)
1833 int err, offset = skb->len;
1835 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1838 __skb_trim(skb, offset);
1843 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1844 struct sk_buff *skb,
1850 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1856 skb->data_len += copy;
1857 skb->truesize += copy;
1858 sk->sk_wmem_queued += copy;
1859 sk_mem_charge(sk, copy);
1863 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1864 struct sk_buff *skb, struct page *page,
1867 if (skb->ip_summed == CHECKSUM_NONE) {
1869 __wsum csum = csum_and_copy_from_user(from,
1870 page_address(page) + off,
1874 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1875 } else if (copy_from_user(page_address(page) + off, from, copy))
1879 skb->data_len += copy;
1880 skb->truesize += copy;
1881 sk->sk_wmem_queued += copy;
1882 sk_mem_charge(sk, copy);
1887 * sk_wmem_alloc_get - returns write allocations
1890 * Returns sk_wmem_alloc minus initial offset of one
1892 static inline int sk_wmem_alloc_get(const struct sock *sk)
1894 return atomic_read(&sk->sk_wmem_alloc) - 1;
1898 * sk_rmem_alloc_get - returns read allocations
1901 * Returns sk_rmem_alloc
1903 static inline int sk_rmem_alloc_get(const struct sock *sk)
1905 return atomic_read(&sk->sk_rmem_alloc);
1909 * sk_has_allocations - check if allocations are outstanding
1912 * Returns true if socket has write or read allocations
1914 static inline bool sk_has_allocations(const struct sock *sk)
1916 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1920 * wq_has_sleeper - check if there are any waiting processes
1921 * @wq: struct socket_wq
1923 * Returns true if socket_wq has waiting processes
1925 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1926 * barrier call. They were added due to the race found within the tcp code.
1928 * Consider following tcp code paths:
1932 * sys_select receive packet
1934 * __add_wait_queue update tp->rcv_nxt
1936 * tp->rcv_nxt check sock_def_readable
1938 * schedule rcu_read_lock();
1939 * wq = rcu_dereference(sk->sk_wq);
1940 * if (wq && waitqueue_active(&wq->wait))
1941 * wake_up_interruptible(&wq->wait)
1945 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1946 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1947 * could then endup calling schedule and sleep forever if there are no more
1948 * data on the socket.
1951 static inline bool wq_has_sleeper(struct socket_wq *wq)
1953 /* We need to be sure we are in sync with the
1954 * add_wait_queue modifications to the wait queue.
1956 * This memory barrier is paired in the sock_poll_wait.
1959 return wq && waitqueue_active(&wq->wait);
1963 * sock_poll_wait - place memory barrier behind the poll_wait call.
1965 * @wait_address: socket wait queue
1968 * See the comments in the wq_has_sleeper function.
1970 static inline void sock_poll_wait(struct file *filp,
1971 wait_queue_head_t *wait_address, poll_table *p)
1973 if (!poll_does_not_wait(p) && wait_address) {
1974 poll_wait(filp, wait_address, p);
1975 /* We need to be sure we are in sync with the
1976 * socket flags modification.
1978 * This memory barrier is paired in the wq_has_sleeper.
1985 * Queue a received datagram if it will fit. Stream and sequenced
1986 * protocols can't normally use this as they need to fit buffers in
1987 * and play with them.
1989 * Inlined as it's very short and called for pretty much every
1990 * packet ever received.
1993 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1997 skb->destructor = sock_wfree;
1999 * We used to take a refcount on sk, but following operation
2000 * is enough to guarantee sk_free() wont free this sock until
2001 * all in-flight packets are completed
2003 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
2006 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2010 skb->destructor = sock_rfree;
2011 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2012 sk_mem_charge(sk, skb->truesize);
2015 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2016 unsigned long expires);
2018 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2020 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2022 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2025 * Recover an error report and clear atomically
2028 static inline int sock_error(struct sock *sk)
2031 if (likely(!sk->sk_err))
2033 err = xchg(&sk->sk_err, 0);
2037 static inline unsigned long sock_wspace(struct sock *sk)
2041 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2042 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2049 static inline void sk_wake_async(struct sock *sk, int how, int band)
2051 if (sock_flag(sk, SOCK_FASYNC))
2052 sock_wake_async(sk->sk_socket, how, band);
2055 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2056 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2057 * Note: for send buffers, TCP works better if we can build two skbs at
2060 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2062 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2063 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2065 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2067 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2068 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2069 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2073 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2076 * sk_page_frag - return an appropriate page_frag
2079 * If socket allocation mode allows current thread to sleep, it means its
2080 * safe to use the per task page_frag instead of the per socket one.
