Loading...
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the AF_INET socket handler.
8 *
9 * Version: @(#)sock.h 1.0.4 05/13/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche <flla@stud.uni-sb.de>
15 *
16 * Fixes:
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
20 * than the reverse.
21 * Alan Cox : Added a zapped field for tcp to note
22 * a socket is reset and must stay shut up
23 * Alan Cox : New fields for options
24 * Pauline Middelink : identd support
25 * Alan Cox : Eliminate low level recv/recvfrom
26 * David S. Miller : New socket lookup architecture.
27 * Steve Whitehouse: Default routines for sock_ops
28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
29 * protinfo be just a void pointer, as the
30 * protocol specific parts were moved to
31 * respective headers and ipv4/v6, etc now
32 * use private slabcaches for its socks
33 * Pedro Hortas : New flags field for socket options
34 */
35#ifndef _SOCK_H
36#define _SOCK_H
37
38#include <linux/hardirq.h>
39#include <linux/kernel.h>
40#include <linux/list.h>
41#include <linux/list_nulls.h>
42#include <linux/timer.h>
43#include <linux/cache.h>
44#include <linux/bitops.h>
45#include <linux/lockdep.h>
46#include <linux/netdevice.h>
47#include <linux/skbuff.h> /* struct sk_buff */
48#include <linux/mm.h>
49#include <linux/security.h>
50#include <linux/slab.h>
51#include <linux/uaccess.h>
52#include <linux/page_counter.h>
53#include <linux/memcontrol.h>
54#include <linux/static_key.h>
55#include <linux/sched.h>
56#include <linux/wait.h>
57#include <linux/cgroup-defs.h>
58#include <linux/rbtree.h>
59#include <linux/filter.h>
60#include <linux/rculist_nulls.h>
61#include <linux/poll.h>
62#include <linux/sockptr.h>
63
64#include <linux/atomic.h>
65#include <linux/refcount.h>
66#include <net/dst.h>
67#include <net/checksum.h>
68#include <net/tcp_states.h>
69#include <linux/net_tstamp.h>
70#include <net/l3mdev.h>
71
72/*
73 * This structure really needs to be cleaned up.
74 * Most of it is for TCP, and not used by any of
75 * the other protocols.
76 */
77
78/* Define this to get the SOCK_DBG debugging facility. */
79#define SOCK_DEBUGGING
80#ifdef SOCK_DEBUGGING
81#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
82 printk(KERN_DEBUG msg); } while (0)
83#else
84/* Validate arguments and do nothing */
85static inline __printf(2, 3)
86void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
87{
88}
89#endif
90
91/* This is the per-socket lock. The spinlock provides a synchronization
92 * between user contexts and software interrupt processing, whereas the
93 * mini-semaphore synchronizes multiple users amongst themselves.
94 */
95typedef struct {
96 spinlock_t slock;
97 int owned;
98 wait_queue_head_t wq;
99 /*
100 * We express the mutex-alike socket_lock semantics
101 * to the lock validator by explicitly managing
102 * the slock as a lock variant (in addition to
103 * the slock itself):
104 */
105#ifdef CONFIG_DEBUG_LOCK_ALLOC
106 struct lockdep_map dep_map;
107#endif
108} socket_lock_t;
109
110struct sock;
111struct proto;
112struct net;
113
114typedef __u32 __bitwise __portpair;
115typedef __u64 __bitwise __addrpair;
116
117/**
118 * struct sock_common - minimal network layer representation of sockets
119 * @skc_daddr: Foreign IPv4 addr
120 * @skc_rcv_saddr: Bound local IPv4 addr
121 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
122 * @skc_hash: hash value used with various protocol lookup tables
123 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
124 * @skc_dport: placeholder for inet_dport/tw_dport
125 * @skc_num: placeholder for inet_num/tw_num
126 * @skc_portpair: __u32 union of @skc_dport & @skc_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_ipv6only: socket is IPV6 only
132 * @skc_net_refcnt: socket is using net ref counting
133 * @skc_bound_dev_if: bound device index if != 0
134 * @skc_bind_node: bind hash linkage for various protocol lookup tables
135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
136 * @skc_prot: protocol handlers inside a network family
137 * @skc_net: reference to the network namespace of this socket
138 * @skc_v6_daddr: IPV6 destination address
139 * @skc_v6_rcv_saddr: IPV6 source address
140 * @skc_cookie: socket's cookie value
141 * @skc_node: main hash linkage for various protocol lookup tables
142 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
143 * @skc_tx_queue_mapping: tx queue number for this connection
144 * @skc_rx_queue_mapping: rx queue number for this connection
145 * @skc_flags: place holder for sk_flags
146 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
147 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
148 * @skc_listener: connection request listener socket (aka rsk_listener)
149 * [union with @skc_flags]
150 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
151 * [union with @skc_flags]
152 * @skc_incoming_cpu: record/match cpu processing incoming packets
153 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
154 * [union with @skc_incoming_cpu]
155 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
156 * [union with @skc_incoming_cpu]
157 * @skc_refcnt: reference count
158 *
159 * This is the minimal network layer representation of sockets, the header
160 * for struct sock and struct inet_timewait_sock.
161 */
162struct sock_common {
163 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
164 * address on 64bit arches : cf INET_MATCH()
165 */
166 union {
167 __addrpair skc_addrpair;
168 struct {
169 __be32 skc_daddr;
170 __be32 skc_rcv_saddr;
171 };
172 };
173 union {
174 unsigned int skc_hash;
175 __u16 skc_u16hashes[2];
176 };
177 /* skc_dport && skc_num must be grouped as well */
178 union {
179 __portpair skc_portpair;
180 struct {
181 __be16 skc_dport;
182 __u16 skc_num;
183 };
184 };
185
186 unsigned short skc_family;
187 volatile unsigned char skc_state;
188 unsigned char skc_reuse:4;
189 unsigned char skc_reuseport:1;
190 unsigned char skc_ipv6only:1;
191 unsigned char skc_net_refcnt:1;
192 int skc_bound_dev_if;
193 union {
194 struct hlist_node skc_bind_node;
195 struct hlist_node skc_portaddr_node;
196 };
197 struct proto *skc_prot;
198 possible_net_t skc_net;
199
200#if IS_ENABLED(CONFIG_IPV6)
201 struct in6_addr skc_v6_daddr;
202 struct in6_addr skc_v6_rcv_saddr;
203#endif
204
205 atomic64_t skc_cookie;
206
207 /* following fields are padding to force
208 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
209 * assuming IPV6 is enabled. We use this padding differently
210 * for different kind of 'sockets'
211 */
212 union {
213 unsigned long skc_flags;
214 struct sock *skc_listener; /* request_sock */
215 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
216 };
217 /*
218 * fields between dontcopy_begin/dontcopy_end
219 * are not copied in sock_copy()
220 */
221 /* private: */
222 int skc_dontcopy_begin[0];
223 /* public: */
224 union {
225 struct hlist_node skc_node;
226 struct hlist_nulls_node skc_nulls_node;
227 };
228 unsigned short skc_tx_queue_mapping;
229#ifdef CONFIG_XPS
230 unsigned short skc_rx_queue_mapping;
231#endif
232 union {
233 int skc_incoming_cpu;
234 u32 skc_rcv_wnd;
235 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
236 };
237
238 refcount_t skc_refcnt;
239 /* private: */
240 int skc_dontcopy_end[0];
241 union {
242 u32 skc_rxhash;
243 u32 skc_window_clamp;
244 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
245 };
246 /* public: */
247};
248
249struct bpf_sk_storage;
250
251/**
252 * struct sock - network layer representation of sockets
253 * @__sk_common: shared layout with inet_timewait_sock
254 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
255 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
256 * @sk_lock: synchronizer
257 * @sk_kern_sock: True if sock is using kernel lock classes
258 * @sk_rcvbuf: size of receive buffer in bytes
259 * @sk_wq: sock wait queue and async head
260 * @sk_rx_dst: receive input route used by early demux
261 * @sk_dst_cache: destination cache
262 * @sk_dst_pending_confirm: need to confirm neighbour
263 * @sk_policy: flow policy
264 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
265 * @sk_receive_queue: incoming packets
