Loading...
1/*
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.
5 *
6 * Definitions for the AF_INET socket handler.
7 *
8 * Version: @(#)sock.h 1.0.4 05/13/93
9 *
10 * Authors: Ross Biro
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>
14 *
15 * Fixes:
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
19 * than the reverse.
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
33 *
34 *
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.
39 */
40#ifndef _SOCK_H
41#define _SOCK_H
42
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/module.h>
50#include <linux/lockdep.h>
51#include <linux/netdevice.h>
52#include <linux/skbuff.h> /* struct sk_buff */
53#include <linux/mm.h>
54#include <linux/security.h>
55#include <linux/slab.h>
56#include <linux/uaccess.h>
57
58#include <linux/filter.h>
59#include <linux/rculist_nulls.h>
60#include <linux/poll.h>
61
62#include <linux/atomic.h>
63#include <net/dst.h>
64#include <net/checksum.h>
65
66/*
67 * This structure really needs to be cleaned up.
68 * Most of it is for TCP, and not used by any of
69 * the other protocols.
70 */
71
72/* Define this to get the SOCK_DBG debugging facility. */
73#define SOCK_DEBUGGING
74#ifdef SOCK_DEBUGGING
75#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
76 printk(KERN_DEBUG msg); } while (0)
77#else
78/* Validate arguments and do nothing */
79static inline void __attribute__ ((format (printf, 2, 3)))
80SOCK_DEBUG(struct sock *sk, const char *msg, ...)
81{
82}
83#endif
84
85/* This is the per-socket lock. The spinlock provides a synchronization
86 * between user contexts and software interrupt processing, whereas the
87 * mini-semaphore synchronizes multiple users amongst themselves.
88 */
89typedef struct {
90 spinlock_t slock;
91 int owned;
92 wait_queue_head_t wq;
93 /*
94 * We express the mutex-alike socket_lock semantics
95 * to the lock validator by explicitly managing
96 * the slock as a lock variant (in addition to
97 * the slock itself):
98 */
99#ifdef CONFIG_DEBUG_LOCK_ALLOC
100 struct lockdep_map dep_map;
101#endif
102} socket_lock_t;
103
104struct sock;
105struct proto;
106struct net;
107
108/**
109 * struct sock_common - minimal network layer representation of sockets
110 * @skc_daddr: Foreign IPv4 addr
111 * @skc_rcv_saddr: Bound local IPv4 addr
112 * @skc_hash: hash value used with various protocol lookup tables
113 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
114 * @skc_family: network address family
115 * @skc_state: Connection state
116 * @skc_reuse: %SO_REUSEADDR setting
117 * @skc_bound_dev_if: bound device index if != 0
118 * @skc_bind_node: bind hash linkage for various protocol lookup tables
119 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
120 * @skc_prot: protocol handlers inside a network family
121 * @skc_net: reference to the network namespace of this socket
122 * @skc_node: main hash linkage for various protocol lookup tables
123 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
124 * @skc_tx_queue_mapping: tx queue number for this connection
125 * @skc_refcnt: reference count
126 *
127 * This is the minimal network layer representation of sockets, the header
128 * for struct sock and struct inet_timewait_sock.
129 */
130struct sock_common {
131 /* skc_daddr and skc_rcv_saddr must be grouped :
132 * cf INET_MATCH() and INET_TW_MATCH()
133 */
134 __be32 skc_daddr;
135 __be32 skc_rcv_saddr;
136
137 union {
138 unsigned int skc_hash;
139 __u16 skc_u16hashes[2];
140 };
141 unsigned short skc_family;
142 volatile unsigned char skc_state;
143 unsigned char skc_reuse;
144 int skc_bound_dev_if;
145 union {
146 struct hlist_node skc_bind_node;
147 struct hlist_nulls_node skc_portaddr_node;
148 };
149 struct proto *skc_prot;
150#ifdef CONFIG_NET_NS
151 struct net *skc_net;
152#endif
153 /*
154 * fields between dontcopy_begin/dontcopy_end
155 * are not copied in sock_copy()
156 */
157 /* private: */
158 int skc_dontcopy_begin[0];
159 /* public: */
160 union {
161 struct hlist_node skc_node;
162 struct hlist_nulls_node skc_nulls_node;
163 };
164 int skc_tx_queue_mapping;
165 atomic_t skc_refcnt;
166 /* private: */
167 int skc_dontcopy_end[0];
168 /* public: */
169};
170
171/**
172 * struct sock - network layer representation of sockets
173 * @__sk_common: shared layout with inet_timewait_sock
174 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
175 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
176 * @sk_lock: synchronizer
177 * @sk_rcvbuf: size of receive buffer in bytes
178 * @sk_wq: sock wait queue and async head
179 * @sk_dst_cache: destination cache
180 * @sk_dst_lock: destination cache lock
181 * @sk_policy: flow policy
182 * @sk_receive_queue: incoming packets
183 * @sk_wmem_alloc: transmit queue bytes committed
184 * @sk_write_queue: Packet sending queue
185 * @sk_async_wait_queue: DMA copied packets
186 * @sk_omem_alloc: "o" is "option" or "other"
187 * @sk_wmem_queued: persistent queue size
188 * @sk_forward_alloc: space allocated forward
189 * @sk_allocation: allocation mode
190 * @sk_sndbuf: size of send buffer in bytes
191 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
192 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
193 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
194 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
195 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
196 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
197 * @sk_gso_max_size: Maximum GSO segment size to build
198 * @sk_lingertime: %SO_LINGER l_linger setting
199 * @sk_backlog: always used with the per-socket spinlock held
200 * @sk_callback_lock: used with the callbacks in the end of this struct
201 * @sk_error_queue: rarely used
202 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
203 * IPV6_ADDRFORM for instance)
204 * @sk_err: last error
205 * @sk_err_soft: errors that don't cause failure but are the cause of a
206 * persistent failure not just 'timed out'
207 * @sk_drops: raw/udp drops counter
208 * @sk_ack_backlog: current listen backlog
209 * @sk_max_ack_backlog: listen backlog set in listen()
210 * @sk_priority: %SO_PRIORITY setting
211 * @sk_type: socket type (%SOCK_STREAM, etc)
212 * @sk_protocol: which protocol this socket belongs in this network family
213 * @sk_peer_pid: &struct pid for this socket's peer
214 * @sk_peer_cred: %SO_PEERCRED setting
215 * @sk_rcvlowat: %SO_RCVLOWAT setting
216 * @sk_rcvtimeo: %SO_RCVTIMEO setting
217 * @sk_sndtimeo: %SO_SNDTIMEO setting
218 * @sk_rxhash: flow hash received from netif layer
219 * @sk_filter: socket filtering instructions
220 * @sk_protinfo: private area, net family specific, when not using slab
221 * @sk_timer: sock cleanup timer
222 * @sk_stamp: time stamp of last packet received
223 * @sk_socket: Identd and reporting IO signals
224 * @sk_user_data: RPC layer private data
225 * @sk_sndmsg_page: cached page for sendmsg
226 * @sk_sndmsg_off: cached offset for sendmsg
227 * @sk_send_head: front of stuff to transmit
228 * @sk_security: used by security modules
229 * @sk_mark: generic packet mark
230 * @sk_classid: this socket's cgroup classid
231 * @sk_write_pending: a write to stream socket waits to start
232 * @sk_state_change: callback to indicate change in the state of the sock
233 * @sk_data_ready: callback to indicate there is data to be processed
234 * @sk_write_space: callback to indicate there is bf sending space available
235 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
236 * @sk_backlog_rcv: callback to process the backlog
237 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
238 */
239struct sock {
240 /*
241 * Now struct inet_timewait_sock also uses sock_common, so please just
242 * don't add nothing before this first member (__sk_common) --acme
243 */
244 struct sock_common __sk_common;
245#define sk_node __sk_common.skc_node
246#define sk_nulls_node __sk_common.skc_nulls_node
247#define sk_refcnt __sk_common.skc_refcnt
248#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
249
250#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
251#define sk_dontcopy_end __sk_common.skc_dontcopy_end
252#define sk_hash __sk_common.skc_hash
253#define sk_family __sk_common.skc_family
254#define sk_state __sk_common.skc_state
255#define sk_reuse __sk_common.skc_reuse
256#define sk_bound_dev_if __sk_common.skc_bound_dev_if
257#define sk_bind_node __sk_common.skc_bind_node
258#define sk_prot __sk_common.skc_prot
259#define sk_net __sk_common.skc_net
260 socket_lock_t sk_lock;
261 struct sk_buff_head sk_receive_queue;
262 /*
263 * The backlog queue is special, it is always used with
264 * the per-socket spinlock held and requires low latency
265 * access. Therefore we special case it's implementation.
266 * Note : rmem_alloc is in this structure to fill a hole
267 * on 64bit arches, not because its logically part of
268 * backlog.
269 */
270 struct {
271 atomic_t rmem_alloc;
272 int len;
273 struct sk_buff *head;
274 struct sk_buff *tail;
275 } sk_backlog;
276#define sk_rmem_alloc sk_backlog.rmem_alloc
277 int sk_forward_alloc;
278#ifdef CONFIG_RPS
279 __u32 sk_rxhash;
280#endif
281 atomic_t sk_drops;
282 int sk_rcvbuf;
283
284 struct sk_filter __rcu *sk_filter;
285 struct socket_wq __rcu *sk_wq;
286
287#ifdef CONFIG_NET_DMA
288 struct sk_buff_head sk_async_wait_queue;
289#endif
290
291#ifdef CONFIG_XFRM
292 struct xfrm_policy *sk_policy[2];
293#endif
294 unsigned long sk_flags;
295 struct dst_entry *sk_dst_cache;
296 spinlock_t sk_dst_lock;
297 atomic_t sk_wmem_alloc;
298 atomic_t sk_omem_alloc;
299 int sk_sndbuf;
300 struct sk_buff_head sk_write_queue;
301 kmemcheck_bitfield_begin(flags);
302 unsigned int sk_shutdown : 2,
303 sk_no_check : 2,
304 sk_userlocks : 4,
305 sk_protocol : 8,
306 sk_type : 16;
307 kmemcheck_bitfield_end(flags);
308 int sk_wmem_queued;
309 gfp_t sk_allocation;
310 int sk_route_caps;
311 int sk_route_nocaps;
312 int sk_gso_type;
313 unsigned int sk_gso_max_size;
314 int sk_rcvlowat;
315 unsigned long sk_lingertime;
316 struct sk_buff_head sk_error_queue;
317 struct proto *sk_prot_creator;
318 rwlock_t sk_callback_lock;
319 int sk_err,
320 sk_err_soft;
321 unsigned short sk_ack_backlog;
322 unsigned short sk_max_ack_backlog;
323 __u32 sk_priority;
324 struct pid *sk_peer_pid;
325 const struct cred *sk_peer_cred;
326 long sk_rcvtimeo;
327 long sk_sndtimeo;
328 void *sk_protinfo;
329 struct timer_list sk_timer;
330 ktime_t sk_stamp;
331 struct socket *sk_socket;
332 void *sk_user_data;
333 struct page *sk_sndmsg_page;
334 struct sk_buff *sk_send_head;
335 __u32 sk_sndmsg_off;
336 int sk_write_pending;
337#ifdef CONFIG_SECURITY
338 void *sk_security;
339#endif
340 __u32 sk_mark;
341 u32 sk_classid;
342 void (*sk_state_change)(struct sock *sk);
343 void (*sk_data_ready)(struct sock *sk, int bytes);
344 void (*sk_write_space)(struct sock *sk);
345 void (*sk_error_report)(struct sock *sk);
346 int (*sk_backlog_rcv)(struct sock *sk,
347 struct sk_buff *skb);
348 void (*sk_destruct)(struct sock *sk);
349};
350
351/*
352 * Hashed lists helper routines
353 */
354static inline struct sock *sk_entry(const struct hlist_node *node)
355{
356 return hlist_entry(node, struct sock, sk_node);
357}
358
359static inline struct sock *__sk_head(const struct hlist_head *head)
360{
361 return hlist_entry(head->first, struct sock, sk_node);
362}
363
364static inline struct sock *sk_head(const struct hlist_head *head)
365{
366 return hlist_empty(head) ? NULL : __sk_head(head);
367}
368
369static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
370{
371 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
372}
373
374static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
375{
376 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
377}
378
379static inline struct sock *sk_next(const struct sock *sk)
380{
381 return sk->sk_node.next ?
382 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
383}
384
385static inline struct sock *sk_nulls_next(const struct sock *sk)
386{
387 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
388 hlist_nulls_entry(sk->sk_nulls_node.next,
389 struct sock, sk_nulls_node) :
390 NULL;
391}
392
393static inline int sk_unhashed(const struct sock *sk)
394{
395 return hlist_unhashed(&sk->sk_node);
396}
397
398static inline int sk_hashed(const struct sock *sk)
399{
400 return !sk_unhashed(sk);
401}
402
403static __inline__ void sk_node_init(struct hlist_node *node)
404{
405 node->pprev = NULL;
406}
407
408static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
409{
410 node->pprev = NULL;
411}
412
413static __inline__ void __sk_del_node(struct sock *sk)
414{
415 __hlist_del(&sk->sk_node);
416}
417
418/* NB: equivalent to hlist_del_init_rcu */
419static __inline__ int __sk_del_node_init(struct sock *sk)
420{
421 if (sk_hashed(sk)) {
422 __sk_del_node(sk);
423 sk_node_init(&sk->sk_node);
424 return 1;
425 }
426 return 0;
427}
428
429/* Grab socket reference count. This operation is valid only
430 when sk is ALREADY grabbed f.e. it is found in hash table
431 or a list and the lookup is made under lock preventing hash table
432 modifications.
433 */
434
435static inline void sock_hold(struct sock *sk)
436{
437 atomic_inc(&sk->sk_refcnt);
438}
439
440/* Ungrab socket in the context, which assumes that socket refcnt
441 cannot hit zero, f.e. it is true in context of any socketcall.
