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
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_refcnt: reference count
155 * This is the minimal network layer representation of sockets, the header
156 * for struct sock and struct inet_timewait_sock.
159 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
160 * address on 64bit arches : cf INET_MATCH()
163 __addrpair skc_addrpair;
166 __be32 skc_rcv_saddr;
170 unsigned int skc_hash;
171 __u16 skc_u16hashes[2];
173 /* skc_dport && skc_num must be grouped as well */
175 __portpair skc_portpair;
182 unsigned short skc_family;
183 volatile unsigned char skc_state;
184 unsigned char skc_reuse:4;
185 unsigned char skc_reuseport:1;
186 unsigned char skc_ipv6only:1;
187 int skc_bound_dev_if;
189 struct hlist_node skc_bind_node;
190 struct hlist_nulls_node skc_portaddr_node;
192 struct proto *skc_prot;
193 possible_net_t skc_net;
195 #if IS_ENABLED(CONFIG_IPV6)
196 struct in6_addr skc_v6_daddr;
197 struct in6_addr skc_v6_rcv_saddr;
200 atomic64_t skc_cookie;
203 * fields between dontcopy_begin/dontcopy_end
204 * are not copied in sock_copy()
207 int skc_dontcopy_begin[0];
210 struct hlist_node skc_node;
211 struct hlist_nulls_node skc_nulls_node;
213 int skc_tx_queue_mapping;
216 int skc_dontcopy_end[0];
222 * struct sock - network layer representation of sockets
223 * @__sk_common: shared layout with inet_timewait_sock
224 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
225 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
226 * @sk_lock: synchronizer
227 * @sk_rcvbuf: size of receive buffer in bytes
228 * @sk_wq: sock wait queue and async head
229 * @sk_rx_dst: receive input route used by early demux
230 * @sk_dst_cache: destination cache
231 * @sk_dst_lock: destination cache lock
232 * @sk_policy: flow policy
233 * @sk_receive_queue: incoming packets
234 * @sk_wmem_alloc: transmit queue bytes committed
235 * @sk_write_queue: Packet sending queue
236 * @sk_omem_alloc: "o" is "option" or "other"
237 * @sk_wmem_queued: persistent queue size
238 * @sk_forward_alloc: space allocated forward
239 * @sk_napi_id: id of the last napi context to receive data for sk
240 * @sk_ll_usec: usecs to busypoll when there is no data
241 * @sk_allocation: allocation mode
242 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
243 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
244 * @sk_sndbuf: size of send buffer in bytes
245 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
246 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
247 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
248 * @sk_no_check_rx: allow zero checksum in RX packets
249 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
250 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
251 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
252 * @sk_gso_max_size: Maximum GSO segment size to build
253 * @sk_gso_max_segs: Maximum number of GSO segments
254 * @sk_lingertime: %SO_LINGER l_linger setting
255 * @sk_backlog: always used with the per-socket spinlock held
256 * @sk_callback_lock: used with the callbacks in the end of this struct
257 * @sk_error_queue: rarely used
258 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
259 * IPV6_ADDRFORM for instance)
260 * @sk_err: last error
261 * @sk_err_soft: errors that don't cause failure but are the cause of a
262 * persistent failure not just 'timed out'
263 * @sk_drops: raw/udp drops counter
264 * @sk_ack_backlog: current listen backlog
265 * @sk_max_ack_backlog: listen backlog set in listen()
266 * @sk_priority: %SO_PRIORITY setting
267 * @sk_cgrp_prioidx: socket group's priority map index
268 * @sk_type: socket type (%SOCK_STREAM, etc)
269 * @sk_protocol: which protocol this socket belongs in this network family
270 * @sk_peer_pid: &struct pid for this socket's peer
271 * @sk_peer_cred: %SO_PEERCRED setting
272 * @sk_rcvlowat: %SO_RCVLOWAT setting
273 * @sk_rcvtimeo: %SO_RCVTIMEO setting
274 * @sk_sndtimeo: %SO_SNDTIMEO setting
275 * @sk_rxhash: flow hash received from netif layer
276 * @sk_incoming_cpu: record cpu processing incoming packets
277 * @sk_txhash: computed flow hash for use on transmit
278 * @sk_filter: socket filtering instructions
279 * @sk_protinfo: private area, net family specific, when not using slab
280 * @sk_timer: sock cleanup timer
281 * @sk_stamp: time stamp of last packet received
282 * @sk_tsflags: SO_TIMESTAMPING socket options
283 * @sk_tskey: counter to disambiguate concurrent tstamp requests
284 * @sk_socket: Identd and reporting IO signals
285 * @sk_user_data: RPC layer private data
286 * @sk_frag: cached page frag
287 * @sk_peek_off: current peek_offset value
288 * @sk_send_head: front of stuff to transmit
289 * @sk_security: used by security modules
290 * @sk_mark: generic packet mark
291 * @sk_classid: this socket's cgroup classid
292 * @sk_cgrp: this socket's cgroup-specific proto data
293 * @sk_write_pending: a write to stream socket waits to start
294 * @sk_state_change: callback to indicate change in the state of the sock
295 * @sk_data_ready: callback to indicate there is data to be processed
296 * @sk_write_space: callback to indicate there is bf sending space available
297 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
298 * @sk_backlog_rcv: callback to process the backlog
299 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
303 * Now struct inet_timewait_sock also uses sock_common, so please just
304 * don't add nothing before this first member (__sk_common) --acme
306 struct sock_common __sk_common;
307 #define sk_node __sk_common.skc_node
308 #define sk_nulls_node __sk_common.skc_nulls_node
309 #define sk_refcnt __sk_common.skc_refcnt
310 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
312 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
313 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
314 #define sk_hash __sk_common.skc_hash
315 #define sk_portpair __sk_common.skc_portpair
316 #define sk_num __sk_common.skc_num
317 #define sk_dport __sk_common.skc_dport
318 #define sk_addrpair __sk_common.skc_addrpair
319 #define sk_daddr __sk_common.skc_daddr
320 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
321 #define sk_family __sk_common.skc_family
322 #define sk_state __sk_common.skc_state
323 #define sk_reuse __sk_common.skc_reuse
324 #define sk_reuseport __sk_common.skc_reuseport
325 #define sk_ipv6only __sk_common.skc_ipv6only
326 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
327 #define sk_bind_node __sk_common.skc_bind_node
328 #define sk_prot __sk_common.skc_prot
329 #define sk_net __sk_common.skc_net
330 #define sk_v6_daddr __sk_common.skc_v6_daddr
331 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
332 #define sk_cookie __sk_common.skc_cookie
334 socket_lock_t sk_lock;
335 struct sk_buff_head sk_receive_queue;
337 * The backlog queue is special, it is always used with
338 * the per-socket spinlock held and requires low latency
339 * access. Therefore we special case it's implementation.
