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/page_counter.h>
58 #include <linux/memcontrol.h>
59 #include <linux/static_key.h>
60 #include <linux/sched.h>
62 #include <linux/filter.h>
63 #include <linux/rculist_nulls.h>
64 #include <linux/poll.h>
66 #include <linux/atomic.h>
68 #include <net/checksum.h>
69 #include <net/tcp_states.h>
70 #include <linux/net_tstamp.h>
75 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
76 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
79 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
84 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
89 * This structure really needs to be cleaned up.
90 * Most of it is for TCP, and not used by any of
91 * the other protocols.
94 /* Define this to get the SOCK_DBG debugging facility. */
95 #define SOCK_DEBUGGING
97 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
98 printk(KERN_DEBUG msg); } while (0)
100 /* Validate arguments and do nothing */
101 static inline __printf(2, 3)
102 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
107 /* This is the per-socket lock. The spinlock provides a synchronization
108 * between user contexts and software interrupt processing, whereas the
109 * mini-semaphore synchronizes multiple users amongst themselves.
114 wait_queue_head_t wq;
116 * We express the mutex-alike socket_lock semantics
117 * to the lock validator by explicitly managing
118 * the slock as a lock variant (in addition to
121 #ifdef CONFIG_DEBUG_LOCK_ALLOC
122 struct lockdep_map dep_map;
130 typedef __u32 __bitwise __portpair;
131 typedef __u64 __bitwise __addrpair;
134 * struct sock_common - minimal network layer representation of sockets
135 * @skc_daddr: Foreign IPv4 addr
136 * @skc_rcv_saddr: Bound local IPv4 addr
137 * @skc_hash: hash value used with various protocol lookup tables
138 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
139 * @skc_dport: placeholder for inet_dport/tw_dport
140 * @skc_num: placeholder for inet_num/tw_num
141 * @skc_family: network address family
142 * @skc_state: Connection state
143 * @skc_reuse: %SO_REUSEADDR setting
144 * @skc_reuseport: %SO_REUSEPORT setting
145 * @skc_bound_dev_if: bound device index if != 0
146 * @skc_bind_node: bind hash linkage for various protocol lookup tables
147 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
148 * @skc_prot: protocol handlers inside a network family
149 * @skc_net: reference to the network namespace of this socket
150 * @skc_node: main hash linkage for various protocol lookup tables
151 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
152 * @skc_tx_queue_mapping: tx queue number for this connection
153 * @skc_flags: place holder for sk_flags
154 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
155 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
156 * @skc_incoming_cpu: record/match cpu processing incoming packets
157 * @skc_refcnt: reference count
159 * This is the minimal network layer representation of sockets, the header
160 * for struct sock and struct inet_timewait_sock.
163 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
164 * address on 64bit arches : cf INET_MATCH()
167 __addrpair skc_addrpair;
170 __be32 skc_rcv_saddr;
174 unsigned int skc_hash;
175 __u16 skc_u16hashes[2];
177 /* skc_dport && skc_num must be grouped as well */
179 __portpair skc_portpair;
186 unsigned short skc_family;
187 volatile unsigned char skc_state;
188 unsigned char skc_reuse:4;
189 unsigned char skc_reuseport:1;
190 unsigned char skc_ipv6only:1;
191 unsigned char skc_net_refcnt:1;
192 int skc_bound_dev_if;
194 struct hlist_node skc_bind_node;
195 struct hlist_nulls_node skc_portaddr_node;
197 struct proto *skc_prot;
198 possible_net_t skc_net;
200 #if IS_ENABLED(CONFIG_IPV6)
201 struct in6_addr skc_v6_daddr;
202 struct in6_addr skc_v6_rcv_saddr;
205 atomic64_t skc_cookie;
207 /* following fields are padding to force
208 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
209 * assuming IPV6 is enabled. We use this padding differently
210 * for different kind of 'sockets'
213 unsigned long skc_flags;
214 struct sock *skc_listener; /* request_sock */
215 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
218 * fields between dontcopy_begin/dontcopy_end
219 * are not copied in sock_copy()
222 int skc_dontcopy_begin[0];
225 struct hlist_node skc_node;
226 struct hlist_nulls_node skc_nulls_node;
228 int skc_tx_queue_mapping;
230 int skc_incoming_cpu;
232 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
237 int skc_dontcopy_end[0];
240 u32 skc_window_clamp;
241 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
248 * struct sock - network layer representation of sockets
249 * @__sk_common: shared layout with inet_timewait_sock
250 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
251 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
252 * @sk_lock: synchronizer
253 * @sk_rcvbuf: size of receive buffer in bytes
254 * @sk_wq: sock wait queue and async head
255 * @sk_rx_dst: receive input route used by early demux
256 * @sk_dst_cache: destination cache
257 * @sk_dst_lock: destination cache lock
258 * @sk_policy: flow policy
259 * @sk_receive_queue: incoming packets
260 * @sk_wmem_alloc: transmit queue bytes committed
261 * @sk_write_queue: Packet sending queue
262 * @sk_omem_alloc: "o" is "option" or "other"
263 * @sk_wmem_queued: persistent queue size
264 * @sk_forward_alloc: space allocated forward
265 * @sk_napi_id: id of the last napi context to receive data for sk
266 * @sk_ll_usec: usecs to busypoll when there is no data
267 * @sk_allocation: allocation mode
268 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
269 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
270 * @sk_sndbuf: size of send buffer in bytes
271 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
272 * @sk_no_check_rx: allow zero checksum in RX packets
273 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
274 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
275 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
276 * @sk_gso_max_size: Maximum GSO segment size to build
277 * @sk_gso_max_segs: Maximum number of GSO segments
278 * @sk_lingertime: %SO_LINGER l_linger setting
279 * @sk_backlog: always used with the per-socket spinlock held
280 * @sk_callback_lock: used with the callbacks in the end of this struct
281 * @sk_error_queue: rarely used
282 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
283 * IPV6_ADDRFORM for instance)
284 * @sk_err: last error
285 * @sk_err_soft: errors that don't cause failure but are the cause of a
286 * persistent failure not just 'timed out'
287 * @sk_drops: raw/udp drops counter
288 * @sk_ack_backlog: current listen backlog
289 * @sk_max_ack_backlog: listen backlog set in listen()
290 * @sk_priority: %SO_PRIORITY setting
291 * @sk_cgrp_prioidx: socket group's priority map index