442 */
443static inline void __sock_put(struct sock *sk)
444{
445 atomic_dec(&sk->sk_refcnt);
446}
447
448static __inline__ int sk_del_node_init(struct sock *sk)
449{
450 int rc = __sk_del_node_init(sk);
451
452 if (rc) {
453 /* paranoid for a while -acme */
454 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
455 __sock_put(sk);
456 }
457 return rc;
458}
459#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
460
461static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
462{
463 if (sk_hashed(sk)) {
464 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
465 return 1;
466 }
467 return 0;
468}
469
470static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
471{
472 int rc = __sk_nulls_del_node_init_rcu(sk);
473
474 if (rc) {
475 /* paranoid for a while -acme */
476 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
477 __sock_put(sk);
478 }
479 return rc;
480}
481
482static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
483{
484 hlist_add_head(&sk->sk_node, list);
485}
486
487static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
488{
489 sock_hold(sk);
490 __sk_add_node(sk, list);
491}
492
493static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
494{
495 sock_hold(sk);
496 hlist_add_head_rcu(&sk->sk_node, list);
497}
498
499static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
500{
501 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
502}
503
504static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
505{
506 sock_hold(sk);
507 __sk_nulls_add_node_rcu(sk, list);
508}
509
510static __inline__ void __sk_del_bind_node(struct sock *sk)
511{
512 __hlist_del(&sk->sk_bind_node);
513}
514
515static __inline__ void sk_add_bind_node(struct sock *sk,
516 struct hlist_head *list)
517{
518 hlist_add_head(&sk->sk_bind_node, list);
519}
520
521#define sk_for_each(__sk, node, list) \
522 hlist_for_each_entry(__sk, node, list, sk_node)
523#define sk_for_each_rcu(__sk, node, list) \
524 hlist_for_each_entry_rcu(__sk, node, list, sk_node)
525#define sk_nulls_for_each(__sk, node, list) \
526 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
527#define sk_nulls_for_each_rcu(__sk, node, list) \
528 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
529#define sk_for_each_from(__sk, node) \
530 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
531 hlist_for_each_entry_from(__sk, node, sk_node)
532#define sk_nulls_for_each_from(__sk, node) \
533 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
534 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
535#define sk_for_each_safe(__sk, node, tmp, list) \
536 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
537#define sk_for_each_bound(__sk, node, list) \
538 hlist_for_each_entry(__sk, node, list, sk_bind_node)
539
540/* Sock flags */
541enum sock_flags {
542 SOCK_DEAD,
543 SOCK_DONE,
544 SOCK_URGINLINE,
545 SOCK_KEEPOPEN,
546 SOCK_LINGER,
547 SOCK_DESTROY,
548 SOCK_BROADCAST,
549 SOCK_TIMESTAMP,
550 SOCK_ZAPPED,
551 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
552 SOCK_DBG, /* %SO_DEBUG setting */
553 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
554 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
555 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
556 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
557 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
558 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
559 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
560 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
561 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
562 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
563 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
564 SOCK_FASYNC, /* fasync() active */
565 SOCK_RXQ_OVFL,
566 SOCK_ZEROCOPY, /* buffers from userspace */
567};
568
569static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
570{
571 nsk->sk_flags = osk->sk_flags;
572}
573
574static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
575{
576 __set_bit(flag, &sk->sk_flags);
577}
578
579static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
580{
581 __clear_bit(flag, &sk->sk_flags);
582}
583
584static inline int sock_flag(struct sock *sk, enum sock_flags flag)
585{
586 return test_bit(flag, &sk->sk_flags);
587}
588
589static inline void sk_acceptq_removed(struct sock *sk)
590{
591 sk->sk_ack_backlog--;
592}
593
594static inline void sk_acceptq_added(struct sock *sk)
595{
596 sk->sk_ack_backlog++;
597}
598
599static inline int sk_acceptq_is_full(struct sock *sk)
600{
601 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
602}
603
604/*
605 * Compute minimal free write space needed to queue new packets.
606 */
607static inline int sk_stream_min_wspace(struct sock *sk)
608{
609 return sk->sk_wmem_queued >> 1;
610}
611
612static inline int sk_stream_wspace(struct sock *sk)
613{
614 return sk->sk_sndbuf - sk->sk_wmem_queued;
615}
616
617extern void sk_stream_write_space(struct sock *sk);
618
619static inline int sk_stream_memory_free(struct sock *sk)
620{
621 return sk->sk_wmem_queued < sk->sk_sndbuf;
622}
623
624/* OOB backlog add */
625static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
626{
627 /* dont let skb dst not refcounted, we are going to leave rcu lock */
628 skb_dst_force(skb);
629
630 if (!sk->sk_backlog.tail)
631 sk->sk_backlog.head = skb;
632 else
633 sk->sk_backlog.tail->next = skb;
634
635 sk->sk_backlog.tail = skb;
636 skb->next = NULL;
637}
638
639/*
640 * Take into account size of receive queue and backlog queue
641 */
642static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
643{
644 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
645
646 return qsize + skb->truesize > sk->sk_rcvbuf;
647}
648
649/* The per-socket spinlock must be held here. */
650static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
651{
652 if (sk_rcvqueues_full(sk, skb))
653 return -ENOBUFS;
654
655 __sk_add_backlog(sk, skb);
656 sk->sk_backlog.len += skb->truesize;
657 return 0;
658}
659
660static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
661{
662 return sk->sk_backlog_rcv(sk, skb);
663}
664
665static inline void sock_rps_record_flow(const struct sock *sk)
666{
667#ifdef CONFIG_RPS
668 struct rps_sock_flow_table *sock_flow_table;
669
670 rcu_read_lock();
671 sock_flow_table = rcu_dereference(rps_sock_flow_table);
672 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
673 rcu_read_unlock();
674#endif
675}
676
677static inline void sock_rps_reset_flow(const struct sock *sk)
678{
679#ifdef CONFIG_RPS
680 struct rps_sock_flow_table *sock_flow_table;
681
682 rcu_read_lock();
683 sock_flow_table = rcu_dereference(rps_sock_flow_table);
684 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
685 rcu_read_unlock();
686#endif
687}
688
689static inline void sock_rps_save_rxhash(struct sock *sk, u32 rxhash)
690{
691#ifdef CONFIG_RPS
692 if (unlikely(sk->sk_rxhash != rxhash)) {
693 sock_rps_reset_flow(sk);
694 sk->sk_rxhash = rxhash;
695 }
696#endif
697}
698
699#define sk_wait_event(__sk, __timeo, __condition) \
700 ({ int __rc; \
701 release_sock(__sk); \
702 __rc = __condition; \
703 if (!__rc) { \
704 *(__timeo) = schedule_timeout(*(__timeo)); \
705 } \
706 lock_sock(__sk); \
707 __rc = __condition; \
708 __rc; \
709 })
710
711extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
712extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
713extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
714extern int sk_stream_error(struct sock *sk, int flags, int err);
715extern void sk_stream_kill_queues(struct sock *sk);
716
717extern int sk_wait_data(struct sock *sk, long *timeo);
718
719struct request_sock_ops;
720struct timewait_sock_ops;
721struct inet_hashinfo;
722struct raw_hashinfo;
723
724/* Networking protocol blocks we attach to sockets.
725 * socket layer -> transport layer interface
726 * transport -> network interface is defined by struct inet_proto
727 */
728struct proto {
729 void (*close)(struct sock *sk,
730 long timeout);
731 int (*connect)(struct sock *sk,
732 struct sockaddr *uaddr,
733 int addr_len);
734 int (*disconnect)(struct sock *sk, int flags);
735
736 struct sock * (*accept) (struct sock *sk, int flags, int *err);
737
738 int (*ioctl)(struct sock *sk, int cmd,
739 unsigned long arg);
740 int (*init)(struct sock *sk);
741 void (*destroy)(struct sock *sk);
742 void (*shutdown)(struct sock *sk, int how);
743 int (*setsockopt)(struct sock *sk, int level,
744 int optname, char __user *optval,
745 unsigned int optlen);
746 int (*getsockopt)(struct sock *sk, int level,
747 int optname, char __user *optval,
748 int __user *option);
749#ifdef CONFIG_COMPAT
750 int (*compat_setsockopt)(struct sock *sk,
751 int level,
752 int optname, char __user *optval,
753 unsigned int optlen);
754 int (*compat_getsockopt)(struct sock *sk,
755 int level,
756 int optname, char __user *optval,
757 int __user *option);
758 int (*compat_ioctl)(struct sock *sk,
759 unsigned int cmd, unsigned long arg);
760#endif
761 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
762 struct msghdr *msg, size_t len);
763 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
764 struct msghdr *msg,
765 size_t len, int noblock, int flags,
766 int *addr_len);
767 int (*sendpage)(struct sock *sk, struct page *page,
768 int offset, size_t size, int flags);
769 int (*bind)(struct sock *sk,
770 struct sockaddr *uaddr, int addr_len);
771
772 int (*backlog_rcv) (struct sock *sk,
773 struct sk_buff *skb);
774
775 /* Keeping track of sk's, looking them up, and port selection methods. */
776 void (*hash)(struct sock *sk);
777 void (*unhash)(struct sock *sk);
778 void (*rehash)(struct sock *sk);
779 int (*get_port)(struct sock *sk, unsigned short snum);
780 void (*clear_sk)(struct sock *sk, int size);
781
782 /* Keeping track of sockets in use */
783#ifdef CONFIG_PROC_FS
784 unsigned int inuse_idx;
785#endif
786
787 /* Memory pressure */
788 void (*enter_memory_pressure)(struct sock *sk);
789 atomic_long_t *memory_allocated; /* Current allocated memory. */
790 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
791 /*
792 * Pressure flag: try to collapse.
793 * Technical note: it is used by multiple contexts non atomically.
794 * All the __sk_mem_schedule() is of this nature: accounting
795 * is strict, actions are advisory and have some latency.
796 */
797 int *memory_pressure;
798 long *sysctl_mem;
799 int *sysctl_wmem;
800 int *sysctl_rmem;
801 int max_header;
802 bool no_autobind;
803
804 struct kmem_cache *slab;
805 unsigned int obj_size;
806 int slab_flags;
807
808 struct percpu_counter *orphan_count;
809
810 struct request_sock_ops *rsk_prot;
811 struct timewait_sock_ops *twsk_prot;
812
813 union {
814 struct inet_hashinfo *hashinfo;
815 struct udp_table *udp_table;
816 struct raw_hashinfo *raw_hash;
817 } h;
818
819 struct module *owner;
820
821 char name[32];
822
823 struct list_head node;
824#ifdef SOCK_REFCNT_DEBUG
825 atomic_t socks;
826#endif
827};
828
829extern int proto_register(struct proto *prot, int alloc_slab);
830extern void proto_unregister(struct proto *prot);
831
832#ifdef SOCK_REFCNT_DEBUG
833static inline void sk_refcnt_debug_inc(struct sock *sk)
834{
835 atomic_inc(&sk->sk_prot->socks);
836}
837
838static inline void sk_refcnt_debug_dec(struct sock *sk)
839{
840 atomic_dec(&sk->sk_prot->socks);
841 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
842 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
843}
844
845static inline void sk_refcnt_debug_release(const struct sock *sk)
846{
847 if (atomic_read(&sk->sk_refcnt) != 1)
848 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
849 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
850}
851#else /* SOCK_REFCNT_DEBUG */
852#define sk_refcnt_debug_inc(sk) do { } while (0)
853#define sk_refcnt_debug_dec(sk) do { } while (0)
854#define sk_refcnt_debug_release(sk) do { } while (0)
855#endif /* SOCK_REFCNT_DEBUG */
856
857
858#ifdef CONFIG_PROC_FS
859/* Called with local bh disabled */
860extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
861extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
862#else
863static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
864 int inc)
865{
866}
867#endif
868
869
870/* With per-bucket locks this operation is not-atomic, so that
871 * this version is not worse.
872 */
873static inline void __sk_prot_rehash(struct sock *sk)
874{
875 sk->sk_prot->unhash(sk);
876 sk->sk_prot->hash(sk);
877}
878
879void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
880
881/* About 10 seconds */
882#define SOCK_DESTROY_TIME (10*HZ)
883
884/* Sockets 0-1023 can't be bound to unless you are superuser */
885#define PROT_SOCK 1024
886
887#define SHUTDOWN_MASK 3
888#define RCV_SHUTDOWN 1
889#define SEND_SHUTDOWN 2
890
891#define SOCK_SNDBUF_LOCK 1
892#define SOCK_RCVBUF_LOCK 2
893#define SOCK_BINDADDR_LOCK 4
894#define SOCK_BINDPORT_LOCK 8
895
896/* sock_iocb: used to kick off async processing of socket ios */
897struct sock_iocb {
898 struct list_head list;
899
900 int flags;
901 int size;
902 struct socket *sock;
903 struct sock *sk;
904 struct scm_cookie *scm;
905 struct msghdr *msg, async_msg;
906 struct kiocb *kiocb;
907};
908
909static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
910{
911 return (struct sock_iocb *)iocb->private;
912}
913
914static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
915{
916 return si->kiocb;
917}
918
919struct socket_alloc {
920 struct socket socket;
921 struct inode vfs_inode;
922};
923
924static inline struct socket *SOCKET_I(struct inode *inode)
925{
926 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
927}
928
929static inline struct inode *SOCK_INODE(struct socket *socket)
930{
931 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
932}
933
934/*
935 * Functions for memory accounting
936 */
937extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
938extern void __sk_mem_reclaim(struct sock *sk);
939
940#define SK_MEM_QUANTUM ((int)PAGE_SIZE)
941#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
942#define SK_MEM_SEND 0
943#define SK_MEM_RECV 1
944
945static inline int sk_mem_pages(int amt)
946{
947 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
948}
949
950static inline int sk_has_account(struct sock *sk)
951{
952 /* return true if protocol supports memory accounting */
953 return !!sk->sk_prot->memory_allocated;
954}
955
956static inline int sk_wmem_schedule(struct sock *sk, int size)
957{
958 if (!sk_has_account(sk))
959 return 1;
960 return size <= sk->sk_forward_alloc ||
961 __sk_mem_schedule(sk, size, SK_MEM_SEND);
962}
963
964static inline int sk_rmem_schedule(struct sock *sk, int size)
965{
966 if (!sk_has_account(sk))
967 return 1;
968 return size <= sk->sk_forward_alloc ||
969 __sk_mem_schedule(sk, size, SK_MEM_RECV);
970}
971
972static inline void sk_mem_reclaim(struct sock *sk)
973{
974 if (!sk_has_account(sk))
975 return;
976 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
977 __sk_mem_reclaim(sk);
978}
979
980static inline void sk_mem_reclaim_partial(struct sock *sk)
981{
982 if (!sk_has_account(sk))
983 return;
984 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
985 __sk_mem_reclaim(sk);
986}
987
988static inline void sk_mem_charge(struct sock *sk, int size)
989{
990 if (!sk_has_account(sk))
991 return;
992 sk->sk_forward_alloc -= size;
993}
994
995static inline void sk_mem_uncharge(struct sock *sk, int size)
996{
997 if (!sk_has_account(sk))
998 return;
999 sk->sk_forward_alloc += size;
1000}
1001
1002static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1003{
1004 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1005 sk->sk_wmem_queued -= skb->truesize;
1006 sk_mem_uncharge(sk, skb->truesize);
1007 __kfree_skb(skb);
1008}
1009
1010/* Used by processes to "lock" a socket state, so that
1011 * interrupts and bottom half handlers won't change it
1012 * from under us. It essentially blocks any incoming
1013 * packets, so that we won't get any new data or any
1014 * packets that change the state of the socket.
1015 *
1016 * While locked, BH processing will add new packets to
1017 * the backlog queue. This queue is processed by the
1018 * owner of the socket lock right before it is released.
1019 *
1020 * Since ~2.3.5 it is also exclusive sleep lock serializing
1021 * accesses from user process context.
1022 */
1023#define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1024
1025/*
1026 * Macro so as to not evaluate some arguments when
1027 * lockdep is not enabled.
1028 *
1029 * Mark both the sk_lock and the sk_lock.slock as a
1030 * per-address-family lock class.
