2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the AF_INET socket handler.
8 * Version: @(#)sock.h 1.0.4 05/13/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 * David S. Miller : New socket lookup architecture.
26 * Steve Whitehouse: Default routines for sock_ops
27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
28 * protinfo be just a void pointer, as the
29 * protocol specific parts were moved to
30 * respective headers and ipv4/v6, etc now
31 * use private slabcaches for its socks
32 * Pedro Hortas : New flags field for socket options
35 * This program is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU General Public License
37 * as published by the Free Software Foundation; either version
38 * 2 of the License, or (at your option) any later version.
43 #include <linux/hardirq.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/list_nulls.h>
47 #include <linux/timer.h>
48 #include <linux/cache.h>
49 #include <linux/bitops.h>
50 #include <linux/lockdep.h>
51 #include <linux/netdevice.h>
52 #include <linux/skbuff.h> /* struct sk_buff */
54 #include <linux/security.h>
55 #include <linux/slab.h>
56 #include <linux/uaccess.h>
57 #include <linux/memcontrol.h>
58 #include <linux/res_counter.h>
59 #include <linux/static_key.h>
60 #include <linux/aio.h>
61 #include <linux/sched.h>
63 #include <linux/filter.h>
64 #include <linux/rculist_nulls.h>
65 #include <linux/poll.h>
67 #include <linux/atomic.h>
69 #include <net/checksum.h>
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;
197 #if IS_ENABLED(CONFIG_IPV6)
198 struct in6_addr skc_v6_daddr;
199 struct in6_addr skc_v6_rcv_saddr;
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_txhash: computed flow hash for use on transmit
277 * @sk_filter: socket filtering instructions
278 * @sk_protinfo: private area, net family specific, when not using slab
279 * @sk_timer: sock cleanup timer
280 * @sk_stamp: time stamp of last packet received
281 * @sk_tsflags: SO_TIMESTAMPING socket options
282 * @sk_tskey: counter to disambiguate concurrent tstamp requests
283 * @sk_socket: Identd and reporting IO signals
284 * @sk_user_data: RPC layer private data
285 * @sk_frag: cached page frag
286 * @sk_peek_off: current peek_offset value
287 * @sk_send_head: front of stuff to transmit
288 * @sk_security: used by security modules
289 * @sk_mark: generic packet mark
290 * @sk_classid: this socket's cgroup classid
291 * @sk_cgrp: this socket's cgroup-specific proto data
292 * @sk_write_pending: a write to stream socket waits to start
293 * @sk_state_change: callback to indicate change in the state of the sock
294 * @sk_data_ready: callback to indicate there is data to be processed
295 * @sk_write_space: callback to indicate there is bf sending space available
296 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
297 * @sk_backlog_rcv: callback to process the backlog
298 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
302 * Now struct inet_timewait_sock also uses sock_common, so please just
303 * don't add nothing before this first member (__sk_common) --acme
305 struct sock_common __sk_common;
306 #define sk_node __sk_common.skc_node
307 #define sk_nulls_node __sk_common.skc_nulls_node
308 #define sk_refcnt __sk_common.skc_refcnt
309 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
311 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
312 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
313 #define sk_hash __sk_common.skc_hash
314 #define sk_portpair __sk_common.skc_portpair
315 #define sk_num __sk_common.skc_num
316 #define sk_dport __sk_common.skc_dport
317 #define sk_addrpair __sk_common.skc_addrpair
318 #define sk_daddr __sk_common.skc_daddr
319 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
320 #define sk_family __sk_common.skc_family
321 #define sk_state __sk_common.skc_state
322 #define sk_reuse __sk_common.skc_reuse
323 #define sk_reuseport __sk_common.skc_reuseport
324 #define sk_ipv6only __sk_common.skc_ipv6only
325 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
326 #define sk_bind_node __sk_common.skc_bind_node
327 #define sk_prot __sk_common.skc_prot
328 #define sk_net __sk_common.skc_net
329 #define sk_v6_daddr __sk_common.skc_v6_daddr
330 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
332 socket_lock_t sk_lock;
333 struct sk_buff_head sk_receive_queue;
335 * The backlog queue is special, it is always used with
336 * the per-socket spinlock held and requires low latency
337 * access. Therefore we special case it's implementation.
338 * Note : rmem_alloc is in this structure to fill a hole
339 * on 64bit arches, not because its logically part of
345 struct sk_buff *head;
346 struct sk_buff *tail;
348 #define sk_rmem_alloc sk_backlog.rmem_alloc
349 int sk_forward_alloc;
354 #ifdef CONFIG_NET_RX_BUSY_POLL
355 unsigned int sk_napi_id;
356 unsigned int sk_ll_usec;
361 struct sk_filter __rcu *sk_filter;
362 struct socket_wq __rcu *sk_wq;
365 struct xfrm_policy *sk_policy[2];
367 unsigned long sk_flags;
368 struct dst_entry *sk_rx_dst;
369 struct dst_entry __rcu *sk_dst_cache;
370 spinlock_t sk_dst_lock;
371 atomic_t sk_wmem_alloc;
372 atomic_t sk_omem_alloc;
374 struct sk_buff_head sk_write_queue;
375 kmemcheck_bitfield_begin(flags);
376 unsigned int sk_shutdown : 2,
382 kmemcheck_bitfield_end(flags);
385 u32 sk_pacing_rate; /* bytes per second */
386 u32 sk_max_pacing_rate;
387 netdev_features_t sk_route_caps;
388 netdev_features_t sk_route_nocaps;
390 unsigned int sk_gso_max_size;
393 unsigned long sk_lingertime;
394 struct sk_buff_head sk_error_queue;
395 struct proto *sk_prot_creator;
396 rwlock_t sk_callback_lock;
399 unsigned short sk_ack_backlog;
400 unsigned short sk_max_ack_backlog;
402 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
403 __u32 sk_cgrp_prioidx;
405 struct pid *sk_peer_pid;
406 const struct cred *sk_peer_cred;
410 struct timer_list sk_timer;
414 struct socket *sk_socket;
416 struct page_frag sk_frag;
417 struct sk_buff *sk_send_head;
419 int sk_write_pending;
420 #ifdef CONFIG_SECURITY
425 struct cg_proto *sk_cgrp;
426 void (*sk_state_change)(struct sock *sk);
427 void (*sk_data_ready)(struct sock *sk);
428 void (*sk_write_space)(struct sock *sk);
429 void (*sk_error_report)(struct sock *sk);
430 int (*sk_backlog_rcv)(struct sock *sk,
431 struct sk_buff *skb);
432 void (*sk_destruct)(struct sock *sk);
435 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
437 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
438 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
441 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
442 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
443 * on a socket means that the socket will reuse everybody else's port
444 * without looking at the other's sk_reuse value.