292 * @sk_type: socket type (%SOCK_STREAM, etc)
293 * @sk_protocol: which protocol this socket belongs in this network family
294 * @sk_peer_pid: &struct pid for this socket's peer
295 * @sk_peer_cred: %SO_PEERCRED setting
296 * @sk_rcvlowat: %SO_RCVLOWAT setting
297 * @sk_rcvtimeo: %SO_RCVTIMEO setting
298 * @sk_sndtimeo: %SO_SNDTIMEO setting
299 * @sk_txhash: computed flow hash for use on transmit
300 * @sk_filter: socket filtering instructions
301 * @sk_timer: sock cleanup timer
302 * @sk_stamp: time stamp of last packet received
303 * @sk_tsflags: SO_TIMESTAMPING socket options
304 * @sk_tskey: counter to disambiguate concurrent tstamp requests
305 * @sk_socket: Identd and reporting IO signals
306 * @sk_user_data: RPC layer private data
307 * @sk_frag: cached page frag
308 * @sk_peek_off: current peek_offset value
309 * @sk_send_head: front of stuff to transmit
310 * @sk_security: used by security modules
311 * @sk_mark: generic packet mark
312 * @sk_classid: this socket's cgroup classid
313 * @sk_cgrp: this socket's cgroup-specific proto data
314 * @sk_write_pending: a write to stream socket waits to start
315 * @sk_state_change: callback to indicate change in the state of the sock
316 * @sk_data_ready: callback to indicate there is data to be processed
317 * @sk_write_space: callback to indicate there is bf sending space available
318 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
319 * @sk_backlog_rcv: callback to process the backlog
320 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
324 * Now struct inet_timewait_sock also uses sock_common, so please just
325 * don't add nothing before this first member (__sk_common) --acme
327 struct sock_common __sk_common;
328 #define sk_node __sk_common.skc_node
329 #define sk_nulls_node __sk_common.skc_nulls_node
330 #define sk_refcnt __sk_common.skc_refcnt
331 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
333 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
334 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
335 #define sk_hash __sk_common.skc_hash
336 #define sk_portpair __sk_common.skc_portpair
337 #define sk_num __sk_common.skc_num
338 #define sk_dport __sk_common.skc_dport
339 #define sk_addrpair __sk_common.skc_addrpair
340 #define sk_daddr __sk_common.skc_daddr
341 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
342 #define sk_family __sk_common.skc_family
343 #define sk_state __sk_common.skc_state
344 #define sk_reuse __sk_common.skc_reuse
345 #define sk_reuseport __sk_common.skc_reuseport
346 #define sk_ipv6only __sk_common.skc_ipv6only
347 #define sk_net_refcnt __sk_common.skc_net_refcnt
348 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
349 #define sk_bind_node __sk_common.skc_bind_node
350 #define sk_prot __sk_common.skc_prot
351 #define sk_net __sk_common.skc_net
352 #define sk_v6_daddr __sk_common.skc_v6_daddr
353 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
354 #define sk_cookie __sk_common.skc_cookie
355 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
356 #define sk_flags __sk_common.skc_flags
357 #define sk_rxhash __sk_common.skc_rxhash
359 socket_lock_t sk_lock;
360 struct sk_buff_head sk_receive_queue;
362 * The backlog queue is special, it is always used with
363 * the per-socket spinlock held and requires low latency
364 * access. Therefore we special case it's implementation.
365 * Note : rmem_alloc is in this structure to fill a hole
366 * on 64bit arches, not because its logically part of
372 struct sk_buff *head;
373 struct sk_buff *tail;
375 #define sk_rmem_alloc sk_backlog.rmem_alloc
376 int sk_forward_alloc;
379 #ifdef CONFIG_NET_RX_BUSY_POLL
380 unsigned int sk_napi_id;
381 unsigned int sk_ll_usec;
386 struct sk_filter __rcu *sk_filter;
387 struct socket_wq __rcu *sk_wq;
390 struct xfrm_policy *sk_policy[2];
392 struct dst_entry *sk_rx_dst;
393 struct dst_entry __rcu *sk_dst_cache;
394 spinlock_t sk_dst_lock;
395 atomic_t sk_wmem_alloc;
396 atomic_t sk_omem_alloc;
398 struct sk_buff_head sk_write_queue;
399 kmemcheck_bitfield_begin(flags);
400 unsigned int sk_shutdown : 2,
406 kmemcheck_bitfield_end(flags);
409 u32 sk_pacing_rate; /* bytes per second */
410 u32 sk_max_pacing_rate;
411 netdev_features_t sk_route_caps;
412 netdev_features_t sk_route_nocaps;
414 unsigned int sk_gso_max_size;
417 unsigned long sk_lingertime;
418 struct sk_buff_head sk_error_queue;
419 struct proto *sk_prot_creator;
420 rwlock_t sk_callback_lock;
424 u32 sk_max_ack_backlog;
426 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
427 __u32 sk_cgrp_prioidx;
429 struct pid *sk_peer_pid;
430 const struct cred *sk_peer_cred;
433 struct timer_list sk_timer;
437 struct socket *sk_socket;
439 struct page_frag sk_frag;
440 struct sk_buff *sk_send_head;
442 int sk_write_pending;
443 #ifdef CONFIG_SECURITY
447 #ifdef CONFIG_CGROUP_NET_CLASSID
450 struct cg_proto *sk_cgrp;
451 void (*sk_state_change)(struct sock *sk);
452 void (*sk_data_ready)(struct sock *sk);
453 void (*sk_write_space)(struct sock *sk);
454 void (*sk_error_report)(struct sock *sk);
455 int (*sk_backlog_rcv)(struct sock *sk,
456 struct sk_buff *skb);
457 void (*sk_destruct)(struct sock *sk);
460 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
462 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
463 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
466 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
467 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
468 * on a socket means that the socket will reuse everybody else's port
469 * without looking at the other's sk_reuse value.
472 #define SK_NO_REUSE 0
473 #define SK_CAN_REUSE 1
474 #define SK_FORCE_REUSE 2
476 static inline int sk_peek_offset(struct sock *sk, int flags)
478 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
479 return sk->sk_peek_off;
484 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
486 if (sk->sk_peek_off >= 0) {
487 if (sk->sk_peek_off >= val)
488 sk->sk_peek_off -= val;
494 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
496 if (sk->sk_peek_off >= 0)
497 sk->sk_peek_off += val;
501 * Hashed lists helper routines
503 static inline struct sock *sk_entry(const struct hlist_node *node)
505 return hlist_entry(node, struct sock, sk_node);
508 static inline struct sock *__sk_head(const struct hlist_head *head)
510 return hlist_entry(head->first, struct sock, sk_node);
513 static inline struct sock *sk_head(const struct hlist_head *head)
515 return hlist_empty(head) ? NULL : __sk_head(head);
518 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
520 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
523 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
525 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
528 static inline struct sock *sk_next(const struct sock *sk)
530 return sk->sk_node.next ?
