2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the AF_INET socket handler.
8 * Version: @(#)sock.h 1.0.4 05/13/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 * David S. Miller : New socket lookup architecture.
26 * Steve Whitehouse: Default routines for sock_ops
27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
28 * protinfo be just a void pointer, as the
29 * protocol specific parts were moved to
30 * respective headers and ipv4/v6, etc now
31 * use private slabcaches for its socks
32 * Pedro Hortas : New flags field for socket options
35 * This program is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU General Public License
37 * as published by the Free Software Foundation; either version
38 * 2 of the License, or (at your option) any later version.
43 #include <linux/hardirq.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/list_nulls.h>
47 #include <linux/timer.h>
48 #include <linux/cache.h>
49 #include <linux/bitops.h>
50 #include <linux/lockdep.h>
51 #include <linux/netdevice.h>
52 #include <linux/skbuff.h> /* struct sk_buff */
54 #include <linux/security.h>
55 #include <linux/slab.h>
56 #include <linux/uaccess.h>
57 #include <linux/memcontrol.h>
58 #include <linux/res_counter.h>
59 #include <linux/static_key.h>
60 #include <linux/aio.h>
61 #include <linux/sched.h>
63 #include <linux/filter.h>
64 #include <linux/rculist_nulls.h>
65 #include <linux/poll.h>
67 #include <linux/atomic.h>
69 #include <net/checksum.h>
74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
88 * This structure really needs to be cleaned up.
89 * Most of it is for TCP, and not used by any of
90 * the other protocols.
93 /* Define this to get the SOCK_DBG debugging facility. */
94 #define SOCK_DEBUGGING
96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
97 printk(KERN_DEBUG msg); } while (0)
99 /* Validate arguments and do nothing */
100 static inline __printf(2, 3)
101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
106 /* This is the per-socket lock. The spinlock provides a synchronization
107 * between user contexts and software interrupt processing, whereas the
108 * mini-semaphore synchronizes multiple users amongst themselves.
113 wait_queue_head_t wq;
115 * We express the mutex-alike socket_lock semantics
116 * to the lock validator by explicitly managing
117 * the slock as a lock variant (in addition to
120 #ifdef CONFIG_DEBUG_LOCK_ALLOC
121 struct lockdep_map dep_map;
129 typedef __u32 __bitwise __portpair;
130 typedef __u64 __bitwise __addrpair;
133 * struct sock_common - minimal network layer representation of sockets
134 * @skc_daddr: Foreign IPv4 addr
135 * @skc_rcv_saddr: Bound local IPv4 addr
136 * @skc_hash: hash value used with various protocol lookup tables
137 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
138 * @skc_dport: placeholder for inet_dport/tw_dport
139 * @skc_num: placeholder for inet_num/tw_num
140 * @skc_family: network address family
141 * @skc_state: Connection state
142 * @skc_reuse: %SO_REUSEADDR setting
143 * @skc_reuseport: %SO_REUSEPORT setting
144 * @skc_bound_dev_if: bound device index if != 0
145 * @skc_bind_node: bind hash linkage for various protocol lookup tables
146 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
147 * @skc_prot: protocol handlers inside a network family
148 * @skc_net: reference to the network namespace of this socket
149 * @skc_node: main hash linkage for various protocol lookup tables
150 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
151 * @skc_tx_queue_mapping: tx queue number for this connection
152 * @skc_refcnt: reference count
154 * This is the minimal network layer representation of sockets, the header
155 * for struct sock and struct inet_timewait_sock.
158 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
159 * address on 64bit arches : cf INET_MATCH() and INET_TW_MATCH()
162 __addrpair skc_addrpair;
165 __be32 skc_rcv_saddr;
169 unsigned int skc_hash;
170 __u16 skc_u16hashes[2];
172 /* skc_dport && skc_num must be grouped as well */
174 __portpair skc_portpair;
181 unsigned short skc_family;
182 volatile unsigned char skc_state;
183 unsigned char skc_reuse:4;
184 unsigned char skc_reuseport:4;
185 int skc_bound_dev_if;
187 struct hlist_node skc_bind_node;
188 struct hlist_nulls_node skc_portaddr_node;
190 struct proto *skc_prot;
195 * fields between dontcopy_begin/dontcopy_end
196 * are not copied in sock_copy()
199 int skc_dontcopy_begin[0];
202 struct hlist_node skc_node;
203 struct hlist_nulls_node skc_nulls_node;
205 int skc_tx_queue_mapping;
208 int skc_dontcopy_end[0];
214 * struct sock - network layer representation of sockets
215 * @__sk_common: shared layout with inet_timewait_sock
216 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
217 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
218 * @sk_lock: synchronizer
219 * @sk_rcvbuf: size of receive buffer in bytes
220 * @sk_wq: sock wait queue and async head
221 * @sk_rx_dst: receive input route used by early tcp demux
222 * @sk_dst_cache: destination cache
223 * @sk_dst_lock: destination cache lock
224 * @sk_policy: flow policy
225 * @sk_receive_queue: incoming packets
226 * @sk_wmem_alloc: transmit queue bytes committed
227 * @sk_write_queue: Packet sending queue
228 * @sk_async_wait_queue: DMA copied packets
229 * @sk_omem_alloc: "o" is "option" or "other"
230 * @sk_wmem_queued: persistent queue size
231 * @sk_forward_alloc: space allocated forward
232 * @sk_allocation: allocation mode
233 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
234 * @sk_sndbuf: size of send buffer in bytes
235 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
236 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
237 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets
238 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
239 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
240 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
241 * @sk_gso_max_size: Maximum GSO segment size to build
242 * @sk_gso_max_segs: Maximum number of GSO segments
243 * @sk_lingertime: %SO_LINGER l_linger setting
244 * @sk_backlog: always used with the per-socket spinlock held
245 * @sk_callback_lock: used with the callbacks in the end of this struct
246 * @sk_error_queue: rarely used
247 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