1031 */
1032#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1033do { \
1034 sk->sk_lock.owned = 0; \
1035 init_waitqueue_head(&sk->sk_lock.wq); \
1036 spin_lock_init(&(sk)->sk_lock.slock); \
1037 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1038 sizeof((sk)->sk_lock)); \
1039 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1040 (skey), (sname)); \
1041 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1042} while (0)
1043
1044extern void lock_sock_nested(struct sock *sk, int subclass);
1045
1046static inline void lock_sock(struct sock *sk)
1047{
1048 lock_sock_nested(sk, 0);
1049}
1050
1051extern void release_sock(struct sock *sk);
1052
1053/* BH context may only use the following locking interface. */
1054#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1055#define bh_lock_sock_nested(__sk) \
1056 spin_lock_nested(&((__sk)->sk_lock.slock), \
1057 SINGLE_DEPTH_NESTING)
1058#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1059
1060extern bool lock_sock_fast(struct sock *sk);
1061/**
1062 * unlock_sock_fast - complement of lock_sock_fast
1063 * @sk: socket
1064 * @slow: slow mode
1065 *
1066 * fast unlock socket for user context.
1067 * If slow mode is on, we call regular release_sock()
1068 */
1069static inline void unlock_sock_fast(struct sock *sk, bool slow)
1070{
1071 if (slow)
1072 release_sock(sk);
1073 else
1074 spin_unlock_bh(&sk->sk_lock.slock);
1075}
1076
1077
1078extern struct sock *sk_alloc(struct net *net, int family,
1079 gfp_t priority,
1080 struct proto *prot);
1081extern void sk_free(struct sock *sk);
1082extern void sk_release_kernel(struct sock *sk);
1083extern struct sock *sk_clone(const struct sock *sk,
1084 const gfp_t priority);
1085
1086extern struct sk_buff *sock_wmalloc(struct sock *sk,
1087 unsigned long size, int force,
1088 gfp_t priority);
1089extern struct sk_buff *sock_rmalloc(struct sock *sk,
1090 unsigned long size, int force,
1091 gfp_t priority);
1092extern void sock_wfree(struct sk_buff *skb);
1093extern void sock_rfree(struct sk_buff *skb);
1094
1095extern int sock_setsockopt(struct socket *sock, int level,
1096 int op, char __user *optval,
1097 unsigned int optlen);
1098
1099extern int sock_getsockopt(struct socket *sock, int level,
1100 int op, char __user *optval,
1101 int __user *optlen);
1102extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1103 unsigned long size,
1104 int noblock,
1105 int *errcode);
1106extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1107 unsigned long header_len,
1108 unsigned long data_len,
1109 int noblock,
1110 int *errcode);
1111extern void *sock_kmalloc(struct sock *sk, int size,
1112 gfp_t priority);
1113extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1114extern void sk_send_sigurg(struct sock *sk);
1115
1116#ifdef CONFIG_CGROUPS
1117extern void sock_update_classid(struct sock *sk);
1118#else
1119static inline void sock_update_classid(struct sock *sk)
1120{
1121}
1122#endif
1123
1124/*
1125 * Functions to fill in entries in struct proto_ops when a protocol
1126 * does not implement a particular function.
1127 */
1128extern int sock_no_bind(struct socket *,
1129 struct sockaddr *, int);
1130extern int sock_no_connect(struct socket *,
1131 struct sockaddr *, int, int);
1132extern int sock_no_socketpair(struct socket *,
1133 struct socket *);
1134extern int sock_no_accept(struct socket *,
1135 struct socket *, int);
1136extern int sock_no_getname(struct socket *,
1137 struct sockaddr *, int *, int);
1138extern unsigned int sock_no_poll(struct file *, struct socket *,
1139 struct poll_table_struct *);
1140extern int sock_no_ioctl(struct socket *, unsigned int,
1141 unsigned long);
1142extern int sock_no_listen(struct socket *, int);
1143extern int sock_no_shutdown(struct socket *, int);
1144extern int sock_no_getsockopt(struct socket *, int , int,
1145 char __user *, int __user *);
1146extern int sock_no_setsockopt(struct socket *, int, int,
1147 char __user *, unsigned int);
1148extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1149 struct msghdr *, size_t);
1150extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1151 struct msghdr *, size_t, int);
1152extern int sock_no_mmap(struct file *file,
1153 struct socket *sock,
1154 struct vm_area_struct *vma);
1155extern ssize_t sock_no_sendpage(struct socket *sock,
1156 struct page *page,
1157 int offset, size_t size,
1158 int flags);
1159
1160/*
1161 * Functions to fill in entries in struct proto_ops when a protocol
1162 * uses the inet style.
1163 */
1164extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1165 char __user *optval, int __user *optlen);
1166extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1167 struct msghdr *msg, size_t size, int flags);
1168extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1169 char __user *optval, unsigned int optlen);
1170extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1171 int optname, char __user *optval, int __user *optlen);
1172extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1173 int optname, char __user *optval, unsigned int optlen);
1174
1175extern void sk_common_release(struct sock *sk);
1176
1177/*
1178 * Default socket callbacks and setup code
1179 */
1180
1181/* Initialise core socket variables */
1182extern void sock_init_data(struct socket *sock, struct sock *sk);
1183
1184extern void sk_filter_release_rcu(struct rcu_head *rcu);
1185
1186/**
1187 * sk_filter_release - release a socket filter
1188 * @fp: filter to remove
1189 *
1190 * Remove a filter from a socket and release its resources.
1191 */
1192
1193static inline void sk_filter_release(struct sk_filter *fp)
1194{
1195 if (atomic_dec_and_test(&fp->refcnt))
1196 call_rcu(&fp->rcu, sk_filter_release_rcu);
1197}
1198
1199static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1200{
1201 unsigned int size = sk_filter_len(fp);
1202
1203 atomic_sub(size, &sk->sk_omem_alloc);
1204 sk_filter_release(fp);
1205}
1206
1207static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1208{
1209 atomic_inc(&fp->refcnt);
1210 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1211}
1212
1213/*
1214 * Socket reference counting postulates.
1215 *
1216 * * Each user of socket SHOULD hold a reference count.
1217 * * Each access point to socket (an hash table bucket, reference from a list,
1218 * running timer, skb in flight MUST hold a reference count.
1219 * * When reference count hits 0, it means it will never increase back.
1220 * * When reference count hits 0, it means that no references from
1221 * outside exist to this socket and current process on current CPU
1222 * is last user and may/should destroy this socket.
1223 * * sk_free is called from any context: process, BH, IRQ. When
1224 * it is called, socket has no references from outside -> sk_free
1225 * may release descendant resources allocated by the socket, but
1226 * to the time when it is called, socket is NOT referenced by any
1227 * hash tables, lists etc.
1228 * * Packets, delivered from outside (from network or from another process)
1229 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1230 * when they sit in queue. Otherwise, packets will leak to hole, when
1231 * socket is looked up by one cpu and unhasing is made by another CPU.
1232 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1233 * (leak to backlog). Packet socket does all the processing inside
1234 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1235 * use separate SMP lock, so that they are prone too.
1236 */
1237
1238/* Ungrab socket and destroy it, if it was the last reference. */
1239static inline void sock_put(struct sock *sk)
1240{
1241 if (atomic_dec_and_test(&sk->sk_refcnt))
1242 sk_free(sk);
1243}
1244
1245extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1246 const int nested);
1247
1248static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1249{
1250 sk->sk_tx_queue_mapping = tx_queue;
1251}
1252
1253static inline void sk_tx_queue_clear(struct sock *sk)
1254{
1255 sk->sk_tx_queue_mapping = -1;
1256}
1257
1258static inline int sk_tx_queue_get(const struct sock *sk)
1259{
1260 return sk ? sk->sk_tx_queue_mapping : -1;
1261}
1262
1263static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1264{
1265 sk_tx_queue_clear(sk);
1266 sk->sk_socket = sock;
1267}
1268
1269static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1270{
1271 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1272 return &rcu_dereference_raw(sk->sk_wq)->wait;
1273}
1274/* Detach socket from process context.
1275 * Announce socket dead, detach it from wait queue and inode.
1276 * Note that parent inode held reference count on this struct sock,
1277 * we do not release it in this function, because protocol
1278 * probably wants some additional cleanups or even continuing
1279 * to work with this socket (TCP).
1280 */
1281static inline void sock_orphan(struct sock *sk)
1282{
1283 write_lock_bh(&sk->sk_callback_lock);
1284 sock_set_flag(sk, SOCK_DEAD);
1285 sk_set_socket(sk, NULL);
1286 sk->sk_wq = NULL;
1287 write_unlock_bh(&sk->sk_callback_lock);
1288}
1289
1290static inline void sock_graft(struct sock *sk, struct socket *parent)
1291{
1292 write_lock_bh(&sk->sk_callback_lock);
1293 sk->sk_wq = parent->wq;
1294 parent->sk = sk;
1295 sk_set_socket(sk, parent);
1296 security_sock_graft(sk, parent);
1297 write_unlock_bh(&sk->sk_callback_lock);
1298}
1299
1300extern int sock_i_uid(struct sock *sk);
1301extern unsigned long sock_i_ino(struct sock *sk);
1302
1303static inline struct dst_entry *
1304__sk_dst_get(struct sock *sk)
1305{
1306 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1307 lockdep_is_held(&sk->sk_lock.slock));
1308}
1309
1310static inline struct dst_entry *
1311sk_dst_get(struct sock *sk)
1312{
1313 struct dst_entry *dst;
1314
1315 rcu_read_lock();
1316 dst = rcu_dereference(sk->sk_dst_cache);
1317 if (dst)
1318 dst_hold(dst);
1319 rcu_read_unlock();
1320 return dst;
1321}
1322
1323extern void sk_reset_txq(struct sock *sk);
1324
1325static inline void dst_negative_advice(struct sock *sk)
1326{
1327 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1328
1329 if (dst && dst->ops->negative_advice) {
1330 ndst = dst->ops->negative_advice(dst);
1331
1332 if (ndst != dst) {
1333 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1334 sk_reset_txq(sk);
1335 }
1336 }
1337}
1338
1339static inline void
1340__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1341{
1342 struct dst_entry *old_dst;
1343
1344 sk_tx_queue_clear(sk);
1345 /*
1346 * This can be called while sk is owned by the caller only,
1347 * with no state that can be checked in a rcu_dereference_check() cond
1348 */
1349 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1350 rcu_assign_pointer(sk->sk_dst_cache, dst);
1351 dst_release(old_dst);
1352}
1353
1354static inline void
1355sk_dst_set(struct sock *sk, struct dst_entry *dst)
1356{
1357 spin_lock(&sk->sk_dst_lock);
1358 __sk_dst_set(sk, dst);
1359 spin_unlock(&sk->sk_dst_lock);
1360}
1361
1362static inline void
1363__sk_dst_reset(struct sock *sk)
1364{
1365 __sk_dst_set(sk, NULL);
1366}
1367
1368static inline void
1369sk_dst_reset(struct sock *sk)
1370{
1371 spin_lock(&sk->sk_dst_lock);
1372 __sk_dst_reset(sk);
1373 spin_unlock(&sk->sk_dst_lock);
1374}
1375
1376extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1377
1378extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1379
1380static inline int sk_can_gso(const struct sock *sk)
1381{
1382 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1383}
1384
1385extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1386
1387static inline void sk_nocaps_add(struct sock *sk, int flags)
1388{
1389 sk->sk_route_nocaps |= flags;
1390 sk->sk_route_caps &= ~flags;
1391}
1392
1393static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1394 char __user *from, char *to,
1395 int copy, int offset)
1396{
1397 if (skb->ip_summed == CHECKSUM_NONE) {
1398 int err = 0;
1399 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1400 if (err)
1401 return err;
1402 skb->csum = csum_block_add(skb->csum, csum, offset);
1403 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1404 if (!access_ok(VERIFY_READ, from, copy) ||
1405 __copy_from_user_nocache(to, from, copy))
1406 return -EFAULT;
1407 } else if (copy_from_user(to, from, copy))
1408 return -EFAULT;
1409
1410 return 0;
1411}
1412
1413static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1414 char __user *from, int copy)
1415{
1416 int err, offset = skb->len;
1417
1418 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1419 copy, offset);
1420 if (err)
1421 __skb_trim(skb, offset);
1422
1423 return err;
1424}
1425
1426static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1427 struct sk_buff *skb,
1428 struct page *page,
1429 int off, int copy)
1430{
1431 int err;
1432
1433 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1434 copy, skb->len);
1435 if (err)
1436 return err;
1437
1438 skb->len += copy;
1439 skb->data_len += copy;
1440 skb->truesize += copy;
1441 sk->sk_wmem_queued += copy;
1442 sk_mem_charge(sk, copy);
1443 return 0;
1444}
1445
1446static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1447 struct sk_buff *skb, struct page *page,
1448 int off, int copy)
1449{
1450 if (skb->ip_summed == CHECKSUM_NONE) {
1451 int err = 0;
1452 __wsum csum = csum_and_copy_from_user(from,
1453 page_address(page) + off,
1454 copy, 0, &err);
1455 if (err)
1456 return err;
1457 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1458 } else if (copy_from_user(page_address(page) + off, from, copy))
1459 return -EFAULT;
1460
1461 skb->len += copy;
1462 skb->data_len += copy;
1463 skb->truesize += copy;
1464 sk->sk_wmem_queued += copy;
1465 sk_mem_charge(sk, copy);
1466 return 0;
1467}
1468
1469/**
1470 * sk_wmem_alloc_get - returns write allocations
1471 * @sk: socket
1472 *
1473 * Returns sk_wmem_alloc minus initial offset of one
1474 */
1475static inline int sk_wmem_alloc_get(const struct sock *sk)
1476{
1477 return atomic_read(&sk->sk_wmem_alloc) - 1;
1478}
1479
1480/**
1481 * sk_rmem_alloc_get - returns read allocations
1482 * @sk: socket
1483 *
1484 * Returns sk_rmem_alloc
1485 */
1486static inline int sk_rmem_alloc_get(const struct sock *sk)
1487{
1488 return atomic_read(&sk->sk_rmem_alloc);
1489}
1490
1491/**
1492 * sk_has_allocations - check if allocations are outstanding
1493 * @sk: socket
1494 *
1495 * Returns true if socket has write or read allocations
1496 */
1497static inline int sk_has_allocations(const struct sock *sk)
1498{
1499 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1500}
1501
1502/**
1503 * wq_has_sleeper - check if there are any waiting processes
1504 * @wq: struct socket_wq
1505 *
1506 * Returns true if socket_wq has waiting processes
1507 *
1508 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1509 * barrier call. They were added due to the race found within the tcp code.
1510 *
1511 * Consider following tcp code paths:
1512 *
1513 * CPU1 CPU2
1514 *
1515 * sys_select receive packet
1516 * ... ...
1517 * __add_wait_queue update tp->rcv_nxt
1518 * ... ...
1519 * tp->rcv_nxt check sock_def_readable
1520 * ... {
1521 * schedule rcu_read_lock();
1522 * wq = rcu_dereference(sk->sk_wq);
1523 * if (wq && waitqueue_active(&wq->wait))
1524 * wake_up_interruptible(&wq->wait)
1525 * ...
1526 * }
1527 *
1528 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1529 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1530 * could then endup calling schedule and sleep forever if there are no more
1531 * data on the socket.