531 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
534 static inline struct sock *sk_nulls_next(const struct sock *sk)
536 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
537 hlist_nulls_entry(sk->sk_nulls_node.next,
538 struct sock, sk_nulls_node) :
542 static inline bool sk_unhashed(const struct sock *sk)
544 return hlist_unhashed(&sk->sk_node);
547 static inline bool sk_hashed(const struct sock *sk)
549 return !sk_unhashed(sk);
552 static inline void sk_node_init(struct hlist_node *node)
557 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
562 static inline void __sk_del_node(struct sock *sk)
564 __hlist_del(&sk->sk_node);
567 /* NB: equivalent to hlist_del_init_rcu */
568 static inline bool __sk_del_node_init(struct sock *sk)
572 sk_node_init(&sk->sk_node);
578 /* Grab socket reference count. This operation is valid only
579 when sk is ALREADY grabbed f.e. it is found in hash table
580 or a list and the lookup is made under lock preventing hash table
584 static inline void sock_hold(struct sock *sk)
586 atomic_inc(&sk->sk_refcnt);
589 /* Ungrab socket in the context, which assumes that socket refcnt
590 cannot hit zero, f.e. it is true in context of any socketcall.
592 static inline void __sock_put(struct sock *sk)
594 atomic_dec(&sk->sk_refcnt);
597 static inline bool sk_del_node_init(struct sock *sk)
599 bool rc = __sk_del_node_init(sk);
602 /* paranoid for a while -acme */
603 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
608 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
610 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
613 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
619 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
621 bool rc = __sk_nulls_del_node_init_rcu(sk);
624 /* paranoid for a while -acme */
625 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
631 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
633 hlist_add_head(&sk->sk_node, list);
636 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
639 __sk_add_node(sk, list);
642 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
645 hlist_add_head_rcu(&sk->sk_node, list);
648 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
650 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
653 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
656 __sk_nulls_add_node_rcu(sk, list);
659 static inline void __sk_del_bind_node(struct sock *sk)
661 __hlist_del(&sk->sk_bind_node);
664 static inline void sk_add_bind_node(struct sock *sk,
665 struct hlist_head *list)
667 hlist_add_head(&sk->sk_bind_node, list);
670 #define sk_for_each(__sk, list) \
671 hlist_for_each_entry(__sk, list, sk_node)
672 #define sk_for_each_rcu(__sk, list) \
673 hlist_for_each_entry_rcu(__sk, list, sk_node)
674 #define sk_nulls_for_each(__sk, node, list) \
675 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
676 #define sk_nulls_for_each_rcu(__sk, node, list) \
677 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
678 #define sk_for_each_from(__sk) \
679 hlist_for_each_entry_from(__sk, sk_node)
680 #define sk_nulls_for_each_from(__sk, node) \
681 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
682 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
683 #define sk_for_each_safe(__sk, tmp, list) \
684 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
685 #define sk_for_each_bound(__sk, list) \
686 hlist_for_each_entry(__sk, list, sk_bind_node)
689 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
690 * @tpos: the type * to use as a loop cursor.
691 * @pos: the &struct hlist_node to use as a loop cursor.
692 * @head: the head for your list.
693 * @offset: offset of hlist_node within the struct.
696 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
697 for (pos = (head)->first; \
698 (!is_a_nulls(pos)) && \
699 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
702 static inline struct user_namespace *sk_user_ns(struct sock *sk)
704 /* Careful only use this in a context where these parameters
705 * can not change and must all be valid, such as recvmsg from
708 return sk->sk_socket->file->f_cred->user_ns;
722 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
723 SOCK_DBG, /* %SO_DEBUG setting */
724 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
725 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
726 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
727 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
728 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
729 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
730 SOCK_FASYNC, /* fasync() active */
732 SOCK_ZEROCOPY, /* buffers from userspace */
733 SOCK_WIFI_STATUS, /* push wifi status to userspace */
734 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
735 * Will use last 4 bytes of packet sent from
736 * user-space instead.
738 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
739 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
742 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
744 nsk->sk_flags = osk->sk_flags;
747 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
749 __set_bit(flag, &sk->sk_flags);
752 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
754 __clear_bit(flag, &sk->sk_flags);
757 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
759 return test_bit(flag, &sk->sk_flags);
763 extern struct static_key memalloc_socks;
764 static inline int sk_memalloc_socks(void)
766 return static_key_false(&memalloc_socks);
770 static inline int sk_memalloc_socks(void)
777 static inline gfp_t sk_gfp_atomic(const struct sock *sk, gfp_t gfp_mask)
779 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
782 static inline void sk_acceptq_removed(struct sock *sk)
784 sk->sk_ack_backlog--;
787 static inline void sk_acceptq_added(struct sock *sk)
789 sk->sk_ack_backlog++;
792 static inline bool sk_acceptq_is_full(const struct sock *sk)
794 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
798 * Compute minimal free write space needed to queue new packets.
800 static inline int sk_stream_min_wspace(const struct sock *sk)
802 return sk->sk_wmem_queued >> 1;
805 static inline int sk_stream_wspace(const struct sock *sk)
807 return sk->sk_sndbuf - sk->sk_wmem_queued;
810 void sk_stream_write_space(struct sock *sk);
812 /* OOB backlog add */
813 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
815 /* dont let skb dst not refcounted, we are going to leave rcu lock */
818 if (!sk->sk_backlog.tail)
819 sk->sk_backlog.head = skb;
821 sk->sk_backlog.tail->next = skb;
823 sk->sk_backlog.tail = skb;
828 * Take into account size of receive queue and backlog queue
829 * Do not take into account this skb truesize,
830 * to allow even a single big packet to come.