248 * IPV6_ADDRFORM for instance)
249 * @sk_err: last error
250 * @sk_err_soft: errors that don't cause failure but are the cause of a
251 * persistent failure not just 'timed out'
252 * @sk_drops: raw/udp drops counter
253 * @sk_ack_backlog: current listen backlog
254 * @sk_max_ack_backlog: listen backlog set in listen()
255 * @sk_priority: %SO_PRIORITY setting
256 * @sk_cgrp_prioidx: socket group's priority map index
257 * @sk_type: socket type (%SOCK_STREAM, etc)
258 * @sk_protocol: which protocol this socket belongs in this network family
259 * @sk_peer_pid: &struct pid for this socket's peer
260 * @sk_peer_cred: %SO_PEERCRED setting
261 * @sk_rcvlowat: %SO_RCVLOWAT setting
262 * @sk_rcvtimeo: %SO_RCVTIMEO setting
263 * @sk_sndtimeo: %SO_SNDTIMEO setting
264 * @sk_rxhash: flow hash received from netif layer
265 * @sk_filter: socket filtering instructions
266 * @sk_protinfo: private area, net family specific, when not using slab
267 * @sk_timer: sock cleanup timer
268 * @sk_stamp: time stamp of last packet received
269 * @sk_socket: Identd and reporting IO signals
270 * @sk_user_data: RPC layer private data
271 * @sk_frag: cached page frag
272 * @sk_peek_off: current peek_offset value
273 * @sk_send_head: front of stuff to transmit
274 * @sk_security: used by security modules
275 * @sk_mark: generic packet mark
276 * @sk_classid: this socket's cgroup classid
277 * @sk_cgrp: this socket's cgroup-specific proto data
278 * @sk_write_pending: a write to stream socket waits to start
279 * @sk_state_change: callback to indicate change in the state of the sock
280 * @sk_data_ready: callback to indicate there is data to be processed
281 * @sk_write_space: callback to indicate there is bf sending space available
282 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
283 * @sk_backlog_rcv: callback to process the backlog
284 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
288 * Now struct inet_timewait_sock also uses sock_common, so please just
289 * don't add nothing before this first member (__sk_common) --acme
291 struct sock_common __sk_common;
292 #define sk_node __sk_common.skc_node
293 #define sk_nulls_node __sk_common.skc_nulls_node
294 #define sk_refcnt __sk_common.skc_refcnt
295 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
297 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
298 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
299 #define sk_hash __sk_common.skc_hash
300 #define sk_family __sk_common.skc_family
301 #define sk_state __sk_common.skc_state
302 #define sk_reuse __sk_common.skc_reuse
303 #define sk_reuseport __sk_common.skc_reuseport
304 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
305 #define sk_bind_node __sk_common.skc_bind_node
306 #define sk_prot __sk_common.skc_prot
307 #define sk_net __sk_common.skc_net
308 socket_lock_t sk_lock;
309 struct sk_buff_head sk_receive_queue;
311 * The backlog queue is special, it is always used with
312 * the per-socket spinlock held and requires low latency
313 * access. Therefore we special case it's implementation.
314 * Note : rmem_alloc is in this structure to fill a hole
315 * on 64bit arches, not because its logically part of
321 struct sk_buff *head;
322 struct sk_buff *tail;
324 #define sk_rmem_alloc sk_backlog.rmem_alloc
325 int sk_forward_alloc;
332 struct sk_filter __rcu *sk_filter;
333 struct socket_wq __rcu *sk_wq;
335 #ifdef CONFIG_NET_DMA
336 struct sk_buff_head sk_async_wait_queue;
340 struct xfrm_policy *sk_policy[2];
342 unsigned long sk_flags;
343 struct dst_entry *sk_rx_dst;
344 struct dst_entry __rcu *sk_dst_cache;
345 spinlock_t sk_dst_lock;
346 atomic_t sk_wmem_alloc;
347 atomic_t sk_omem_alloc;
349 struct sk_buff_head sk_write_queue;
350 kmemcheck_bitfield_begin(flags);
351 unsigned int sk_shutdown : 2,
356 kmemcheck_bitfield_end(flags);
359 u32 sk_pacing_rate; /* bytes per second */
360 netdev_features_t sk_route_caps;
361 netdev_features_t sk_route_nocaps;
363 unsigned int sk_gso_max_size;
366 unsigned long sk_lingertime;
367 struct sk_buff_head sk_error_queue;
368 struct proto *sk_prot_creator;
369 rwlock_t sk_callback_lock;
372 unsigned short sk_ack_backlog;
373 unsigned short sk_max_ack_backlog;
375 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
376 __u32 sk_cgrp_prioidx;
378 struct pid *sk_peer_pid;
379 const struct cred *sk_peer_cred;
383 struct timer_list sk_timer;
385 struct socket *sk_socket;
387 struct page_frag sk_frag;
388 struct sk_buff *sk_send_head;
390 int sk_write_pending;
391 #ifdef CONFIG_SECURITY
396 struct cg_proto *sk_cgrp;
397 void (*sk_state_change)(struct sock *sk);
398 void (*sk_data_ready)(struct sock *sk, int bytes);
399 void (*sk_write_space)(struct sock *sk);
400 void (*sk_error_report)(struct sock *sk);
401 int (*sk_backlog_rcv)(struct sock *sk,
402 struct sk_buff *skb);
403 void (*sk_destruct)(struct sock *sk);
407 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
408 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
409 * on a socket means that the socket will reuse everybody else's port
410 * without looking at the other's sk_reuse value.
413 #define SK_NO_REUSE 0
414 #define SK_CAN_REUSE 1
415 #define SK_FORCE_REUSE 2
417 static inline int sk_peek_offset(struct sock *sk, int flags)
419 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
420 return sk->sk_peek_off;
425 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
427 if (sk->sk_peek_off >= 0) {
428 if (sk->sk_peek_off >= val)
429 sk->sk_peek_off -= val;
435 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
437 if (sk->sk_peek_off >= 0)
438 sk->sk_peek_off += val;
442 * Hashed lists helper routines
444 static inline struct sock *sk_entry(const struct hlist_node *node)
446 return hlist_entry(node, struct sock, sk_node);
449 static inline struct sock *__sk_head(const struct hlist_head *head)
451 return hlist_entry(head->first, struct sock, sk_node);
454 static inline struct sock *sk_head(const struct hlist_head *head)
456 return hlist_empty(head) ? NULL : __sk_head(head);
459 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
461 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
464 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
466 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
469 static inline struct sock *sk_next(const struct sock *sk)
471 return sk->sk_node.next ?