1532 *
1533 */
1534static inline bool wq_has_sleeper(struct socket_wq *wq)
1535{
1536
1537 /*
1538 * We need to be sure we are in sync with the
1539 * add_wait_queue modifications to the wait queue.
1540 *
1541 * This memory barrier is paired in the sock_poll_wait.
1542 */
1543 smp_mb();
1544 return wq && waitqueue_active(&wq->wait);
1545}
1546
1547/**
1548 * sock_poll_wait - place memory barrier behind the poll_wait call.
1549 * @filp: file
1550 * @wait_address: socket wait queue
1551 * @p: poll_table
1552 *
1553 * See the comments in the wq_has_sleeper function.
1554 */
1555static inline void sock_poll_wait(struct file *filp,
1556 wait_queue_head_t *wait_address, poll_table *p)
1557{
1558 if (p && wait_address) {
1559 poll_wait(filp, wait_address, p);
1560 /*
1561 * We need to be sure we are in sync with the
1562 * socket flags modification.
1563 *
1564 * This memory barrier is paired in the wq_has_sleeper.
1565 */
1566 smp_mb();
1567 }
1568}
1569
1570/*
1571 * Queue a received datagram if it will fit. Stream and sequenced
1572 * protocols can't normally use this as they need to fit buffers in
1573 * and play with them.
1574 *
1575 * Inlined as it's very short and called for pretty much every
1576 * packet ever received.
1577 */
1578
1579static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1580{
1581 skb_orphan(skb);
1582 skb->sk = sk;
1583 skb->destructor = sock_wfree;
1584 /*
1585 * We used to take a refcount on sk, but following operation
1586 * is enough to guarantee sk_free() wont free this sock until
1587 * all in-flight packets are completed
1588 */
1589 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1590}
1591
1592static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1593{
1594 skb_orphan(skb);
1595 skb->sk = sk;
1596 skb->destructor = sock_rfree;
1597 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1598 sk_mem_charge(sk, skb->truesize);
1599}
1600
1601extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1602 unsigned long expires);
1603
1604extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1605
1606extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1607
1608extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1609
1610/*
1611 * Recover an error report and clear atomically
1612 */
1613
1614static inline int sock_error(struct sock *sk)
1615{
1616 int err;
1617 if (likely(!sk->sk_err))
1618 return 0;
1619 err = xchg(&sk->sk_err, 0);
1620 return -err;
1621}
1622
1623static inline unsigned long sock_wspace(struct sock *sk)
1624{
1625 int amt = 0;
1626
1627 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1628 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1629 if (amt < 0)
1630 amt = 0;
1631 }
1632 return amt;
1633}
1634
1635static inline void sk_wake_async(struct sock *sk, int how, int band)
1636{
1637 if (sock_flag(sk, SOCK_FASYNC))
1638 sock_wake_async(sk->sk_socket, how, band);
1639}
1640
1641#define SOCK_MIN_SNDBUF 2048
1642/*
1643 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1644 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1645 */
1646#define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1647
1648static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1649{
1650 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1651 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1652 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1653 }
1654}
1655
1656struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1657
1658static inline struct page *sk_stream_alloc_page(struct sock *sk)
1659{
1660 struct page *page = NULL;
1661
1662 page = alloc_pages(sk->sk_allocation, 0);
1663 if (!page) {
1664 sk->sk_prot->enter_memory_pressure(sk);
1665 sk_stream_moderate_sndbuf(sk);
1666 }
1667 return page;
1668}
1669
1670/*
1671 * Default write policy as shown to user space via poll/select/SIGIO
1672 */
1673static inline int sock_writeable(const struct sock *sk)
1674{
1675 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1676}
1677
1678static inline gfp_t gfp_any(void)
1679{
1680 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1681}
1682
1683static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1684{
1685 return noblock ? 0 : sk->sk_rcvtimeo;
1686}
1687
1688static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1689{
1690 return noblock ? 0 : sk->sk_sndtimeo;
1691}
1692
1693static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1694{
1695 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1696}
1697
1698/* Alas, with timeout socket operations are not restartable.
1699 * Compare this to poll().
1700 */
1701static inline int sock_intr_errno(long timeo)
1702{
1703 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1704}
1705
1706extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1707 struct sk_buff *skb);
1708
1709static __inline__ void
1710sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1711{
1712 ktime_t kt = skb->tstamp;
1713 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1714
1715 /*
1716 * generate control messages if
1717 * - receive time stamping in software requested (SOCK_RCVTSTAMP
1718 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
1719 * - software time stamp available and wanted
1720 * (SOCK_TIMESTAMPING_SOFTWARE)
1721 * - hardware time stamps available and wanted
1722 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
1723 * SOCK_TIMESTAMPING_RAW_HARDWARE)
1724 */
1725 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
1726 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
1727 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
1728 (hwtstamps->hwtstamp.tv64 &&
1729 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
1730 (hwtstamps->syststamp.tv64 &&
1731 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
1732 __sock_recv_timestamp(msg, sk, skb);
1733 else
1734 sk->sk_stamp = kt;
1735}
1736
1737extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
1738 struct sk_buff *skb);
1739
1740static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
1741 struct sk_buff *skb)
1742{
1743#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
1744 (1UL << SOCK_RCVTSTAMP) | \
1745 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
1746 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
1747 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
1748 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
1749
1750 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
1751 __sock_recv_ts_and_drops(msg, sk, skb);
1752 else
1753 sk->sk_stamp = skb->tstamp;
1754}
1755
1756/**
1757 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
1758 * @sk: socket sending this packet
1759 * @tx_flags: filled with instructions for time stamping
1760 *
1761 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
1762 * parameters are invalid.
1763 */
1764extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
1765
1766/**
1767 * sk_eat_skb - Release a skb if it is no longer needed
1768 * @sk: socket to eat this skb from
1769 * @skb: socket buffer to eat
1770 * @copied_early: flag indicating whether DMA operations copied this data early
1771 *
1772 * This routine must be called with interrupts disabled or with the socket
1773 * locked so that the sk_buff queue operation is ok.
1774*/
1775#ifdef CONFIG_NET_DMA
1776static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1777{
1778 __skb_unlink(skb, &sk->sk_receive_queue);
1779 if (!copied_early)
1780 __kfree_skb(skb);
1781 else
1782 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
1783}
1784#else
1785static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1786{
1787 __skb_unlink(skb, &sk->sk_receive_queue);
1788 __kfree_skb(skb);
1789}
1790#endif
1791
1792static inline
1793struct net *sock_net(const struct sock *sk)
1794{
1795 return read_pnet(&sk->sk_net);
1796}
1797
1798static inline
1799void sock_net_set(struct sock *sk, struct net *net)
1800{
1801 write_pnet(&sk->sk_net, net);
1802}
1803
1804/*
1805 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
1806 * They should not hold a reference to a namespace in order to allow
1807 * to stop it.
1808 * Sockets after sk_change_net should be released using sk_release_kernel
1809 */
1810static inline void sk_change_net(struct sock *sk, struct net *net)
1811{
1812 put_net(sock_net(sk));
1813 sock_net_set(sk, hold_net(net));
1814}
1815
1816static inline struct sock *skb_steal_sock(struct sk_buff *skb)
1817{
1818 if (unlikely(skb->sk)) {
1819 struct sock *sk = skb->sk;
1820
1821 skb->destructor = NULL;
1822 skb->sk = NULL;
1823 return sk;
1824 }
1825 return NULL;
1826}
1827
1828extern void sock_enable_timestamp(struct sock *sk, int flag);
1829extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1830extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1831
1832/*
1833 * Enable debug/info messages
1834 */
1835extern int net_msg_warn;
1836#define NETDEBUG(fmt, args...) \
1837 do { if (net_msg_warn) printk(fmt,##args); } while (0)
1838
1839#define LIMIT_NETDEBUG(fmt, args...) \
1840 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1841
1842extern __u32 sysctl_wmem_max;
1843extern __u32 sysctl_rmem_max;
1844
1845extern void sk_init(void);
1846
1847extern int sysctl_optmem_max;
1848
1849extern __u32 sysctl_wmem_default;
1850extern __u32 sysctl_rmem_default;
1851
1852#endif /* _SOCK_H */
1/*
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.
5 *
6 * Definitions for the AF_INET socket handler.
7 *
8 * Version: @(#)sock.h 1.0.4 05/13/93
9 *
10 * Authors: Ross Biro
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>
14 *
15 * Fixes:
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
19 * than the reverse.
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
33 *
34 *
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.
39 */
40#ifndef _SOCK_H
41#define _SOCK_H
42
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 */
53#include <linux/mm.h>
54#include <linux/security.h>
55#include <linux/slab.h>
56#include <linux/uaccess.h>
57#include <linux/page_counter.h>
58#include <linux/memcontrol.h>
59#include <linux/static_key.h>
60#include <linux/sched.h>
61#include <linux/wait.h>
62#include <linux/cgroup-defs.h>
63
64#include <linux/filter.h>
65#include <linux/rculist_nulls.h>
66#include <linux/poll.h>
67
68#include <linux/atomic.h>
69#include <net/dst.h>
70#include <net/checksum.h>
71#include <net/tcp_states.h>
72#include <linux/net_tstamp.h>
73
74/*
75 * This structure really needs to be cleaned up.
76 * Most of it is for TCP, and not used by any of
77 * the other protocols.
78 */
79
80/* Define this to get the SOCK_DBG debugging facility. */
81#define SOCK_DEBUGGING
82#ifdef SOCK_DEBUGGING
83#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
84 printk(KERN_DEBUG msg); } while (0)
85#else
86/* Validate arguments and do nothing */
87static inline __printf(2, 3)
88void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
89{
90}
91#endif
92
93/* This is the per-socket lock. The spinlock provides a synchronization
94 * between user contexts and software interrupt processing, whereas the
95 * mini-semaphore synchronizes multiple users amongst themselves.
96 */
97typedef struct {
98 spinlock_t slock;
99 int owned;
100 wait_queue_head_t wq;
101 /*
102 * We express the mutex-alike socket_lock semantics
103 * to the lock validator by explicitly managing
104 * the slock as a lock variant (in addition to
105 * the slock itself):
106 */
107#ifdef CONFIG_DEBUG_LOCK_ALLOC
108 struct lockdep_map dep_map;
109#endif
110} socket_lock_t;
111
112struct sock;
113struct proto;
114struct net;
115
116typedef __u32 __bitwise __portpair;
117typedef __u64 __bitwise __addrpair;
118
119/**
120 * struct sock_common - minimal network layer representation of sockets
121 * @skc_daddr: Foreign IPv4 addr
122 * @skc_rcv_saddr: Bound local IPv4 addr
123 * @skc_hash: hash value used with various protocol lookup tables
124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
125 * @skc_dport: placeholder for inet_dport/tw_dport
126 * @skc_num: placeholder for inet_num/tw_num
127 * @skc_family: network address family
128 * @skc_state: Connection state
129 * @skc_reuse: %SO_REUSEADDR setting
130 * @skc_reuseport: %SO_REUSEPORT setting
131 * @skc_bound_dev_if: bound device index if != 0
132 * @skc_bind_node: bind hash linkage for various protocol lookup tables
133 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
134 * @skc_prot: protocol handlers inside a network family
135 * @skc_net: reference to the network namespace of this socket
136 * @skc_node: main hash linkage for various protocol lookup tables
137 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
138 * @skc_tx_queue_mapping: tx queue number for this connection
139 * @skc_flags: place holder for sk_flags
140 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
141 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
142 * @skc_incoming_cpu: record/match cpu processing incoming packets
143 * @skc_refcnt: reference count
144 *
145 * This is the minimal network layer representation of sockets, the header
146 * for struct sock and struct inet_timewait_sock.