447 #define SK_NO_REUSE 0
448 #define SK_CAN_REUSE 1
449 #define SK_FORCE_REUSE 2
451 static inline int sk_peek_offset(struct sock *sk, int flags)
453 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
454 return sk->sk_peek_off;
459 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
461 if (sk->sk_peek_off >= 0) {
462 if (sk->sk_peek_off >= val)
463 sk->sk_peek_off -= val;
469 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
471 if (sk->sk_peek_off >= 0)
472 sk->sk_peek_off += val;
476 * Hashed lists helper routines
478 static inline struct sock *sk_entry(const struct hlist_node *node)
480 return hlist_entry(node, struct sock, sk_node);
483 static inline struct sock *__sk_head(const struct hlist_head *head)
485 return hlist_entry(head->first, struct sock, sk_node);
488 static inline struct sock *sk_head(const struct hlist_head *head)
490 return hlist_empty(head) ? NULL : __sk_head(head);
493 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
495 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
498 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
500 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
503 static inline struct sock *sk_next(const struct sock *sk)
505 return sk->sk_node.next ?
506 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
509 static inline struct sock *sk_nulls_next(const struct sock *sk)
511 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
512 hlist_nulls_entry(sk->sk_nulls_node.next,
513 struct sock, sk_nulls_node) :
517 static inline bool sk_unhashed(const struct sock *sk)
519 return hlist_unhashed(&sk->sk_node);
522 static inline bool sk_hashed(const struct sock *sk)
524 return !sk_unhashed(sk);
527 static inline void sk_node_init(struct hlist_node *node)
532 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
537 static inline void __sk_del_node(struct sock *sk)
539 __hlist_del(&sk->sk_node);
542 /* NB: equivalent to hlist_del_init_rcu */
543 static inline bool __sk_del_node_init(struct sock *sk)
547 sk_node_init(&sk->sk_node);
553 /* Grab socket reference count. This operation is valid only
554 when sk is ALREADY grabbed f.e. it is found in hash table
555 or a list and the lookup is made under lock preventing hash table
559 static inline void sock_hold(struct sock *sk)
561 atomic_inc(&sk->sk_refcnt);
564 /* Ungrab socket in the context, which assumes that socket refcnt
565 cannot hit zero, f.e. it is true in context of any socketcall.
567 static inline void __sock_put(struct sock *sk)
569 atomic_dec(&sk->sk_refcnt);
572 static inline bool sk_del_node_init(struct sock *sk)
574 bool rc = __sk_del_node_init(sk);
577 /* paranoid for a while -acme */
578 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
583 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
585 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
588 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
594 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
596 bool rc = __sk_nulls_del_node_init_rcu(sk);
599 /* paranoid for a while -acme */
600 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
606 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
608 hlist_add_head(&sk->sk_node, list);
611 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
614 __sk_add_node(sk, list);
617 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
620 hlist_add_head_rcu(&sk->sk_node, list);
623 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
625 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
628 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
631 __sk_nulls_add_node_rcu(sk, list);
634 static inline void __sk_del_bind_node(struct sock *sk)
636 __hlist_del(&sk->sk_bind_node);
639 static inline void sk_add_bind_node(struct sock *sk,
640 struct hlist_head *list)
642 hlist_add_head(&sk->sk_bind_node, list);
645 #define sk_for_each(__sk, list) \
646 hlist_for_each_entry(__sk, list, sk_node)
647 #define sk_for_each_rcu(__sk, list) \
648 hlist_for_each_entry_rcu(__sk, list, sk_node)
649 #define sk_nulls_for_each(__sk, node, list) \
650 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
651 #define sk_nulls_for_each_rcu(__sk, node, list) \
652 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
653 #define sk_for_each_from(__sk) \
654 hlist_for_each_entry_from(__sk, sk_node)
655 #define sk_nulls_for_each_from(__sk, node) \
656 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
657 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
658 #define sk_for_each_safe(__sk, tmp, list) \
659 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
660 #define sk_for_each_bound(__sk, list) \
661 hlist_for_each_entry(__sk, list, sk_bind_node)
664 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
665 * @tpos: the type * to use as a loop cursor.
666 * @pos: the &struct hlist_node to use as a loop cursor.
667 * @head: the head for your list.
668 * @offset: offset of hlist_node within the struct.
671 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
672 for (pos = (head)->first; \
673 (!is_a_nulls(pos)) && \
674 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
677 static inline struct user_namespace *sk_user_ns(struct sock *sk)
679 /* Careful only use this in a context where these parameters
680 * can not change and must all be valid, such as recvmsg from
683 return sk->sk_socket->file->f_cred->user_ns;
697 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
698 SOCK_DBG, /* %SO_DEBUG setting */
699 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
700 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
701 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
702 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
703 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
704 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
705 SOCK_FASYNC, /* fasync() active */
707 SOCK_ZEROCOPY, /* buffers from userspace */
708 SOCK_WIFI_STATUS, /* push wifi status to userspace */
709 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
710 * Will use last 4 bytes of packet sent from
711 * user-space instead.