266 * @sk_wmem_alloc: transmit queue bytes committed
267 * @sk_tsq_flags: TCP Small Queues flags
268 * @sk_write_queue: Packet sending queue
269 * @sk_omem_alloc: "o" is "option" or "other"
270 * @sk_wmem_queued: persistent queue size
271 * @sk_forward_alloc: space allocated forward
272 * @sk_napi_id: id of the last napi context to receive data for sk
273 * @sk_ll_usec: usecs to busypoll when there is no data
274 * @sk_allocation: allocation mode
275 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
276 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
277 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
278 * @sk_sndbuf: size of send buffer in bytes
279 * @__sk_flags_offset: empty field used to determine location of bitfield
280 * @sk_padding: unused element for alignment
281 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
282 * @sk_no_check_rx: allow zero checksum in RX packets
283 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
284 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
285 * @sk_route_forced_caps: static, forced route capabilities
286 * (set in tcp_init_sock())
287 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
288 * @sk_gso_max_size: Maximum GSO segment size to build
289 * @sk_gso_max_segs: Maximum number of GSO segments
290 * @sk_pacing_shift: scaling factor for TCP Small Queues
291 * @sk_lingertime: %SO_LINGER l_linger setting
292 * @sk_backlog: always used with the per-socket spinlock held
293 * @sk_callback_lock: used with the callbacks in the end of this struct
294 * @sk_error_queue: rarely used
295 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
296 * IPV6_ADDRFORM for instance)
297 * @sk_err: last error
298 * @sk_err_soft: errors that don't cause failure but are the cause of a
299 * persistent failure not just 'timed out'
300 * @sk_drops: raw/udp drops counter
301 * @sk_ack_backlog: current listen backlog
302 * @sk_max_ack_backlog: listen backlog set in listen()
303 * @sk_uid: user id of owner
304 * @sk_priority: %SO_PRIORITY setting
305 * @sk_type: socket type (%SOCK_STREAM, etc)
306 * @sk_protocol: which protocol this socket belongs in this network family
307 * @sk_peer_pid: &struct pid for this socket's peer
308 * @sk_peer_cred: %SO_PEERCRED setting
309 * @sk_rcvlowat: %SO_RCVLOWAT setting
310 * @sk_rcvtimeo: %SO_RCVTIMEO setting
311 * @sk_sndtimeo: %SO_SNDTIMEO setting
312 * @sk_txhash: computed flow hash for use on transmit
313 * @sk_filter: socket filtering instructions
314 * @sk_timer: sock cleanup timer
315 * @sk_stamp: time stamp of last packet received
316 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
317 * @sk_tsflags: SO_TIMESTAMPING socket options
318 * @sk_tskey: counter to disambiguate concurrent tstamp requests
319 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
320 * @sk_socket: Identd and reporting IO signals
321 * @sk_user_data: RPC layer private data
322 * @sk_frag: cached page frag
323 * @sk_peek_off: current peek_offset value
324 * @sk_send_head: front of stuff to transmit
325 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
326 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
327 * @sk_security: used by security modules
328 * @sk_mark: generic packet mark
329 * @sk_cgrp_data: cgroup data for this cgroup
330 * @sk_memcg: this socket's memory cgroup association
331 * @sk_write_pending: a write to stream socket waits to start
332 * @sk_state_change: callback to indicate change in the state of the sock
333 * @sk_data_ready: callback to indicate there is data to be processed
334 * @sk_write_space: callback to indicate there is bf sending space available
335 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
336 * @sk_backlog_rcv: callback to process the backlog
337 * @sk_validate_xmit_skb: ptr to an optional validate function
338 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
339 * @sk_reuseport_cb: reuseport group container
340 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
341 * @sk_rcu: used during RCU grace period
342 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
343 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
344 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
345 * @sk_txtime_unused: unused txtime flags
346 */
347struct sock {
348 /*
349 * Now struct inet_timewait_sock also uses sock_common, so please just
350 * don't add nothing before this first member (__sk_common) --acme
351 */
352 struct sock_common __sk_common;
353#define sk_node __sk_common.skc_node
354#define sk_nulls_node __sk_common.skc_nulls_node
355#define sk_refcnt __sk_common.skc_refcnt
356#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
357#ifdef CONFIG_XPS
358#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
359#endif
360
361#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
362#define sk_dontcopy_end __sk_common.skc_dontcopy_end
363#define sk_hash __sk_common.skc_hash
364#define sk_portpair __sk_common.skc_portpair
365#define sk_num __sk_common.skc_num
366#define sk_dport __sk_common.skc_dport
367#define sk_addrpair __sk_common.skc_addrpair
368#define sk_daddr __sk_common.skc_daddr
369#define sk_rcv_saddr __sk_common.skc_rcv_saddr
370#define sk_family __sk_common.skc_family
371#define sk_state __sk_common.skc_state
372#define sk_reuse __sk_common.skc_reuse
373#define sk_reuseport __sk_common.skc_reuseport
374#define sk_ipv6only __sk_common.skc_ipv6only
375#define sk_net_refcnt __sk_common.skc_net_refcnt
376#define sk_bound_dev_if __sk_common.skc_bound_dev_if
377#define sk_bind_node __sk_common.skc_bind_node
378#define sk_prot __sk_common.skc_prot
379#define sk_net __sk_common.skc_net
380#define sk_v6_daddr __sk_common.skc_v6_daddr
381#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
382#define sk_cookie __sk_common.skc_cookie
383#define sk_incoming_cpu __sk_common.skc_incoming_cpu
384#define sk_flags __sk_common.skc_flags
385#define sk_rxhash __sk_common.skc_rxhash
386
387 socket_lock_t sk_lock;
388 atomic_t sk_drops;
389 int sk_rcvlowat;
390 struct sk_buff_head sk_error_queue;
391 struct sk_buff *sk_rx_skb_cache;
392 struct sk_buff_head sk_receive_queue;
393 /*
394 * The backlog queue is special, it is always used with
395 * the per-socket spinlock held and requires low latency
396 * access. Therefore we special case it's implementation.
397 * Note : rmem_alloc is in this structure to fill a hole
398 * on 64bit arches, not because its logically part of
399 * backlog.
400 */
401 struct {
402 atomic_t rmem_alloc;
403 int len;
404 struct sk_buff *head;
405 struct sk_buff *tail;
406 } sk_backlog;
407#define sk_rmem_alloc sk_backlog.rmem_alloc
408
409 int sk_forward_alloc;
410#ifdef CONFIG_NET_RX_BUSY_POLL
411 unsigned int sk_ll_usec;
412 /* ===== mostly read cache line ===== */
413 unsigned int sk_napi_id;
414#endif
415 int sk_rcvbuf;
416
417 struct sk_filter __rcu *sk_filter;
418 union {
419 struct socket_wq __rcu *sk_wq;
420 /* private: */
421 struct socket_wq *sk_wq_raw;
422 /* public: */
423 };
424#ifdef CONFIG_XFRM
425 struct xfrm_policy __rcu *sk_policy[2];
426#endif
427 struct dst_entry *sk_rx_dst;
428 struct dst_entry __rcu *sk_dst_cache;
429 atomic_t sk_omem_alloc;
430 int sk_sndbuf;
431
432 /* ===== cache line for TX ===== */
433 int sk_wmem_queued;
434 refcount_t sk_wmem_alloc;
435 unsigned long sk_tsq_flags;
436 union {
437 struct sk_buff *sk_send_head;
438 struct rb_root tcp_rtx_queue;
439 };
440 struct sk_buff *sk_tx_skb_cache;
441 struct sk_buff_head sk_write_queue;
442 __s32 sk_peek_off;
443 int sk_write_pending;
444 __u32 sk_dst_pending_confirm;
445 u32 sk_pacing_status; /* see enum sk_pacing */
446 long sk_sndtimeo;
447 struct timer_list sk_timer;
448 __u32 sk_priority;
449 __u32 sk_mark;
450 unsigned long sk_pacing_rate; /* bytes per second */
451 unsigned long sk_max_pacing_rate;
452 struct page_frag sk_frag;
453 netdev_features_t sk_route_caps;
454 netdev_features_t sk_route_nocaps;
455 netdev_features_t sk_route_forced_caps;
456 int sk_gso_type;
457 unsigned int sk_gso_max_size;
458 gfp_t sk_allocation;
459 __u32 sk_txhash;
460
461 /*
462 * Because of non atomicity rules, all
463 * changes are protected by socket lock.