147 */
148struct sock_common {
149 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
150 * address on 64bit arches : cf INET_MATCH()
151 */
152 union {
153 __addrpair skc_addrpair;
154 struct {
155 __be32 skc_daddr;
156 __be32 skc_rcv_saddr;
157 };
158 };
159 union {
160 unsigned int skc_hash;
161 __u16 skc_u16hashes[2];
162 };
163 /* skc_dport && skc_num must be grouped as well */
164 union {
165 __portpair skc_portpair;
166 struct {
167 __be16 skc_dport;
168 __u16 skc_num;
169 };
170 };
171
172 unsigned short skc_family;
173 volatile unsigned char skc_state;
174 unsigned char skc_reuse:4;
175 unsigned char skc_reuseport:1;
176 unsigned char skc_ipv6only:1;
177 unsigned char skc_net_refcnt:1;
178 int skc_bound_dev_if;
179 union {
180 struct hlist_node skc_bind_node;
181 struct hlist_node skc_portaddr_node;
182 };
183 struct proto *skc_prot;
184 possible_net_t skc_net;
185
186#if IS_ENABLED(CONFIG_IPV6)
187 struct in6_addr skc_v6_daddr;
188 struct in6_addr skc_v6_rcv_saddr;
189#endif
190
191 atomic64_t skc_cookie;
192
193 /* following fields are padding to force
194 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
195 * assuming IPV6 is enabled. We use this padding differently
196 * for different kind of 'sockets'
197 */
198 union {
199 unsigned long skc_flags;
200 struct sock *skc_listener; /* request_sock */
201 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
202 };
203 /*
204 * fields between dontcopy_begin/dontcopy_end
205 * are not copied in sock_copy()
206 */
207 /* private: */
208 int skc_dontcopy_begin[0];
209 /* public: */
210 union {
211 struct hlist_node skc_node;
212 struct hlist_nulls_node skc_nulls_node;
213 };
214 int skc_tx_queue_mapping;
215 union {
216 int skc_incoming_cpu;
217 u32 skc_rcv_wnd;
218 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
219 };
220
221 atomic_t skc_refcnt;
222 /* private: */
223 int skc_dontcopy_end[0];
224 union {
225 u32 skc_rxhash;
226 u32 skc_window_clamp;
227 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
228 };
229 /* public: */
230};
231
232/**
233 * struct sock - network layer representation of sockets
234 * @__sk_common: shared layout with inet_timewait_sock
235 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
236 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
237 * @sk_lock: synchronizer
238 * @sk_rcvbuf: size of receive buffer in bytes
239 * @sk_wq: sock wait queue and async head
240 * @sk_rx_dst: receive input route used by early demux
241 * @sk_dst_cache: destination cache
242 * @sk_policy: flow policy
243 * @sk_receive_queue: incoming packets
244 * @sk_wmem_alloc: transmit queue bytes committed
245 * @sk_write_queue: Packet sending queue
246 * @sk_omem_alloc: "o" is "option" or "other"
247 * @sk_wmem_queued: persistent queue size
248 * @sk_forward_alloc: space allocated forward
249 * @sk_napi_id: id of the last napi context to receive data for sk
250 * @sk_ll_usec: usecs to busypoll when there is no data
251 * @sk_allocation: allocation mode
252 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
253 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
254 * @sk_sndbuf: size of send buffer in bytes
255 * @sk_padding: unused element for alignment
256 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
257 * @sk_no_check_rx: allow zero checksum in RX packets
258 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
259 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
260 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
261 * @sk_gso_max_size: Maximum GSO segment size to build
262 * @sk_gso_max_segs: Maximum number of GSO segments
263 * @sk_lingertime: %SO_LINGER l_linger setting
264 * @sk_backlog: always used with the per-socket spinlock held
265 * @sk_callback_lock: used with the callbacks in the end of this struct
266 * @sk_error_queue: rarely used
267 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
268 * IPV6_ADDRFORM for instance)
269 * @sk_err: last error
270 * @sk_err_soft: errors that don't cause failure but are the cause of a
271 * persistent failure not just 'timed out'
272 * @sk_drops: raw/udp drops counter
273 * @sk_ack_backlog: current listen backlog
274 * @sk_max_ack_backlog: listen backlog set in listen()
275 * @sk_priority: %SO_PRIORITY setting
276 * @sk_type: socket type (%SOCK_STREAM, etc)
277 * @sk_protocol: which protocol this socket belongs in this network family
278 * @sk_peer_pid: &struct pid for this socket's peer
279 * @sk_peer_cred: %SO_PEERCRED setting
280 * @sk_rcvlowat: %SO_RCVLOWAT setting
281 * @sk_rcvtimeo: %SO_RCVTIMEO setting
282 * @sk_sndtimeo: %SO_SNDTIMEO setting
283 * @sk_txhash: computed flow hash for use on transmit
284 * @sk_filter: socket filtering instructions
285 * @sk_timer: sock cleanup timer
286 * @sk_stamp: time stamp of last packet received
287 * @sk_tsflags: SO_TIMESTAMPING socket options
288 * @sk_tskey: counter to disambiguate concurrent tstamp requests
289 * @sk_socket: Identd and reporting IO signals
290 * @sk_user_data: RPC layer private data
291 * @sk_frag: cached page frag
292 * @sk_peek_off: current peek_offset value
293 * @sk_send_head: front of stuff to transmit
294 * @sk_security: used by security modules
295 * @sk_mark: generic packet mark
296 * @sk_cgrp_data: cgroup data for this cgroup
297 * @sk_memcg: this socket's memory cgroup association
298 * @sk_write_pending: a write to stream socket waits to start
299 * @sk_state_change: callback to indicate change in the state of the sock
300 * @sk_data_ready: callback to indicate there is data to be processed
301 * @sk_write_space: callback to indicate there is bf sending space available
302 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
303 * @sk_backlog_rcv: callback to process the backlog
304 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
305 * @sk_reuseport_cb: reuseport group container
306 * @sk_rcu: used during RCU grace period
307 */
308struct sock {
309 /*
310 * Now struct inet_timewait_sock also uses sock_common, so please just
311 * don't add nothing before this first member (__sk_common) --acme
312 */
313 struct sock_common __sk_common;
314#define sk_node __sk_common.skc_node
315#define sk_nulls_node __sk_common.skc_nulls_node
316#define sk_refcnt __sk_common.skc_refcnt
317#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
318
319#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
320#define sk_dontcopy_end __sk_common.skc_dontcopy_end
321#define sk_hash __sk_common.skc_hash
322#define sk_portpair __sk_common.skc_portpair
323#define sk_num __sk_common.skc_num
324#define sk_dport __sk_common.skc_dport
325#define sk_addrpair __sk_common.skc_addrpair
326#define sk_daddr __sk_common.skc_daddr
327#define sk_rcv_saddr __sk_common.skc_rcv_saddr
328#define sk_family __sk_common.skc_family
329#define sk_state __sk_common.skc_state
330#define sk_reuse __sk_common.skc_reuse
331#define sk_reuseport __sk_common.skc_reuseport
332#define sk_ipv6only __sk_common.skc_ipv6only
333#define sk_net_refcnt __sk_common.skc_net_refcnt
334#define sk_bound_dev_if __sk_common.skc_bound_dev_if
335#define sk_bind_node __sk_common.skc_bind_node
336#define sk_prot __sk_common.skc_prot
337#define sk_net __sk_common.skc_net
338#define sk_v6_daddr __sk_common.skc_v6_daddr
339#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
340#define sk_cookie __sk_common.skc_cookie
341#define sk_incoming_cpu __sk_common.skc_incoming_cpu
342#define sk_flags __sk_common.skc_flags
343#define sk_rxhash __sk_common.skc_rxhash
344
345 socket_lock_t sk_lock;
346 atomic_t sk_drops;
347 int sk_rcvlowat;
348 struct sk_buff_head sk_error_queue;
349 struct sk_buff_head sk_receive_queue;
350 /*
351 * The backlog queue is special, it is always used with
352 * the per-socket spinlock held and requires low latency
353 * access. Therefore we special case it's implementation.
354 * Note : rmem_alloc is in this structure to fill a hole
355 * on 64bit arches, not because its logically part of
356 * backlog.
357 */
358 struct {
359 atomic_t rmem_alloc;
360 int len;
361 struct sk_buff *head;
362 struct sk_buff *tail;
363 } sk_backlog;
364#define sk_rmem_alloc sk_backlog.rmem_alloc
365
366 int sk_forward_alloc;
367#ifdef CONFIG_NET_RX_BUSY_POLL
368 unsigned int sk_ll_usec;
369 /* ===== mostly read cache line ===== */
370 unsigned int sk_napi_id;
371#endif
372 int sk_rcvbuf;
373
374 struct sk_filter __rcu *sk_filter;
375 union {
376 struct socket_wq __rcu *sk_wq;
377 struct socket_wq *sk_wq_raw;
378 };
379#ifdef CONFIG_XFRM
380 struct xfrm_policy __rcu *sk_policy[2];
381#endif
382 struct dst_entry *sk_rx_dst;
383 struct dst_entry __rcu *sk_dst_cache;
384 atomic_t sk_omem_alloc;
385 int sk_sndbuf;
386
387 /* ===== cache line for TX ===== */
388 int sk_wmem_queued;
389 atomic_t sk_wmem_alloc;
390 unsigned long sk_tsq_flags;
391 struct sk_buff *sk_send_head;
392 struct sk_buff_head sk_write_queue;
393 __s32 sk_peek_off;
394 int sk_write_pending;
395 long sk_sndtimeo;
396 struct timer_list sk_timer;
397 __u32 sk_priority;
398 __u32 sk_mark;
399 u32 sk_pacing_rate; /* bytes per second */
400 u32 sk_max_pacing_rate;
401 struct page_frag sk_frag;
402 netdev_features_t sk_route_caps;
403 netdev_features_t sk_route_nocaps;
404 int sk_gso_type;
405 unsigned int sk_gso_max_size;
406 gfp_t sk_allocation;
407 __u32 sk_txhash;
408
409 /*
410 * Because of non atomicity rules, all
411 * changes are protected by socket lock.
412 */
413 unsigned int __sk_flags_offset[0];
414#ifdef __BIG_ENDIAN_BITFIELD
415#define SK_FL_PROTO_SHIFT 16
416#define SK_FL_PROTO_MASK 0x00ff0000
417
418#define SK_FL_TYPE_SHIFT 0
419#define SK_FL_TYPE_MASK 0x0000ffff
420#else
421#define SK_FL_PROTO_SHIFT 8
422#define SK_FL_PROTO_MASK 0x0000ff00
423
424#define SK_FL_TYPE_SHIFT 16
425#define SK_FL_TYPE_MASK 0xffff0000
426#endif
427
428 kmemcheck_bitfield_begin(flags);
429 unsigned int sk_padding : 2,
430 sk_no_check_tx : 1,
431 sk_no_check_rx : 1,
432 sk_userlocks : 4,
433 sk_protocol : 8,
434 sk_type : 16;
435#define SK_PROTOCOL_MAX U8_MAX
436 kmemcheck_bitfield_end(flags);
437
438 u16 sk_gso_max_segs;
439 unsigned long sk_lingertime;
440 struct proto *sk_prot_creator;
441 rwlock_t sk_callback_lock;
442 int sk_err,
443 sk_err_soft;
444 u32 sk_ack_backlog;
445 u32 sk_max_ack_backlog;
446 kuid_t sk_uid;
447 struct pid *sk_peer_pid;
448 const struct cred *sk_peer_cred;
449 long sk_rcvtimeo;
450 ktime_t sk_stamp;
451 u16 sk_tsflags;
452 u8 sk_shutdown;
453 u32 sk_tskey;
454 struct socket *sk_socket;
455 void *sk_user_data;
456#ifdef CONFIG_SECURITY
457 void *sk_security;
458#endif
459 struct sock_cgroup_data sk_cgrp_data;
460 struct mem_cgroup *sk_memcg;
461 void (*sk_state_change)(struct sock *sk);
462 void (*sk_data_ready)(struct sock *sk);
463 void (*sk_write_space)(struct sock *sk);
464 void (*sk_error_report)(struct sock *sk);
465 int (*sk_backlog_rcv)(struct sock *sk,
466 struct sk_buff *skb);
467 void (*sk_destruct)(struct sock *sk);
468 struct sock_reuseport __rcu *sk_reuseport_cb;
469 struct rcu_head sk_rcu;
470};
471
472#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
473
474#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
475#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
476
477/*
478 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
479 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
480 * on a socket means that the socket will reuse everybody else's port
481 * without looking at the other's sk_reuse value.
482 */
483
484#define SK_NO_REUSE 0
485#define SK_CAN_REUSE 1
486#define SK_FORCE_REUSE 2
487
488int sk_set_peek_off(struct sock *sk, int val);
489
490static inline int sk_peek_offset(struct sock *sk, int flags)
491{
492 if (unlikely(flags & MSG_PEEK)) {
493 s32 off = READ_ONCE(sk->sk_peek_off);
494 if (off >= 0)
495 return off;
496 }
497
498 return 0;
499}
500
501static inline void sk_peek_offset_bwd(struct sock *sk, int val)
502{
503 s32 off = READ_ONCE(sk->sk_peek_off);
504
505 if (unlikely(off >= 0)) {
506 off = max_t(s32, off - val, 0);
507 WRITE_ONCE(sk->sk_peek_off, off);
508 }
509}
510
511static inline void sk_peek_offset_fwd(struct sock *sk, int val)
512{
513 sk_peek_offset_bwd(sk, -val);
514}
515
516/*
517 * Hashed lists helper routines
518 */
519static inline struct sock *sk_entry(const struct hlist_node *node)
520{
521 return hlist_entry(node, struct sock, sk_node);
522}
523
524static inline struct sock *__sk_head(const struct hlist_head *head)
525{
526 return hlist_entry(head->first, struct sock, sk_node);
527}
528
529static inline struct sock *sk_head(const struct hlist_head *head)
530{
531 return hlist_empty(head) ? NULL : __sk_head(head);
532}
533
534static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
535{
536 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
537}
538
539static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
540{
541 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
542}
543
544static inline struct sock *sk_next(const struct sock *sk)
545{
546 return sk->sk_node.next ?
547 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
548}
549
550static inline struct sock *sk_nulls_next(const struct sock *sk)
551{
552 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
553 hlist_nulls_entry(sk->sk_nulls_node.next,
554 struct sock, sk_nulls_node) :
555 NULL;
556}
557
558static inline bool sk_unhashed(const struct sock *sk)
559{
560 return hlist_unhashed(&sk->sk_node);
561}
562
563static inline bool sk_hashed(const struct sock *sk)
564{
565 return !sk_unhashed(sk);
566}
567
568static inline void sk_node_init(struct hlist_node *node)
569{
570 node->pprev = NULL;
571}
572
573static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
574{
575 node->pprev = NULL;
576}
577
578static inline void __sk_del_node(struct sock *sk)
579{
580 __hlist_del(&sk->sk_node);
581}
582
583/* NB: equivalent to hlist_del_init_rcu */
584static inline bool __sk_del_node_init(struct sock *sk)
585{
586 if (sk_hashed(sk)) {
587 __sk_del_node(sk);
588 sk_node_init(&sk->sk_node);
589 return true;
590 }
591 return false;
592}
593
594/* Grab socket reference count. This operation is valid only
595 when sk is ALREADY grabbed f.e. it is found in hash table
596 or a list and the lookup is made under lock preventing hash table
597 modifications.
598 */
599
600static __always_inline void sock_hold(struct sock *sk)
601{
602 atomic_inc(&sk->sk_refcnt);
603}
604
605/* Ungrab socket in the context, which assumes that socket refcnt
606 cannot hit zero, f.e. it is true in context of any socketcall.
607 */
608static __always_inline void __sock_put(struct sock *sk)
609{
610 atomic_dec(&sk->sk_refcnt);
611}
612
613static inline bool sk_del_node_init(struct sock *sk)
614{
615 bool rc = __sk_del_node_init(sk);
616
617 if (rc) {
618 /* paranoid for a while -acme */
619 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
620 __sock_put(sk);
621 }
622 return rc;
623}
624#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
625
626static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
627{
628 if (sk_hashed(sk)) {
629 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
630 return true;
631 }
632 return false;
633}
634
635static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
636{
637 bool rc = __sk_nulls_del_node_init_rcu(sk);
638
639 if (rc) {
640 /* paranoid for a while -acme */
641 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
642 __sock_put(sk);
643 }
644 return rc;
645}
646
647static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
648{
649 hlist_add_head(&sk->sk_node, list);
650}
651
652static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
653{
654 sock_hold(sk);
655 __sk_add_node(sk, list);
656}
657
658static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
659{
660 sock_hold(sk);
661 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
662 sk->sk_family == AF_INET6)
663 hlist_add_tail_rcu(&sk->sk_node, list);
664 else
665 hlist_add_head_rcu(&sk->sk_node, list);
666}
667
668static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
669{
670 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
671 sk->sk_family == AF_INET6)
672 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
673 else
674 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
675}
676
677static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
678{
679 sock_hold(sk);
680 __sk_nulls_add_node_rcu(sk, list);
681}
682
683static inline void __sk_del_bind_node(struct sock *sk)
684{
685 __hlist_del(&sk->sk_bind_node);
686}
687
688static inline void sk_add_bind_node(struct sock *sk,
689 struct hlist_head *list)
690{
691 hlist_add_head(&sk->sk_bind_node, list);
692}
693
694#define sk_for_each(__sk, list) \
695 hlist_for_each_entry(__sk, list, sk_node)
696#define sk_for_each_rcu(__sk, list) \
697 hlist_for_each_entry_rcu(__sk, list, sk_node)
698#define sk_nulls_for_each(__sk, node, list) \
699 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
700#define sk_nulls_for_each_rcu(__sk, node, list) \
701 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
702#define sk_for_each_from(__sk) \
703 hlist_for_each_entry_from(__sk, sk_node)
704#define sk_nulls_for_each_from(__sk, node) \
705 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
706 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
707#define sk_for_each_safe(__sk, tmp, list) \
708 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
709#define sk_for_each_bound(__sk, list) \
710 hlist_for_each_entry(__sk, list, sk_bind_node)
711
712/**
713 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
714 * @tpos: the type * to use as a loop cursor.
715 * @pos: the &struct hlist_node to use as a loop cursor.
716 * @head: the head for your list.
717 * @offset: offset of hlist_node within the struct.