832 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
834 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
836 return qsize > limit;
839 /* The per-socket spinlock must be held here. */
840 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
843 if (sk_rcvqueues_full(sk, limit))
847 * If the skb was allocated from pfmemalloc reserves, only
848 * allow SOCK_MEMALLOC sockets to use it as this socket is
849 * helping free memory
851 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
854 __sk_add_backlog(sk, skb);
855 sk->sk_backlog.len += skb->truesize;
859 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
861 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
863 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
864 return __sk_backlog_rcv(sk, skb);
866 return sk->sk_backlog_rcv(sk, skb);
869 static inline void sk_incoming_cpu_update(struct sock *sk)
871 sk->sk_incoming_cpu = raw_smp_processor_id();
874 static inline void sock_rps_record_flow_hash(__u32 hash)
877 struct rps_sock_flow_table *sock_flow_table;
880 sock_flow_table = rcu_dereference(rps_sock_flow_table);
881 rps_record_sock_flow(sock_flow_table, hash);
886 static inline void sock_rps_record_flow(const struct sock *sk)
889 sock_rps_record_flow_hash(sk->sk_rxhash);
893 static inline void sock_rps_save_rxhash(struct sock *sk,
894 const struct sk_buff *skb)
897 if (unlikely(sk->sk_rxhash != skb->hash))
898 sk->sk_rxhash = skb->hash;
902 static inline void sock_rps_reset_rxhash(struct sock *sk)
909 #define sk_wait_event(__sk, __timeo, __condition) \
911 release_sock(__sk); \
912 __rc = __condition; \
914 *(__timeo) = schedule_timeout(*(__timeo)); \
916 sched_annotate_sleep(); \
918 __rc = __condition; \
922 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
923 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
924 void sk_stream_wait_close(struct sock *sk, long timeo_p);
925 int sk_stream_error(struct sock *sk, int flags, int err);
926 void sk_stream_kill_queues(struct sock *sk);
927 void sk_set_memalloc(struct sock *sk);
928 void sk_clear_memalloc(struct sock *sk);
930 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
932 struct request_sock_ops;
933 struct timewait_sock_ops;
934 struct inet_hashinfo;
939 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
940 * un-modified. Special care is taken when initializing object to zero.
942 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
944 if (offsetof(struct sock, sk_node.next) != 0)
945 memset(sk, 0, offsetof(struct sock, sk_node.next));
946 memset(&sk->sk_node.pprev, 0,
947 size - offsetof(struct sock, sk_node.pprev));
950 /* Networking protocol blocks we attach to sockets.
951 * socket layer -> transport layer interface
954 void (*close)(struct sock *sk,
956 int (*connect)(struct sock *sk,
957 struct sockaddr *uaddr,
959 int (*disconnect)(struct sock *sk, int flags);
961 struct sock * (*accept)(struct sock *sk, int flags, int *err);
963 int (*ioctl)(struct sock *sk, int cmd,
965 int (*init)(struct sock *sk);
966 void (*destroy)(struct sock *sk);
967 void (*shutdown)(struct sock *sk, int how);
968 int (*setsockopt)(struct sock *sk, int level,
969 int optname, char __user *optval,
970 unsigned int optlen);
971 int (*getsockopt)(struct sock *sk, int level,
972 int optname, char __user *optval,
975 int (*compat_setsockopt)(struct sock *sk,
977 int optname, char __user *optval,
978 unsigned int optlen);
979 int (*compat_getsockopt)(struct sock *sk,
981 int optname, char __user *optval,
983 int (*compat_ioctl)(struct sock *sk,
984 unsigned int cmd, unsigned long arg);
986 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
988 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
989 size_t len, int noblock, int flags,
991 int (*sendpage)(struct sock *sk, struct page *page,
992 int offset, size_t size, int flags);
993 int (*bind)(struct sock *sk,
994 struct sockaddr *uaddr, int addr_len);
996 int (*backlog_rcv) (struct sock *sk,
997 struct sk_buff *skb);
999 void (*release_cb)(struct sock *sk);
1001 /* Keeping track of sk's, looking them up, and port selection methods. */
1002 void (*hash)(struct sock *sk);
1003 void (*unhash)(struct sock *sk);
1004 void (*rehash)(struct sock *sk);
1005 int (*get_port)(struct sock *sk, unsigned short snum);
1006 void (*clear_sk)(struct sock *sk, int size);
1008 /* Keeping track of sockets in use */
1009 #ifdef CONFIG_PROC_FS
1010 unsigned int inuse_idx;
1013 bool (*stream_memory_free)(const struct sock *sk);
1014 /* Memory pressure */
1015 void (*enter_memory_pressure)(struct sock *sk);
1016 atomic_long_t *memory_allocated; /* Current allocated memory. */
1017 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1019 * Pressure flag: try to collapse.
1020 * Technical note: it is used by multiple contexts non atomically.
1021 * All the __sk_mem_schedule() is of this nature: accounting
1022 * is strict, actions are advisory and have some latency.
1024 int *memory_pressure;
1031 struct kmem_cache *slab;
1032 unsigned int obj_size;
1035 struct percpu_counter *orphan_count;
1037 struct request_sock_ops *rsk_prot;
1038 struct timewait_sock_ops *twsk_prot;
1041 struct inet_hashinfo *hashinfo;
1042 struct udp_table *udp_table;
1043 struct raw_hashinfo *raw_hash;
1046 struct module *owner;
1050 struct list_head node;
1051 #ifdef SOCK_REFCNT_DEBUG
1054 #ifdef CONFIG_MEMCG_KMEM
1056 * cgroup specific init/deinit functions. Called once for all
1057 * protocols that implement it, from cgroups populate function.
1058 * This function has to setup any files the protocol want to
1059 * appear in the kmem cgroup filesystem.
1061 int (*init_cgroup)(struct mem_cgroup *memcg,
1062 struct cgroup_subsys *ss);
1063 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1064 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1068 int proto_register(struct proto *prot, int alloc_slab);
1069 void proto_unregister(struct proto *prot);
1071 #ifdef SOCK_REFCNT_DEBUG
1072 static inline void sk_refcnt_debug_inc(struct sock *sk)
1074 atomic_inc(&sk->sk_prot->socks);
1077 static inline void sk_refcnt_debug_dec(struct sock *sk)
1079 atomic_dec(&sk->sk_prot->socks);
1080 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1081 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1084 static inline void sk_refcnt_debug_release(const struct sock *sk)
1086 if (atomic_read(&sk->sk_refcnt) != 1)
1087 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1088 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1090 #else /* SOCK_REFCNT_DEBUG */
1091 #define sk_refcnt_debug_inc(sk) do { } while (0)
1092 #define sk_refcnt_debug_dec(sk) do { } while (0)
1093 #define sk_refcnt_debug_release(sk) do { } while (0)
1094 #endif /* SOCK_REFCNT_DEBUG */
1096 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1097 extern struct static_key memcg_socket_limit_enabled;
1098 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1099 struct cg_proto *cg_proto)
1101 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1103 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1105 #define mem_cgroup_sockets_enabled 0
1106 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1107 struct cg_proto *cg_proto)
1113 static inline bool sk_stream_memory_free(const struct sock *sk)
1115 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1118 return sk->sk_prot->stream_memory_free ?