464 */
465 u8 sk_padding : 1,
466 sk_kern_sock : 1,
467 sk_no_check_tx : 1,
468 sk_no_check_rx : 1,
469 sk_userlocks : 4;
470 u8 sk_pacing_shift;
471 u16 sk_type;
472 u16 sk_protocol;
473 u16 sk_gso_max_segs;
474 unsigned long sk_lingertime;
475 struct proto *sk_prot_creator;
476 rwlock_t sk_callback_lock;
477 int sk_err,
478 sk_err_soft;
479 u32 sk_ack_backlog;
480 u32 sk_max_ack_backlog;
481 kuid_t sk_uid;
482 struct pid *sk_peer_pid;
483 const struct cred *sk_peer_cred;
484 long sk_rcvtimeo;
485 ktime_t sk_stamp;
486#if BITS_PER_LONG==32
487 seqlock_t sk_stamp_seq;
488#endif
489 u16 sk_tsflags;
490 u8 sk_shutdown;
491 u32 sk_tskey;
492 atomic_t sk_zckey;
493
494 u8 sk_clockid;
495 u8 sk_txtime_deadline_mode : 1,
496 sk_txtime_report_errors : 1,
497 sk_txtime_unused : 6;
498
499 struct socket *sk_socket;
500 void *sk_user_data;
501#ifdef CONFIG_SECURITY
502 void *sk_security;
503#endif
504 struct sock_cgroup_data sk_cgrp_data;
505 struct mem_cgroup *sk_memcg;
506 void (*sk_state_change)(struct sock *sk);
507 void (*sk_data_ready)(struct sock *sk);
508 void (*sk_write_space)(struct sock *sk);
509 void (*sk_error_report)(struct sock *sk);
510 int (*sk_backlog_rcv)(struct sock *sk,
511 struct sk_buff *skb);
512#ifdef CONFIG_SOCK_VALIDATE_XMIT
513 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
514 struct net_device *dev,
515 struct sk_buff *skb);
516#endif
517 void (*sk_destruct)(struct sock *sk);
518 struct sock_reuseport __rcu *sk_reuseport_cb;
519#ifdef CONFIG_BPF_SYSCALL
520 struct bpf_sk_storage __rcu *sk_bpf_storage;
521#endif
522 struct rcu_head sk_rcu;
523};
524
525enum sk_pacing {
526 SK_PACING_NONE = 0,
527 SK_PACING_NEEDED = 1,
528 SK_PACING_FQ = 2,
529};
530
531/* Pointer stored in sk_user_data might not be suitable for copying
532 * when cloning the socket. For instance, it can point to a reference
533 * counted object. sk_user_data bottom bit is set if pointer must not
534 * be copied.
535 */
536#define SK_USER_DATA_NOCOPY 1UL
537#define SK_USER_DATA_BPF 2UL /* Managed by BPF */
538#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF)
539
540/**
541 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
542 * @sk: socket
543 */
544static inline bool sk_user_data_is_nocopy(const struct sock *sk)
545{
546 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
547}
548
549#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
550
551#define rcu_dereference_sk_user_data(sk) \
552({ \
553 void *__tmp = rcu_dereference(__sk_user_data((sk))); \
554 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \
555})
556#define rcu_assign_sk_user_data(sk, ptr) \
557({ \
558 uintptr_t __tmp = (uintptr_t)(ptr); \
559 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
560 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \
561})
562#define rcu_assign_sk_user_data_nocopy(sk, ptr) \
563({ \
564 uintptr_t __tmp = (uintptr_t)(ptr); \
565 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \
566 rcu_assign_pointer(__sk_user_data((sk)), \
567 __tmp | SK_USER_DATA_NOCOPY); \
568})
569
570/*
571 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
572 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
573 * on a socket means that the socket will reuse everybody else's port
574 * without looking at the other's sk_reuse value.
575 */
576
577#define SK_NO_REUSE 0
578#define SK_CAN_REUSE 1
579#define SK_FORCE_REUSE 2
580
581int sk_set_peek_off(struct sock *sk, int val);
582
583static inline int sk_peek_offset(struct sock *sk, int flags)
584{
585 if (unlikely(flags & MSG_PEEK)) {
586 return READ_ONCE(sk->sk_peek_off);
587 }
588
589 return 0;
590}
591
592static inline void sk_peek_offset_bwd(struct sock *sk, int val)
593{
594 s32 off = READ_ONCE(sk->sk_peek_off);
595
596 if (unlikely(off >= 0)) {
597 off = max_t(s32, off - val, 0);
598 WRITE_ONCE(sk->sk_peek_off, off);
599 }
600}
601
602static inline void sk_peek_offset_fwd(struct sock *sk, int val)
603{
604 sk_peek_offset_bwd(sk, -val);
605}
606
607/*
608 * Hashed lists helper routines
609 */
610static inline struct sock *sk_entry(const struct hlist_node *node)
611{
612 return hlist_entry(node, struct sock, sk_node);
613}
614
615static inline struct sock *__sk_head(const struct hlist_head *head)
616{
617 return hlist_entry(head->first, struct sock, sk_node);
618}
619
620static inline struct sock *sk_head(const struct hlist_head *head)
621{
622 return hlist_empty(head) ? NULL : __sk_head(head);
623}
624
625static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
626{
627 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
628}
629
630static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
631{
632 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
633}
634
635static inline struct sock *sk_next(const struct sock *sk)
636{
637 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
638}
639
640static inline struct sock *sk_nulls_next(const struct sock *sk)
641{
642 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
643 hlist_nulls_entry(sk->sk_nulls_node.next,
644 struct sock, sk_nulls_node) :
645 NULL;
646}
647
648static inline bool sk_unhashed(const struct sock *sk)
649{
650 return hlist_unhashed(&sk->sk_node);
651}
652
653static inline bool sk_hashed(const struct sock *sk)
654{
655 return !sk_unhashed(sk);
656}
657
658static inline void sk_node_init(struct hlist_node *node)
659{
660 node->pprev = NULL;
661}
662
663static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
664{
665 node->pprev = NULL;
666}
667
668static inline void __sk_del_node(struct sock *sk)
669{
670 __hlist_del(&sk->sk_node);
671}
672
673/* NB: equivalent to hlist_del_init_rcu */
674static inline bool __sk_del_node_init(struct sock *sk)
675{
676 if (sk_hashed(sk)) {
677 __sk_del_node(sk);
678 sk_node_init(&sk->sk_node);
679 return true;
680 }
681 return false;
682}
683
684/* Grab socket reference count. This operation is valid only
685 when sk is ALREADY grabbed f.e. it is found in hash table
686 or a list and the lookup is made under lock preventing hash table
687 modifications.
688 */
689
690static __always_inline void sock_hold(struct sock *sk)
691{
692 refcount_inc(&sk->sk_refcnt);
693}
694
695/* Ungrab socket in the context, which assumes that socket refcnt
696 cannot hit zero, f.e. it is true in context of any socketcall.