472 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
475 static inline struct sock *sk_nulls_next(const struct sock *sk)
477 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
478 hlist_nulls_entry(sk->sk_nulls_node.next,
479 struct sock, sk_nulls_node) :
483 static inline bool sk_unhashed(const struct sock *sk)
485 return hlist_unhashed(&sk->sk_node);
488 static inline bool sk_hashed(const struct sock *sk)
490 return !sk_unhashed(sk);
493 static inline void sk_node_init(struct hlist_node *node)
498 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
503 static inline void __sk_del_node(struct sock *sk)
505 __hlist_del(&sk->sk_node);
508 /* NB: equivalent to hlist_del_init_rcu */
509 static inline bool __sk_del_node_init(struct sock *sk)
513 sk_node_init(&sk->sk_node);
519 /* Grab socket reference count. This operation is valid only
520 when sk is ALREADY grabbed f.e. it is found in hash table
521 or a list and the lookup is made under lock preventing hash table
525 static inline void sock_hold(struct sock *sk)
527 atomic_inc(&sk->sk_refcnt);
530 /* Ungrab socket in the context, which assumes that socket refcnt
531 cannot hit zero, f.e. it is true in context of any socketcall.
533 static inline void __sock_put(struct sock *sk)
535 atomic_dec(&sk->sk_refcnt);
538 static inline bool sk_del_node_init(struct sock *sk)
540 bool rc = __sk_del_node_init(sk);
543 /* paranoid for a while -acme */
544 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
549 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
551 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
554 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
560 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
562 bool rc = __sk_nulls_del_node_init_rcu(sk);
565 /* paranoid for a while -acme */
566 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
572 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
574 hlist_add_head(&sk->sk_node, list);
577 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
580 __sk_add_node(sk, list);
583 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
586 hlist_add_head_rcu(&sk->sk_node, list);
589 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
591 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
594 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
597 __sk_nulls_add_node_rcu(sk, list);
600 static inline void __sk_del_bind_node(struct sock *sk)
602 __hlist_del(&sk->sk_bind_node);
605 static inline void sk_add_bind_node(struct sock *sk,
606 struct hlist_head *list)
608 hlist_add_head(&sk->sk_bind_node, list);
611 #define sk_for_each(__sk, list) \
612 hlist_for_each_entry(__sk, list, sk_node)
613 #define sk_for_each_rcu(__sk, list) \
614 hlist_for_each_entry_rcu(__sk, list, sk_node)
615 #define sk_nulls_for_each(__sk, node, list) \
616 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
617 #define sk_nulls_for_each_rcu(__sk, node, list) \
618 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
619 #define sk_for_each_from(__sk) \
620 hlist_for_each_entry_from(__sk, sk_node)
621 #define sk_nulls_for_each_from(__sk, node) \
622 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
623 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
624 #define sk_for_each_safe(__sk, tmp, list) \
625 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
626 #define sk_for_each_bound(__sk, list) \
627 hlist_for_each_entry(__sk, list, sk_bind_node)
629 static inline struct user_namespace *sk_user_ns(struct sock *sk)
631 /* Careful only use this in a context where these parameters
632 * can not change and must all be valid, such as recvmsg from
635 return sk->sk_socket->file->f_cred->user_ns;
649 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
650 SOCK_DBG, /* %SO_DEBUG setting */
651 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
652 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
653 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
654 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
655 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
656 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
657 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
658 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
659 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
660 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
661 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
662 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
663 SOCK_FASYNC, /* fasync() active */
665 SOCK_ZEROCOPY, /* buffers from userspace */
666 SOCK_WIFI_STATUS, /* push wifi status to userspace */
667 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
668 * Will use last 4 bytes of packet sent from
669 * user-space instead.
671 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
672 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
675 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
677 nsk->sk_flags = osk->sk_flags;
680 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
682 __set_bit(flag, &sk->sk_flags);
685 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
687 __clear_bit(flag, &sk->sk_flags);
690 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
692 return test_bit(flag, &sk->sk_flags);
696 extern struct static_key memalloc_socks;
697 static inline int sk_memalloc_socks(void)
699 return static_key_false(&memalloc_socks);
703 static inline int sk_memalloc_socks(void)
710 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
712 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
715 static inline void sk_acceptq_removed(struct sock *sk)
717 sk->sk_ack_backlog--;
720 static inline void sk_acceptq_added(struct sock *sk)
722 sk->sk_ack_backlog++;
725 static inline bool sk_acceptq_is_full(const struct sock *sk)
727 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
731 * Compute minimal free write space needed to queue new packets.
733 static inline int sk_stream_min_wspace(const struct sock *sk)
735 return sk->sk_wmem_queued >> 1;
738 static inline int sk_stream_wspace(const struct sock *sk)
740 return sk->sk_sndbuf - sk->sk_wmem_queued;
743 extern void sk_stream_write_space(struct sock *sk);
745 static inline bool sk_stream_memory_free(const struct sock *sk)
747 return sk->sk_wmem_queued < sk->sk_sndbuf;
750 /* OOB backlog add */
751 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
753 /* dont let skb dst not refcounted, we are going to leave rcu lock */
756 if (!sk->sk_backlog.tail)
757 sk->sk_backlog.head = skb;
759 sk->sk_backlog.tail->next = skb;
761 sk->sk_backlog.tail = skb;
766 * Take into account size of receive queue and backlog queue
767 * Do not take into account this skb truesize,
768 * to allow even a single big packet to come.
770 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
773 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
775 return qsize > limit;
778 /* The per-socket spinlock must be held here. */
779 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
782 if (sk_rcvqueues_full(sk, skb, limit))
785 __sk_add_backlog(sk, skb);
786 sk->sk_backlog.len += skb->truesize;
790 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
792 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
794 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
795 return __sk_backlog_rcv(sk, skb);
797 return sk->sk_backlog_rcv(sk, skb);
800 static inline void sock_rps_record_flow(const struct sock *sk)
803 struct rps_sock_flow_table *sock_flow_table;
806 sock_flow_table = rcu_dereference(rps_sock_flow_table);
807 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
812 static inline void sock_rps_reset_flow(const struct sock *sk)
815 struct rps_sock_flow_table *sock_flow_table;
818 sock_flow_table = rcu_dereference(rps_sock_flow_table);
819 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
824 static inline void sock_rps_save_rxhash(struct sock *sk,
825 const struct sk_buff *skb)
828 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
829 sock_rps_reset_flow(sk);
830 sk->sk_rxhash = skb->rxhash;
835 static inline void sock_rps_reset_rxhash(struct sock *sk)
838 sock_rps_reset_flow(sk);
843 #define sk_wait_event(__sk, __timeo, __condition) \
845 release_sock(__sk); \
846 __rc = __condition; \
848 *(__timeo) = schedule_timeout(*(__timeo)); \
851 __rc = __condition; \
855 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
856 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
857 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
858 extern int sk_stream_error(struct sock *sk, int flags, int err);
859 extern void sk_stream_kill_queues(struct sock *sk);
860 extern void sk_set_memalloc(struct sock *sk);
861 extern void sk_clear_memalloc(struct sock *sk);
863 extern int sk_wait_data(struct sock *sk, long *timeo);
865 struct request_sock_ops;
866 struct timewait_sock_ops;
867 struct inet_hashinfo;
872 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
873 * un-modified. Special care is taken when initializing object to zero.