1119 sk->sk_prot->stream_memory_free(sk) : true;
1122 static inline bool sk_stream_is_writeable(const struct sock *sk)
1124 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1125 sk_stream_memory_free(sk);
1129 static inline bool sk_has_memory_pressure(const struct sock *sk)
1131 return sk->sk_prot->memory_pressure != NULL;
1134 static inline bool sk_under_memory_pressure(const struct sock *sk)
1136 if (!sk->sk_prot->memory_pressure)
1139 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1140 return !!sk->sk_cgrp->memory_pressure;
1142 return !!*sk->sk_prot->memory_pressure;
1145 static inline void sk_leave_memory_pressure(struct sock *sk)
1147 int *memory_pressure = sk->sk_prot->memory_pressure;
1149 if (!memory_pressure)
1152 if (*memory_pressure)
1153 *memory_pressure = 0;
1155 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1156 struct cg_proto *cg_proto = sk->sk_cgrp;
1157 struct proto *prot = sk->sk_prot;
1159 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1160 cg_proto->memory_pressure = 0;
1165 static inline void sk_enter_memory_pressure(struct sock *sk)
1167 if (!sk->sk_prot->enter_memory_pressure)
1170 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1171 struct cg_proto *cg_proto = sk->sk_cgrp;
1172 struct proto *prot = sk->sk_prot;
1174 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1175 cg_proto->memory_pressure = 1;
1178 sk->sk_prot->enter_memory_pressure(sk);
1181 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1183 long *prot = sk->sk_prot->sysctl_mem;
1184 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1185 prot = sk->sk_cgrp->sysctl_mem;
1189 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1193 page_counter_charge(&prot->memory_allocated, amt);
1195 if (page_counter_read(&prot->memory_allocated) >
1196 prot->memory_allocated.limit)
1197 *parent_status = OVER_LIMIT;
1200 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1203 page_counter_uncharge(&prot->memory_allocated, amt);
1207 sk_memory_allocated(const struct sock *sk)
1209 struct proto *prot = sk->sk_prot;
1211 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1212 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1214 return atomic_long_read(prot->memory_allocated);
1218 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1220 struct proto *prot = sk->sk_prot;
1222 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1223 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1224 /* update the root cgroup regardless */
1225 atomic_long_add_return(amt, prot->memory_allocated);
1226 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1229 return atomic_long_add_return(amt, prot->memory_allocated);
1233 sk_memory_allocated_sub(struct sock *sk, int amt)
1235 struct proto *prot = sk->sk_prot;
1237 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1238 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1240 atomic_long_sub(amt, prot->memory_allocated);
1243 static inline void sk_sockets_allocated_dec(struct sock *sk)
1245 struct proto *prot = sk->sk_prot;
1247 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1248 struct cg_proto *cg_proto = sk->sk_cgrp;
1250 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1251 percpu_counter_dec(&cg_proto->sockets_allocated);
1254 percpu_counter_dec(prot->sockets_allocated);
1257 static inline void sk_sockets_allocated_inc(struct sock *sk)
1259 struct proto *prot = sk->sk_prot;
1261 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1262 struct cg_proto *cg_proto = sk->sk_cgrp;
1264 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1265 percpu_counter_inc(&cg_proto->sockets_allocated);
1268 percpu_counter_inc(prot->sockets_allocated);
1272 sk_sockets_allocated_read_positive(struct sock *sk)
1274 struct proto *prot = sk->sk_prot;
1276 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1277 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1279 return percpu_counter_read_positive(prot->sockets_allocated);
1283 proto_sockets_allocated_sum_positive(struct proto *prot)
1285 return percpu_counter_sum_positive(prot->sockets_allocated);
1289 proto_memory_allocated(struct proto *prot)
1291 return atomic_long_read(prot->memory_allocated);
1295 proto_memory_pressure(struct proto *prot)
1297 if (!prot->memory_pressure)
1299 return !!*prot->memory_pressure;
1303 #ifdef CONFIG_PROC_FS
1304 /* Called with local bh disabled */
1305 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1306 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1308 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1315 /* With per-bucket locks this operation is not-atomic, so that
1316 * this version is not worse.
1318 static inline void __sk_prot_rehash(struct sock *sk)
1320 sk->sk_prot->unhash(sk);
1321 sk->sk_prot->hash(sk);
1324 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1326 /* About 10 seconds */
1327 #define SOCK_DESTROY_TIME (10*HZ)
1329 /* Sockets 0-1023 can't be bound to unless you are superuser */
1330 #define PROT_SOCK 1024
1332 #define SHUTDOWN_MASK 3
1333 #define RCV_SHUTDOWN 1
1334 #define SEND_SHUTDOWN 2
1336 #define SOCK_SNDBUF_LOCK 1
1337 #define SOCK_RCVBUF_LOCK 2
1338 #define SOCK_BINDADDR_LOCK 4
1339 #define SOCK_BINDPORT_LOCK 8
1341 struct socket_alloc {
1342 struct socket socket;
1343 struct inode vfs_inode;
1346 static inline struct socket *SOCKET_I(struct inode *inode)
1348 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1351 static inline struct inode *SOCK_INODE(struct socket *socket)
1353 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1357 * Functions for memory accounting
1359 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1360 void __sk_mem_reclaim(struct sock *sk, int amount);
1362 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1363 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1364 #define SK_MEM_SEND 0
1365 #define SK_MEM_RECV 1
1367 static inline int sk_mem_pages(int amt)
1369 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1372 static inline bool sk_has_account(struct sock *sk)
1374 /* return true if protocol supports memory accounting */
1375 return !!sk->sk_prot->memory_allocated;
1378 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1380 if (!sk_has_account(sk))
1382 return size <= sk->sk_forward_alloc ||
1383 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1387 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1389 if (!sk_has_account(sk))
1391 return size<= sk->sk_forward_alloc ||
1392 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1393 skb_pfmemalloc(skb);
1396 static inline void sk_mem_reclaim(struct sock *sk)
1398 if (!sk_has_account(sk))
1400 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1401 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1404 static inline void sk_mem_reclaim_partial(struct sock *sk)
1406 if (!sk_has_account(sk))
1408 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1409 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1412 static inline void sk_mem_charge(struct sock *sk, int size)
1414 if (!sk_has_account(sk))
1416 sk->sk_forward_alloc -= size;
1419 static inline void sk_mem_uncharge(struct sock *sk, int size)
1421 if (!sk_has_account(sk))
1423 sk->sk_forward_alloc += size;
1426 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1428 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1429 sk->sk_wmem_queued -= skb->truesize;
1430 sk_mem_uncharge(sk, skb->truesize);
1434 /* Used by processes to "lock" a socket state, so that
1435 * interrupts and bottom half handlers won't change it
1436 * from under us. It essentially blocks any incoming
1437 * packets, so that we won't get any new data or any
1438 * packets that change the state of the socket.