340 * Note : rmem_alloc is in this structure to fill a hole
341 * on 64bit arches, not because its logically part of
347 struct sk_buff *head;
348 struct sk_buff *tail;
350 #define sk_rmem_alloc sk_backlog.rmem_alloc
351 int sk_forward_alloc;
357 * Warned : sk_incoming_cpu can be set from softirq,
358 * Do not use this hole without fully understanding possible issues.
362 #ifdef CONFIG_NET_RX_BUSY_POLL
363 unsigned int sk_napi_id;
364 unsigned int sk_ll_usec;
369 struct sk_filter __rcu *sk_filter;
370 struct socket_wq __rcu *sk_wq;
373 struct xfrm_policy *sk_policy[2];
375 unsigned long sk_flags;
376 struct dst_entry *sk_rx_dst;
377 struct dst_entry __rcu *sk_dst_cache;
378 spinlock_t sk_dst_lock;
379 atomic_t sk_wmem_alloc;
380 atomic_t sk_omem_alloc;
382 struct sk_buff_head sk_write_queue;
383 kmemcheck_bitfield_begin(flags);
384 unsigned int sk_shutdown : 2,
390 kmemcheck_bitfield_end(flags);
393 u32 sk_pacing_rate; /* bytes per second */
394 u32 sk_max_pacing_rate;
395 netdev_features_t sk_route_caps;
396 netdev_features_t sk_route_nocaps;
398 unsigned int sk_gso_max_size;
401 unsigned long sk_lingertime;
402 struct sk_buff_head sk_error_queue;
403 struct proto *sk_prot_creator;
404 rwlock_t sk_callback_lock;
408 u32 sk_max_ack_backlog;
410 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
411 __u32 sk_cgrp_prioidx;
413 struct pid *sk_peer_pid;
414 const struct cred *sk_peer_cred;
418 struct timer_list sk_timer;
422 struct socket *sk_socket;
424 struct page_frag sk_frag;
425 struct sk_buff *sk_send_head;
427 int sk_write_pending;
428 #ifdef CONFIG_SECURITY
433 struct cg_proto *sk_cgrp;
434 void (*sk_state_change)(struct sock *sk);
435 void (*sk_data_ready)(struct sock *sk);
436 void (*sk_write_space)(struct sock *sk);
437 void (*sk_error_report)(struct sock *sk);
438 int (*sk_backlog_rcv)(struct sock *sk,
439 struct sk_buff *skb);
440 void (*sk_destruct)(struct sock *sk);
443 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
445 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
446 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
449 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
450 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
451 * on a socket means that the socket will reuse everybody else's port
452 * without looking at the other's sk_reuse value.
455 #define SK_NO_REUSE 0
456 #define SK_CAN_REUSE 1
457 #define SK_FORCE_REUSE 2
459 static inline int sk_peek_offset(struct sock *sk, int flags)
461 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
462 return sk->sk_peek_off;
467 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
469 if (sk->sk_peek_off >= 0) {
470 if (sk->sk_peek_off >= val)
471 sk->sk_peek_off -= val;
477 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
479 if (sk->sk_peek_off >= 0)
480 sk->sk_peek_off += val;
484 * Hashed lists helper routines
486 static inline struct sock *sk_entry(const struct hlist_node *node)
488 return hlist_entry(node, struct sock, sk_node);
491 static inline struct sock *__sk_head(const struct hlist_head *head)
493 return hlist_entry(head->first, struct sock, sk_node);
496 static inline struct sock *sk_head(const struct hlist_head *head)
498 return hlist_empty(head) ? NULL : __sk_head(head);
501 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
503 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
506 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
508 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
511 static inline struct sock *sk_next(const struct sock *sk)
513 return sk->sk_node.next ?
514 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
517 static inline struct sock *sk_nulls_next(const struct sock *sk)
519 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
520 hlist_nulls_entry(sk->sk_nulls_node.next,
521 struct sock, sk_nulls_node) :
525 static inline bool sk_unhashed(const struct sock *sk)
527 return hlist_unhashed(&sk->sk_node);
530 static inline bool sk_hashed(const struct sock *sk)
532 return !sk_unhashed(sk);
535 static inline void sk_node_init(struct hlist_node *node)
540 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
545 static inline void __sk_del_node(struct sock *sk)
547 __hlist_del(&sk->sk_node);
550 /* NB: equivalent to hlist_del_init_rcu */
551 static inline bool __sk_del_node_init(struct sock *sk)
555 sk_node_init(&sk->sk_node);
561 /* Grab socket reference count. This operation is valid only
562 when sk is ALREADY grabbed f.e. it is found in hash table
563 or a list and the lookup is made under lock preventing hash table
567 static inline void sock_hold(struct sock *sk)
569 atomic_inc(&sk->sk_refcnt);
572 /* Ungrab socket in the context, which assumes that socket refcnt
573 cannot hit zero, f.e. it is true in context of any socketcall.