713 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
714 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
717 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
719 nsk->sk_flags = osk->sk_flags;
722 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
724 __set_bit(flag, &sk->sk_flags);
727 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
729 __clear_bit(flag, &sk->sk_flags);
732 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
734 return test_bit(flag, &sk->sk_flags);
738 extern struct static_key memalloc_socks;
739 static inline int sk_memalloc_socks(void)
741 return static_key_false(&memalloc_socks);
745 static inline int sk_memalloc_socks(void)
752 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
754 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
757 static inline void sk_acceptq_removed(struct sock *sk)
759 sk->sk_ack_backlog--;
762 static inline void sk_acceptq_added(struct sock *sk)
764 sk->sk_ack_backlog++;
767 static inline bool sk_acceptq_is_full(const struct sock *sk)
769 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
773 * Compute minimal free write space needed to queue new packets.
775 static inline int sk_stream_min_wspace(const struct sock *sk)
777 return sk->sk_wmem_queued >> 1;
780 static inline int sk_stream_wspace(const struct sock *sk)
782 return sk->sk_sndbuf - sk->sk_wmem_queued;
785 void sk_stream_write_space(struct sock *sk);
787 /* OOB backlog add */
788 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
790 /* dont let skb dst not refcounted, we are going to leave rcu lock */
793 if (!sk->sk_backlog.tail)
794 sk->sk_backlog.head = skb;
796 sk->sk_backlog.tail->next = skb;
798 sk->sk_backlog.tail = skb;
803 * Take into account size of receive queue and backlog queue
804 * Do not take into account this skb truesize,
805 * to allow even a single big packet to come.
807 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
809 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
811 return qsize > limit;
814 /* The per-socket spinlock must be held here. */
815 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
818 if (sk_rcvqueues_full(sk, limit))
821 __sk_add_backlog(sk, skb);
822 sk->sk_backlog.len += skb->truesize;
826 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
828 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
830 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
831 return __sk_backlog_rcv(sk, skb);
833 return sk->sk_backlog_rcv(sk, skb);
836 static inline void sock_rps_record_flow_hash(__u32 hash)
839 struct rps_sock_flow_table *sock_flow_table;
842 sock_flow_table = rcu_dereference(rps_sock_flow_table);
843 rps_record_sock_flow(sock_flow_table, hash);
848 static inline void sock_rps_reset_flow_hash(__u32 hash)
851 struct rps_sock_flow_table *sock_flow_table;
854 sock_flow_table = rcu_dereference(rps_sock_flow_table);
855 rps_reset_sock_flow(sock_flow_table, hash);
860 static inline void sock_rps_record_flow(const struct sock *sk)
863 sock_rps_record_flow_hash(sk->sk_rxhash);
867 static inline void sock_rps_reset_flow(const struct sock *sk)
870 sock_rps_reset_flow_hash(sk->sk_rxhash);
874 static inline void sock_rps_save_rxhash(struct sock *sk,
875 const struct sk_buff *skb)
878 if (unlikely(sk->sk_rxhash != skb->hash)) {
879 sock_rps_reset_flow(sk);
880 sk->sk_rxhash = skb->hash;
885 static inline void sock_rps_reset_rxhash(struct sock *sk)
888 sock_rps_reset_flow(sk);
893 #define sk_wait_event(__sk, __timeo, __condition) \
895 release_sock(__sk); \
896 __rc = __condition; \
898 *(__timeo) = schedule_timeout(*(__timeo)); \
901 __rc = __condition; \
905 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
906 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
907 void sk_stream_wait_close(struct sock *sk, long timeo_p);
908 int sk_stream_error(struct sock *sk, int flags, int err);
909 void sk_stream_kill_queues(struct sock *sk);
910 void sk_set_memalloc(struct sock *sk);
911 void sk_clear_memalloc(struct sock *sk);
913 int sk_wait_data(struct sock *sk, long *timeo);
915 struct request_sock_ops;
916 struct timewait_sock_ops;
917 struct inet_hashinfo;
922 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
923 * un-modified. Special care is taken when initializing object to zero.
925 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
927 if (offsetof(struct sock, sk_node.next) != 0)
928 memset(sk, 0, offsetof(struct sock, sk_node.next));
929 memset(&sk->sk_node.pprev, 0,
930 size - offsetof(struct sock, sk_node.pprev));
933 /* Networking protocol blocks we attach to sockets.
934 * socket layer -> transport layer interface
935 * transport -> network interface is defined by struct inet_proto
938 void (*close)(struct sock *sk,
940 int (*connect)(struct sock *sk,
941 struct sockaddr *uaddr,
943 int (*disconnect)(struct sock *sk, int flags);
945 struct sock * (*accept)(struct sock *sk, int flags, int *err);
947 int (*ioctl)(struct sock *sk, int cmd,
949 int (*init)(struct sock *sk);
950 void (*destroy)(struct sock *sk);
951 void (*shutdown)(struct sock *sk, int how);
952 int (*setsockopt)(struct sock *sk, int level,
953 int optname, char __user *optval,
954 unsigned int optlen);
955 int (*getsockopt)(struct sock *sk, int level,
956 int optname, char __user *optval,
959 int (*compat_setsockopt)(struct sock *sk,
961 int optname, char __user *optval,
962 unsigned int optlen);
963 int (*compat_getsockopt)(struct sock *sk,
965 int optname, char __user *optval,
967 int (*compat_ioctl)(struct sock *sk,
968 unsigned int cmd, unsigned long arg);
970 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
971 struct msghdr *msg, size_t len);
972 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
974 size_t len, int noblock, int flags,
976 int (*sendpage)(struct sock *sk, struct page *page,
977 int offset, size_t size, int flags);
978 int (*bind)(struct sock *sk,
979 struct sockaddr *uaddr, int addr_len);
981 int (*backlog_rcv) (struct sock *sk,
982 struct sk_buff *skb);
984 void (*release_cb)(struct sock *sk);
986 /* Keeping track of sk's, looking them up, and port selection methods. */
987 void (*hash)(struct sock *sk);
988 void (*unhash)(struct sock *sk);
989 void (*rehash)(struct sock *sk);
990 int (*get_port)(struct sock *sk, unsigned short snum);
991 void (*clear_sk)(struct sock *sk, int size);
993 /* Keeping track of sockets in use */
994 #ifdef CONFIG_PROC_FS
995 unsigned int inuse_idx;
998 bool (*stream_memory_free)(const struct sock *sk);
999 /* Memory pressure */
1000 void (*enter_memory_pressure)(struct sock *sk);
1001 atomic_long_t *memory_allocated; /* Current allocated memory. */
1002 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1004 * Pressure flag: try to collapse.