2082 static inline struct page_frag *sk_page_frag(struct sock *sk)
2084 if (sk->sk_allocation & __GFP_WAIT)
2085 return ¤t->task_frag;
2087 return &sk->sk_frag;
2090 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2093 * Default write policy as shown to user space via poll/select/SIGIO
2095 static inline bool sock_writeable(const struct sock *sk)
2097 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2100 static inline gfp_t gfp_any(void)
2102 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2105 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2107 return noblock ? 0 : sk->sk_rcvtimeo;
2110 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2112 return noblock ? 0 : sk->sk_sndtimeo;
2115 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2117 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2120 /* Alas, with timeout socket operations are not restartable.
2121 * Compare this to poll().
2123 static inline int sock_intr_errno(long timeo)
2125 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2128 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2129 struct sk_buff *skb);
2130 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2131 struct sk_buff *skb);
2134 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2136 ktime_t kt = skb->tstamp;
2137 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2140 * generate control messages if
2141 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2142 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2143 * - software time stamp available and wanted
2144 * (SOCK_TIMESTAMPING_SOFTWARE)
2145 * - hardware time stamps available and wanted
2146 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2147 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2149 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2150 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2151 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2152 (hwtstamps->hwtstamp.tv64 &&
2153 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2154 (hwtstamps->syststamp.tv64 &&
2155 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2156 __sock_recv_timestamp(msg, sk, skb);
2160 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2161 __sock_recv_wifi_status(msg, sk, skb);
2164 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2165 struct sk_buff *skb);
2167 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2168 struct sk_buff *skb)
2170 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2171 (1UL << SOCK_RCVTSTAMP) | \
2172 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2173 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2174 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2175 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2177 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2178 __sock_recv_ts_and_drops(msg, sk, skb);
2180 sk->sk_stamp = skb->tstamp;
2184 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2185 * @sk: socket sending this packet
2186 * @tx_flags: filled with instructions for time stamping
2188 * Currently only depends on SOCK_TIMESTAMPING* flags.
2190 void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2193 * sk_eat_skb - Release a skb if it is no longer needed
2194 * @sk: socket to eat this skb from
2195 * @skb: socket buffer to eat
2196 * @copied_early: flag indicating whether DMA operations copied this data early
2198 * This routine must be called with interrupts disabled or with the socket
2199 * locked so that the sk_buff queue operation is ok.
2201 #ifdef CONFIG_NET_DMA
2202 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2204 __skb_unlink(skb, &sk->sk_receive_queue);
2208 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2211 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2213 __skb_unlink(skb, &sk->sk_receive_queue);
2219 struct net *sock_net(const struct sock *sk)
2221 return read_pnet(&sk->sk_net);
2225 void sock_net_set(struct sock *sk, struct net *net)
2227 write_pnet(&sk->sk_net, net);
2231 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2232 * They should not hold a reference to a namespace in order to allow
2234 * Sockets after sk_change_net should be released using sk_release_kernel
2236 static inline void sk_change_net(struct sock *sk, struct net *net)
2238 put_net(sock_net(sk));
2239 sock_net_set(sk, hold_net(net));
2242 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2245 struct sock *sk = skb->sk;
2247 skb->destructor = NULL;
2254 void sock_enable_timestamp(struct sock *sk, int flag);
2255 int sock_get_timestamp(struct sock *, struct timeval __user *);
2256 int sock_get_timestampns(struct sock *, struct timespec __user *);
2257 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2261 * Enable debug/info messages
2263 extern int net_msg_warn;
2264 #define NETDEBUG(fmt, args...) \
2265 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2267 #define LIMIT_NETDEBUG(fmt, args...) \
2268 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2270 extern __u32 sysctl_wmem_max;
2271 extern __u32 sysctl_rmem_max;
2273 extern int sysctl_optmem_max;
2275 extern __u32 sysctl_wmem_default;
2276 extern __u32 sysctl_rmem_default;
2278 #endif /* _SOCK_H */