697 */
698static __always_inline void __sock_put(struct sock *sk)
699{
700 refcount_dec(&sk->sk_refcnt);
701}
702
703static inline bool sk_del_node_init(struct sock *sk)
704{
705 bool rc = __sk_del_node_init(sk);
706
707 if (rc) {
708 /* paranoid for a while -acme */
709 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
710 __sock_put(sk);
711 }
712 return rc;
713}
714#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
715
716static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
717{
718 if (sk_hashed(sk)) {
719 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
720 return true;
721 }
722 return false;
723}
724
725static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
726{
727 bool rc = __sk_nulls_del_node_init_rcu(sk);
728
729 if (rc) {
730 /* paranoid for a while -acme */
731 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
732 __sock_put(sk);
733 }
734 return rc;
735}
736
737static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
738{
739 hlist_add_head(&sk->sk_node, list);
740}
741
742static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
743{
744 sock_hold(sk);
745 __sk_add_node(sk, list);
746}
747
748static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
749{
750 sock_hold(sk);
751 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
752 sk->sk_family == AF_INET6)
753 hlist_add_tail_rcu(&sk->sk_node, list);
754 else
755 hlist_add_head_rcu(&sk->sk_node, list);
756}
757
758static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
759{
760 sock_hold(sk);
761 hlist_add_tail_rcu(&sk->sk_node, list);
762}
763
764static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
765{
766 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
767}
768
769static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
770{
771 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
772}
773
774static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
775{
776 sock_hold(sk);
777 __sk_nulls_add_node_rcu(sk, list);
778}
779
780static inline void __sk_del_bind_node(struct sock *sk)
781{
782 __hlist_del(&sk->sk_bind_node);
783}
784
785static inline void sk_add_bind_node(struct sock *sk,
786 struct hlist_head *list)
787{
788 hlist_add_head(&sk->sk_bind_node, list);
789}
790
791#define sk_for_each(__sk, list) \
792 hlist_for_each_entry(__sk, list, sk_node)
793#define sk_for_each_rcu(__sk, list) \
794 hlist_for_each_entry_rcu(__sk, list, sk_node)
795#define sk_nulls_for_each(__sk, node, list) \
796 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
797#define sk_nulls_for_each_rcu(__sk, node, list) \
798 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
799#define sk_for_each_from(__sk) \
800 hlist_for_each_entry_from(__sk, sk_node)
801#define sk_nulls_for_each_from(__sk, node) \
802 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
803 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
804#define sk_for_each_safe(__sk, tmp, list) \
805 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
806#define sk_for_each_bound(__sk, list) \
807 hlist_for_each_entry(__sk, list, sk_bind_node)
808
809/**
810 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
811 * @tpos: the type * to use as a loop cursor.
812 * @pos: the &struct hlist_node to use as a loop cursor.
813 * @head: the head for your list.
814 * @offset: offset of hlist_node within the struct.
815 *
816 */
817#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
818 for (pos = rcu_dereference(hlist_first_rcu(head)); \
819 pos != NULL && \
820 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
821 pos = rcu_dereference(hlist_next_rcu(pos)))
822
823static inline struct user_namespace *sk_user_ns(struct sock *sk)
824{
825 /* Careful only use this in a context where these parameters
826 * can not change and must all be valid, such as recvmsg from
827 * userspace.
828 */
829 return sk->sk_socket->file->f_cred->user_ns;
830}
831
832/* Sock flags */
833enum sock_flags {
834 SOCK_DEAD,
835 SOCK_DONE,
836 SOCK_URGINLINE,
837 SOCK_KEEPOPEN,
838 SOCK_LINGER,
839 SOCK_DESTROY,
840 SOCK_BROADCAST,
841 SOCK_TIMESTAMP,
842 SOCK_ZAPPED,
843 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
844 SOCK_DBG, /* %SO_DEBUG setting */
845 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
846 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
847 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
848 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
849 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
850 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
851 SOCK_FASYNC, /* fasync() active */
852 SOCK_RXQ_OVFL,
853 SOCK_ZEROCOPY, /* buffers from userspace */
854 SOCK_WIFI_STATUS, /* push wifi status to userspace */
855 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
856 * Will use last 4 bytes of packet sent from
857 * user-space instead.
858 */
859 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
860 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
861 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
862 SOCK_TXTIME,
863 SOCK_XDP, /* XDP is attached */
864 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
865};
866
867#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
868
869static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
870{
871 nsk->sk_flags = osk->sk_flags;
872}
873
874static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
875{
876 __set_bit(flag, &sk->sk_flags);
877}
878
879static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
880{
881 __clear_bit(flag, &sk->sk_flags);
882}
883
884static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
885 int valbool)
886{
887 if (valbool)
888 sock_set_flag(sk, bit);
889 else
890 sock_reset_flag(sk, bit);
891}
892
893static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
894{
895 return test_bit(flag, &sk->sk_flags);
896}
897
898#ifdef CONFIG_NET
899DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
900static inline int sk_memalloc_socks(void)
901{
902 return static_branch_unlikely(&memalloc_socks_key);
903}
904
905void __receive_sock(struct file *file);
906#else
907
908static inline int sk_memalloc_socks(void)
909{
910 return 0;
911}
912
913static inline void __receive_sock(struct file *file)
914{ }
915#endif
916
917static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
918{
919 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
920}
921
922static inline void sk_acceptq_removed(struct sock *sk)
923{
924 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
925}
926
927static inline void sk_acceptq_added(struct sock *sk)
928{
929 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
930}
931
932static inline bool sk_acceptq_is_full(const struct sock *sk)
933{
934 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
935}
936
937/*
938 * Compute minimal free write space needed to queue new packets.
939 */
940static inline int sk_stream_min_wspace(const struct sock *sk)
941{
942 return READ_ONCE(sk->sk_wmem_queued) >> 1;
943}
944
945static inline int sk_stream_wspace(const struct sock *sk)
946{
947 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
948}
949
950static inline void sk_wmem_queued_add(struct sock *sk, int val)
951{
952 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
953}
954
955void sk_stream_write_space(struct sock *sk);
956
957/* OOB backlog add */
958static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
959{
960 /* dont let skb dst not refcounted, we are going to leave rcu lock */
961 skb_dst_force(skb);
962
963 if (!sk->sk_backlog.tail)
964 WRITE_ONCE(sk->sk_backlog.head, skb);
965 else
966 sk->sk_backlog.tail->next = skb;
967
968 WRITE_ONCE(sk->sk_backlog.tail, skb);
969 skb->next = NULL;
970}
971
972/*
973 * Take into account size of receive queue and backlog queue
974 * Do not take into account this skb truesize,
975 * to allow even a single big packet to come.
976 */
977static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
978{
979 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
980
981 return qsize > limit;
982}
983
984/* The per-socket spinlock must be held here. */
985static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
986 unsigned int limit)
987{
988 if (sk_rcvqueues_full(sk, limit))
989 return -ENOBUFS;
990
991 /*
992 * If the skb was allocated from pfmemalloc reserves, only
993 * allow SOCK_MEMALLOC sockets to use it as this socket is
994 * helping free memory
995 */
996 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
997 return -ENOMEM;
998
999 __sk_add_backlog(sk, skb);
1000 sk->sk_backlog.len += skb->truesize;
1001 return 0;
1002}
1003
1004int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1005
1006static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1007{
1008 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1009 return __sk_backlog_rcv(sk, skb);
1010
1011 return sk->sk_backlog_rcv(sk, skb);
1012}
1013
1014static inline void sk_incoming_cpu_update(struct sock *sk)
1015{
1016 int cpu = raw_smp_processor_id();
1017
1018 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1019 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1020}
1021
1022static inline void sock_rps_record_flow_hash(__u32 hash)
1023{
1024#ifdef CONFIG_RPS
1025 struct rps_sock_flow_table *sock_flow_table;
1026
1027 rcu_read_lock();
1028 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1029 rps_record_sock_flow(sock_flow_table, hash);
1030 rcu_read_unlock();
1031#endif
1032}
1033
1034static inline void sock_rps_record_flow(const struct sock *sk)
1035{
1036#ifdef CONFIG_RPS
1037 if (static_branch_unlikely(&rfs_needed)) {
1038 /* Reading sk->sk_rxhash might incur an expensive cache line
1039 * miss.