1005 * Technical note: it is used by multiple contexts non atomically.
1006 * All the __sk_mem_schedule() is of this nature: accounting
1007 * is strict, actions are advisory and have some latency.
1009 int *memory_pressure;
1016 struct kmem_cache *slab;
1017 unsigned int obj_size;
1020 struct percpu_counter *orphan_count;
1022 struct request_sock_ops *rsk_prot;
1023 struct timewait_sock_ops *twsk_prot;
1026 struct inet_hashinfo *hashinfo;
1027 struct udp_table *udp_table;
1028 struct raw_hashinfo *raw_hash;
1031 struct module *owner;
1035 struct list_head node;
1036 #ifdef SOCK_REFCNT_DEBUG
1039 #ifdef CONFIG_MEMCG_KMEM
1041 * cgroup specific init/deinit functions. Called once for all
1042 * protocols that implement it, from cgroups populate function.
1043 * This function has to setup any files the protocol want to
1044 * appear in the kmem cgroup filesystem.
1046 int (*init_cgroup)(struct mem_cgroup *memcg,
1047 struct cgroup_subsys *ss);
1048 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1049 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1054 * Bits in struct cg_proto.flags
1056 enum cg_proto_flags {
1057 /* Currently active and new sockets should be assigned to cgroups */
1059 /* It was ever activated; we must disarm static keys on destruction */
1060 MEMCG_SOCK_ACTIVATED,
1064 struct res_counter memory_allocated; /* Current allocated memory. */
1065 struct percpu_counter sockets_allocated; /* Current number of sockets. */
1066 int memory_pressure;
1068 unsigned long flags;
1070 * memcg field is used to find which memcg we belong directly
1071 * Each memcg struct can hold more than one cg_proto, so container_of
1074 * The elegant solution would be having an inverse function to
1075 * proto_cgroup in struct proto, but that means polluting the structure
1076 * for everybody, instead of just for memcg users.
1078 struct mem_cgroup *memcg;
1081 int proto_register(struct proto *prot, int alloc_slab);
1082 void proto_unregister(struct proto *prot);
1084 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1086 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1089 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1091 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1094 #ifdef SOCK_REFCNT_DEBUG
1095 static inline void sk_refcnt_debug_inc(struct sock *sk)
1097 atomic_inc(&sk->sk_prot->socks);
1100 static inline void sk_refcnt_debug_dec(struct sock *sk)
1102 atomic_dec(&sk->sk_prot->socks);
1103 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1104 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1107 static inline void sk_refcnt_debug_release(const struct sock *sk)
1109 if (atomic_read(&sk->sk_refcnt) != 1)
1110 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1111 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1113 #else /* SOCK_REFCNT_DEBUG */
1114 #define sk_refcnt_debug_inc(sk) do { } while (0)
1115 #define sk_refcnt_debug_dec(sk) do { } while (0)
1116 #define sk_refcnt_debug_release(sk) do { } while (0)
1117 #endif /* SOCK_REFCNT_DEBUG */
1119 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1120 extern struct static_key memcg_socket_limit_enabled;
1121 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1122 struct cg_proto *cg_proto)
1124 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1126 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1128 #define mem_cgroup_sockets_enabled 0
1129 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1130 struct cg_proto *cg_proto)
1136 static inline bool sk_stream_memory_free(const struct sock *sk)
1138 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1141 return sk->sk_prot->stream_memory_free ?
1142 sk->sk_prot->stream_memory_free(sk) : true;
1145 static inline bool sk_stream_is_writeable(const struct sock *sk)
1147 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1148 sk_stream_memory_free(sk);
1152 static inline bool sk_has_memory_pressure(const struct sock *sk)
1154 return sk->sk_prot->memory_pressure != NULL;
1157 static inline bool sk_under_memory_pressure(const struct sock *sk)
1159 if (!sk->sk_prot->memory_pressure)
1162 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1163 return !!sk->sk_cgrp->memory_pressure;
1165 return !!*sk->sk_prot->memory_pressure;
1168 static inline void sk_leave_memory_pressure(struct sock *sk)
1170 int *memory_pressure = sk->sk_prot->memory_pressure;
1172 if (!memory_pressure)
1175 if (*memory_pressure)
1176 *memory_pressure = 0;
1178 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1179 struct cg_proto *cg_proto = sk->sk_cgrp;
1180 struct proto *prot = sk->sk_prot;
1182 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1183 cg_proto->memory_pressure = 0;
1188 static inline void sk_enter_memory_pressure(struct sock *sk)
1190 if (!sk->sk_prot->enter_memory_pressure)
1193 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1194 struct cg_proto *cg_proto = sk->sk_cgrp;
1195 struct proto *prot = sk->sk_prot;
1197 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1198 cg_proto->memory_pressure = 1;
1201 sk->sk_prot->enter_memory_pressure(sk);
1204 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1206 long *prot = sk->sk_prot->sysctl_mem;
1207 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1208 prot = sk->sk_cgrp->sysctl_mem;
1212 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1216 struct res_counter *fail;
1219 ret = res_counter_charge_nofail(&prot->memory_allocated,
1220 amt << PAGE_SHIFT, &fail);
1222 *parent_status = OVER_LIMIT;
1225 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1228 res_counter_uncharge(&prot->memory_allocated, amt << PAGE_SHIFT);
1231 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1234 ret = res_counter_read_u64(&prot->memory_allocated, RES_USAGE);
1235 return ret >> PAGE_SHIFT;
1239 sk_memory_allocated(const struct sock *sk)
1241 struct proto *prot = sk->sk_prot;
1242 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1243 return memcg_memory_allocated_read(sk->sk_cgrp);
1245 return atomic_long_read(prot->memory_allocated);