575 static inline void __sock_put(struct sock *sk)
577 atomic_dec(&sk->sk_refcnt);
580 static inline bool sk_del_node_init(struct sock *sk)
582 bool rc = __sk_del_node_init(sk);
585 /* paranoid for a while -acme */
586 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
591 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
593 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
596 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
602 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
604 bool rc = __sk_nulls_del_node_init_rcu(sk);
607 /* paranoid for a while -acme */
608 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
614 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
616 hlist_add_head(&sk->sk_node, list);
619 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
622 __sk_add_node(sk, list);
625 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
628 hlist_add_head_rcu(&sk->sk_node, list);
631 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
633 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
636 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
639 __sk_nulls_add_node_rcu(sk, list);
642 static inline void __sk_del_bind_node(struct sock *sk)
644 __hlist_del(&sk->sk_bind_node);
647 static inline void sk_add_bind_node(struct sock *sk,
648 struct hlist_head *list)
650 hlist_add_head(&sk->sk_bind_node, list);
653 #define sk_for_each(__sk, list) \
654 hlist_for_each_entry(__sk, list, sk_node)
655 #define sk_for_each_rcu(__sk, list) \
656 hlist_for_each_entry_rcu(__sk, list, sk_node)
657 #define sk_nulls_for_each(__sk, node, list) \
658 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
659 #define sk_nulls_for_each_rcu(__sk, node, list) \
660 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
661 #define sk_for_each_from(__sk) \
662 hlist_for_each_entry_from(__sk, sk_node)
663 #define sk_nulls_for_each_from(__sk, node) \
664 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
665 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
666 #define sk_for_each_safe(__sk, tmp, list) \
667 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
668 #define sk_for_each_bound(__sk, list) \
669 hlist_for_each_entry(__sk, list, sk_bind_node)
672 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
673 * @tpos: the type * to use as a loop cursor.
674 * @pos: the &struct hlist_node to use as a loop cursor.
675 * @head: the head for your list.
676 * @offset: offset of hlist_node within the struct.
679 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
680 for (pos = (head)->first; \
681 (!is_a_nulls(pos)) && \
682 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
685 static inline struct user_namespace *sk_user_ns(struct sock *sk)
687 /* Careful only use this in a context where these parameters
688 * can not change and must all be valid, such as recvmsg from
691 return sk->sk_socket->file->f_cred->user_ns;
705 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
706 SOCK_DBG, /* %SO_DEBUG setting */
707 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
708 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
709 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
710 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
711 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
712 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
713 SOCK_FASYNC, /* fasync() active */
715 SOCK_ZEROCOPY, /* buffers from userspace */
716 SOCK_WIFI_STATUS, /* push wifi status to userspace */
717 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
718 * Will use last 4 bytes of packet sent from
719 * user-space instead.
721 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
722 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
725 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
727 nsk->sk_flags = osk->sk_flags;
730 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
732 __set_bit(flag, &sk->sk_flags);
735 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
737 __clear_bit(flag, &sk->sk_flags);
740 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
742 return test_bit(flag, &sk->sk_flags);
746 extern struct static_key memalloc_socks;
747 static inline int sk_memalloc_socks(void)
749 return static_key_false(&memalloc_socks);
753 static inline int sk_memalloc_socks(void)
760 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
762 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
765 static inline void sk_acceptq_removed(struct sock *sk)
767 sk->sk_ack_backlog--;
770 static inline void sk_acceptq_added(struct sock *sk)
772 sk->sk_ack_backlog++;
775 static inline bool sk_acceptq_is_full(const struct sock *sk)
777 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
781 * Compute minimal free write space needed to queue new packets.
783 static inline int sk_stream_min_wspace(const struct sock *sk)
785 return sk->sk_wmem_queued >> 1;
788 static inline int sk_stream_wspace(const struct sock *sk)
790 return sk->sk_sndbuf - sk->sk_wmem_queued;
793 void sk_stream_write_space(struct sock *sk);
795 /* OOB backlog add */
796 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
798 /* dont let skb dst not refcounted, we are going to leave rcu lock */
801 if (!sk->sk_backlog.tail)
802 sk->sk_backlog.head = skb;
804 sk->sk_backlog.tail->next = skb;
806 sk->sk_backlog.tail = skb;
811 * Take into account size of receive queue and backlog queue
812 * Do not take into account this skb truesize,
813 * to allow even a single big packet to come.
815 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
817 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
819 return qsize > limit;
822 /* The per-socket spinlock must be held here. */
823 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
826 if (sk_rcvqueues_full(sk, limit))
829 __sk_add_backlog(sk, skb);
830 sk->sk_backlog.len += skb->truesize;
834 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
836 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
838 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
839 return __sk_backlog_rcv(sk, skb);
841 return sk->sk_backlog_rcv(sk, skb);
844 static inline void sk_incoming_cpu_update(struct sock *sk)
846 sk->sk_incoming_cpu = raw_smp_processor_id();
849 static inline void sock_rps_record_flow_hash(__u32 hash)
852 struct rps_sock_flow_table *sock_flow_table;
855 sock_flow_table = rcu_dereference(rps_sock_flow_table);
856 rps_record_sock_flow(sock_flow_table, hash);
861 static inline void sock_rps_record_flow(const struct sock *sk)
864 sock_rps_record_flow_hash(sk->sk_rxhash);
868 static inline void sock_rps_save_rxhash(struct sock *sk,
869 const struct sk_buff *skb)
872 if (unlikely(sk->sk_rxhash != skb->hash))
873 sk->sk_rxhash = skb->hash;
877 static inline void sock_rps_reset_rxhash(struct sock *sk)
884 #define sk_wait_event(__sk, __timeo, __condition) \
886 release_sock(__sk); \
887 __rc = __condition; \
889 *(__timeo) = schedule_timeout(*(__timeo)); \
891 sched_annotate_sleep(); \
893 __rc = __condition; \
897 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
898 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
899 void sk_stream_wait_close(struct sock *sk, long timeo_p);
900 int sk_stream_error(struct sock *sk, int flags, int err);
901 void sk_stream_kill_queues(struct sock *sk);
902 void sk_set_memalloc(struct sock *sk);
903 void sk_clear_memalloc(struct sock *sk);
905 int sk_wait_data(struct sock *sk, long *timeo);
907 struct request_sock_ops;
908 struct timewait_sock_ops;
909 struct inet_hashinfo;
914 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
915 * un-modified. Special care is taken when initializing object to zero.
917 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
919 if (offsetof(struct sock, sk_node.next) != 0)
920 memset(sk, 0, offsetof(struct sock, sk_node.next));
921 memset(&sk->sk_node.pprev, 0,
922 size - offsetof(struct sock, sk_node.pprev));
925 /* Networking protocol blocks we attach to sockets.