1440 * While locked, BH processing will add new packets to
1441 * the backlog queue. This queue is processed by the
1442 * owner of the socket lock right before it is released.
1444 * Since ~2.3.5 it is also exclusive sleep lock serializing
1445 * accesses from user process context.
1447 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1449 static inline void sock_release_ownership(struct sock *sk)
1451 sk->sk_lock.owned = 0;
1455 * Macro so as to not evaluate some arguments when
1456 * lockdep is not enabled.
1458 * Mark both the sk_lock and the sk_lock.slock as a
1459 * per-address-family lock class.
1461 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1463 sk->sk_lock.owned = 0; \
1464 init_waitqueue_head(&sk->sk_lock.wq); \
1465 spin_lock_init(&(sk)->sk_lock.slock); \
1466 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1467 sizeof((sk)->sk_lock)); \
1468 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1470 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1473 void lock_sock_nested(struct sock *sk, int subclass);
1475 static inline void lock_sock(struct sock *sk)
1477 lock_sock_nested(sk, 0);
1480 void release_sock(struct sock *sk);
1482 /* BH context may only use the following locking interface. */
1483 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1484 #define bh_lock_sock_nested(__sk) \
1485 spin_lock_nested(&((__sk)->sk_lock.slock), \
1486 SINGLE_DEPTH_NESTING)
1487 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1489 bool lock_sock_fast(struct sock *sk);
1491 * unlock_sock_fast - complement of lock_sock_fast
1495 * fast unlock socket for user context.
1496 * If slow mode is on, we call regular release_sock()
1498 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1503 spin_unlock_bh(&sk->sk_lock.slock);
1507 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1508 struct proto *prot, int kern);
1509 void sk_free(struct sock *sk);
1510 void sk_destruct(struct sock *sk);
1511 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1513 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1515 void sock_wfree(struct sk_buff *skb);
1516 void skb_orphan_partial(struct sk_buff *skb);
1517 void sock_rfree(struct sk_buff *skb);
1518 void sock_efree(struct sk_buff *skb);
1520 void sock_edemux(struct sk_buff *skb);
1522 #define sock_edemux(skb) sock_efree(skb)
1525 int sock_setsockopt(struct socket *sock, int level, int op,
1526 char __user *optval, unsigned int optlen);
1528 int sock_getsockopt(struct socket *sock, int level, int op,
1529 char __user *optval, int __user *optlen);
1530 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1531 int noblock, int *errcode);
1532 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1533 unsigned long data_len, int noblock,
1534 int *errcode, int max_page_order);
1535 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1536 void sock_kfree_s(struct sock *sk, void *mem, int size);
1537 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1538 void sk_send_sigurg(struct sock *sk);
1540 struct sockcm_cookie {
1544 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1545 struct sockcm_cookie *sockc);
1548 * Functions to fill in entries in struct proto_ops when a protocol
1549 * does not implement a particular function.
1551 int sock_no_bind(struct socket *, struct sockaddr *, int);
1552 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1553 int sock_no_socketpair(struct socket *, struct socket *);
1554 int sock_no_accept(struct socket *, struct socket *, int);
1555 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1556 unsigned int sock_no_poll(struct file *, struct socket *,
1557 struct poll_table_struct *);
1558 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1559 int sock_no_listen(struct socket *, int);
1560 int sock_no_shutdown(struct socket *, int);
1561 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1562 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1563 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1564 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1565 int sock_no_mmap(struct file *file, struct socket *sock,
1566 struct vm_area_struct *vma);
1567 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1568 size_t size, int flags);
1571 * Functions to fill in entries in struct proto_ops when a protocol
1572 * uses the inet style.
1574 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1575 char __user *optval, int __user *optlen);
1576 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1578 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1579 char __user *optval, unsigned int optlen);
1580 int compat_sock_common_getsockopt(struct socket *sock, int level,
1581 int optname, char __user *optval, int __user *optlen);
1582 int compat_sock_common_setsockopt(struct socket *sock, int level,
1583 int optname, char __user *optval, unsigned int optlen);
1585 void sk_common_release(struct sock *sk);
1588 * Default socket callbacks and setup code
1591 /* Initialise core socket variables */
1592 void sock_init_data(struct socket *sock, struct sock *sk);
1595 * Socket reference counting postulates.
1597 * * Each user of socket SHOULD hold a reference count.
1598 * * Each access point to socket (an hash table bucket, reference from a list,
1599 * running timer, skb in flight MUST hold a reference count.
1600 * * When reference count hits 0, it means it will never increase back.
1601 * * When reference count hits 0, it means that no references from
1602 * outside exist to this socket and current process on current CPU
1603 * is last user and may/should destroy this socket.
1604 * * sk_free is called from any context: process, BH, IRQ. When
1605 * it is called, socket has no references from outside -> sk_free
1606 * may release descendant resources allocated by the socket, but
1607 * to the time when it is called, socket is NOT referenced by any
1608 * hash tables, lists etc.
1609 * * Packets, delivered from outside (from network or from another process)
1610 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1611 * when they sit in queue. Otherwise, packets will leak to hole, when
1612 * socket is looked up by one cpu and unhasing is made by another CPU.
1613 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1614 * (leak to backlog). Packet socket does all the processing inside
1615 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1616 * use separate SMP lock, so that they are prone too.