1249 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1251 struct proto *prot = sk->sk_prot;
1253 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1254 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1255 /* update the root cgroup regardless */
1256 atomic_long_add_return(amt, prot->memory_allocated);
1257 return memcg_memory_allocated_read(sk->sk_cgrp);
1260 return atomic_long_add_return(amt, prot->memory_allocated);
1264 sk_memory_allocated_sub(struct sock *sk, int amt)
1266 struct proto *prot = sk->sk_prot;
1268 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1269 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1271 atomic_long_sub(amt, prot->memory_allocated);
1274 static inline void sk_sockets_allocated_dec(struct sock *sk)
1276 struct proto *prot = sk->sk_prot;
1278 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1279 struct cg_proto *cg_proto = sk->sk_cgrp;
1281 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1282 percpu_counter_dec(&cg_proto->sockets_allocated);
1285 percpu_counter_dec(prot->sockets_allocated);
1288 static inline void sk_sockets_allocated_inc(struct sock *sk)
1290 struct proto *prot = sk->sk_prot;
1292 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1293 struct cg_proto *cg_proto = sk->sk_cgrp;
1295 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1296 percpu_counter_inc(&cg_proto->sockets_allocated);
1299 percpu_counter_inc(prot->sockets_allocated);
1303 sk_sockets_allocated_read_positive(struct sock *sk)
1305 struct proto *prot = sk->sk_prot;
1307 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1308 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1310 return percpu_counter_read_positive(prot->sockets_allocated);
1314 proto_sockets_allocated_sum_positive(struct proto *prot)
1316 return percpu_counter_sum_positive(prot->sockets_allocated);
1320 proto_memory_allocated(struct proto *prot)
1322 return atomic_long_read(prot->memory_allocated);
1326 proto_memory_pressure(struct proto *prot)
1328 if (!prot->memory_pressure)
1330 return !!*prot->memory_pressure;
1334 #ifdef CONFIG_PROC_FS
1335 /* Called with local bh disabled */
1336 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1337 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1339 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1346 /* With per-bucket locks this operation is not-atomic, so that
1347 * this version is not worse.
1349 static inline void __sk_prot_rehash(struct sock *sk)
1351 sk->sk_prot->unhash(sk);
1352 sk->sk_prot->hash(sk);
1355 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1357 /* About 10 seconds */
1358 #define SOCK_DESTROY_TIME (10*HZ)
1360 /* Sockets 0-1023 can't be bound to unless you are superuser */
1361 #define PROT_SOCK 1024
1363 #define SHUTDOWN_MASK 3
1364 #define RCV_SHUTDOWN 1
1365 #define SEND_SHUTDOWN 2
1367 #define SOCK_SNDBUF_LOCK 1
1368 #define SOCK_RCVBUF_LOCK 2
1369 #define SOCK_BINDADDR_LOCK 4
1370 #define SOCK_BINDPORT_LOCK 8
1372 /* sock_iocb: used to kick off async processing of socket ios */
1374 struct list_head list;
1378 struct socket *sock;
1380 struct scm_cookie *scm;
1381 struct msghdr *msg, async_msg;
1382 struct kiocb *kiocb;
1385 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1387 return (struct sock_iocb *)iocb->private;
1390 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1395 struct socket_alloc {
1396 struct socket socket;
1397 struct inode vfs_inode;
1400 static inline struct socket *SOCKET_I(struct inode *inode)
1402 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1405 static inline struct inode *SOCK_INODE(struct socket *socket)
1407 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1411 * Functions for memory accounting
1413 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1414 void __sk_mem_reclaim(struct sock *sk);
1416 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1417 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1418 #define SK_MEM_SEND 0
1419 #define SK_MEM_RECV 1
1421 static inline int sk_mem_pages(int amt)
1423 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1426 static inline bool sk_has_account(struct sock *sk)
1428 /* return true if protocol supports memory accounting */
1429 return !!sk->sk_prot->memory_allocated;
1432 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1434 if (!sk_has_account(sk))
1436 return size <= sk->sk_forward_alloc ||
1437 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1441 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1443 if (!sk_has_account(sk))
1445 return size<= sk->sk_forward_alloc ||
1446 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1447 skb_pfmemalloc(skb);
1450 static inline void sk_mem_reclaim(struct sock *sk)
1452 if (!sk_has_account(sk))
1454 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1455 __sk_mem_reclaim(sk);
1458 static inline void sk_mem_reclaim_partial(struct sock *sk)
1460 if (!sk_has_account(sk))
1462 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1463 __sk_mem_reclaim(sk);
1466 static inline void sk_mem_charge(struct sock *sk, int size)
1468 if (!sk_has_account(sk))
1470 sk->sk_forward_alloc -= size;
1473 static inline void sk_mem_uncharge(struct sock *sk, int size)
1475 if (!sk_has_account(sk))
1477 sk->sk_forward_alloc += size;
1480 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1482 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1483 sk->sk_wmem_queued -= skb->truesize;
1484 sk_mem_uncharge(sk, skb->truesize);
1488 /* Used by processes to "lock" a socket state, so that
1489 * interrupts and bottom half handlers won't change it
1490 * from under us. It essentially blocks any incoming
1491 * packets, so that we won't get any new data or any
1492 * packets that change the state of the socket.
1494 * While locked, BH processing will add new packets to
1495 * the backlog queue. This queue is processed by the
1496 * owner of the socket lock right before it is released.
1498 * Since ~2.3.5 it is also exclusive sleep lock serializing
1499 * accesses from user process context.