875 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
877 if (offsetof(struct sock, sk_node.next) != 0)
878 memset(sk, 0, offsetof(struct sock, sk_node.next));
879 memset(&sk->sk_node.pprev, 0,
880 size - offsetof(struct sock, sk_node.pprev));
883 /* Networking protocol blocks we attach to sockets.
884 * socket layer -> transport layer interface
885 * transport -> network interface is defined by struct inet_proto
888 void (*close)(struct sock *sk,
890 int (*connect)(struct sock *sk,
891 struct sockaddr *uaddr,
893 int (*disconnect)(struct sock *sk, int flags);
895 struct sock * (*accept)(struct sock *sk, int flags, int *err);
897 int (*ioctl)(struct sock *sk, int cmd,
899 int (*init)(struct sock *sk);
900 void (*destroy)(struct sock *sk);
901 void (*shutdown)(struct sock *sk, int how);
902 int (*setsockopt)(struct sock *sk, int level,
903 int optname, char __user *optval,
904 unsigned int optlen);
905 int (*getsockopt)(struct sock *sk, int level,
906 int optname, char __user *optval,
909 int (*compat_setsockopt)(struct sock *sk,
911 int optname, char __user *optval,
912 unsigned int optlen);
913 int (*compat_getsockopt)(struct sock *sk,
915 int optname, char __user *optval,
917 int (*compat_ioctl)(struct sock *sk,
918 unsigned int cmd, unsigned long arg);
920 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
921 struct msghdr *msg, size_t len);
922 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
924 size_t len, int noblock, int flags,
926 int (*sendpage)(struct sock *sk, struct page *page,
927 int offset, size_t size, int flags);
928 int (*bind)(struct sock *sk,
929 struct sockaddr *uaddr, int addr_len);
931 int (*backlog_rcv) (struct sock *sk,
932 struct sk_buff *skb);
934 void (*release_cb)(struct sock *sk);
936 /* Keeping track of sk's, looking them up, and port selection methods. */
937 void (*hash)(struct sock *sk);
938 void (*unhash)(struct sock *sk);
939 void (*rehash)(struct sock *sk);
940 int (*get_port)(struct sock *sk, unsigned short snum);
941 void (*clear_sk)(struct sock *sk, int size);
943 /* Keeping track of sockets in use */
944 #ifdef CONFIG_PROC_FS
945 unsigned int inuse_idx;
948 /* Memory pressure */
949 void (*enter_memory_pressure)(struct sock *sk);
950 atomic_long_t *memory_allocated; /* Current allocated memory. */
951 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
953 * Pressure flag: try to collapse.
954 * Technical note: it is used by multiple contexts non atomically.
955 * All the __sk_mem_schedule() is of this nature: accounting
956 * is strict, actions are advisory and have some latency.
958 int *memory_pressure;
965 struct kmem_cache *slab;
966 unsigned int obj_size;
969 struct percpu_counter *orphan_count;
971 struct request_sock_ops *rsk_prot;
972 struct timewait_sock_ops *twsk_prot;
975 struct inet_hashinfo *hashinfo;
976 struct udp_table *udp_table;
977 struct raw_hashinfo *raw_hash;
980 struct module *owner;
984 struct list_head node;
985 #ifdef SOCK_REFCNT_DEBUG
988 #ifdef CONFIG_MEMCG_KMEM
990 * cgroup specific init/deinit functions. Called once for all
991 * protocols that implement it, from cgroups populate function.
992 * This function has to setup any files the protocol want to
993 * appear in the kmem cgroup filesystem.
995 int (*init_cgroup)(struct mem_cgroup *memcg,
996 struct cgroup_subsys *ss);
997 void (*destroy_cgroup)(struct mem_cgroup *memcg);
998 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1003 * Bits in struct cg_proto.flags
1005 enum cg_proto_flags {
1006 /* Currently active and new sockets should be assigned to cgroups */
1008 /* It was ever activated; we must disarm static keys on destruction */
1009 MEMCG_SOCK_ACTIVATED,
1013 void (*enter_memory_pressure)(struct sock *sk);
1014 struct res_counter *memory_allocated; /* Current allocated memory. */
1015 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1016 int *memory_pressure;
1018 unsigned long flags;
1020 * memcg field is used to find which memcg we belong directly
1021 * Each memcg struct can hold more than one cg_proto, so container_of
1024 * The elegant solution would be having an inverse function to
1025 * proto_cgroup in struct proto, but that means polluting the structure
1026 * for everybody, instead of just for memcg users.