1619 /* Ungrab socket and destroy it, if it was the last reference. */
1620 static inline void sock_put(struct sock *sk)
1622 if (atomic_dec_and_test(&sk->sk_refcnt))
1625 /* Generic version of sock_put(), dealing with all sockets
1626 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1628 void sock_gen_put(struct sock *sk);
1630 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1632 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1634 sk->sk_tx_queue_mapping = tx_queue;
1637 static inline void sk_tx_queue_clear(struct sock *sk)
1639 sk->sk_tx_queue_mapping = -1;
1642 static inline int sk_tx_queue_get(const struct sock *sk)
1644 return sk ? sk->sk_tx_queue_mapping : -1;
1647 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1649 sk_tx_queue_clear(sk);
1650 sk->sk_socket = sock;
1653 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1655 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1656 return &rcu_dereference_raw(sk->sk_wq)->wait;
1658 /* Detach socket from process context.
1659 * Announce socket dead, detach it from wait queue and inode.
1660 * Note that parent inode held reference count on this struct sock,
1661 * we do not release it in this function, because protocol
1662 * probably wants some additional cleanups or even continuing
1663 * to work with this socket (TCP).
1665 static inline void sock_orphan(struct sock *sk)
1667 write_lock_bh(&sk->sk_callback_lock);
1668 sock_set_flag(sk, SOCK_DEAD);
1669 sk_set_socket(sk, NULL);
1671 write_unlock_bh(&sk->sk_callback_lock);
1674 static inline void sock_graft(struct sock *sk, struct socket *parent)
1676 write_lock_bh(&sk->sk_callback_lock);
1677 sk->sk_wq = parent->wq;
1679 sk_set_socket(sk, parent);
1680 security_sock_graft(sk, parent);
1681 write_unlock_bh(&sk->sk_callback_lock);
1684 kuid_t sock_i_uid(struct sock *sk);
1685 unsigned long sock_i_ino(struct sock *sk);
1687 static inline u32 net_tx_rndhash(void)
1689 u32 v = prandom_u32();
1694 static inline void sk_set_txhash(struct sock *sk)
1696 sk->sk_txhash = net_tx_rndhash();
1699 static inline void sk_rethink_txhash(struct sock *sk)
1705 static inline struct dst_entry *
1706 __sk_dst_get(struct sock *sk)
1708 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1709 lockdep_is_held(&sk->sk_lock.slock));
1712 static inline struct dst_entry *
1713 sk_dst_get(struct sock *sk)
1715 struct dst_entry *dst;
1718 dst = rcu_dereference(sk->sk_dst_cache);
1719 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1725 static inline void dst_negative_advice(struct sock *sk)
1727 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1729 sk_rethink_txhash(sk);
1731 if (dst && dst->ops->negative_advice) {
1732 ndst = dst->ops->negative_advice(dst);
1735 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1736 sk_tx_queue_clear(sk);
1742 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1744 struct dst_entry *old_dst;
1746 sk_tx_queue_clear(sk);
1748 * This can be called while sk is owned by the caller only,
1749 * with no state that can be checked in a rcu_dereference_check() cond
1751 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1752 rcu_assign_pointer(sk->sk_dst_cache, dst);
1753 dst_release(old_dst);
1757 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1759 struct dst_entry *old_dst;
1761 sk_tx_queue_clear(sk);
1762 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1763 dst_release(old_dst);
1767 __sk_dst_reset(struct sock *sk)
1769 __sk_dst_set(sk, NULL);
1773 sk_dst_reset(struct sock *sk)
1775 sk_dst_set(sk, NULL);
1778 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1780 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1782 bool sk_mc_loop(struct sock *sk);
1784 static inline bool sk_can_gso(const struct sock *sk)
1786 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1789 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1791 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1793 sk->sk_route_nocaps |= flags;
1794 sk->sk_route_caps &= ~flags;
1797 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1798 struct iov_iter *from, char *to,
1799 int copy, int offset)
1801 if (skb->ip_summed == CHECKSUM_NONE) {
1803 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1805 skb->csum = csum_block_add(skb->csum, csum, offset);
1806 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1807 if (copy_from_iter_nocache(to, copy, from) != copy)
1809 } else if (copy_from_iter(to, copy, from) != copy)
1815 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1816 struct iov_iter *from, int copy)
1818 int err, offset = skb->len;
1820 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1823 __skb_trim(skb, offset);
1828 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1829 struct sk_buff *skb,
1835 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1841 skb->data_len += copy;
1842 skb->truesize += copy;
1843 sk->sk_wmem_queued += copy;
1844 sk_mem_charge(sk, copy);
1849 * sk_wmem_alloc_get - returns write allocations
1852 * Returns sk_wmem_alloc minus initial offset of one
1854 static inline int sk_wmem_alloc_get(const struct sock *sk)
1856 return atomic_read(&sk->sk_wmem_alloc) - 1;
1860 * sk_rmem_alloc_get - returns read allocations
1863 * Returns sk_rmem_alloc
1865 static inline int sk_rmem_alloc_get(const struct sock *sk)
1867 return atomic_read(&sk->sk_rmem_alloc);
1871 * sk_has_allocations - check if allocations are outstanding
1874 * Returns true if socket has write or read allocations
1876 static inline bool sk_has_allocations(const struct sock *sk)
1878 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1882 * wq_has_sleeper - check if there are any waiting processes
1883 * @wq: struct socket_wq
1885 * Returns true if socket_wq has waiting processes
1887 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1888 * barrier call. They were added due to the race found within the tcp code.
1890 * Consider following tcp code paths:
1894 * sys_select receive packet
1896 * __add_wait_queue update tp->rcv_nxt
1898 * tp->rcv_nxt check sock_def_readable
1900 * schedule rcu_read_lock();
1901 * wq = rcu_dereference(sk->sk_wq);
1902 * if (wq && waitqueue_active(&wq->wait))
1903 * wake_up_interruptible(&wq->wait)
1907 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1908 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1909 * could then endup calling schedule and sleep forever if there are no more
1910 * data on the socket.
1913 static inline bool wq_has_sleeper(struct socket_wq *wq)
1915 /* We need to be sure we are in sync with the
1916 * add_wait_queue modifications to the wait queue.
1918 * This memory barrier is paired in the sock_poll_wait.
1921 return wq && waitqueue_active(&wq->wait);
1925 * sock_poll_wait - place memory barrier behind the poll_wait call.
1927 * @wait_address: socket wait queue
1930 * See the comments in the wq_has_sleeper function.