718 *
719 */
720#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
721 for (pos = rcu_dereference((head)->first); \
722 pos != NULL && \
723 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
724 pos = rcu_dereference(pos->next))
725
726static inline struct user_namespace *sk_user_ns(struct sock *sk)
727{
728 /* Careful only use this in a context where these parameters
729 * can not change and must all be valid, such as recvmsg from
730 * userspace.
731 */
732 return sk->sk_socket->file->f_cred->user_ns;
733}
734
735/* Sock flags */
736enum sock_flags {
737 SOCK_DEAD,
738 SOCK_DONE,
739 SOCK_URGINLINE,
740 SOCK_KEEPOPEN,
741 SOCK_LINGER,
742 SOCK_DESTROY,
743 SOCK_BROADCAST,
744 SOCK_TIMESTAMP,
745 SOCK_ZAPPED,
746 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
747 SOCK_DBG, /* %SO_DEBUG setting */
748 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
749 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
750 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
751 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
752 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
753 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
754 SOCK_FASYNC, /* fasync() active */
755 SOCK_RXQ_OVFL,
756 SOCK_ZEROCOPY, /* buffers from userspace */
757 SOCK_WIFI_STATUS, /* push wifi status to userspace */
758 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
759 * Will use last 4 bytes of packet sent from
760 * user-space instead.
761 */
762 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
763 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
764 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
765};
766
767#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
768
769static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
770{
771 nsk->sk_flags = osk->sk_flags;
772}
773
774static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
775{
776 __set_bit(flag, &sk->sk_flags);
777}
778
779static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
780{
781 __clear_bit(flag, &sk->sk_flags);
782}
783
784static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
785{
786 return test_bit(flag, &sk->sk_flags);
787}
788
789#ifdef CONFIG_NET
790extern struct static_key memalloc_socks;
791static inline int sk_memalloc_socks(void)
792{
793 return static_key_false(&memalloc_socks);
794}
795#else
796
797static inline int sk_memalloc_socks(void)
798{
799 return 0;
800}
801
802#endif
803
804static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
805{
806 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
807}
808
809static inline void sk_acceptq_removed(struct sock *sk)
810{
811 sk->sk_ack_backlog--;
812}
813
814static inline void sk_acceptq_added(struct sock *sk)
815{
816 sk->sk_ack_backlog++;
817}
818
819static inline bool sk_acceptq_is_full(const struct sock *sk)
820{
821 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
822}
823
824/*
825 * Compute minimal free write space needed to queue new packets.
826 */
827static inline int sk_stream_min_wspace(const struct sock *sk)
828{
829 return sk->sk_wmem_queued >> 1;
830}
831
832static inline int sk_stream_wspace(const struct sock *sk)
833{
834 return sk->sk_sndbuf - sk->sk_wmem_queued;
835}
836
837void sk_stream_write_space(struct sock *sk);
838
839/* OOB backlog add */
840static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
841{
842 /* dont let skb dst not refcounted, we are going to leave rcu lock */
843 skb_dst_force_safe(skb);
844
845 if (!sk->sk_backlog.tail)
846 sk->sk_backlog.head = skb;
847 else
848 sk->sk_backlog.tail->next = skb;
849
850 sk->sk_backlog.tail = skb;
851 skb->next = NULL;
852}
853
854/*
855 * Take into account size of receive queue and backlog queue
856 * Do not take into account this skb truesize,
857 * to allow even a single big packet to come.
858 */
859static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
860{
861 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
862
863 return qsize > limit;
864}
865
866/* The per-socket spinlock must be held here. */
867static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
868 unsigned int limit)
869{
870 if (sk_rcvqueues_full(sk, limit))
871 return -ENOBUFS;
872
873 /*
874 * If the skb was allocated from pfmemalloc reserves, only
875 * allow SOCK_MEMALLOC sockets to use it as this socket is
876 * helping free memory
877 */
878 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
879 return -ENOMEM;
880
881 __sk_add_backlog(sk, skb);
882 sk->sk_backlog.len += skb->truesize;
883 return 0;
884}
885
886int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
887
888static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
889{
890 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
891 return __sk_backlog_rcv(sk, skb);
892
893 return sk->sk_backlog_rcv(sk, skb);
894}
895
896static inline void sk_incoming_cpu_update(struct sock *sk)
897{
898 sk->sk_incoming_cpu = raw_smp_processor_id();
899}
900
901static inline void sock_rps_record_flow_hash(__u32 hash)
902{
903#ifdef CONFIG_RPS
904 struct rps_sock_flow_table *sock_flow_table;
905
906 rcu_read_lock();
907 sock_flow_table = rcu_dereference(rps_sock_flow_table);
908 rps_record_sock_flow(sock_flow_table, hash);
909 rcu_read_unlock();
910#endif
911}
912
913static inline void sock_rps_record_flow(const struct sock *sk)
914{
915#ifdef CONFIG_RPS
916 if (static_key_false(&rfs_needed)) {
917 /* Reading sk->sk_rxhash might incur an expensive cache line
918 * miss.
919 *
920 * TCP_ESTABLISHED does cover almost all states where RFS
921 * might be useful, and is cheaper [1] than testing :
922 * IPv4: inet_sk(sk)->inet_daddr
923 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
924 * OR an additional socket flag
925 * [1] : sk_state and sk_prot are in the same cache line.
926 */
927 if (sk->sk_state == TCP_ESTABLISHED)
928 sock_rps_record_flow_hash(sk->sk_rxhash);
929 }
930#endif
931}
932
933static inline void sock_rps_save_rxhash(struct sock *sk,
934 const struct sk_buff *skb)
935{
936#ifdef CONFIG_RPS
937 if (unlikely(sk->sk_rxhash != skb->hash))
938 sk->sk_rxhash = skb->hash;
939#endif
940}
941
942static inline void sock_rps_reset_rxhash(struct sock *sk)
943{
944#ifdef CONFIG_RPS
945 sk->sk_rxhash = 0;
946#endif
947}
948
949#define sk_wait_event(__sk, __timeo, __condition, __wait) \
950 ({ int __rc; \
951 release_sock(__sk); \
952 __rc = __condition; \
953 if (!__rc) { \
954 *(__timeo) = wait_woken(__wait, \
955 TASK_INTERRUPTIBLE, \
956 *(__timeo)); \
957 } \
958 sched_annotate_sleep(); \
959 lock_sock(__sk); \
960 __rc = __condition; \
961 __rc; \
962 })
963
964int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
965int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
966void sk_stream_wait_close(struct sock *sk, long timeo_p);
967int sk_stream_error(struct sock *sk, int flags, int err);
968void sk_stream_kill_queues(struct sock *sk);
969void sk_set_memalloc(struct sock *sk);
970void sk_clear_memalloc(struct sock *sk);
971
972void __sk_flush_backlog(struct sock *sk);
973
974static inline bool sk_flush_backlog(struct sock *sk)
975{
976 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
977 __sk_flush_backlog(sk);
978 return true;
979 }
980 return false;
981}
982
983int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
984
985struct request_sock_ops;
986struct timewait_sock_ops;
987struct inet_hashinfo;
988struct raw_hashinfo;
989struct module;
990
991/*
992 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
993 * un-modified. Special care is taken when initializing object to zero.
994 */
995static inline void sk_prot_clear_nulls(struct sock *sk, int size)
996{
997 if (offsetof(struct sock, sk_node.next) != 0)
998 memset(sk, 0, offsetof(struct sock, sk_node.next));
999 memset(&sk->sk_node.pprev, 0,
1000 size - offsetof(struct sock, sk_node.pprev));
1001}
1002
1003/* Networking protocol blocks we attach to sockets.
1004 * socket layer -> transport layer interface
1005 */
1006struct proto {
1007 void (*close)(struct sock *sk,
1008 long timeout);
1009 int (*connect)(struct sock *sk,
1010 struct sockaddr *uaddr,
1011 int addr_len);
1012 int (*disconnect)(struct sock *sk, int flags);
1013
1014 struct sock * (*accept)(struct sock *sk, int flags, int *err);
1015
1016 int (*ioctl)(struct sock *sk, int cmd,
1017 unsigned long arg);
1018 int (*init)(struct sock *sk);
1019 void (*destroy)(struct sock *sk);
1020 void (*shutdown)(struct sock *sk, int how);
1021 int (*setsockopt)(struct sock *sk, int level,
1022 int optname, char __user *optval,
1023 unsigned int optlen);
1024 int (*getsockopt)(struct sock *sk, int level,
1025 int optname, char __user *optval,
1026 int __user *option);
1027#ifdef CONFIG_COMPAT
1028 int (*compat_setsockopt)(struct sock *sk,
1029 int level,
1030 int optname, char __user *optval,
1031 unsigned int optlen);
1032 int (*compat_getsockopt)(struct sock *sk,
1033 int level,
1034 int optname, char __user *optval,
1035 int __user *option);
1036 int (*compat_ioctl)(struct sock *sk,
1037 unsigned int cmd, unsigned long arg);
1038#endif
1039 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1040 size_t len);
1041 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1042 size_t len, int noblock, int flags,
1043 int *addr_len);
1044 int (*sendpage)(struct sock *sk, struct page *page,
1045 int offset, size_t size, int flags);
1046 int (*bind)(struct sock *sk,
1047 struct sockaddr *uaddr, int addr_len);
1048
1049 int (*backlog_rcv) (struct sock *sk,
1050 struct sk_buff *skb);
1051
1052 void (*release_cb)(struct sock *sk);
1053
1054 /* Keeping track of sk's, looking them up, and port selection methods. */
1055 int (*hash)(struct sock *sk);
1056 void (*unhash)(struct sock *sk);
1057 void (*rehash)(struct sock *sk);
1058 int (*get_port)(struct sock *sk, unsigned short snum);
1059
1060 /* Keeping track of sockets in use */
1061#ifdef CONFIG_PROC_FS
1062 unsigned int inuse_idx;
1063#endif
1064
1065 bool (*stream_memory_free)(const struct sock *sk);
1066 /* Memory pressure */
1067 void (*enter_memory_pressure)(struct sock *sk);
1068 atomic_long_t *memory_allocated; /* Current allocated memory. */
1069 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1070 /*
1071 * Pressure flag: try to collapse.
1072 * Technical note: it is used by multiple contexts non atomically.
1073 * All the __sk_mem_schedule() is of this nature: accounting
1074 * is strict, actions are advisory and have some latency.
1075 */
1076 int *memory_pressure;
1077 long *sysctl_mem;
1078 int *sysctl_wmem;
1079 int *sysctl_rmem;
1080 int max_header;
1081 bool no_autobind;
1082
1083 struct kmem_cache *slab;
1084 unsigned int obj_size;
1085 int slab_flags;
1086
1087 struct percpu_counter *orphan_count;
1088
1089 struct request_sock_ops *rsk_prot;
1090 struct timewait_sock_ops *twsk_prot;
1091
1092 union {
1093 struct inet_hashinfo *hashinfo;
1094 struct udp_table *udp_table;
1095 struct raw_hashinfo *raw_hash;
1096 } h;
1097
1098 struct module *owner;
1099
1100 char name[32];
1101
1102 struct list_head node;
1103#ifdef SOCK_REFCNT_DEBUG
1104 atomic_t socks;
1105#endif
1106 int (*diag_destroy)(struct sock *sk, int err);
1107};
1108
1109int proto_register(struct proto *prot, int alloc_slab);
1110void proto_unregister(struct proto *prot);
1111
1112#ifdef SOCK_REFCNT_DEBUG
1113static inline void sk_refcnt_debug_inc(struct sock *sk)
1114{
1115 atomic_inc(&sk->sk_prot->socks);
1116}
1117
1118static inline void sk_refcnt_debug_dec(struct sock *sk)
1119{
1120 atomic_dec(&sk->sk_prot->socks);
1121 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1122 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1123}
1124
1125static inline void sk_refcnt_debug_release(const struct sock *sk)
1126{
1127 if (atomic_read(&sk->sk_refcnt) != 1)
1128 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1129 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1130}
1131#else /* SOCK_REFCNT_DEBUG */
1132#define sk_refcnt_debug_inc(sk) do { } while (0)
1133#define sk_refcnt_debug_dec(sk) do { } while (0)
1134#define sk_refcnt_debug_release(sk) do { } while (0)
1135#endif /* SOCK_REFCNT_DEBUG */
1136
1137static inline bool sk_stream_memory_free(const struct sock *sk)
1138{
1139 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1140 return false;
1141
1142 return sk->sk_prot->stream_memory_free ?
1143 sk->sk_prot->stream_memory_free(sk) : true;
1144}
1145
1146static inline bool sk_stream_is_writeable(const struct sock *sk)
1147{
1148 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1149 sk_stream_memory_free(sk);
1150}
1151
1152static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1153 struct cgroup *ancestor)
1154{
1155#ifdef CONFIG_SOCK_CGROUP_DATA
1156 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1157 ancestor);
1158#else
1159 return -ENOTSUPP;
1160#endif
1161}
1162
1163static inline bool sk_has_memory_pressure(const struct sock *sk)
1164{
1165 return sk->sk_prot->memory_pressure != NULL;
1166}
1167
1168static inline bool sk_under_memory_pressure(const struct sock *sk)
1169{
1170 if (!sk->sk_prot->memory_pressure)
1171 return false;
1172
1173 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1174 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1175 return true;
1176
1177 return !!*sk->sk_prot->memory_pressure;
1178}
1179
1180static inline void sk_leave_memory_pressure(struct sock *sk)
1181{
1182 int *memory_pressure = sk->sk_prot->memory_pressure;
1183
1184 if (!memory_pressure)
1185 return;
1186
1187 if (*memory_pressure)
1188 *memory_pressure = 0;
1189}
1190
1191static inline void sk_enter_memory_pressure(struct sock *sk)
1192{
1193 if (!sk->sk_prot->enter_memory_pressure)
1194 return;
1195
1196 sk->sk_prot->enter_memory_pressure(sk);
1197}
1198
1199static inline long
1200sk_memory_allocated(const struct sock *sk)
1201{
1202 return atomic_long_read(sk->sk_prot->memory_allocated);
1203}
1204
1205static inline long
1206sk_memory_allocated_add(struct sock *sk, int amt)
1207{
1208 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1209}
1210
1211static inline void
1212sk_memory_allocated_sub(struct sock *sk, int amt)
1213{
1214 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1215}
1216
1217static inline void sk_sockets_allocated_dec(struct sock *sk)
1218{
1219 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1220}
1221
1222static inline void sk_sockets_allocated_inc(struct sock *sk)
1223{
1224 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1225}
1226
1227static inline int
1228sk_sockets_allocated_read_positive(struct sock *sk)
1229{
1230 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1231}
1232
1233static inline int
1234proto_sockets_allocated_sum_positive(struct proto *prot)
1235{
1236 return percpu_counter_sum_positive(prot->sockets_allocated);
1237}
1238
1239static inline long
1240proto_memory_allocated(struct proto *prot)
1241{
1242 return atomic_long_read(prot->memory_allocated);
1243}
1244
1245static inline bool
1246proto_memory_pressure(struct proto *prot)
1247{
1248 if (!prot->memory_pressure)
1249 return false;
1250 return !!*prot->memory_pressure;
1251}
1252
1253
1254#ifdef CONFIG_PROC_FS
1255/* Called with local bh disabled */
1256void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1257int sock_prot_inuse_get(struct net *net, struct proto *proto);
1258#else
1259static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1260 int inc)
1261{
1262}
1263#endif
1264
1265
1266/* With per-bucket locks this operation is not-atomic, so that
1267 * this version is not worse.