1028 struct mem_cgroup *memcg;
1031 extern int proto_register(struct proto *prot, int alloc_slab);
1032 extern void proto_unregister(struct proto *prot);
1034 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1036 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1039 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1041 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1044 #ifdef SOCK_REFCNT_DEBUG
1045 static inline void sk_refcnt_debug_inc(struct sock *sk)
1047 atomic_inc(&sk->sk_prot->socks);
1050 static inline void sk_refcnt_debug_dec(struct sock *sk)
1052 atomic_dec(&sk->sk_prot->socks);
1053 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1054 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1057 static inline void sk_refcnt_debug_release(const struct sock *sk)
1059 if (atomic_read(&sk->sk_refcnt) != 1)
1060 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1061 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1063 #else /* SOCK_REFCNT_DEBUG */
1064 #define sk_refcnt_debug_inc(sk) do { } while (0)
1065 #define sk_refcnt_debug_dec(sk) do { } while (0)
1066 #define sk_refcnt_debug_release(sk) do { } while (0)
1067 #endif /* SOCK_REFCNT_DEBUG */
1069 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1070 extern struct static_key memcg_socket_limit_enabled;
1071 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1072 struct cg_proto *cg_proto)
1074 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1076 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1078 #define mem_cgroup_sockets_enabled 0
1079 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1080 struct cg_proto *cg_proto)
1087 static inline bool sk_has_memory_pressure(const struct sock *sk)
1089 return sk->sk_prot->memory_pressure != NULL;
1092 static inline bool sk_under_memory_pressure(const struct sock *sk)
1094 if (!sk->sk_prot->memory_pressure)
1097 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1098 return !!*sk->sk_cgrp->memory_pressure;
1100 return !!*sk->sk_prot->memory_pressure;
1103 static inline void sk_leave_memory_pressure(struct sock *sk)
1105 int *memory_pressure = sk->sk_prot->memory_pressure;
1107 if (!memory_pressure)
1110 if (*memory_pressure)
1111 *memory_pressure = 0;
1113 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1114 struct cg_proto *cg_proto = sk->sk_cgrp;
1115 struct proto *prot = sk->sk_prot;
1117 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1118 if (*cg_proto->memory_pressure)
1119 *cg_proto->memory_pressure = 0;
1124 static inline void sk_enter_memory_pressure(struct sock *sk)
1126 if (!sk->sk_prot->enter_memory_pressure)
1129 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1130 struct cg_proto *cg_proto = sk->sk_cgrp;
1131 struct proto *prot = sk->sk_prot;
1133 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1134 cg_proto->enter_memory_pressure(sk);
1137 sk->sk_prot->enter_memory_pressure(sk);
1140 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1142 long *prot = sk->sk_prot->sysctl_mem;
1143 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1144 prot = sk->sk_cgrp->sysctl_mem;
1148 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1152 struct res_counter *fail;
1155 ret = res_counter_charge_nofail(prot->memory_allocated,
1156 amt << PAGE_SHIFT, &fail);
1158 *parent_status = OVER_LIMIT;
1161 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1164 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1167 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1170 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1171 return ret >> PAGE_SHIFT;
1175 sk_memory_allocated(const struct sock *sk)
1177 struct proto *prot = sk->sk_prot;
1178 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1179 return memcg_memory_allocated_read(sk->sk_cgrp);
1181 return atomic_long_read(prot->memory_allocated);
1185 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1187 struct proto *prot = sk->sk_prot;
1189 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1190 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1191 /* update the root cgroup regardless */
1192 atomic_long_add_return(amt, prot->memory_allocated);
1193 return memcg_memory_allocated_read(sk->sk_cgrp);
1196 return atomic_long_add_return(amt, prot->memory_allocated);
1200 sk_memory_allocated_sub(struct sock *sk, int amt)
1202 struct proto *prot = sk->sk_prot;
1204 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1205 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1207 atomic_long_sub(amt, prot->memory_allocated);
1210 static inline void sk_sockets_allocated_dec(struct sock *sk)
1212 struct proto *prot = sk->sk_prot;
1214 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1215 struct cg_proto *cg_proto = sk->sk_cgrp;
1217 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1218 percpu_counter_dec(cg_proto->sockets_allocated);
1221 percpu_counter_dec(prot->sockets_allocated);
1224 static inline void sk_sockets_allocated_inc(struct sock *sk)
1226 struct proto *prot = sk->sk_prot;
1228 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1229 struct cg_proto *cg_proto = sk->sk_cgrp;
1231 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1232 percpu_counter_inc(cg_proto->sockets_allocated);
1235 percpu_counter_inc(prot->sockets_allocated);
1239 sk_sockets_allocated_read_positive(struct sock *sk)
1241 struct proto *prot = sk->sk_prot;
1243 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1244 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1246 return percpu_counter_read_positive(prot->sockets_allocated);
1250 proto_sockets_allocated_sum_positive(struct proto *prot)
1252 return percpu_counter_sum_positive(prot->sockets_allocated);
1256 proto_memory_allocated(struct proto *prot)
1258 return atomic_long_read(prot->memory_allocated);
1262 proto_memory_pressure(struct proto *prot)
1264 if (!prot->memory_pressure)
1266 return !!*prot->memory_pressure;
1270 #ifdef CONFIG_PROC_FS
1271 /* Called with local bh disabled */
1272 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1273 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1275 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1282 /* With per-bucket locks this operation is not-atomic, so that
1283 * this version is not worse.
1285 static inline void __sk_prot_rehash(struct sock *sk)
1287 sk->sk_prot->unhash(sk);
1288 sk->sk_prot->hash(sk);
1291 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1293 /* About 10 seconds */
1294 #define SOCK_DESTROY_TIME (10*HZ)
1296 /* Sockets 0-1023 can't be bound to unless you are superuser */
1297 #define PROT_SOCK 1024
1299 #define SHUTDOWN_MASK 3
1300 #define RCV_SHUTDOWN 1
1301 #define SEND_SHUTDOWN 2
1303 #define SOCK_SNDBUF_LOCK 1
1304 #define SOCK_RCVBUF_LOCK 2
1305 #define SOCK_BINDADDR_LOCK 4
1306 #define SOCK_BINDPORT_LOCK 8
1308 /* sock_iocb: used to kick off async processing of socket ios */
1310 struct list_head list;
1314 struct socket *sock;
1316 struct scm_cookie *scm;
1317 struct msghdr *msg, async_msg;
1318 struct kiocb *kiocb;
1321 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1323 return (struct sock_iocb *)iocb->private;
1326 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1331 struct socket_alloc {
1332 struct socket socket;
1333 struct inode vfs_inode;
1336 static inline struct socket *SOCKET_I(struct inode *inode)
1338 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1341 static inline struct inode *SOCK_INODE(struct socket *socket)
1343 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1347 * Functions for memory accounting
1349 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1350 extern void __sk_mem_reclaim(struct sock *sk);
1352 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1353 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1354 #define SK_MEM_SEND 0
1355 #define SK_MEM_RECV 1
1357 static inline int sk_mem_pages(int amt)
1359 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1362 static inline bool sk_has_account(struct sock *sk)
1364 /* return true if protocol supports memory accounting */
1365 return !!sk->sk_prot->memory_allocated;
1368 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1370 if (!sk_has_account(sk))
1372 return size <= sk->sk_forward_alloc ||
1373 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1377 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1379 if (!sk_has_account(sk))
1381 return size<= sk->sk_forward_alloc ||
1382 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1383 skb_pfmemalloc(skb);
1386 static inline void sk_mem_reclaim(struct sock *sk)
1388 if (!sk_has_account(sk))
1390 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1391 __sk_mem_reclaim(sk);
1394 static inline void sk_mem_reclaim_partial(struct sock *sk)
1396 if (!sk_has_account(sk))
1398 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1399 __sk_mem_reclaim(sk);
1402 static inline void sk_mem_charge(struct sock *sk, int size)
1404 if (!sk_has_account(sk))
1406 sk->sk_forward_alloc -= size;
1409 static inline void sk_mem_uncharge(struct sock *sk, int size)
1411 if (!sk_has_account(sk))
1413 sk->sk_forward_alloc += size;
1416 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1418 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1419 sk->sk_wmem_queued -= skb->truesize;
1420 sk_mem_uncharge(sk, skb->truesize);
1424 /* Used by processes to "lock" a socket state, so that
1425 * interrupts and bottom half handlers won't change it
1426 * from under us. It essentially blocks any incoming
1427 * packets, so that we won't get any new data or any
1428 * packets that change the state of the socket.