926 * socket layer -> transport layer interface
927 * transport -> network interface is defined by struct inet_proto
930 void (*close)(struct sock *sk,
932 int (*connect)(struct sock *sk,
933 struct sockaddr *uaddr,
935 int (*disconnect)(struct sock *sk, int flags);
937 struct sock * (*accept)(struct sock *sk, int flags, int *err);
939 int (*ioctl)(struct sock *sk, int cmd,
941 int (*init)(struct sock *sk);
942 void (*destroy)(struct sock *sk);
943 void (*shutdown)(struct sock *sk, int how);
944 int (*setsockopt)(struct sock *sk, int level,
945 int optname, char __user *optval,
946 unsigned int optlen);
947 int (*getsockopt)(struct sock *sk, int level,
948 int optname, char __user *optval,
951 int (*compat_setsockopt)(struct sock *sk,
953 int optname, char __user *optval,
954 unsigned int optlen);
955 int (*compat_getsockopt)(struct sock *sk,
957 int optname, char __user *optval,
959 int (*compat_ioctl)(struct sock *sk,
960 unsigned int cmd, unsigned long arg);
962 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
964 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
965 size_t len, int noblock, int flags,
967 int (*sendpage)(struct sock *sk, struct page *page,
968 int offset, size_t size, int flags);
969 int (*bind)(struct sock *sk,
970 struct sockaddr *uaddr, int addr_len);
972 int (*backlog_rcv) (struct sock *sk,
973 struct sk_buff *skb);
975 void (*release_cb)(struct sock *sk);
977 /* Keeping track of sk's, looking them up, and port selection methods. */
978 void (*hash)(struct sock *sk);
979 void (*unhash)(struct sock *sk);
980 void (*rehash)(struct sock *sk);
981 int (*get_port)(struct sock *sk, unsigned short snum);
982 void (*clear_sk)(struct sock *sk, int size);
984 /* Keeping track of sockets in use */
985 #ifdef CONFIG_PROC_FS
986 unsigned int inuse_idx;
989 bool (*stream_memory_free)(const struct sock *sk);
990 /* Memory pressure */
991 void (*enter_memory_pressure)(struct sock *sk);
992 atomic_long_t *memory_allocated; /* Current allocated memory. */
993 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
995 * Pressure flag: try to collapse.
996 * Technical note: it is used by multiple contexts non atomically.
997 * All the __sk_mem_schedule() is of this nature: accounting
998 * is strict, actions are advisory and have some latency.
1000 int *memory_pressure;
1007 struct kmem_cache *slab;
1008 unsigned int obj_size;
1011 struct percpu_counter *orphan_count;
1013 struct request_sock_ops *rsk_prot;
1014 struct timewait_sock_ops *twsk_prot;
1017 struct inet_hashinfo *hashinfo;
1018 struct udp_table *udp_table;
1019 struct raw_hashinfo *raw_hash;
1022 struct module *owner;
1026 struct list_head node;
1027 #ifdef SOCK_REFCNT_DEBUG
1030 #ifdef CONFIG_MEMCG_KMEM
1032 * cgroup specific init/deinit functions. Called once for all
1033 * protocols that implement it, from cgroups populate function.
1034 * This function has to setup any files the protocol want to
1035 * appear in the kmem cgroup filesystem.
1037 int (*init_cgroup)(struct mem_cgroup *memcg,
1038 struct cgroup_subsys *ss);
1039 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1040 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1045 * Bits in struct cg_proto.flags
1047 enum cg_proto_flags {
1048 /* Currently active and new sockets should be assigned to cgroups */
1050 /* It was ever activated; we must disarm static keys on destruction */
1051 MEMCG_SOCK_ACTIVATED,
1055 struct page_counter memory_allocated; /* Current allocated memory. */
1056 struct percpu_counter sockets_allocated; /* Current number of sockets. */
1057 int memory_pressure;
1059 unsigned long flags;
1061 * memcg field is used to find which memcg we belong directly
1062 * Each memcg struct can hold more than one cg_proto, so container_of
1065 * The elegant solution would be having an inverse function to
1066 * proto_cgroup in struct proto, but that means polluting the structure
1067 * for everybody, instead of just for memcg users.
1069 struct mem_cgroup *memcg;
1072 int proto_register(struct proto *prot, int alloc_slab);
1073 void proto_unregister(struct proto *prot);
1075 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1077 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1080 #ifdef SOCK_REFCNT_DEBUG
1081 static inline void sk_refcnt_debug_inc(struct sock *sk)
1083 atomic_inc(&sk->sk_prot->socks);
1086 static inline void sk_refcnt_debug_dec(struct sock *sk)
1088 atomic_dec(&sk->sk_prot->socks);
1089 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1090 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1093 static inline void sk_refcnt_debug_release(const struct sock *sk)
1095 if (atomic_read(&sk->sk_refcnt) != 1)
1096 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1097 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1099 #else /* SOCK_REFCNT_DEBUG */
1100 #define sk_refcnt_debug_inc(sk) do { } while (0)
1101 #define sk_refcnt_debug_dec(sk) do { } while (0)
1102 #define sk_refcnt_debug_release(sk) do { } while (0)
1103 #endif /* SOCK_REFCNT_DEBUG */
1105 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1106 extern struct static_key memcg_socket_limit_enabled;
1107 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1108 struct cg_proto *cg_proto)
1110 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1112 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1114 #define mem_cgroup_sockets_enabled 0
1115 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1116 struct cg_proto *cg_proto)
1122 static inline bool sk_stream_memory_free(const struct sock *sk)
1124 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1127 return sk->sk_prot->stream_memory_free ?