1268 */
1269static inline int __sk_prot_rehash(struct sock *sk)
1270{
1271 sk->sk_prot->unhash(sk);
1272 return sk->sk_prot->hash(sk);
1273}
1274
1275/* About 10 seconds */
1276#define SOCK_DESTROY_TIME (10*HZ)
1277
1278/* Sockets 0-1023 can't be bound to unless you are superuser */
1279#define PROT_SOCK 1024
1280
1281#define SHUTDOWN_MASK 3
1282#define RCV_SHUTDOWN 1
1283#define SEND_SHUTDOWN 2
1284
1285#define SOCK_SNDBUF_LOCK 1
1286#define SOCK_RCVBUF_LOCK 2
1287#define SOCK_BINDADDR_LOCK 4
1288#define SOCK_BINDPORT_LOCK 8
1289
1290struct socket_alloc {
1291 struct socket socket;
1292 struct inode vfs_inode;
1293};
1294
1295static inline struct socket *SOCKET_I(struct inode *inode)
1296{
1297 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1298}
1299
1300static inline struct inode *SOCK_INODE(struct socket *socket)
1301{
1302 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1303}
1304
1305/*
1306 * Functions for memory accounting
1307 */
1308int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1309int __sk_mem_schedule(struct sock *sk, int size, int kind);
1310void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1311void __sk_mem_reclaim(struct sock *sk, int amount);
1312
1313/* We used to have PAGE_SIZE here, but systems with 64KB pages
1314 * do not necessarily have 16x time more memory than 4KB ones.
1315 */
1316#define SK_MEM_QUANTUM 4096
1317#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1318#define SK_MEM_SEND 0
1319#define SK_MEM_RECV 1
1320
1321/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1322static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1323{
1324 long val = sk->sk_prot->sysctl_mem[index];
1325
1326#if PAGE_SIZE > SK_MEM_QUANTUM
1327 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1328#elif PAGE_SIZE < SK_MEM_QUANTUM
1329 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1330#endif
1331 return val;
1332}
1333
1334static inline int sk_mem_pages(int amt)
1335{
1336 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1337}
1338
1339static inline bool sk_has_account(struct sock *sk)
1340{
1341 /* return true if protocol supports memory accounting */
1342 return !!sk->sk_prot->memory_allocated;
1343}
1344
1345static inline bool sk_wmem_schedule(struct sock *sk, int size)
1346{
1347 if (!sk_has_account(sk))
1348 return true;
1349 return size <= sk->sk_forward_alloc ||
1350 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1351}
1352
1353static inline bool
1354sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1355{
1356 if (!sk_has_account(sk))
1357 return true;
1358 return size<= sk->sk_forward_alloc ||
1359 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1360 skb_pfmemalloc(skb);
1361}
1362
1363static inline void sk_mem_reclaim(struct sock *sk)
1364{
1365 if (!sk_has_account(sk))
1366 return;
1367 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1368 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1369}
1370
1371static inline void sk_mem_reclaim_partial(struct sock *sk)
1372{
1373 if (!sk_has_account(sk))
1374 return;
1375 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1376 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1377}
1378
1379static inline void sk_mem_charge(struct sock *sk, int size)
1380{
1381 if (!sk_has_account(sk))
1382 return;
1383 sk->sk_forward_alloc -= size;
1384}
1385
1386static inline void sk_mem_uncharge(struct sock *sk, int size)
1387{
1388 if (!sk_has_account(sk))
1389 return;
1390 sk->sk_forward_alloc += size;
1391
1392 /* Avoid a possible overflow.
1393 * TCP send queues can make this happen, if sk_mem_reclaim()
1394 * is not called and more than 2 GBytes are released at once.
1395 *
1396 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1397 * no need to hold that much forward allocation anyway.
1398 */
1399 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1400 __sk_mem_reclaim(sk, 1 << 20);
1401}
1402
1403static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1404{
1405 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1406 sk->sk_wmem_queued -= skb->truesize;
1407 sk_mem_uncharge(sk, skb->truesize);
1408 __kfree_skb(skb);
1409}
1410
1411static inline void sock_release_ownership(struct sock *sk)
1412{
1413 if (sk->sk_lock.owned) {
1414 sk->sk_lock.owned = 0;
1415
1416 /* The sk_lock has mutex_unlock() semantics: */
1417 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1418 }
1419}
1420
1421/*
1422 * Macro so as to not evaluate some arguments when
1423 * lockdep is not enabled.
1424 *
1425 * Mark both the sk_lock and the sk_lock.slock as a
1426 * per-address-family lock class.
1427 */
1428#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1429do { \
1430 sk->sk_lock.owned = 0; \
1431 init_waitqueue_head(&sk->sk_lock.wq); \
1432 spin_lock_init(&(sk)->sk_lock.slock); \
1433 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1434 sizeof((sk)->sk_lock)); \
1435 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1436 (skey), (sname)); \
1437 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1438} while (0)
1439
1440#ifdef CONFIG_LOCKDEP
1441static inline bool lockdep_sock_is_held(const struct sock *csk)
1442{
1443 struct sock *sk = (struct sock *)csk;
1444
1445 return lockdep_is_held(&sk->sk_lock) ||
1446 lockdep_is_held(&sk->sk_lock.slock);
1447}
1448#endif
1449
1450void lock_sock_nested(struct sock *sk, int subclass);
1451
1452static inline void lock_sock(struct sock *sk)
1453{
1454 lock_sock_nested(sk, 0);
1455}
1456
1457void release_sock(struct sock *sk);
1458
1459/* BH context may only use the following locking interface. */
1460#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1461#define bh_lock_sock_nested(__sk) \
1462 spin_lock_nested(&((__sk)->sk_lock.slock), \
1463 SINGLE_DEPTH_NESTING)
1464#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1465
1466bool lock_sock_fast(struct sock *sk);
1467/**
1468 * unlock_sock_fast - complement of lock_sock_fast
1469 * @sk: socket
1470 * @slow: slow mode
1471 *
1472 * fast unlock socket for user context.
1473 * If slow mode is on, we call regular release_sock()
1474 */
1475static inline void unlock_sock_fast(struct sock *sk, bool slow)
1476{
1477 if (slow)
1478 release_sock(sk);
1479 else
1480 spin_unlock_bh(&sk->sk_lock.slock);
1481}
1482
1483/* Used by processes to "lock" a socket state, so that
1484 * interrupts and bottom half handlers won't change it
1485 * from under us. It essentially blocks any incoming
1486 * packets, so that we won't get any new data or any
1487 * packets that change the state of the socket.
1488 *
1489 * While locked, BH processing will add new packets to
1490 * the backlog queue. This queue is processed by the
1491 * owner of the socket lock right before it is released.
1492 *
1493 * Since ~2.3.5 it is also exclusive sleep lock serializing
1494 * accesses from user process context.
1495 */
1496
1497static inline void sock_owned_by_me(const struct sock *sk)
1498{
1499#ifdef CONFIG_LOCKDEP
1500 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1501#endif
1502}
1503
1504static inline bool sock_owned_by_user(const struct sock *sk)
1505{
1506 sock_owned_by_me(sk);
1507 return sk->sk_lock.owned;
1508}
1509
1510/* no reclassification while locks are held */
1511static inline bool sock_allow_reclassification(const struct sock *csk)
1512{
1513 struct sock *sk = (struct sock *)csk;
1514
1515 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1516}
1517
1518struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1519 struct proto *prot, int kern);
1520void sk_free(struct sock *sk);
1521void sk_destruct(struct sock *sk);
1522struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1523
1524struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1525 gfp_t priority);
1526void __sock_wfree(struct sk_buff *skb);
1527void sock_wfree(struct sk_buff *skb);
1528void skb_orphan_partial(struct sk_buff *skb);
1529void sock_rfree(struct sk_buff *skb);
1530void sock_efree(struct sk_buff *skb);
1531#ifdef CONFIG_INET
1532void sock_edemux(struct sk_buff *skb);
1533#else
1534#define sock_edemux(skb) sock_efree(skb)
1535#endif
1536
1537int sock_setsockopt(struct socket *sock, int level, int op,
1538 char __user *optval, unsigned int optlen);
1539
1540int sock_getsockopt(struct socket *sock, int level, int op,
1541 char __user *optval, int __user *optlen);
1542struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1543 int noblock, int *errcode);
1544struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1545 unsigned long data_len, int noblock,
1546 int *errcode, int max_page_order);
1547void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1548void sock_kfree_s(struct sock *sk, void *mem, int size);
1549void sock_kzfree_s(struct sock *sk, void *mem, int size);
1550void sk_send_sigurg(struct sock *sk);
1551
1552struct sockcm_cookie {
1553 u32 mark;
1554 u16 tsflags;
1555};
1556
1557int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1558 struct sockcm_cookie *sockc);
1559int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1560 struct sockcm_cookie *sockc);
1561
1562/*
1563 * Functions to fill in entries in struct proto_ops when a protocol
1564 * does not implement a particular function.
1565 */
1566int sock_no_bind(struct socket *, struct sockaddr *, int);
1567int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1568int sock_no_socketpair(struct socket *, struct socket *);
1569int sock_no_accept(struct socket *, struct socket *, int);
1570int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1571unsigned int sock_no_poll(struct file *, struct socket *,
1572 struct poll_table_struct *);
1573int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1574int sock_no_listen(struct socket *, int);
1575int sock_no_shutdown(struct socket *, int);
1576int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1577int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1578int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1579int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1580int sock_no_mmap(struct file *file, struct socket *sock,
1581 struct vm_area_struct *vma);
1582ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1583 size_t size, int flags);
1584
1585/*
1586 * Functions to fill in entries in struct proto_ops when a protocol
1587 * uses the inet style.
1588 */
1589int sock_common_getsockopt(struct socket *sock, int level, int optname,
1590 char __user *optval, int __user *optlen);
1591int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1592 int flags);
1593int sock_common_setsockopt(struct socket *sock, int level, int optname,
1594 char __user *optval, unsigned int optlen);
1595int compat_sock_common_getsockopt(struct socket *sock, int level,
1596 int optname, char __user *optval, int __user *optlen);
1597int compat_sock_common_setsockopt(struct socket *sock, int level,
1598 int optname, char __user *optval, unsigned int optlen);
1599
1600void sk_common_release(struct sock *sk);
1601
1602/*
1603 * Default socket callbacks and setup code
1604 */
1605
1606/* Initialise core socket variables */
1607void sock_init_data(struct socket *sock, struct sock *sk);
1608
1609/*
1610 * Socket reference counting postulates.
1611 *
1612 * * Each user of socket SHOULD hold a reference count.
1613 * * Each access point to socket (an hash table bucket, reference from a list,
1614 * running timer, skb in flight MUST hold a reference count.
1615 * * When reference count hits 0, it means it will never increase back.
1616 * * When reference count hits 0, it means that no references from
1617 * outside exist to this socket and current process on current CPU
1618 * is last user and may/should destroy this socket.
1619 * * sk_free is called from any context: process, BH, IRQ. When
1620 * it is called, socket has no references from outside -> sk_free
1621 * may release descendant resources allocated by the socket, but
1622 * to the time when it is called, socket is NOT referenced by any
1623 * hash tables, lists etc.
1624 * * Packets, delivered from outside (from network or from another process)
1625 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1626 * when they sit in queue. Otherwise, packets will leak to hole, when
1627 * socket is looked up by one cpu and unhasing is made by another CPU.
1628 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1629 * (leak to backlog). Packet socket does all the processing inside
1630 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1631 * use separate SMP lock, so that they are prone too.
1632 */
1633
1634/* Ungrab socket and destroy it, if it was the last reference. */
1635static inline void sock_put(struct sock *sk)
1636{
1637 if (atomic_dec_and_test(&sk->sk_refcnt))
1638 sk_free(sk);
1639}
1640/* Generic version of sock_put(), dealing with all sockets
1641 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1642 */
1643void sock_gen_put(struct sock *sk);
1644
1645int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1646 unsigned int trim_cap, bool refcounted);
1647static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1648 const int nested)
1649{
1650 return __sk_receive_skb(sk, skb, nested, 1, true);
1651}
1652
1653static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1654{
1655 sk->sk_tx_queue_mapping = tx_queue;
1656}
1657
1658static inline void sk_tx_queue_clear(struct sock *sk)
1659{
1660 sk->sk_tx_queue_mapping = -1;
1661}
1662
1663static inline int sk_tx_queue_get(const struct sock *sk)
1664{
1665 return sk ? sk->sk_tx_queue_mapping : -1;
1666}
1667
1668static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1669{
1670 sk_tx_queue_clear(sk);
1671 sk->sk_socket = sock;
1672}
1673
1674static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1675{
1676 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1677 return &rcu_dereference_raw(sk->sk_wq)->wait;
1678}
1679/* Detach socket from process context.
1680 * Announce socket dead, detach it from wait queue and inode.
1681 * Note that parent inode held reference count on this struct sock,
1682 * we do not release it in this function, because protocol
1683 * probably wants some additional cleanups or even continuing
1684 * to work with this socket (TCP).