1430 * While locked, BH processing will add new packets to
1431 * the backlog queue. This queue is processed by the
1432 * owner of the socket lock right before it is released.
1434 * Since ~2.3.5 it is also exclusive sleep lock serializing
1435 * accesses from user process context.
1437 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1439 static inline void sock_release_ownership(struct sock *sk)
1441 sk->sk_lock.owned = 0;
1445 * Macro so as to not evaluate some arguments when
1446 * lockdep is not enabled.
1448 * Mark both the sk_lock and the sk_lock.slock as a
1449 * per-address-family lock class.
1451 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1453 sk->sk_lock.owned = 0; \
1454 init_waitqueue_head(&sk->sk_lock.wq); \
1455 spin_lock_init(&(sk)->sk_lock.slock); \
1456 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1457 sizeof((sk)->sk_lock)); \
1458 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1460 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1463 extern void lock_sock_nested(struct sock *sk, int subclass);
1465 static inline void lock_sock(struct sock *sk)
1467 lock_sock_nested(sk, 0);
1470 extern void release_sock(struct sock *sk);
1472 /* BH context may only use the following locking interface. */
1473 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1474 #define bh_lock_sock_nested(__sk) \
1475 spin_lock_nested(&((__sk)->sk_lock.slock), \
1476 SINGLE_DEPTH_NESTING)
1477 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1479 extern bool lock_sock_fast(struct sock *sk);
1481 * unlock_sock_fast - complement of lock_sock_fast
1485 * fast unlock socket for user context.
1486 * If slow mode is on, we call regular release_sock()
1488 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1493 spin_unlock_bh(&sk->sk_lock.slock);
1497 extern struct sock *sk_alloc(struct net *net, int family,
1499 struct proto *prot);
1500 extern void sk_free(struct sock *sk);
1501 extern void sk_release_kernel(struct sock *sk);
1502 extern struct sock *sk_clone_lock(const struct sock *sk,
1503 const gfp_t priority);
1505 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1506 unsigned long size, int force,
1508 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1509 unsigned long size, int force,
1511 extern void sock_wfree(struct sk_buff *skb);
1512 extern void sock_rfree(struct sk_buff *skb);
1513 extern void sock_edemux(struct sk_buff *skb);
1515 extern int sock_setsockopt(struct socket *sock, int level,
1516 int op, char __user *optval,
1517 unsigned int optlen);
1519 extern int sock_getsockopt(struct socket *sock, int level,
1520 int op, char __user *optval,
1521 int __user *optlen);
1522 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1526 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1527 unsigned long header_len,
1528 unsigned long data_len,
1531 extern void *sock_kmalloc(struct sock *sk, int size,
1533 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1534 extern void sk_send_sigurg(struct sock *sk);
1537 * Functions to fill in entries in struct proto_ops when a protocol
1538 * does not implement a particular function.
1540 extern int sock_no_bind(struct socket *,
1541 struct sockaddr *, int);
1542 extern int sock_no_connect(struct socket *,
1543 struct sockaddr *, int, int);
1544 extern int sock_no_socketpair(struct socket *,
1546 extern int sock_no_accept(struct socket *,
1547 struct socket *, int);
1548 extern int sock_no_getname(struct socket *,
1549 struct sockaddr *, int *, int);
1550 extern unsigned int sock_no_poll(struct file *, struct socket *,
1551 struct poll_table_struct *);
1552 extern int sock_no_ioctl(struct socket *, unsigned int,
1554 extern int sock_no_listen(struct socket *, int);
1555 extern int sock_no_shutdown(struct socket *, int);
1556 extern int sock_no_getsockopt(struct socket *, int , int,
1557 char __user *, int __user *);
1558 extern int sock_no_setsockopt(struct socket *, int, int,
1559 char __user *, unsigned int);
1560 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1561 struct msghdr *, size_t);
1562 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1563 struct msghdr *, size_t, int);
1564 extern int sock_no_mmap(struct file *file,
1565 struct socket *sock,
1566 struct vm_area_struct *vma);
1567 extern ssize_t sock_no_sendpage(struct socket *sock,
1569 int offset, size_t size,
1573 * Functions to fill in entries in struct proto_ops when a protocol
1574 * uses the inet style.
1576 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1577 char __user *optval, int __user *optlen);
1578 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1579 struct msghdr *msg, size_t size, int flags);
1580 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1581 char __user *optval, unsigned int optlen);
1582 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1583 int optname, char __user *optval, int __user *optlen);
1584 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1585 int optname, char __user *optval, unsigned int optlen);
1587 extern void sk_common_release(struct sock *sk);
1590 * Default socket callbacks and setup code
1593 /* Initialise core socket variables */
1594 extern void sock_init_data(struct socket *sock, struct sock *sk);
1596 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1599 * sk_filter_release - release a socket filter
1600 * @fp: filter to remove
1602 * Remove a filter from a socket and release its resources.
1605 static inline void sk_filter_release(struct sk_filter *fp)
1607 if (atomic_dec_and_test(&fp->refcnt))
1608 call_rcu(&fp->rcu, sk_filter_release_rcu);
1611 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1613 unsigned int size = sk_filter_len(fp);
1615 atomic_sub(size, &sk->sk_omem_alloc);
1616 sk_filter_release(fp);
1619 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1621 atomic_inc(&fp->refcnt);
1622 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1626 * Socket reference counting postulates.
1628 * * Each user of socket SHOULD hold a reference count.
1629 * * Each access point to socket (an hash table bucket, reference from a list,
1630 * running timer, skb in flight MUST hold a reference count.
1631 * * When reference count hits 0, it means it will never increase back.
1632 * * When reference count hits 0, it means that no references from
1633 * outside exist to this socket and current process on current CPU
1634 * is last user and may/should destroy this socket.