1501 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1503 static inline void sock_release_ownership(struct sock *sk)
1505 sk->sk_lock.owned = 0;
1509 * Macro so as to not evaluate some arguments when
1510 * lockdep is not enabled.
1512 * Mark both the sk_lock and the sk_lock.slock as a
1513 * per-address-family lock class.
1515 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1517 sk->sk_lock.owned = 0; \
1518 init_waitqueue_head(&sk->sk_lock.wq); \
1519 spin_lock_init(&(sk)->sk_lock.slock); \
1520 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1521 sizeof((sk)->sk_lock)); \
1522 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1524 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1527 void lock_sock_nested(struct sock *sk, int subclass);
1529 static inline void lock_sock(struct sock *sk)
1531 lock_sock_nested(sk, 0);
1534 void release_sock(struct sock *sk);
1536 /* BH context may only use the following locking interface. */
1537 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1538 #define bh_lock_sock_nested(__sk) \
1539 spin_lock_nested(&((__sk)->sk_lock.slock), \
1540 SINGLE_DEPTH_NESTING)
1541 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1543 bool lock_sock_fast(struct sock *sk);
1545 * unlock_sock_fast - complement of lock_sock_fast
1549 * fast unlock socket for user context.
1550 * If slow mode is on, we call regular release_sock()
1552 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1557 spin_unlock_bh(&sk->sk_lock.slock);
1561 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1562 struct proto *prot);
1563 void sk_free(struct sock *sk);
1564 void sk_release_kernel(struct sock *sk);
1565 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1567 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1569 void sock_wfree(struct sk_buff *skb);
1570 void skb_orphan_partial(struct sk_buff *skb);
1571 void sock_rfree(struct sk_buff *skb);
1572 void sock_efree(struct sk_buff *skb);
1574 void sock_edemux(struct sk_buff *skb);
1576 #define sock_edemux(skb) sock_efree(skb)
1579 int sock_setsockopt(struct socket *sock, int level, int op,
1580 char __user *optval, unsigned int optlen);
1582 int sock_getsockopt(struct socket *sock, int level, int op,
1583 char __user *optval, int __user *optlen);
1584 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1585 int noblock, int *errcode);
1586 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1587 unsigned long data_len, int noblock,
1588 int *errcode, int max_page_order);
1589 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1590 void sock_kfree_s(struct sock *sk, void *mem, int size);
1591 void sk_send_sigurg(struct sock *sk);
1594 * Functions to fill in entries in struct proto_ops when a protocol
1595 * does not implement a particular function.
1597 int sock_no_bind(struct socket *, struct sockaddr *, int);
1598 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1599 int sock_no_socketpair(struct socket *, struct socket *);
1600 int sock_no_accept(struct socket *, struct socket *, int);
1601 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1602 unsigned int sock_no_poll(struct file *, struct socket *,
1603 struct poll_table_struct *);
1604 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1605 int sock_no_listen(struct socket *, int);
1606 int sock_no_shutdown(struct socket *, int);
1607 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1608 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1609 int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t);
1610 int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t,
1612 int sock_no_mmap(struct file *file, struct socket *sock,
1613 struct vm_area_struct *vma);
1614 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1615 size_t size, int flags);
1618 * Functions to fill in entries in struct proto_ops when a protocol
1619 * uses the inet style.
1621 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1622 char __user *optval, int __user *optlen);
1623 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1624 struct msghdr *msg, size_t size, int flags);
1625 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1626 char __user *optval, unsigned int optlen);
1627 int compat_sock_common_getsockopt(struct socket *sock, int level,
1628 int optname, char __user *optval, int __user *optlen);
1629 int compat_sock_common_setsockopt(struct socket *sock, int level,
1630 int optname, char __user *optval, unsigned int optlen);
1632 void sk_common_release(struct sock *sk);
1635 * Default socket callbacks and setup code
1638 /* Initialise core socket variables */
1639 void sock_init_data(struct socket *sock, struct sock *sk);
1642 * Socket reference counting postulates.
1644 * * Each user of socket SHOULD hold a reference count.
1645 * * Each access point to socket (an hash table bucket, reference from a list,
1646 * running timer, skb in flight MUST hold a reference count.
1647 * * When reference count hits 0, it means it will never increase back.
1648 * * When reference count hits 0, it means that no references from
1649 * outside exist to this socket and current process on current CPU
1650 * is last user and may/should destroy this socket.
1651 * * sk_free is called from any context: process, BH, IRQ. When
1652 * it is called, socket has no references from outside -> sk_free
1653 * may release descendant resources allocated by the socket, but
1654 * to the time when it is called, socket is NOT referenced by any
1655 * hash tables, lists etc.
1656 * * Packets, delivered from outside (from network or from another process)
1657 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1658 * when they sit in queue. Otherwise, packets will leak to hole, when
1659 * socket is looked up by one cpu and unhasing is made by another CPU.
1660 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1661 * (leak to backlog). Packet socket does all the processing inside
1662 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1663 * use separate SMP lock, so that they are prone too.
1666 /* Ungrab socket and destroy it, if it was the last reference. */
1667 static inline void sock_put(struct sock *sk)
1669 if (atomic_dec_and_test(&sk->sk_refcnt))
1672 /* Generic version of sock_put(), dealing with all sockets
1673 * (TCP_TIMEWAIT, ESTABLISHED...)
1675 void sock_gen_put(struct sock *sk);
1677 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1679 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1681 sk->sk_tx_queue_mapping = tx_queue;
1684 static inline void sk_tx_queue_clear(struct sock *sk)
1686 sk->sk_tx_queue_mapping = -1;
1689 static inline int sk_tx_queue_get(const struct sock *sk)
1691 return sk ? sk->sk_tx_queue_mapping : -1;
1694 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1696 sk_tx_queue_clear(sk);
1697 sk->sk_socket = sock;
1700 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1702 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1703 return &rcu_dereference_raw(sk->sk_wq)->wait;
1705 /* Detach socket from process context.