1128 sk->sk_prot->stream_memory_free(sk) : true;
1131 static inline bool sk_stream_is_writeable(const struct sock *sk)
1133 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1134 sk_stream_memory_free(sk);
1138 static inline bool sk_has_memory_pressure(const struct sock *sk)
1140 return sk->sk_prot->memory_pressure != NULL;
1143 static inline bool sk_under_memory_pressure(const struct sock *sk)
1145 if (!sk->sk_prot->memory_pressure)
1148 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1149 return !!sk->sk_cgrp->memory_pressure;
1151 return !!*sk->sk_prot->memory_pressure;
1154 static inline void sk_leave_memory_pressure(struct sock *sk)
1156 int *memory_pressure = sk->sk_prot->memory_pressure;
1158 if (!memory_pressure)
1161 if (*memory_pressure)
1162 *memory_pressure = 0;
1164 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1165 struct cg_proto *cg_proto = sk->sk_cgrp;
1166 struct proto *prot = sk->sk_prot;
1168 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1169 cg_proto->memory_pressure = 0;
1174 static inline void sk_enter_memory_pressure(struct sock *sk)
1176 if (!sk->sk_prot->enter_memory_pressure)
1179 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1180 struct cg_proto *cg_proto = sk->sk_cgrp;
1181 struct proto *prot = sk->sk_prot;
1183 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1184 cg_proto->memory_pressure = 1;
1187 sk->sk_prot->enter_memory_pressure(sk);
1190 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1192 long *prot = sk->sk_prot->sysctl_mem;
1193 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1194 prot = sk->sk_cgrp->sysctl_mem;
1198 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1202 page_counter_charge(&prot->memory_allocated, amt);
1204 if (page_counter_read(&prot->memory_allocated) >
1205 prot->memory_allocated.limit)
1206 *parent_status = OVER_LIMIT;
1209 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1212 page_counter_uncharge(&prot->memory_allocated, amt);
1216 sk_memory_allocated(const struct sock *sk)
1218 struct proto *prot = sk->sk_prot;
1220 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1221 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1223 return atomic_long_read(prot->memory_allocated);
1227 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1229 struct proto *prot = sk->sk_prot;
1231 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1232 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1233 /* update the root cgroup regardless */
1234 atomic_long_add_return(amt, prot->memory_allocated);
1235 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1238 return atomic_long_add_return(amt, prot->memory_allocated);
1242 sk_memory_allocated_sub(struct sock *sk, int amt)
1244 struct proto *prot = sk->sk_prot;
1246 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1247 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1249 atomic_long_sub(amt, prot->memory_allocated);
1252 static inline void sk_sockets_allocated_dec(struct sock *sk)
1254 struct proto *prot = sk->sk_prot;
1256 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1257 struct cg_proto *cg_proto = sk->sk_cgrp;
1259 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1260 percpu_counter_dec(&cg_proto->sockets_allocated);
1263 percpu_counter_dec(prot->sockets_allocated);
1266 static inline void sk_sockets_allocated_inc(struct sock *sk)
1268 struct proto *prot = sk->sk_prot;
1270 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1271 struct cg_proto *cg_proto = sk->sk_cgrp;
1273 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1274 percpu_counter_inc(&cg_proto->sockets_allocated);
1277 percpu_counter_inc(prot->sockets_allocated);
1281 sk_sockets_allocated_read_positive(struct sock *sk)
1283 struct proto *prot = sk->sk_prot;
1285 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1286 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1288 return percpu_counter_read_positive(prot->sockets_allocated);
1292 proto_sockets_allocated_sum_positive(struct proto *prot)
1294 return percpu_counter_sum_positive(prot->sockets_allocated);
1298 proto_memory_allocated(struct proto *prot)
1300 return atomic_long_read(prot->memory_allocated);
1304 proto_memory_pressure(struct proto *prot)
1306 if (!prot->memory_pressure)
1308 return !!*prot->memory_pressure;
1312 #ifdef CONFIG_PROC_FS
1313 /* Called with local bh disabled */
1314 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1315 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1317 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1324 /* With per-bucket locks this operation is not-atomic, so that
1325 * this version is not worse.
1327 static inline void __sk_prot_rehash(struct sock *sk)
1329 sk->sk_prot->unhash(sk);
1330 sk->sk_prot->hash(sk);
1333 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1335 /* About 10 seconds */
1336 #define SOCK_DESTROY_TIME (10*HZ)
1338 /* Sockets 0-1023 can't be bound to unless you are superuser */
1339 #define PROT_SOCK 1024
1341 #define SHUTDOWN_MASK 3
1342 #define RCV_SHUTDOWN 1
1343 #define SEND_SHUTDOWN 2
1345 #define SOCK_SNDBUF_LOCK 1
1346 #define SOCK_RCVBUF_LOCK 2
1347 #define SOCK_BINDADDR_LOCK 4
1348 #define SOCK_BINDPORT_LOCK 8
1350 struct socket_alloc {
1351 struct socket socket;
1352 struct inode vfs_inode;
1355 static inline struct socket *SOCKET_I(struct inode *inode)
1357 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1360 static inline struct inode *SOCK_INODE(struct socket *socket)
1362 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1366 * Functions for memory accounting
1368 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1369 void __sk_mem_reclaim(struct sock *sk);
1371 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1372 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1373 #define SK_MEM_SEND 0
1374 #define SK_MEM_RECV 1
1376 static inline int sk_mem_pages(int amt)
1378 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1381 static inline bool sk_has_account(struct sock *sk)
1383 /* return true if protocol supports memory accounting */
1384 return !!sk->sk_prot->memory_allocated;
1387 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1389 if (!sk_has_account(sk))
1391 return size <= sk->sk_forward_alloc ||
1392 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1396 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1398 if (!sk_has_account(sk))
1400 return size<= sk->sk_forward_alloc ||
1401 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1402 skb_pfmemalloc(skb);
1405 static inline void sk_mem_reclaim(struct sock *sk)
1407 if (!sk_has_account(sk))
1409 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1410 __sk_mem_reclaim(sk);
1413 static inline void sk_mem_reclaim_partial(struct sock *sk)
1415 if (!sk_has_account(sk))
1417 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1418 __sk_mem_reclaim(sk);
1421 static inline void sk_mem_charge(struct sock *sk, int size)
1423 if (!sk_has_account(sk))
1425 sk->sk_forward_alloc -= size;
1428 static inline void sk_mem_uncharge(struct sock *sk, int size)
1430 if (!sk_has_account(sk))
1432 sk->sk_forward_alloc += size;
1435 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1437 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1438 sk->sk_wmem_queued -= skb->truesize;
1439 sk_mem_uncharge(sk, skb->truesize);
1443 /* Used by processes to "lock" a socket state, so that
1444 * interrupts and bottom half handlers won't change it
1445 * from under us. It essentially blocks any incoming
1446 * packets, so that we won't get any new data or any
1447 * packets that change the state of the socket.