1635 * * sk_free is called from any context: process, BH, IRQ. When
1636 * it is called, socket has no references from outside -> sk_free
1637 * may release descendant resources allocated by the socket, but
1638 * to the time when it is called, socket is NOT referenced by any
1639 * hash tables, lists etc.
1640 * * Packets, delivered from outside (from network or from another process)
1641 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1642 * when they sit in queue. Otherwise, packets will leak to hole, when
1643 * socket is looked up by one cpu and unhasing is made by another CPU.
1644 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1645 * (leak to backlog). Packet socket does all the processing inside
1646 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1647 * use separate SMP lock, so that they are prone too.
1650 /* Ungrab socket and destroy it, if it was the last reference. */
1651 static inline void sock_put(struct sock *sk)
1653 if (atomic_dec_and_test(&sk->sk_refcnt))
1657 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1660 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1662 sk->sk_tx_queue_mapping = tx_queue;
1665 static inline void sk_tx_queue_clear(struct sock *sk)
1667 sk->sk_tx_queue_mapping = -1;
1670 static inline int sk_tx_queue_get(const struct sock *sk)
1672 return sk ? sk->sk_tx_queue_mapping : -1;
1675 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1677 sk_tx_queue_clear(sk);
1678 sk->sk_socket = sock;
1681 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1683 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1684 return &rcu_dereference_raw(sk->sk_wq)->wait;
1686 /* Detach socket from process context.
1687 * Announce socket dead, detach it from wait queue and inode.
1688 * Note that parent inode held reference count on this struct sock,
1689 * we do not release it in this function, because protocol
1690 * probably wants some additional cleanups or even continuing
1691 * to work with this socket (TCP).
1693 static inline void sock_orphan(struct sock *sk)
1695 write_lock_bh(&sk->sk_callback_lock);
1696 sock_set_flag(sk, SOCK_DEAD);
1697 sk_set_socket(sk, NULL);
1699 write_unlock_bh(&sk->sk_callback_lock);
1702 static inline void sock_graft(struct sock *sk, struct socket *parent)
1704 write_lock_bh(&sk->sk_callback_lock);
1705 sk->sk_wq = parent->wq;
1707 sk_set_socket(sk, parent);
1708 security_sock_graft(sk, parent);
1709 write_unlock_bh(&sk->sk_callback_lock);
1712 extern kuid_t sock_i_uid(struct sock *sk);
1713 extern unsigned long sock_i_ino(struct sock *sk);
1715 static inline struct dst_entry *
1716 __sk_dst_get(struct sock *sk)
1718 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1719 lockdep_is_held(&sk->sk_lock.slock));
1722 static inline struct dst_entry *
1723 sk_dst_get(struct sock *sk)
1725 struct dst_entry *dst;
1728 dst = rcu_dereference(sk->sk_dst_cache);
1729 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1735 extern void sk_reset_txq(struct sock *sk);
1737 static inline void dst_negative_advice(struct sock *sk)
1739 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1741 if (dst && dst->ops->negative_advice) {
1742 ndst = dst->ops->negative_advice(dst);
1745 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1752 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1754 struct dst_entry *old_dst;
1756 sk_tx_queue_clear(sk);
1758 * This can be called while sk is owned by the caller only,
1759 * with no state that can be checked in a rcu_dereference_check() cond
1761 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1762 rcu_assign_pointer(sk->sk_dst_cache, dst);
1763 dst_release(old_dst);
1767 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1769 struct dst_entry *old_dst;
1771 sk_tx_queue_clear(sk);
1772 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1773 dst_release(old_dst);
1777 __sk_dst_reset(struct sock *sk)
1779 __sk_dst_set(sk, NULL);
1783 sk_dst_reset(struct sock *sk)
1785 sk_dst_set(sk, NULL);
1788 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1790 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1792 static inline bool sk_can_gso(const struct sock *sk)
1794 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1797 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1799 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1801 sk->sk_route_nocaps |= flags;
1802 sk->sk_route_caps &= ~flags;
1805 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1806 char __user *from, char *to,
1807 int copy, int offset)
1809 if (skb->ip_summed == CHECKSUM_NONE) {
1811 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1814 skb->csum = csum_block_add(skb->csum, csum, offset);
1815 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1816 if (!access_ok(VERIFY_READ, from, copy) ||
1817 __copy_from_user_nocache(to, from, copy))
1819 } else if (copy_from_user(to, from, copy))
1825 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1826 char __user *from, int copy)
1828 int err, offset = skb->len;
1830 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1833 __skb_trim(skb, offset);
1838 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1839 struct sk_buff *skb,
1845 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1851 skb->data_len += copy;
1852 skb->truesize += copy;
1853 sk->sk_wmem_queued += copy;
1854 sk_mem_charge(sk, copy);
1858 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1859 struct sk_buff *skb, struct page *page,
1862 if (skb->ip_summed == CHECKSUM_NONE) {
1864 __wsum csum = csum_and_copy_from_user(from,
1865 page_address(page) + off,
1869 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1870 } else if (copy_from_user(page_address(page) + off, from, copy))
1874 skb->data_len += copy;
1875 skb->truesize += copy;
1876 sk->sk_wmem_queued += copy;
1877 sk_mem_charge(sk, copy);
1882 * sk_wmem_alloc_get - returns write allocations
1885 * Returns sk_wmem_alloc minus initial offset of one
1887 static inline int sk_wmem_alloc_get(const struct sock *sk)
1889 return atomic_read(&sk->sk_wmem_alloc) - 1;
1893 * sk_rmem_alloc_get - returns read allocations
1896 * Returns sk_rmem_alloc
1898 static inline int sk_rmem_alloc_get(const struct sock *sk)
1900 return atomic_read(&sk->sk_rmem_alloc);
1904 * sk_has_allocations - check if allocations are outstanding
1907 * Returns true if socket has write or read allocations
1909 static inline bool sk_has_allocations(const struct sock *sk)
1911 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1915 * wq_has_sleeper - check if there are any waiting processes
1916 * @wq: struct socket_wq
1918 * Returns true if socket_wq has waiting processes
1920 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1921 * barrier call. They were added due to the race found within the tcp code.
1923 * Consider following tcp code paths:
1927 * sys_select receive packet
1929 * __add_wait_queue update tp->rcv_nxt
1931 * tp->rcv_nxt check sock_def_readable
1933 * schedule rcu_read_lock();
1934 * wq = rcu_dereference(sk->sk_wq);
1935 * if (wq && waitqueue_active(&wq->wait))
1936 * wake_up_interruptible(&wq->wait)
1940 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1941 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1942 * could then endup calling schedule and sleep forever if there are no more
1943 * data on the socket.