1040 *
1041 * TCP_ESTABLISHED does cover almost all states where RFS
1042 * might be useful, and is cheaper [1] than testing :
1043 * IPv4: inet_sk(sk)->inet_daddr
1044 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1045 * OR an additional socket flag
1046 * [1] : sk_state and sk_prot are in the same cache line.
1047 */
1048 if (sk->sk_state == TCP_ESTABLISHED)
1049 sock_rps_record_flow_hash(sk->sk_rxhash);
1050 }
1051#endif
1052}
1053
1054static inline void sock_rps_save_rxhash(struct sock *sk,
1055 const struct sk_buff *skb)
1056{
1057#ifdef CONFIG_RPS
1058 if (unlikely(sk->sk_rxhash != skb->hash))
1059 sk->sk_rxhash = skb->hash;
1060#endif
1061}
1062
1063static inline void sock_rps_reset_rxhash(struct sock *sk)
1064{
1065#ifdef CONFIG_RPS
1066 sk->sk_rxhash = 0;
1067#endif
1068}
1069
1070#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1071 ({ int __rc; \
1072 release_sock(__sk); \
1073 __rc = __condition; \
1074 if (!__rc) { \
1075 *(__timeo) = wait_woken(__wait, \
1076 TASK_INTERRUPTIBLE, \
1077 *(__timeo)); \
1078 } \
1079 sched_annotate_sleep(); \
1080 lock_sock(__sk); \
1081 __rc = __condition; \
1082 __rc; \
1083 })
1084
1085int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1086int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1087void sk_stream_wait_close(struct sock *sk, long timeo_p);
1088int sk_stream_error(struct sock *sk, int flags, int err);
1089void sk_stream_kill_queues(struct sock *sk);
1090void sk_set_memalloc(struct sock *sk);
1091void sk_clear_memalloc(struct sock *sk);
1092
1093void __sk_flush_backlog(struct sock *sk);
1094
1095static inline bool sk_flush_backlog(struct sock *sk)
1096{
1097 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1098 __sk_flush_backlog(sk);
1099 return true;
1100 }
1101 return false;
1102}
1103
1104int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1105
1106struct request_sock_ops;
1107struct timewait_sock_ops;
1108struct inet_hashinfo;
1109struct raw_hashinfo;
1110struct smc_hashinfo;
1111struct module;
1112
1113/*
1114 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1115 * un-modified. Special care is taken when initializing object to zero.
1116 */
1117static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1118{
1119 if (offsetof(struct sock, sk_node.next) != 0)
1120 memset(sk, 0, offsetof(struct sock, sk_node.next));
1121 memset(&sk->sk_node.pprev, 0,
1122 size - offsetof(struct sock, sk_node.pprev));
1123}
1124
1125/* Networking protocol blocks we attach to sockets.
1126 * socket layer -> transport layer interface
1127 */
1128struct proto {
1129 void (*close)(struct sock *sk,
1130 long timeout);
1131 int (*pre_connect)(struct sock *sk,
1132 struct sockaddr *uaddr,
1133 int addr_len);
1134 int (*connect)(struct sock *sk,
1135 struct sockaddr *uaddr,
1136 int addr_len);
1137 int (*disconnect)(struct sock *sk, int flags);
1138
1139 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1140 bool kern);
1141
1142 int (*ioctl)(struct sock *sk, int cmd,
1143 unsigned long arg);
1144 int (*init)(struct sock *sk);
1145 void (*destroy)(struct sock *sk);
1146 void (*shutdown)(struct sock *sk, int how);
1147 int (*setsockopt)(struct sock *sk, int level,
1148 int optname, sockptr_t optval,
1149 unsigned int optlen);
1150 int (*getsockopt)(struct sock *sk, int level,
1151 int optname, char __user *optval,
1152 int __user *option);
1153 void (*keepalive)(struct sock *sk, int valbool);
1154#ifdef CONFIG_COMPAT
1155 int (*compat_ioctl)(struct sock *sk,
1156 unsigned int cmd, unsigned long arg);
1157#endif
1158 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1159 size_t len);
1160 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1161 size_t len, int noblock, int flags,
1162 int *addr_len);
1163 int (*sendpage)(struct sock *sk, struct page *page,
1164 int offset, size_t size, int flags);
1165 int (*bind)(struct sock *sk,
1166 struct sockaddr *addr, int addr_len);
1167 int (*bind_add)(struct sock *sk,
1168 struct sockaddr *addr, int addr_len);
1169
1170 int (*backlog_rcv) (struct sock *sk,
1171 struct sk_buff *skb);
1172
1173 void (*release_cb)(struct sock *sk);
1174
1175 /* Keeping track of sk's, looking them up, and port selection methods. */
1176 int (*hash)(struct sock *sk);
1177 void (*unhash)(struct sock *sk);
1178 void (*rehash)(struct sock *sk);
1179 int (*get_port)(struct sock *sk, unsigned short snum);
1180
1181 /* Keeping track of sockets in use */
1182#ifdef CONFIG_PROC_FS
1183 unsigned int inuse_idx;
1184#endif
1185
1186 bool (*stream_memory_free)(const struct sock *sk, int wake);
1187 bool (*stream_memory_read)(const struct sock *sk);
1188 /* Memory pressure */
1189 void (*enter_memory_pressure)(struct sock *sk);
1190 void (*leave_memory_pressure)(struct sock *sk);
1191 atomic_long_t *memory_allocated; /* Current allocated memory. */
1192 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1193 /*
1194 * Pressure flag: try to collapse.
1195 * Technical note: it is used by multiple contexts non atomically.
1196 * All the __sk_mem_schedule() is of this nature: accounting
1197 * is strict, actions are advisory and have some latency.
1198 */
1199 unsigned long *memory_pressure;
1200 long *sysctl_mem;
1201
1202 int *sysctl_wmem;
1203 int *sysctl_rmem;
1204 u32 sysctl_wmem_offset;
1205 u32 sysctl_rmem_offset;
1206
1207 int max_header;
1208 bool no_autobind;
1209
1210 struct kmem_cache *slab;
1211 unsigned int obj_size;
1212 slab_flags_t slab_flags;
1213 unsigned int useroffset; /* Usercopy region offset */
1214 unsigned int usersize; /* Usercopy region size */
1215
1216 struct percpu_counter *orphan_count;
1217
1218 struct request_sock_ops *rsk_prot;
1219 struct timewait_sock_ops *twsk_prot;
1220
1221 union {
1222 struct inet_hashinfo *hashinfo;
1223 struct udp_table *udp_table;
1224 struct raw_hashinfo *raw_hash;
1225 struct smc_hashinfo *smc_hash;
1226 } h;
1227
1228 struct module *owner;
1229
1230 char name[32];
1231
1232 struct list_head node;
1233#ifdef SOCK_REFCNT_DEBUG
1234 atomic_t socks;
1235#endif
1236 int (*diag_destroy)(struct sock *sk, int err);
1237} __randomize_layout;
1238
1239int proto_register(struct proto *prot, int alloc_slab);
1240void proto_unregister(struct proto *prot);
1241int sock_load_diag_module(int family, int protocol);
1242
1243#ifdef SOCK_REFCNT_DEBUG
1244static inline void sk_refcnt_debug_inc(struct sock *sk)
1245{
1246 atomic_inc(&sk->sk_prot->socks);
1247}
1248
1249static inline void sk_refcnt_debug_dec(struct sock *sk)
1250{
1251 atomic_dec(&sk->sk_prot->socks);
1252 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1253 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1254}
1255
1256static inline void sk_refcnt_debug_release(const struct sock *sk)
1257{
1258 if (refcount_read(&sk->sk_refcnt) != 1)
1259 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1260 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1261}
1262#else /* SOCK_REFCNT_DEBUG */
1263#define sk_refcnt_debug_inc(sk) do { } while (0)
1264#define sk_refcnt_debug_dec(sk) do { } while (0)
1265#define sk_refcnt_debug_release(sk) do { } while (0)
1266#endif /* SOCK_REFCNT_DEBUG */
1267
1268static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1269{
1270 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1271 return false;
1272
1273 return sk->sk_prot->stream_memory_free ?