1449 * While locked, BH processing will add new packets to
1450 * the backlog queue. This queue is processed by the
1451 * owner of the socket lock right before it is released.
1453 * Since ~2.3.5 it is also exclusive sleep lock serializing
1454 * accesses from user process context.
1456 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1458 static inline void sock_release_ownership(struct sock *sk)
1460 sk->sk_lock.owned = 0;
1464 * Macro so as to not evaluate some arguments when
1465 * lockdep is not enabled.
1467 * Mark both the sk_lock and the sk_lock.slock as a
1468 * per-address-family lock class.
1470 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1472 sk->sk_lock.owned = 0; \
1473 init_waitqueue_head(&sk->sk_lock.wq); \
1474 spin_lock_init(&(sk)->sk_lock.slock); \
1475 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1476 sizeof((sk)->sk_lock)); \
1477 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1479 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1482 void lock_sock_nested(struct sock *sk, int subclass);
1484 static inline void lock_sock(struct sock *sk)
1486 lock_sock_nested(sk, 0);
1489 void release_sock(struct sock *sk);
1491 /* BH context may only use the following locking interface. */
1492 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1493 #define bh_lock_sock_nested(__sk) \
1494 spin_lock_nested(&((__sk)->sk_lock.slock), \
1495 SINGLE_DEPTH_NESTING)
1496 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1498 bool lock_sock_fast(struct sock *sk);
1500 * unlock_sock_fast - complement of lock_sock_fast
1504 * fast unlock socket for user context.
1505 * If slow mode is on, we call regular release_sock()
1507 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1512 spin_unlock_bh(&sk->sk_lock.slock);
1516 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1517 struct proto *prot);
1518 void sk_free(struct sock *sk);
1519 void sk_release_kernel(struct sock *sk);
1520 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1522 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1524 void sock_wfree(struct sk_buff *skb);
1525 void skb_orphan_partial(struct sk_buff *skb);
1526 void sock_rfree(struct sk_buff *skb);
1527 void sock_efree(struct sk_buff *skb);
1529 void sock_edemux(struct sk_buff *skb);
1531 #define sock_edemux(skb) sock_efree(skb)
1534 int sock_setsockopt(struct socket *sock, int level, int op,
1535 char __user *optval, unsigned int optlen);
1537 int sock_getsockopt(struct socket *sock, int level, int op,
1538 char __user *optval, int __user *optlen);
1539 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1540 int noblock, int *errcode);
1541 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1542 unsigned long data_len, int noblock,
1543 int *errcode, int max_page_order);
1544 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1545 void sock_kfree_s(struct sock *sk, void *mem, int size);
1546 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1547 void sk_send_sigurg(struct sock *sk);
1550 * Functions to fill in entries in struct proto_ops when a protocol
1551 * does not implement a particular function.
1553 int sock_no_bind(struct socket *, struct sockaddr *, int);
1554 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1555 int sock_no_socketpair(struct socket *, struct socket *);
1556 int sock_no_accept(struct socket *, struct socket *, int);
1557 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1558 unsigned int sock_no_poll(struct file *, struct socket *,
1559 struct poll_table_struct *);
1560 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1561 int sock_no_listen(struct socket *, int);
1562 int sock_no_shutdown(struct socket *, int);
1563 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1564 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1565 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1566 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1567 int sock_no_mmap(struct file *file, struct socket *sock,
1568 struct vm_area_struct *vma);
1569 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1570 size_t size, int flags);
1573 * Functions to fill in entries in struct proto_ops when a protocol
1574 * uses the inet style.
1576 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1577 char __user *optval, int __user *optlen);
1578 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1580 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1581 char __user *optval, unsigned int optlen);
1582 int compat_sock_common_getsockopt(struct socket *sock, int level,
1583 int optname, char __user *optval, int __user *optlen);
1584 int compat_sock_common_setsockopt(struct socket *sock, int level,
1585 int optname, char __user *optval, unsigned int optlen);
1587 void sk_common_release(struct sock *sk);
1590 * Default socket callbacks and setup code
1593 /* Initialise core socket variables */
1594 void sock_init_data(struct socket *sock, struct sock *sk);
1597 * Socket reference counting postulates.
1599 * * Each user of socket SHOULD hold a reference count.
1600 * * Each access point to socket (an hash table bucket, reference from a list,
1601 * running timer, skb in flight MUST hold a reference count.
1602 * * When reference count hits 0, it means it will never increase back.
1603 * * When reference count hits 0, it means that no references from
1604 * outside exist to this socket and current process on current CPU
1605 * is last user and may/should destroy this socket.
1606 * * sk_free is called from any context: process, BH, IRQ. When
1607 * it is called, socket has no references from outside -> sk_free
1608 * may release descendant resources allocated by the socket, but
1609 * to the time when it is called, socket is NOT referenced by any
1610 * hash tables, lists etc.
1611 * * Packets, delivered from outside (from network or from another process)
1612 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1613 * when they sit in queue. Otherwise, packets will leak to hole, when
1614 * socket is looked up by one cpu and unhasing is made by another CPU.
1615 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1616 * (leak to backlog). Packet socket does all the processing inside
1617 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1618 * use separate SMP lock, so that they are prone too.
1621 /* Ungrab socket and destroy it, if it was the last reference. */
1622 static inline void sock_put(struct sock *sk)
1624 if (atomic_dec_and_test(&sk->sk_refcnt))
1627 /* Generic version of sock_put(), dealing with all sockets
1628 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1630 void sock_gen_put(struct sock *sk);
1632 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1634 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1636 sk->sk_tx_queue_mapping = tx_queue;
1639 static inline void sk_tx_queue_clear(struct sock *sk)
1641 sk->sk_tx_queue_mapping = -1;
1644 static inline int sk_tx_queue_get(const struct sock *sk)
1646 return sk ? sk->sk_tx_queue_mapping : -1;
1649 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1651 sk_tx_queue_clear(sk);
1652 sk->sk_socket = sock;
1655 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1657 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1658 return &rcu_dereference_raw(sk->sk_wq)->wait;
1660 /* Detach socket from process context.