1946 static inline bool wq_has_sleeper(struct socket_wq *wq)
1948 /* We need to be sure we are in sync with the
1949 * add_wait_queue modifications to the wait queue.
1951 * This memory barrier is paired in the sock_poll_wait.
1954 return wq && waitqueue_active(&wq->wait);
1958 * sock_poll_wait - place memory barrier behind the poll_wait call.
1960 * @wait_address: socket wait queue
1963 * See the comments in the wq_has_sleeper function.
1965 static inline void sock_poll_wait(struct file *filp,
1966 wait_queue_head_t *wait_address, poll_table *p)
1968 if (!poll_does_not_wait(p) && wait_address) {
1969 poll_wait(filp, wait_address, p);
1970 /* We need to be sure we are in sync with the
1971 * socket flags modification.
1973 * This memory barrier is paired in the wq_has_sleeper.
1980 * Queue a received datagram if it will fit. Stream and sequenced
1981 * protocols can't normally use this as they need to fit buffers in
1982 * and play with them.
1984 * Inlined as it's very short and called for pretty much every
1985 * packet ever received.
1988 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1992 skb->destructor = sock_wfree;
1994 * We used to take a refcount on sk, but following operation
1995 * is enough to guarantee sk_free() wont free this sock until
1996 * all in-flight packets are completed
1998 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
2001 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2005 skb->destructor = sock_rfree;
2006 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2007 sk_mem_charge(sk, skb->truesize);
2010 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2011 unsigned long expires);
2013 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2015 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2017 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2020 * Recover an error report and clear atomically
2023 static inline int sock_error(struct sock *sk)
2026 if (likely(!sk->sk_err))
2028 err = xchg(&sk->sk_err, 0);
2032 static inline unsigned long sock_wspace(struct sock *sk)
2036 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2037 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2044 static inline void sk_wake_async(struct sock *sk, int how, int band)
2046 if (sock_flag(sk, SOCK_FASYNC))
2047 sock_wake_async(sk->sk_socket, how, band);
2050 #define SOCK_MIN_SNDBUF 2048
2052 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
2053 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
2055 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
2057 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2059 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2060 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2061 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2065 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2068 * sk_page_frag - return an appropriate page_frag
2071 * If socket allocation mode allows current thread to sleep, it means its
2072 * safe to use the per task page_frag instead of the per socket one.
2074 static inline struct page_frag *sk_page_frag(struct sock *sk)
2076 if (sk->sk_allocation & __GFP_WAIT)
2077 return ¤t->task_frag;
2079 return &sk->sk_frag;
2082 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2085 * Default write policy as shown to user space via poll/select/SIGIO
2087 static inline bool sock_writeable(const struct sock *sk)
2089 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2092 static inline gfp_t gfp_any(void)
2094 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2097 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2099 return noblock ? 0 : sk->sk_rcvtimeo;
2102 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2104 return noblock ? 0 : sk->sk_sndtimeo;
2107 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2109 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2112 /* Alas, with timeout socket operations are not restartable.
2113 * Compare this to poll().
2115 static inline int sock_intr_errno(long timeo)
2117 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2120 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2121 struct sk_buff *skb);
2122 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2123 struct sk_buff *skb);
2126 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2128 ktime_t kt = skb->tstamp;
2129 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2132 * generate control messages if
2133 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2134 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2135 * - software time stamp available and wanted
2136 * (SOCK_TIMESTAMPING_SOFTWARE)
2137 * - hardware time stamps available and wanted
2138 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2139 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2141 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2142 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2143 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2144 (hwtstamps->hwtstamp.tv64 &&
2145 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2146 (hwtstamps->syststamp.tv64 &&
2147 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2148 __sock_recv_timestamp(msg, sk, skb);
2152 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2153 __sock_recv_wifi_status(msg, sk, skb);
2156 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2157 struct sk_buff *skb);
2159 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2160 struct sk_buff *skb)
2162 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2163 (1UL << SOCK_RCVTSTAMP) | \
2164 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2165 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2166 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2167 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2169 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2170 __sock_recv_ts_and_drops(msg, sk, skb);
2172 sk->sk_stamp = skb->tstamp;
2176 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2177 * @sk: socket sending this packet
2178 * @tx_flags: filled with instructions for time stamping
2180 * Currently only depends on SOCK_TIMESTAMPING* flags.
2182 extern void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2185 * sk_eat_skb - Release a skb if it is no longer needed
2186 * @sk: socket to eat this skb from
2187 * @skb: socket buffer to eat
2188 * @copied_early: flag indicating whether DMA operations copied this data early
2190 * This routine must be called with interrupts disabled or with the socket
2191 * locked so that the sk_buff queue operation is ok.
2193 #ifdef CONFIG_NET_DMA
2194 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2196 __skb_unlink(skb, &sk->sk_receive_queue);
2200 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2203 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2205 __skb_unlink(skb, &sk->sk_receive_queue);
2211 struct net *sock_net(const struct sock *sk)
2213 return read_pnet(&sk->sk_net);
2217 void sock_net_set(struct sock *sk, struct net *net)
2219 write_pnet(&sk->sk_net, net);
2223 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2224 * They should not hold a reference to a namespace in order to allow
2226 * Sockets after sk_change_net should be released using sk_release_kernel
2228 static inline void sk_change_net(struct sock *sk, struct net *net)
2230 put_net(sock_net(sk));
2231 sock_net_set(sk, hold_net(net));
2234 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2237 struct sock *sk = skb->sk;
2239 skb->destructor = NULL;
2246 extern void sock_enable_timestamp(struct sock *sk, int flag);
2247 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2248 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2250 bool sk_ns_capable(const struct sock *sk,
2251 struct user_namespace *user_ns, int cap);
2252 bool sk_capable(const struct sock *sk, int cap);
2253 bool sk_net_capable(const struct sock *sk, int cap);
2256 * Enable debug/info messages
2258 extern int net_msg_warn;
2259 #define NETDEBUG(fmt, args...) \
2260 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2262 #define LIMIT_NETDEBUG(fmt, args...) \
2263 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2265 extern __u32 sysctl_wmem_max;
2266 extern __u32 sysctl_rmem_max;
2268 extern int sysctl_optmem_max;
2270 extern __u32 sysctl_wmem_default;
2271 extern __u32 sysctl_rmem_default;
2273 #endif /* _SOCK_H */