1274 sk->sk_prot->stream_memory_free(sk, wake) : true;
1275}
1276
1277static inline bool sk_stream_memory_free(const struct sock *sk)
1278{
1279 return __sk_stream_memory_free(sk, 0);
1280}
1281
1282static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1283{
1284 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1285 __sk_stream_memory_free(sk, wake);
1286}
1287
1288static inline bool sk_stream_is_writeable(const struct sock *sk)
1289{
1290 return __sk_stream_is_writeable(sk, 0);
1291}
1292
1293static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1294 struct cgroup *ancestor)
1295{
1296#ifdef CONFIG_SOCK_CGROUP_DATA
1297 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1298 ancestor);
1299#else
1300 return -ENOTSUPP;
1301#endif
1302}
1303
1304static inline bool sk_has_memory_pressure(const struct sock *sk)
1305{
1306 return sk->sk_prot->memory_pressure != NULL;
1307}
1308
1309static inline bool sk_under_memory_pressure(const struct sock *sk)
1310{
1311 if (!sk->sk_prot->memory_pressure)
1312 return false;
1313
1314 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1315 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1316 return true;
1317
1318 return !!*sk->sk_prot->memory_pressure;
1319}
1320
1321static inline long
1322sk_memory_allocated(const struct sock *sk)
1323{
1324 return atomic_long_read(sk->sk_prot->memory_allocated);
1325}
1326
1327static inline long
1328sk_memory_allocated_add(struct sock *sk, int amt)
1329{
1330 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1331}
1332
1333static inline void
1334sk_memory_allocated_sub(struct sock *sk, int amt)
1335{
1336 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1337}
1338
1339static inline void sk_sockets_allocated_dec(struct sock *sk)
1340{
1341 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1342}
1343
1344static inline void sk_sockets_allocated_inc(struct sock *sk)
1345{
1346 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1347}
1348
1349static inline u64
1350sk_sockets_allocated_read_positive(struct sock *sk)
1351{
1352 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1353}
1354
1355static inline int
1356proto_sockets_allocated_sum_positive(struct proto *prot)
1357{
1358 return percpu_counter_sum_positive(prot->sockets_allocated);
1359}
1360
1361static inline long
1362proto_memory_allocated(struct proto *prot)
1363{
1364 return atomic_long_read(prot->memory_allocated);
1365}
1366
1367static inline bool
1368proto_memory_pressure(struct proto *prot)
1369{
1370 if (!prot->memory_pressure)
1371 return false;
1372 return !!*prot->memory_pressure;
1373}
1374
1375
1376#ifdef CONFIG_PROC_FS
1377/* Called with local bh disabled */
1378void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1379int sock_prot_inuse_get(struct net *net, struct proto *proto);
1380int sock_inuse_get(struct net *net);
1381#else
1382static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1383 int inc)
1384{
1385}
1386#endif
1387
1388
1389/* With per-bucket locks this operation is not-atomic, so that
1390 * this version is not worse.
1391 */
1392static inline int __sk_prot_rehash(struct sock *sk)
1393{
1394 sk->sk_prot->unhash(sk);
1395 return sk->sk_prot->hash(sk);
1396}
1397
1398/* About 10 seconds */
1399#define SOCK_DESTROY_TIME (10*HZ)
1400
1401/* Sockets 0-1023 can't be bound to unless you are superuser */
1402#define PROT_SOCK 1024
1403
1404#define SHUTDOWN_MASK 3
1405#define RCV_SHUTDOWN 1
1406#define SEND_SHUTDOWN 2
1407
1408#define SOCK_SNDBUF_LOCK 1
1409#define SOCK_RCVBUF_LOCK 2
1410#define SOCK_BINDADDR_LOCK 4
1411#define SOCK_BINDPORT_LOCK 8
1412
1413struct socket_alloc {
1414 struct socket socket;
1415 struct inode vfs_inode;
1416};
1417
1418static inline struct socket *SOCKET_I(struct inode *inode)
1419{
1420 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1421}
1422
1423static inline struct inode *SOCK_INODE(struct socket *socket)
1424{
1425 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1426}
1427
1428/*
1429 * Functions for memory accounting
1430 */
1431int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1432int __sk_mem_schedule(struct sock *sk, int size, int kind);
1433void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1434void __sk_mem_reclaim(struct sock *sk, int amount);
1435
1436/* We used to have PAGE_SIZE here, but systems with 64KB pages
1437 * do not necessarily have 16x time more memory than 4KB ones.
1438 */
1439#define SK_MEM_QUANTUM 4096
1440#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1441#define SK_MEM_SEND 0
1442#define SK_MEM_RECV 1
1443
1444/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1445static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1446{
1447 long val = sk->sk_prot->sysctl_mem[index];
1448
1449#if PAGE_SIZE > SK_MEM_QUANTUM
1450 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1451#elif PAGE_SIZE < SK_MEM_QUANTUM
1452 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1453#endif
1454 return val;
1455}
1456
1457static inline int sk_mem_pages(int amt)
1458{
1459 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1460}
1461
1462static inline bool sk_has_account(struct sock *sk)
1463{
1464 /* return true if protocol supports memory accounting */
1465 return !!sk->sk_prot->memory_allocated;
1466}
1467
1468static inline bool sk_wmem_schedule(struct sock *sk, int size)
1469{
1470 if (!sk_has_account(sk))
1471 return true;
1472 return size <= sk->sk_forward_alloc ||
1473 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1474}
1475
1476static inline bool
1477sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1478{
1479 if (!sk_has_account(sk))
1480 return true;
1481 return size<= sk->sk_forward_alloc ||
1482 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1483 skb_pfmemalloc(skb);
1484}
1485
1486static inline void sk_mem_reclaim(struct sock *sk)
1487{
1488 if (!sk_has_account(sk))
1489 return;
1490 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1491 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1492}
1493
1494static inline void sk_mem_reclaim_partial(struct sock *sk)
1495{
1496 if (!sk_has_account(sk))
1497 return;
1498 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1499 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1500}
1501
1502static inline void sk_mem_charge(struct sock *sk, int size)
1503{
1504 if (!sk_has_account(sk))
1505 return;
1506 sk->sk_forward_alloc -= size;
1507}
1508
1509static inline void sk_mem_uncharge(struct sock *sk, int size)
1510{
1511 if (!sk_has_account(sk))
1512 return;
1513 sk->sk_forward_alloc += size;
1514
1515 /* Avoid a possible overflow.
1516 * TCP send queues can make this happen, if sk_mem_reclaim()
1517 * is not called and more than 2 GBytes are released at once.
1518 *
1519 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1520 * no need to hold that much forward allocation anyway.
1521 */
1522 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1523 __sk_mem_reclaim(sk, 1 << 20);
1524}
1525
1526DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1527static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1528{
1529 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1530 sk_wmem_queued_add(sk, -skb->truesize);
1531 sk_mem_uncharge(sk, skb->truesize);
1532 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1533 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1534 skb_ext_reset(skb);
1535 skb_zcopy_clear(skb, true);
1536 sk->sk_tx_skb_cache = skb;
1537 return;
1538 }
1539 __kfree_skb(skb);
1540}
1541
1542static inline void sock_release_ownership(struct sock *sk)
1543{
1544 if (sk->sk_lock.owned) {
1545 sk->sk_lock.owned = 0;
1546
1547 /* The sk_lock has mutex_unlock() semantics: */
1548 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1549 }
1550}
1551
1552/*
1553 * Macro so as to not evaluate some arguments when
1554 * lockdep is not enabled.