1661 * Announce socket dead, detach it from wait queue and inode.
1662 * Note that parent inode held reference count on this struct sock,
1663 * we do not release it in this function, because protocol
1664 * probably wants some additional cleanups or even continuing
1665 * to work with this socket (TCP).
1667 static inline void sock_orphan(struct sock *sk)
1669 write_lock_bh(&sk->sk_callback_lock);
1670 sock_set_flag(sk, SOCK_DEAD);
1671 sk_set_socket(sk, NULL);
1673 write_unlock_bh(&sk->sk_callback_lock);
1676 static inline void sock_graft(struct sock *sk, struct socket *parent)
1678 write_lock_bh(&sk->sk_callback_lock);
1679 sk->sk_wq = parent->wq;
1681 sk_set_socket(sk, parent);
1682 security_sock_graft(sk, parent);
1683 write_unlock_bh(&sk->sk_callback_lock);
1686 kuid_t sock_i_uid(struct sock *sk);
1687 unsigned long sock_i_ino(struct sock *sk);
1689 static inline struct dst_entry *
1690 __sk_dst_get(struct sock *sk)
1692 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1693 lockdep_is_held(&sk->sk_lock.slock));
1696 static inline struct dst_entry *
1697 sk_dst_get(struct sock *sk)
1699 struct dst_entry *dst;
1702 dst = rcu_dereference(sk->sk_dst_cache);
1703 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1709 static inline void dst_negative_advice(struct sock *sk)
1711 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1713 if (dst && dst->ops->negative_advice) {
1714 ndst = dst->ops->negative_advice(dst);
1717 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1718 sk_tx_queue_clear(sk);
1724 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1726 struct dst_entry *old_dst;
1728 sk_tx_queue_clear(sk);
1730 * This can be called while sk is owned by the caller only,
1731 * with no state that can be checked in a rcu_dereference_check() cond
1733 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1734 rcu_assign_pointer(sk->sk_dst_cache, dst);
1735 dst_release(old_dst);
1739 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1741 struct dst_entry *old_dst;
1743 sk_tx_queue_clear(sk);
1744 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1745 dst_release(old_dst);
1749 __sk_dst_reset(struct sock *sk)
1751 __sk_dst_set(sk, NULL);
1755 sk_dst_reset(struct sock *sk)
1757 sk_dst_set(sk, NULL);
1760 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1762 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1764 bool sk_mc_loop(struct sock *sk);
1766 static inline bool sk_can_gso(const struct sock *sk)
1768 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1771 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1773 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1775 sk->sk_route_nocaps |= flags;
1776 sk->sk_route_caps &= ~flags;
1779 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1780 struct iov_iter *from, char *to,
1781 int copy, int offset)
1783 if (skb->ip_summed == CHECKSUM_NONE) {
1785 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1787 skb->csum = csum_block_add(skb->csum, csum, offset);
1788 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1789 if (copy_from_iter_nocache(to, copy, from) != copy)
1791 } else if (copy_from_iter(to, copy, from) != copy)
1797 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1798 struct iov_iter *from, int copy)
1800 int err, offset = skb->len;
1802 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1805 __skb_trim(skb, offset);
1810 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1811 struct sk_buff *skb,
1817 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1823 skb->data_len += copy;
1824 skb->truesize += copy;
1825 sk->sk_wmem_queued += copy;
1826 sk_mem_charge(sk, copy);
1831 * sk_wmem_alloc_get - returns write allocations
1834 * Returns sk_wmem_alloc minus initial offset of one
1836 static inline int sk_wmem_alloc_get(const struct sock *sk)
1838 return atomic_read(&sk->sk_wmem_alloc) - 1;
1842 * sk_rmem_alloc_get - returns read allocations
1845 * Returns sk_rmem_alloc
1847 static inline int sk_rmem_alloc_get(const struct sock *sk)
1849 return atomic_read(&sk->sk_rmem_alloc);
1853 * sk_has_allocations - check if allocations are outstanding
1856 * Returns true if socket has write or read allocations
1858 static inline bool sk_has_allocations(const struct sock *sk)
1860 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1864 * wq_has_sleeper - check if there are any waiting processes
1865 * @wq: struct socket_wq
1867 * Returns true if socket_wq has waiting processes
1869 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1870 * barrier call. They were added due to the race found within the tcp code.
1872 * Consider following tcp code paths:
1876 * sys_select receive packet
1878 * __add_wait_queue update tp->rcv_nxt
1880 * tp->rcv_nxt check sock_def_readable
1882 * schedule rcu_read_lock();
1883 * wq = rcu_dereference(sk->sk_wq);
1884 * if (wq && waitqueue_active(&wq->wait))
1885 * wake_up_interruptible(&wq->wait)
1889 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1890 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1891 * could then endup calling schedule and sleep forever if there are no more
1892 * data on the socket.
1895 static inline bool wq_has_sleeper(struct socket_wq *wq)
1897 /* We need to be sure we are in sync with the
1898 * add_wait_queue modifications to the wait queue.
1900 * This memory barrier is paired in the sock_poll_wait.
1903 return wq && waitqueue_active(&wq->wait);
1907 * sock_poll_wait - place memory barrier behind the poll_wait call.
1909 * @wait_address: socket wait queue
1912 * See the comments in the wq_has_sleeper function.
1914 static inline void sock_poll_wait(struct file *filp,
1915 wait_queue_head_t *wait_address, poll_table *p)
1917 if (!poll_does_not_wait(p) && wait_address) {
1918 poll_wait(filp, wait_address, p);
1919 /* We need to be sure we are in sync with the
1920 * socket flags modification.
1922 * This memory barrier is paired in the wq_has_sleeper.