1706 * Announce socket dead, detach it from wait queue and inode.
1707 * Note that parent inode held reference count on this struct sock,
1708 * we do not release it in this function, because protocol
1709 * probably wants some additional cleanups or even continuing
1710 * to work with this socket (TCP).
1712 static inline void sock_orphan(struct sock *sk)
1714 write_lock_bh(&sk->sk_callback_lock);
1715 sock_set_flag(sk, SOCK_DEAD);
1716 sk_set_socket(sk, NULL);
1718 write_unlock_bh(&sk->sk_callback_lock);
1721 static inline void sock_graft(struct sock *sk, struct socket *parent)
1723 write_lock_bh(&sk->sk_callback_lock);
1724 sk->sk_wq = parent->wq;
1726 sk_set_socket(sk, parent);
1727 security_sock_graft(sk, parent);
1728 write_unlock_bh(&sk->sk_callback_lock);
1731 kuid_t sock_i_uid(struct sock *sk);
1732 unsigned long sock_i_ino(struct sock *sk);
1734 static inline struct dst_entry *
1735 __sk_dst_get(struct sock *sk)
1737 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1738 lockdep_is_held(&sk->sk_lock.slock));
1741 static inline struct dst_entry *
1742 sk_dst_get(struct sock *sk)
1744 struct dst_entry *dst;
1747 dst = rcu_dereference(sk->sk_dst_cache);
1748 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1754 static inline void dst_negative_advice(struct sock *sk)
1756 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1758 if (dst && dst->ops->negative_advice) {
1759 ndst = dst->ops->negative_advice(dst);
1762 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1763 sk_tx_queue_clear(sk);
1769 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1771 struct dst_entry *old_dst;
1773 sk_tx_queue_clear(sk);
1775 * This can be called while sk is owned by the caller only,
1776 * with no state that can be checked in a rcu_dereference_check() cond
1778 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1779 rcu_assign_pointer(sk->sk_dst_cache, dst);
1780 dst_release(old_dst);
1784 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1786 struct dst_entry *old_dst;
1788 sk_tx_queue_clear(sk);
1789 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1790 dst_release(old_dst);
1794 __sk_dst_reset(struct sock *sk)
1796 __sk_dst_set(sk, NULL);
1800 sk_dst_reset(struct sock *sk)
1802 sk_dst_set(sk, NULL);
1805 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1807 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1809 static inline bool sk_can_gso(const struct sock *sk)
1811 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1814 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1816 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1818 sk->sk_route_nocaps |= flags;
1819 sk->sk_route_caps &= ~flags;
1822 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1823 char __user *from, char *to,
1824 int copy, int offset)
1826 if (skb->ip_summed == CHECKSUM_NONE) {
1828 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1831 skb->csum = csum_block_add(skb->csum, csum, offset);
1832 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1833 if (!access_ok(VERIFY_READ, from, copy) ||
1834 __copy_from_user_nocache(to, from, copy))
1836 } else if (copy_from_user(to, from, copy))
1842 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1843 char __user *from, int copy)
1845 int err, offset = skb->len;
1847 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1850 __skb_trim(skb, offset);
1855 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1856 struct sk_buff *skb,
1862 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1868 skb->data_len += copy;
1869 skb->truesize += copy;
1870 sk->sk_wmem_queued += copy;
1871 sk_mem_charge(sk, copy);
1875 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1876 struct sk_buff *skb, struct page *page,
1879 if (skb->ip_summed == CHECKSUM_NONE) {
1881 __wsum csum = csum_and_copy_from_user(from,
1882 page_address(page) + off,
1886 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1887 } else if (copy_from_user(page_address(page) + off, from, copy))
1891 skb->data_len += copy;
1892 skb->truesize += copy;
1893 sk->sk_wmem_queued += copy;
1894 sk_mem_charge(sk, copy);
1899 * sk_wmem_alloc_get - returns write allocations
1902 * Returns sk_wmem_alloc minus initial offset of one
1904 static inline int sk_wmem_alloc_get(const struct sock *sk)
1906 return atomic_read(&sk->sk_wmem_alloc) - 1;
1910 * sk_rmem_alloc_get - returns read allocations
1913 * Returns sk_rmem_alloc
1915 static inline int sk_rmem_alloc_get(const struct sock *sk)
1917 return atomic_read(&sk->sk_rmem_alloc);
1921 * sk_has_allocations - check if allocations are outstanding
1924 * Returns true if socket has write or read allocations
1926 static inline bool sk_has_allocations(const struct sock *sk)
1928 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1932 * wq_has_sleeper - check if there are any waiting processes
1933 * @wq: struct socket_wq
1935 * Returns true if socket_wq has waiting processes
1937 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1938 * barrier call. They were added due to the race found within the tcp code.
1940 * Consider following tcp code paths:
1944 * sys_select receive packet
1946 * __add_wait_queue update tp->rcv_nxt
1948 * tp->rcv_nxt check sock_def_readable
1950 * schedule rcu_read_lock();
1951 * wq = rcu_dereference(sk->sk_wq);
1952 * if (wq && waitqueue_active(&wq->wait))
1953 * wake_up_interruptible(&wq->wait)
1957 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1958 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1959 * could then endup calling schedule and sleep forever if there are no more
1960 * data on the socket.
1963 static inline bool wq_has_sleeper(struct socket_wq *wq)
1965 /* We need to be sure we are in sync with the
1966 * add_wait_queue modifications to the wait queue.
1968 * This memory barrier is paired in the sock_poll_wait.
1971 return wq && waitqueue_active(&wq->wait);
1975 * sock_poll_wait - place memory barrier behind the poll_wait call.
1977 * @wait_address: socket wait queue
1980 * See the comments in the wq_has_sleeper function.
1982 static inline void sock_poll_wait(struct file *filp,
1983 wait_queue_head_t *wait_address, poll_table *p)
1985 if (!poll_does_not_wait(p) && wait_address) {
1986 poll_wait(filp, wait_address, p);
1987 /* We need to be sure we are in sync with the
1988 * socket flags modification.