1555 *
1556 * Mark both the sk_lock and the sk_lock.slock as a
1557 * per-address-family lock class.
1558 */
1559#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1560do { \
1561 sk->sk_lock.owned = 0; \
1562 init_waitqueue_head(&sk->sk_lock.wq); \
1563 spin_lock_init(&(sk)->sk_lock.slock); \
1564 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1565 sizeof((sk)->sk_lock)); \
1566 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1567 (skey), (sname)); \
1568 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1569} while (0)
1570
1571#ifdef CONFIG_LOCKDEP
1572static inline bool lockdep_sock_is_held(const struct sock *sk)
1573{
1574 return lockdep_is_held(&sk->sk_lock) ||
1575 lockdep_is_held(&sk->sk_lock.slock);
1576}
1577#endif
1578
1579void lock_sock_nested(struct sock *sk, int subclass);
1580
1581static inline void lock_sock(struct sock *sk)
1582{
1583 lock_sock_nested(sk, 0);
1584}
1585
1586void __release_sock(struct sock *sk);
1587void release_sock(struct sock *sk);
1588
1589/* BH context may only use the following locking interface. */
1590#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1591#define bh_lock_sock_nested(__sk) \
1592 spin_lock_nested(&((__sk)->sk_lock.slock), \
1593 SINGLE_DEPTH_NESTING)
1594#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1595
1596bool lock_sock_fast(struct sock *sk);
1597/**
1598 * unlock_sock_fast - complement of lock_sock_fast
1599 * @sk: socket
1600 * @slow: slow mode
1601 *
1602 * fast unlock socket for user context.
1603 * If slow mode is on, we call regular release_sock()
1604 */
1605static inline void unlock_sock_fast(struct sock *sk, bool slow)
1606{
1607 if (slow)
1608 release_sock(sk);
1609 else
1610 spin_unlock_bh(&sk->sk_lock.slock);
1611}
1612
1613/* Used by processes to "lock" a socket state, so that
1614 * interrupts and bottom half handlers won't change it
1615 * from under us. It essentially blocks any incoming
1616 * packets, so that we won't get any new data or any
1617 * packets that change the state of the socket.
1618 *
1619 * While locked, BH processing will add new packets to
1620 * the backlog queue. This queue is processed by the
1621 * owner of the socket lock right before it is released.
1622 *
1623 * Since ~2.3.5 it is also exclusive sleep lock serializing
1624 * accesses from user process context.
1625 */
1626
1627static inline void sock_owned_by_me(const struct sock *sk)
1628{
1629#ifdef CONFIG_LOCKDEP
1630 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1631#endif
1632}
1633
1634static inline bool sock_owned_by_user(const struct sock *sk)
1635{
1636 sock_owned_by_me(sk);
1637 return sk->sk_lock.owned;
1638}
1639
1640static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1641{
1642 return sk->sk_lock.owned;
1643}
1644
1645/* no reclassification while locks are held */
1646static inline bool sock_allow_reclassification(const struct sock *csk)
1647{
1648 struct sock *sk = (struct sock *)csk;
1649
1650 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1651}
1652
1653struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1654 struct proto *prot, int kern);
1655void sk_free(struct sock *sk);
1656void sk_destruct(struct sock *sk);
1657struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1658void sk_free_unlock_clone(struct sock *sk);
1659
1660struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1661 gfp_t priority);
1662void __sock_wfree(struct sk_buff *skb);
1663void sock_wfree(struct sk_buff *skb);
1664struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1665 gfp_t priority);
1666void skb_orphan_partial(struct sk_buff *skb);
1667void sock_rfree(struct sk_buff *skb);
1668void sock_efree(struct sk_buff *skb);
1669#ifdef CONFIG_INET
1670void sock_edemux(struct sk_buff *skb);
1671void sock_pfree(struct sk_buff *skb);
1672#else
1673#define sock_edemux sock_efree
1674#endif
1675
1676int sock_setsockopt(struct socket *sock, int level, int op,
1677 sockptr_t optval, unsigned int optlen);
1678
1679int sock_getsockopt(struct socket *sock, int level, int op,
1680 char __user *optval, int __user *optlen);
1681int sock_gettstamp(struct socket *sock, void __user *userstamp,
1682 bool timeval, bool time32);
1683struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1684 int noblock, int *errcode);
1685struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1686 unsigned long data_len, int noblock,
1687 int *errcode, int max_page_order);
1688void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1689void sock_kfree_s(struct sock *sk, void *mem, int size);
1690void sock_kzfree_s(struct sock *sk, void *mem, int size);
1691void sk_send_sigurg(struct sock *sk);
1692
1693struct sockcm_cookie {
1694 u64 transmit_time;
1695 u32 mark;
1696 u16 tsflags;
1697};
1698
1699static inline void sockcm_init(struct sockcm_cookie *sockc,
1700 const struct sock *sk)
1701{
1702 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1703}
1704
1705int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1706 struct sockcm_cookie *sockc);
1707int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1708 struct sockcm_cookie *sockc);
1709
1710/*
1711 * Functions to fill in entries in struct proto_ops when a protocol
1712 * does not implement a particular function.
1713 */
1714int sock_no_bind(struct socket *, struct sockaddr *, int);
1715int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1716int sock_no_socketpair(struct socket *, struct socket *);
1717int sock_no_accept(struct socket *, struct socket *, int, bool);
1718int sock_no_getname(struct socket *, struct sockaddr *, int);
1719int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1720int sock_no_listen(struct socket *, int);
1721int sock_no_shutdown(struct socket *, int);
1722int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1723int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1724int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1725int sock_no_mmap(struct file *file, struct socket *sock,
1726 struct vm_area_struct *vma);
1727ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1728 size_t size, int flags);
1729ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1730 int offset, size_t size, int flags);
1731
1732/*
1733 * Functions to fill in entries in struct proto_ops when a protocol
1734 * uses the inet style.
1735 */
1736int sock_common_getsockopt(struct socket *sock, int level, int optname,
1737 char __user *optval, int __user *optlen);
1738int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1739 int flags);
1740int sock_common_setsockopt(struct socket *sock, int level, int optname,
1741 sockptr_t optval, unsigned int optlen);
1742
1743void sk_common_release(struct sock *sk);
1744
1745/*
1746 * Default socket callbacks and setup code
1747 */
1748
1749/* Initialise core socket variables */
1750void sock_init_data(struct socket *sock, struct sock *sk);
1751
1752/*
1753 * Socket reference counting postulates.
1754 *
1755 * * Each user of socket SHOULD hold a reference count.
1756 * * Each access point to socket (an hash table bucket, reference from a list,
1757 * running timer, skb in flight MUST hold a reference count.
1758 * * When reference count hits 0, it means it will never increase back.
1759 * * When reference count hits 0, it means that no references from
1760 * outside exist to this socket and current process on current CPU
1761 * is last user and may/should destroy this socket.
1762 * * sk_free is called from any context: process, BH, IRQ. When
1763 * it is called, socket has no references from outside -> sk_free
1764 * may release descendant resources allocated by the socket, but
1765 * to the time when it is called, socket is NOT referenced by any
1766 * hash tables, lists etc.
1767 * * Packets, delivered from outside (from network or from another process)
1768 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1769 * when they sit in queue. Otherwise, packets will leak to hole, when
1770 * socket is looked up by one cpu and unhasing is made by another CPU.
1771 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1772 * (leak to backlog). Packet socket does all the processing inside
1773 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1774 * use separate SMP lock, so that they are prone too.