1928 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1930 if (sk->sk_txhash) {
1932 skb->hash = sk->sk_txhash;
1937 * Queue a received datagram if it will fit. Stream and sequenced
1938 * protocols can't normally use this as they need to fit buffers in
1939 * and play with them.
1941 * Inlined as it's very short and called for pretty much every
1942 * packet ever received.
1945 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1949 skb->destructor = sock_wfree;
1950 skb_set_hash_from_sk(skb, sk);
1952 * We used to take a refcount on sk, but following operation
1953 * is enough to guarantee sk_free() wont free this sock until
1954 * all in-flight packets are completed
1956 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1959 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1963 skb->destructor = sock_rfree;
1964 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1965 sk_mem_charge(sk, skb->truesize);
1968 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1969 unsigned long expires);
1971 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1973 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1975 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1976 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1979 * Recover an error report and clear atomically
1982 static inline int sock_error(struct sock *sk)
1985 if (likely(!sk->sk_err))
1987 err = xchg(&sk->sk_err, 0);
1991 static inline unsigned long sock_wspace(struct sock *sk)
1995 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1996 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2003 static inline void sk_wake_async(struct sock *sk, int how, int band)
2005 if (sock_flag(sk, SOCK_FASYNC))
2006 sock_wake_async(sk->sk_socket, how, band);
2009 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2010 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2011 * Note: for send buffers, TCP works better if we can build two skbs at
2014 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2016 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2017 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2019 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2021 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2022 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2023 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2027 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2030 * sk_page_frag - return an appropriate page_frag
2033 * If socket allocation mode allows current thread to sleep, it means its
2034 * safe to use the per task page_frag instead of the per socket one.
2036 static inline struct page_frag *sk_page_frag(struct sock *sk)
2038 if (sk->sk_allocation & __GFP_WAIT)
2039 return ¤t->task_frag;
2041 return &sk->sk_frag;
2044 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2047 * Default write policy as shown to user space via poll/select/SIGIO
2049 static inline bool sock_writeable(const struct sock *sk)
2051 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2054 static inline gfp_t gfp_any(void)
2056 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2059 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2061 return noblock ? 0 : sk->sk_rcvtimeo;
2064 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2066 return noblock ? 0 : sk->sk_sndtimeo;
2069 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2071 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2074 /* Alas, with timeout socket operations are not restartable.
2075 * Compare this to poll().
2077 static inline int sock_intr_errno(long timeo)
2079 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2082 struct sock_skb_cb {
2086 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2087 * using skb->cb[] would keep using it directly and utilize its
2088 * alignement guarantee.
2090 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2091 sizeof(struct sock_skb_cb)))
2093 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2094 SOCK_SKB_CB_OFFSET))
2096 #define sock_skb_cb_check_size(size) \
2097 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2100 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2102 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2105 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2106 struct sk_buff *skb);
2107 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2108 struct sk_buff *skb);
2111 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2113 ktime_t kt = skb->tstamp;
2114 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2117 * generate control messages if
2118 * - receive time stamping in software requested
2119 * - software time stamp available and wanted
2120 * - hardware time stamps available and wanted
2122 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2123 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2124 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2125 (hwtstamps->hwtstamp.tv64 &&
2126 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2127 __sock_recv_timestamp(msg, sk, skb);
2131 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2132 __sock_recv_wifi_status(msg, sk, skb);
2135 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2136 struct sk_buff *skb);
2138 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2139 struct sk_buff *skb)
2141 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2142 (1UL << SOCK_RCVTSTAMP))
2143 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2144 SOF_TIMESTAMPING_RAW_HARDWARE)
2146 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2147 __sock_recv_ts_and_drops(msg, sk, skb);
2149 sk->sk_stamp = skb->tstamp;
2152 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2155 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2156 * @sk: socket sending this packet
2157 * @tx_flags: completed with instructions for time stamping
2159 * Note : callers should take care of initial *tx_flags value (usually 0)
2161 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2163 if (unlikely(sk->sk_tsflags))
2164 __sock_tx_timestamp(sk, tx_flags);
2165 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2166 *tx_flags |= SKBTX_WIFI_STATUS;
2170 * sk_eat_skb - Release a skb if it is no longer needed
2171 * @sk: socket to eat this skb from
2172 * @skb: socket buffer to eat
2174 * This routine must be called with interrupts disabled or with the socket
2175 * locked so that the sk_buff queue operation is ok.
2177 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2179 __skb_unlink(skb, &sk->sk_receive_queue);
2184 struct net *sock_net(const struct sock *sk)
2186 return read_pnet(&sk->sk_net);
2190 void sock_net_set(struct sock *sk, struct net *net)
2192 write_pnet(&sk->sk_net, net);
2196 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2197 * They should not hold a reference to a namespace in order to allow
2199 * Sockets after sk_change_net should be released using sk_release_kernel
2201 static inline void sk_change_net(struct sock *sk, struct net *net)
2203 struct net *current_net = sock_net(sk);
2205 if (!net_eq(current_net, net)) {
2206 put_net(current_net);
2207 sock_net_set(sk, net);
2211 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2214 struct sock *sk = skb->sk;
2216 skb->destructor = NULL;
2223 /* This helper checks if a socket is a full socket,
2224 * ie _not_ a timewait or request socket.
2226 static inline bool sk_fullsock(const struct sock *sk)
2228 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2231 void sock_enable_timestamp(struct sock *sk, int flag);
2232 int sock_get_timestamp(struct sock *, struct timeval __user *);
2233 int sock_get_timestampns(struct sock *, struct timespec __user *);
2234 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2237 bool sk_ns_capable(const struct sock *sk,
2238 struct user_namespace *user_ns, int cap);
2239 bool sk_capable(const struct sock *sk, int cap);
2240 bool sk_net_capable(const struct sock *sk, int cap);
2242 extern __u32 sysctl_wmem_max;
2243 extern __u32 sysctl_rmem_max;
2245 extern int sysctl_tstamp_allow_data;
2246 extern int sysctl_optmem_max;
2248 extern __u32 sysctl_wmem_default;
2249 extern __u32 sysctl_rmem_default;
2251 #endif /* _SOCK_H */