1685 */
1686static inline void sock_orphan(struct sock *sk)
1687{
1688 write_lock_bh(&sk->sk_callback_lock);
1689 sock_set_flag(sk, SOCK_DEAD);
1690 sk_set_socket(sk, NULL);
1691 sk->sk_wq = NULL;
1692 write_unlock_bh(&sk->sk_callback_lock);
1693}
1694
1695static inline void sock_graft(struct sock *sk, struct socket *parent)
1696{
1697 write_lock_bh(&sk->sk_callback_lock);
1698 sk->sk_wq = parent->wq;
1699 parent->sk = sk;
1700 sk_set_socket(sk, parent);
1701 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1702 security_sock_graft(sk, parent);
1703 write_unlock_bh(&sk->sk_callback_lock);
1704}
1705
1706kuid_t sock_i_uid(struct sock *sk);
1707unsigned long sock_i_ino(struct sock *sk);
1708
1709static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1710{
1711 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1712}
1713
1714static inline u32 net_tx_rndhash(void)
1715{
1716 u32 v = prandom_u32();
1717
1718 return v ?: 1;
1719}
1720
1721static inline void sk_set_txhash(struct sock *sk)
1722{
1723 sk->sk_txhash = net_tx_rndhash();
1724}
1725
1726static inline void sk_rethink_txhash(struct sock *sk)
1727{
1728 if (sk->sk_txhash)
1729 sk_set_txhash(sk);
1730}
1731
1732static inline struct dst_entry *
1733__sk_dst_get(struct sock *sk)
1734{
1735 return rcu_dereference_check(sk->sk_dst_cache,
1736 lockdep_sock_is_held(sk));
1737}
1738
1739static inline struct dst_entry *
1740sk_dst_get(struct sock *sk)
1741{
1742 struct dst_entry *dst;
1743
1744 rcu_read_lock();
1745 dst = rcu_dereference(sk->sk_dst_cache);
1746 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1747 dst = NULL;
1748 rcu_read_unlock();
1749 return dst;
1750}
1751
1752static inline void dst_negative_advice(struct sock *sk)
1753{
1754 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1755
1756 sk_rethink_txhash(sk);
1757
1758 if (dst && dst->ops->negative_advice) {
1759 ndst = dst->ops->negative_advice(dst);
1760
1761 if (ndst != dst) {
1762 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1763 sk_tx_queue_clear(sk);
1764 }
1765 }
1766}
1767
1768static inline void
1769__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1770{
1771 struct dst_entry *old_dst;
1772
1773 sk_tx_queue_clear(sk);
1774 /*
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
1777 */
1778 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1779 rcu_assign_pointer(sk->sk_dst_cache, dst);
1780 dst_release(old_dst);
1781}
1782
1783static inline void
1784sk_dst_set(struct sock *sk, struct dst_entry *dst)
1785{
1786 struct dst_entry *old_dst;
1787
1788 sk_tx_queue_clear(sk);
1789 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1790 dst_release(old_dst);
1791}
1792
1793static inline void
1794__sk_dst_reset(struct sock *sk)
1795{
1796 __sk_dst_set(sk, NULL);
1797}
1798
1799static inline void
1800sk_dst_reset(struct sock *sk)
1801{
1802 sk_dst_set(sk, NULL);
1803}
1804
1805struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1806
1807struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1808
1809bool sk_mc_loop(struct sock *sk);
1810
1811static inline bool sk_can_gso(const struct sock *sk)
1812{
1813 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1814}
1815
1816void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1817
1818static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1819{
1820 sk->sk_route_nocaps |= flags;
1821 sk->sk_route_caps &= ~flags;
1822}
1823
1824static inline bool sk_check_csum_caps(struct sock *sk)
1825{
1826 return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1827 (sk->sk_family == PF_INET &&
1828 (sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1829 (sk->sk_family == PF_INET6 &&
1830 (sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1831}
1832
1833static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1834 struct iov_iter *from, char *to,
1835 int copy, int offset)
1836{
1837 if (skb->ip_summed == CHECKSUM_NONE) {
1838 __wsum csum = 0;
1839 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1840 return -EFAULT;
1841 skb->csum = csum_block_add(skb->csum, csum, offset);
1842 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1843 if (!copy_from_iter_full_nocache(to, copy, from))
1844 return -EFAULT;
1845 } else if (!copy_from_iter_full(to, copy, from))
1846 return -EFAULT;
1847
1848 return 0;
1849}
1850
1851static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1852 struct iov_iter *from, int copy)
1853{
1854 int err, offset = skb->len;
1855
1856 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1857 copy, offset);
1858 if (err)
1859 __skb_trim(skb, offset);
1860
1861 return err;
1862}
1863
1864static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1865 struct sk_buff *skb,
1866 struct page *page,
1867 int off, int copy)
1868{
1869 int err;
1870
1871 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1872 copy, skb->len);
1873 if (err)
1874 return err;
1875
1876 skb->len += copy;
1877 skb->data_len += copy;
1878 skb->truesize += copy;
1879 sk->sk_wmem_queued += copy;
1880 sk_mem_charge(sk, copy);
1881 return 0;
1882}
1883
1884/**
1885 * sk_wmem_alloc_get - returns write allocations
1886 * @sk: socket
1887 *
1888 * Returns sk_wmem_alloc minus initial offset of one
1889 */
1890static inline int sk_wmem_alloc_get(const struct sock *sk)
1891{
1892 return atomic_read(&sk->sk_wmem_alloc) - 1;
1893}
1894
1895/**
1896 * sk_rmem_alloc_get - returns read allocations
1897 * @sk: socket
1898 *
1899 * Returns sk_rmem_alloc
1900 */
1901static inline int sk_rmem_alloc_get(const struct sock *sk)
1902{
1903 return atomic_read(&sk->sk_rmem_alloc);
1904}
1905
1906/**
1907 * sk_has_allocations - check if allocations are outstanding
1908 * @sk: socket
1909 *
1910 * Returns true if socket has write or read allocations
1911 */
1912static inline bool sk_has_allocations(const struct sock *sk)
1913{
1914 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1915}
1916
1917/**
1918 * skwq_has_sleeper - check if there are any waiting processes
1919 * @wq: struct socket_wq
1920 *
1921 * Returns true if socket_wq has waiting processes
1922 *
1923 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1924 * barrier call. They were added due to the race found within the tcp code.
1925 *
1926 * Consider following tcp code paths:
1927 *
1928 * CPU1 CPU2
1929 *
1930 * sys_select receive packet
1931 * ... ...
1932 * __add_wait_queue update tp->rcv_nxt
1933 * ... ...
1934 * tp->rcv_nxt check sock_def_readable
1935 * ... {
1936 * schedule rcu_read_lock();
1937 * wq = rcu_dereference(sk->sk_wq);
1938 * if (wq && waitqueue_active(&wq->wait))
1939 * wake_up_interruptible(&wq->wait)
1940 * ...
1941 * }
1942 *
1943 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1944 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1945 * could then endup calling schedule and sleep forever if there are no more
1946 * data on the socket.
1947 *
1948 */
1949static inline bool skwq_has_sleeper(struct socket_wq *wq)
1950{
1951 return wq && wq_has_sleeper(&wq->wait);
1952}
1953
1954/**
1955 * sock_poll_wait - place memory barrier behind the poll_wait call.
1956 * @filp: file
1957 * @wait_address: socket wait queue
1958 * @p: poll_table
1959 *
1960 * See the comments in the wq_has_sleeper function.
1961 */
1962static inline void sock_poll_wait(struct file *filp,
1963 wait_queue_head_t *wait_address, poll_table *p)
1964{
1965 if (!poll_does_not_wait(p) && wait_address) {
1966 poll_wait(filp, wait_address, p);
1967 /* We need to be sure we are in sync with the
1968 * socket flags modification.
1969 *
1970 * This memory barrier is paired in the wq_has_sleeper.
1971 */
1972 smp_mb();
1973 }
1974}
1975
1976static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1977{
1978 if (sk->sk_txhash) {
1979 skb->l4_hash = 1;
1980 skb->hash = sk->sk_txhash;
1981 }
1982}
1983
1984void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
1985
1986/*
1987 * Queue a received datagram if it will fit. Stream and sequenced
1988 * protocols can't normally use this as they need to fit buffers in
1989 * and play with them.
1990 *
1991 * Inlined as it's very short and called for pretty much every
1992 * packet ever received.
1993 */
1994static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1995{
1996 skb_orphan(skb);
1997 skb->sk = sk;
1998 skb->destructor = sock_rfree;
1999 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2000 sk_mem_charge(sk, skb->truesize);
2001}
2002
2003void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2004 unsigned long expires);
2005
2006void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2007
2008int __sk_queue_drop_skb(struct sock *sk, struct sk_buff *skb,
2009 unsigned int flags,
2010 void (*destructor)(struct sock *sk,
2011 struct sk_buff *skb));
2012int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2013int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2014
2015int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2016struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2017
2018/*
2019 * Recover an error report and clear atomically
2020 */
2021
2022static inline int sock_error(struct sock *sk)
2023{
2024 int err;
2025 if (likely(!sk->sk_err))
2026 return 0;
2027 err = xchg(&sk->sk_err, 0);
2028 return -err;
2029}
2030
2031static inline unsigned long sock_wspace(struct sock *sk)
2032{
2033 int amt = 0;
2034
2035 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2036 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2037 if (amt < 0)
2038 amt = 0;
2039 }
2040 return amt;
2041}
2042
2043/* Note:
2044 * We use sk->sk_wq_raw, from contexts knowing this
2045 * pointer is not NULL and cannot disappear/change.
2046 */
2047static inline void sk_set_bit(int nr, struct sock *sk)
2048{
2049 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2050 !sock_flag(sk, SOCK_FASYNC))
2051 return;
2052
2053 set_bit(nr, &sk->sk_wq_raw->flags);
2054}
2055
2056static inline void sk_clear_bit(int nr, struct sock *sk)
2057{
2058 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2059 !sock_flag(sk, SOCK_FASYNC))
2060 return;
2061
2062 clear_bit(nr, &sk->sk_wq_raw->flags);
2063}
2064
2065static inline void sk_wake_async(const struct sock *sk, int how, int band)
2066{
2067 if (sock_flag(sk, SOCK_FASYNC)) {
2068 rcu_read_lock();
2069 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2070 rcu_read_unlock();
2071 }
2072}
2073
2074/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2075 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2076 * Note: for send buffers, TCP works better if we can build two skbs at
2077 * minimum.
2078 */
2079#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2080
2081#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2082#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2083
2084static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2085{
2086 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2087 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2088 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2089 }
2090}
2091
2092struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2093 bool force_schedule);
2094
2095/**
2096 * sk_page_frag - return an appropriate page_frag
2097 * @sk: socket
2098 *
2099 * If socket allocation mode allows current thread to sleep, it means its
2100 * safe to use the per task page_frag instead of the per socket one.
2101 */
2102static inline struct page_frag *sk_page_frag(struct sock *sk)
2103{
2104 if (gfpflags_allow_blocking(sk->sk_allocation))
2105 return ¤t->task_frag;
2106
2107 return &sk->sk_frag;
2108}
2109
2110bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2111
2112/*
2113 * Default write policy as shown to user space via poll/select/SIGIO
2114 */
2115static inline bool sock_writeable(const struct sock *sk)
2116{
2117 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2118}
2119
2120static inline gfp_t gfp_any(void)
2121{
2122 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2123}
2124
2125static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2126{
2127 return noblock ? 0 : sk->sk_rcvtimeo;
2128}
2129
2130static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2131{
2132 return noblock ? 0 : sk->sk_sndtimeo;
2133}
2134
2135static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2136{
2137 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2138}
2139
2140/* Alas, with timeout socket operations are not restartable.
2141 * Compare this to poll().
2142 */
2143static inline int sock_intr_errno(long timeo)
2144{
2145 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2146}
2147
2148struct sock_skb_cb {
2149 u32 dropcount;
2150};
2151
2152/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2153 * using skb->cb[] would keep using it directly and utilize its
2154 * alignement guarantee.
2155 */
2156#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2157 sizeof(struct sock_skb_cb)))
2158
2159#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2160 SOCK_SKB_CB_OFFSET))
2161
2162#define sock_skb_cb_check_size(size) \
2163 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2164
2165static inline void
2166sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2167{
2168 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2169 atomic_read(&sk->sk_drops) : 0;
2170}
2171
2172static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2173{
2174 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2175
2176 atomic_add(segs, &sk->sk_drops);
2177}
2178
2179void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2180 struct sk_buff *skb);
2181void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2182 struct sk_buff *skb);
2183
2184static inline void
2185sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2186{
2187 ktime_t kt = skb->tstamp;
2188 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2189
2190 /*
2191 * generate control messages if
2192 * - receive time stamping in software requested
2193 * - software time stamp available and wanted
2194 * - hardware time stamps available and wanted
2195 */
2196 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2197 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2198 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2199 (hwtstamps->hwtstamp &&
2200 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2201 __sock_recv_timestamp(msg, sk, skb);
2202 else
2203 sk->sk_stamp = kt;
2204
2205 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2206 __sock_recv_wifi_status(msg, sk, skb);
2207}
2208
2209void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2210 struct sk_buff *skb);
2211
2212static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2213 struct sk_buff *skb)
2214{
2215#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2216 (1UL << SOCK_RCVTSTAMP))
2217#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2218 SOF_TIMESTAMPING_RAW_HARDWARE)
2219
2220 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2221 __sock_recv_ts_and_drops(msg, sk, skb);
2222 else
2223 sk->sk_stamp = skb->tstamp;
2224}
2225
2226void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2227
2228/**
2229 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2230 * @sk: socket sending this packet
2231 * @tsflags: timestamping flags to use
2232 * @tx_flags: completed with instructions for time stamping
2233 *
2234 * Note : callers should take care of initial *tx_flags value (usually 0)
2235 */
2236static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2237 __u8 *tx_flags)
2238{
2239 if (unlikely(tsflags))
2240 __sock_tx_timestamp(tsflags, tx_flags);
2241 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2242 *tx_flags |= SKBTX_WIFI_STATUS;
2243}
2244
2245/**
2246 * sk_eat_skb - Release a skb if it is no longer needed
2247 * @sk: socket to eat this skb from
2248 * @skb: socket buffer to eat
2249 *
2250 * This routine must be called with interrupts disabled or with the socket
2251 * locked so that the sk_buff queue operation is ok.
2252*/
2253static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2254{
2255 __skb_unlink(skb, &sk->sk_receive_queue);
2256 __kfree_skb(skb);
2257}
2258
2259static inline
2260struct net *sock_net(const struct sock *sk)
2261{
2262 return read_pnet(&sk->sk_net);
2263}
2264
2265static inline
2266void sock_net_set(struct sock *sk, struct net *net)
2267{
2268 write_pnet(&sk->sk_net, net);
2269}
2270
2271static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2272{
2273 if (skb->sk) {
2274 struct sock *sk = skb->sk;
2275
2276 skb->destructor = NULL;
2277 skb->sk = NULL;
2278 return sk;
2279 }
2280 return NULL;
2281}
2282
2283/* This helper checks if a socket is a full socket,
2284 * ie _not_ a timewait or request socket.
2285 */
2286static inline bool sk_fullsock(const struct sock *sk)
2287{
2288 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2289}
2290
2291/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2292 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2293 */
2294static inline bool sk_listener(const struct sock *sk)
2295{
2296 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2297}
2298
2299/**
2300 * sk_state_load - read sk->sk_state for lockless contexts
2301 * @sk: socket pointer
2302 *
2303 * Paired with sk_state_store(). Used in places we do not hold socket lock :
2304 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2305 */
2306static inline int sk_state_load(const struct sock *sk)
2307{
2308 return smp_load_acquire(&sk->sk_state);
2309}
2310
2311/**
2312 * sk_state_store - update sk->sk_state
2313 * @sk: socket pointer
2314 * @newstate: new state
2315 *
2316 * Paired with sk_state_load(). Should be used in contexts where
2317 * state change might impact lockless readers.
2318 */
2319static inline void sk_state_store(struct sock *sk, int newstate)
2320{
2321 smp_store_release(&sk->sk_state, newstate);
2322}
2323
2324void sock_enable_timestamp(struct sock *sk, int flag);
2325int sock_get_timestamp(struct sock *, struct timeval __user *);
2326int sock_get_timestampns(struct sock *, struct timespec __user *);
2327int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2328 int type);
2329
2330bool sk_ns_capable(const struct sock *sk,
2331 struct user_namespace *user_ns, int cap);
2332bool sk_capable(const struct sock *sk, int cap);
2333bool sk_net_capable(const struct sock *sk, int cap);
2334
2335extern __u32 sysctl_wmem_max;
2336extern __u32 sysctl_rmem_max;
2337
2338extern int sysctl_tstamp_allow_data;
2339extern int sysctl_optmem_max;
2340
2341extern __u32 sysctl_wmem_default;
2342extern __u32 sysctl_rmem_default;
2343
2344#endif /* _SOCK_H */