1932 static inline void sock_poll_wait(struct file *filp,
1933 wait_queue_head_t *wait_address, poll_table *p)
1935 if (!poll_does_not_wait(p) && wait_address) {
1936 poll_wait(filp, wait_address, p);
1937 /* We need to be sure we are in sync with the
1938 * socket flags modification.
1940 * This memory barrier is paired in the wq_has_sleeper.
1946 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1948 if (sk->sk_txhash) {
1950 skb->hash = sk->sk_txhash;
1954 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
1957 * Queue a received datagram if it will fit. Stream and sequenced
1958 * protocols can't normally use this as they need to fit buffers in
1959 * and play with them.
1961 * Inlined as it's very short and called for pretty much every
1962 * packet ever received.
1964 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1968 skb->destructor = sock_rfree;
1969 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1970 sk_mem_charge(sk, skb->truesize);
1973 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1974 unsigned long expires);
1976 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1978 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1980 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1981 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1984 * Recover an error report and clear atomically
1987 static inline int sock_error(struct sock *sk)
1990 if (likely(!sk->sk_err))
1992 err = xchg(&sk->sk_err, 0);
1996 static inline unsigned long sock_wspace(struct sock *sk)
2000 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2001 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2008 static inline void sk_wake_async(struct sock *sk, int how, int band)
2010 if (sock_flag(sk, SOCK_FASYNC))
2011 sock_wake_async(sk->sk_socket, how, band);
2014 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2015 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2016 * Note: for send buffers, TCP works better if we can build two skbs at
2019 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2021 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2022 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2024 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2026 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2027 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2028 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2032 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2033 bool force_schedule);
2036 * sk_page_frag - return an appropriate page_frag
2039 * If socket allocation mode allows current thread to sleep, it means its
2040 * safe to use the per task page_frag instead of the per socket one.
2042 static inline struct page_frag *sk_page_frag(struct sock *sk)
2044 if (gfpflags_allow_blocking(sk->sk_allocation))
2045 return ¤t->task_frag;
2047 return &sk->sk_frag;
2050 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2053 * Default write policy as shown to user space via poll/select/SIGIO
2055 static inline bool sock_writeable(const struct sock *sk)
2057 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2060 static inline gfp_t gfp_any(void)
2062 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2065 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2067 return noblock ? 0 : sk->sk_rcvtimeo;
2070 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2072 return noblock ? 0 : sk->sk_sndtimeo;
2075 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2077 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2080 /* Alas, with timeout socket operations are not restartable.
2081 * Compare this to poll().
2083 static inline int sock_intr_errno(long timeo)
2085 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2088 struct sock_skb_cb {
2092 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2093 * using skb->cb[] would keep using it directly and utilize its
2094 * alignement guarantee.
2096 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2097 sizeof(struct sock_skb_cb)))
2099 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2100 SOCK_SKB_CB_OFFSET))
2102 #define sock_skb_cb_check_size(size) \
2103 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2106 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2108 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2111 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2112 struct sk_buff *skb);
2113 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2114 struct sk_buff *skb);
2117 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2119 ktime_t kt = skb->tstamp;
2120 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2123 * generate control messages if
2124 * - receive time stamping in software requested
2125 * - software time stamp available and wanted
2126 * - hardware time stamps available and wanted
2128 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2129 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2130 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2131 (hwtstamps->hwtstamp.tv64 &&
2132 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2133 __sock_recv_timestamp(msg, sk, skb);
2137 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2138 __sock_recv_wifi_status(msg, sk, skb);
2141 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2142 struct sk_buff *skb);
2144 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2145 struct sk_buff *skb)
2147 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2148 (1UL << SOCK_RCVTSTAMP))
2149 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2150 SOF_TIMESTAMPING_RAW_HARDWARE)
2152 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2153 __sock_recv_ts_and_drops(msg, sk, skb);
2155 sk->sk_stamp = skb->tstamp;
2158 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2161 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2162 * @sk: socket sending this packet
2163 * @tx_flags: completed with instructions for time stamping
2165 * Note : callers should take care of initial *tx_flags value (usually 0)
2167 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2169 if (unlikely(sk->sk_tsflags))
2170 __sock_tx_timestamp(sk, tx_flags);
2171 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2172 *tx_flags |= SKBTX_WIFI_STATUS;
2176 * sk_eat_skb - Release a skb if it is no longer needed
2177 * @sk: socket to eat this skb from
2178 * @skb: socket buffer to eat
2180 * This routine must be called with interrupts disabled or with the socket
2181 * locked so that the sk_buff queue operation is ok.
2183 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2185 __skb_unlink(skb, &sk->sk_receive_queue);
2190 struct net *sock_net(const struct sock *sk)
2192 return read_pnet(&sk->sk_net);
2196 void sock_net_set(struct sock *sk, struct net *net)
2198 write_pnet(&sk->sk_net, net);
2201 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2204 struct sock *sk = skb->sk;
2206 skb->destructor = NULL;
2213 /* This helper checks if a socket is a full socket,
2214 * ie _not_ a timewait or request socket.
2216 static inline bool sk_fullsock(const struct sock *sk)
2218 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2221 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2222 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2224 static inline bool sk_listener(const struct sock *sk)
2226 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2230 * sk_state_load - read sk->sk_state for lockless contexts
2231 * @sk: socket pointer
2233 * Paired with sk_state_store(). Used in places we do not hold socket lock :
2234 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2236 static inline int sk_state_load(const struct sock *sk)
2238 return smp_load_acquire(&sk->sk_state);
2242 * sk_state_store - update sk->sk_state
2243 * @sk: socket pointer
2244 * @newstate: new state
2246 * Paired with sk_state_load(). Should be used in contexts where
2247 * state change might impact lockless readers.
2249 static inline void sk_state_store(struct sock *sk, int newstate)
2251 smp_store_release(&sk->sk_state, newstate);
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,
2260 bool sk_ns_capable(const struct sock *sk,
2261 struct user_namespace *user_ns, int cap);
2262 bool sk_capable(const struct sock *sk, int cap);
2263 bool sk_net_capable(const struct sock *sk, int cap);
2265 extern __u32 sysctl_wmem_max;
2266 extern __u32 sysctl_rmem_max;
2268 extern int sysctl_tstamp_allow_data;
2269 extern int sysctl_optmem_max;
2271 extern __u32 sysctl_wmem_default;
2272 extern __u32 sysctl_rmem_default;
2274 #endif /* _SOCK_H */