1990 * This memory barrier is paired in the wq_has_sleeper.
1996 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1998 if (sk->sk_txhash) {
2000 skb->hash = sk->sk_txhash;
2005 * Queue a received datagram if it will fit. Stream and sequenced
2006 * protocols can't normally use this as they need to fit buffers in
2007 * and play with them.
2009 * Inlined as it's very short and called for pretty much every
2010 * packet ever received.
2013 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2017 skb->destructor = sock_wfree;
2018 skb_set_hash_from_sk(skb, sk);
2020 * We used to take a refcount on sk, but following operation
2021 * is enough to guarantee sk_free() wont free this sock until
2022 * all in-flight packets are completed
2024 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
2027 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2031 skb->destructor = sock_rfree;
2032 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2033 sk_mem_charge(sk, skb->truesize);
2036 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2037 unsigned long expires);
2039 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2041 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2043 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2044 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2047 * Recover an error report and clear atomically
2050 static inline int sock_error(struct sock *sk)
2053 if (likely(!sk->sk_err))
2055 err = xchg(&sk->sk_err, 0);
2059 static inline unsigned long sock_wspace(struct sock *sk)
2063 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2064 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2071 static inline void sk_wake_async(struct sock *sk, int how, int band)
2073 if (sock_flag(sk, SOCK_FASYNC))
2074 sock_wake_async(sk->sk_socket, how, band);
2077 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2078 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2079 * Note: for send buffers, TCP works better if we can build two skbs at
2082 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2084 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2085 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2087 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2089 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2090 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2091 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2095 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2098 * sk_page_frag - return an appropriate page_frag
2101 * If socket allocation mode allows current thread to sleep, it means its
2102 * safe to use the per task page_frag instead of the per socket one.
2104 static inline struct page_frag *sk_page_frag(struct sock *sk)
2106 if (sk->sk_allocation & __GFP_WAIT)
2107 return ¤t->task_frag;
2109 return &sk->sk_frag;
2112 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2115 * Default write policy as shown to user space via poll/select/SIGIO
2117 static inline bool sock_writeable(const struct sock *sk)
2119 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2122 static inline gfp_t gfp_any(void)
2124 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2127 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2129 return noblock ? 0 : sk->sk_rcvtimeo;
2132 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2134 return noblock ? 0 : sk->sk_sndtimeo;
2137 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2139 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2142 /* Alas, with timeout socket operations are not restartable.
2143 * Compare this to poll().
2145 static inline int sock_intr_errno(long timeo)
2147 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2150 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2151 struct sk_buff *skb);
2152 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2153 struct sk_buff *skb);
2156 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2158 ktime_t kt = skb->tstamp;
2159 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2162 * generate control messages if
2163 * - receive time stamping in software requested
2164 * - software time stamp available and wanted
2165 * - hardware time stamps available and wanted
2167 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2168 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2169 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2170 (hwtstamps->hwtstamp.tv64 &&
2171 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2172 __sock_recv_timestamp(msg, sk, skb);
2176 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2177 __sock_recv_wifi_status(msg, sk, skb);
2180 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2181 struct sk_buff *skb);
2183 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2184 struct sk_buff *skb)
2186 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2187 (1UL << SOCK_RCVTSTAMP))
2188 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2189 SOF_TIMESTAMPING_RAW_HARDWARE)
2191 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2192 __sock_recv_ts_and_drops(msg, sk, skb);
2194 sk->sk_stamp = skb->tstamp;
2197 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2200 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2201 * @sk: socket sending this packet
2202 * @tx_flags: completed with instructions for time stamping
2204 * Note : callers should take care of initial *tx_flags value (usually 0)
2206 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2208 if (unlikely(sk->sk_tsflags))
2209 __sock_tx_timestamp(sk, tx_flags);
2210 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2211 *tx_flags |= SKBTX_WIFI_STATUS;
2215 * sk_eat_skb - Release a skb if it is no longer needed
2216 * @sk: socket to eat this skb from
2217 * @skb: socket buffer to eat
2219 * This routine must be called with interrupts disabled or with the socket
2220 * locked so that the sk_buff queue operation is ok.
2222 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2224 __skb_unlink(skb, &sk->sk_receive_queue);
2229 struct net *sock_net(const struct sock *sk)
2231 return read_pnet(&sk->sk_net);
2235 void sock_net_set(struct sock *sk, struct net *net)
2237 write_pnet(&sk->sk_net, net);
2241 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2242 * They should not hold a reference to a namespace in order to allow
2244 * Sockets after sk_change_net should be released using sk_release_kernel
2246 static inline void sk_change_net(struct sock *sk, struct net *net)
2248 struct net *current_net = sock_net(sk);
2250 if (!net_eq(current_net, net)) {
2251 put_net(current_net);
2252 sock_net_set(sk, hold_net(net));
2256 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2259 struct sock *sk = skb->sk;
2261 skb->destructor = NULL;
2268 void sock_enable_timestamp(struct sock *sk, int flag);
2269 int sock_get_timestamp(struct sock *, struct timeval __user *);
2270 int sock_get_timestampns(struct sock *, struct timespec __user *);
2271 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2274 bool sk_ns_capable(const struct sock *sk,
2275 struct user_namespace *user_ns, int cap);
2276 bool sk_capable(const struct sock *sk, int cap);
2277 bool sk_net_capable(const struct sock *sk, int cap);
2280 * Enable debug/info messages
2282 extern int net_msg_warn;
2283 #define NETDEBUG(fmt, args...) \
2284 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2286 #define LIMIT_NETDEBUG(fmt, args...) \
2287 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2289 extern __u32 sysctl_wmem_max;
2290 extern __u32 sysctl_rmem_max;
2292 extern int sysctl_optmem_max;
2294 extern __u32 sysctl_wmem_default;
2295 extern __u32 sysctl_rmem_default;
2297 #endif /* _SOCK_H */