1775 */
1776
1777/* Ungrab socket and destroy it, if it was the last reference. */
1778static inline void sock_put(struct sock *sk)
1779{
1780 if (refcount_dec_and_test(&sk->sk_refcnt))
1781 sk_free(sk);
1782}
1783/* Generic version of sock_put(), dealing with all sockets
1784 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1785 */
1786void sock_gen_put(struct sock *sk);
1787
1788int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1789 unsigned int trim_cap, bool refcounted);
1790static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1791 const int nested)
1792{
1793 return __sk_receive_skb(sk, skb, nested, 1, true);
1794}
1795
1796static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1797{
1798 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1799 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1800 return;
1801 sk->sk_tx_queue_mapping = tx_queue;
1802}
1803
1804#define NO_QUEUE_MAPPING USHRT_MAX
1805
1806static inline void sk_tx_queue_clear(struct sock *sk)
1807{
1808 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1809}
1810
1811static inline int sk_tx_queue_get(const struct sock *sk)
1812{
1813 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1814 return sk->sk_tx_queue_mapping;
1815
1816 return -1;
1817}
1818
1819static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1820{
1821#ifdef CONFIG_XPS
1822 if (skb_rx_queue_recorded(skb)) {
1823 u16 rx_queue = skb_get_rx_queue(skb);
1824
1825 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1826 return;
1827
1828 sk->sk_rx_queue_mapping = rx_queue;
1829 }
1830#endif
1831}
1832
1833static inline void sk_rx_queue_clear(struct sock *sk)
1834{
1835#ifdef CONFIG_XPS
1836 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1837#endif
1838}
1839
1840#ifdef CONFIG_XPS
1841static inline int sk_rx_queue_get(const struct sock *sk)
1842{
1843 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1844 return sk->sk_rx_queue_mapping;
1845
1846 return -1;
1847}
1848#endif
1849
1850static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1851{
1852 sk->sk_socket = sock;
1853}
1854
1855static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1856{
1857 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1858 return &rcu_dereference_raw(sk->sk_wq)->wait;
1859}
1860/* Detach socket from process context.
1861 * Announce socket dead, detach it from wait queue and inode.
1862 * Note that parent inode held reference count on this struct sock,
1863 * we do not release it in this function, because protocol
1864 * probably wants some additional cleanups or even continuing
1865 * to work with this socket (TCP).
1866 */
1867static inline void sock_orphan(struct sock *sk)
1868{
1869 write_lock_bh(&sk->sk_callback_lock);
1870 sock_set_flag(sk, SOCK_DEAD);
1871 sk_set_socket(sk, NULL);
1872 sk->sk_wq = NULL;
1873 write_unlock_bh(&sk->sk_callback_lock);
1874}
1875
1876static inline void sock_graft(struct sock *sk, struct socket *parent)
1877{
1878 WARN_ON(parent->sk);
1879 write_lock_bh(&sk->sk_callback_lock);
1880 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1881 parent->sk = sk;
1882 sk_set_socket(sk, parent);
1883 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1884 security_sock_graft(sk, parent);
1885 write_unlock_bh(&sk->sk_callback_lock);
1886}
1887
1888kuid_t sock_i_uid(struct sock *sk);
1889unsigned long sock_i_ino(struct sock *sk);
1890
1891static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1892{
1893 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1894}
1895
1896static inline u32 net_tx_rndhash(void)
1897{
1898 u32 v = prandom_u32();
1899
1900 return v ?: 1;
1901}
1902
1903static inline void sk_set_txhash(struct sock *sk)
1904{
1905 sk->sk_txhash = net_tx_rndhash();
1906}
1907
1908static inline void sk_rethink_txhash(struct sock *sk)
1909{
1910 if (sk->sk_txhash)
1911 sk_set_txhash(sk);
1912}
1913
1914static inline struct dst_entry *
1915__sk_dst_get(struct sock *sk)
1916{
1917 return rcu_dereference_check(sk->sk_dst_cache,
1918 lockdep_sock_is_held(sk));
1919}
1920
1921static inline struct dst_entry *
1922sk_dst_get(struct sock *sk)
1923{
1924 struct dst_entry *dst;
1925
1926 rcu_read_lock();
1927 dst = rcu_dereference(sk->sk_dst_cache);
1928 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1929 dst = NULL;
1930 rcu_read_unlock();
1931 return dst;
1932}
1933
1934static inline void dst_negative_advice(struct sock *sk)
1935{
1936 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1937
1938 sk_rethink_txhash(sk);
1939
1940 if (dst && dst->ops->negative_advice) {
1941 ndst = dst->ops->negative_advice(dst);
1942
1943 if (ndst != dst) {
1944 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1945 sk_tx_queue_clear(sk);
1946 sk->sk_dst_pending_confirm = 0;
1947 }
1948 }
1949}
1950
1951static inline void
1952__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1953{
1954 struct dst_entry *old_dst;
1955
1956 sk_tx_queue_clear(sk);
1957 sk->sk_dst_pending_confirm = 0;
1958 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1959 lockdep_sock_is_held(sk));
1960 rcu_assign_pointer(sk->sk_dst_cache, dst);
1961 dst_release(old_dst);
1962}
1963
1964static inline void
1965sk_dst_set(struct sock *sk, struct dst_entry *dst)
1966{
1967 struct dst_entry *old_dst;
1968
1969 sk_tx_queue_clear(sk);
1970 sk->sk_dst_pending_confirm = 0;
1971 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1972 dst_release(old_dst);
1973}
1974
1975static inline void
1976__sk_dst_reset(struct sock *sk)
1977{
1978 __sk_dst_set(sk, NULL);
1979}
1980
1981static inline void
1982sk_dst_reset(struct sock *sk)
1983{
1984 sk_dst_set(sk, NULL);
1985}
1986
1987struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1988
1989struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1990
1991static inline void sk_dst_confirm(struct sock *sk)
1992{
1993 if (!READ_ONCE(sk->sk_dst_pending_confirm))
1994 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
1995}
1996
1997static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1998{
1999 if (skb_get_dst_pending_confirm(skb)) {
2000 struct sock *sk = skb->sk;
2001 unsigned long now = jiffies;
2002
2003 /* avoid dirtying neighbour */
2004 if (READ_ONCE(n->confirmed) != now)
2005 WRITE_ONCE(n->confirmed, now);
2006 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2007 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2008 }
2009}
2010
2011bool sk_mc_loop(struct sock *sk);
2012
2013static inline bool sk_can_gso(const struct sock *sk)
2014{
2015 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2016}
2017
2018void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2019
2020static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2021{
2022 sk->sk_route_nocaps |= flags;
2023 sk->sk_route_caps &= ~flags;
2024}
2025
2026static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2027 struct iov_iter *from, char *to,
2028 int copy, int offset)
2029{
2030 if (skb->ip_summed == CHECKSUM_NONE) {
2031 __wsum csum = 0;
2032 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2033 return -EFAULT;
2034 skb->csum = csum_block_add(skb->csum, csum, offset);
2035 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2036 if (!copy_from_iter_full_nocache(to, copy, from))
2037 return -EFAULT;
2038 } else if (!copy_from_iter_full(to, copy, from))
2039 return -EFAULT;
2040
2041 return 0;
2042}
2043
2044static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2045 struct iov_iter *from, int copy)
2046{
2047 int err, offset = skb->len;
2048
2049 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2050 copy, offset);
2051 if (err)
2052 __skb_trim(skb, offset);
2053
2054 return err;
2055}
2056
2057static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2058 struct sk_buff *skb,
2059 struct page *page,
2060 int off, int copy)
2061{
2062 int err;
2063
2064 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2065 copy, skb->len);
2066 if (err)
2067 return err;
2068
2069 skb->len += copy;
2070 skb->data_len += copy;
2071 skb->truesize += copy;
2072 sk_wmem_queued_add(sk, copy);
2073 sk_mem_charge(sk, copy);
2074 return 0;
2075}
2076
2077/**
2078 * sk_wmem_alloc_get - returns write allocations
2079 * @sk: socket
2080 *
2081 * Return: sk_wmem_alloc minus initial offset of one
2082 */
2083static inline int sk_wmem_alloc_get(const struct sock *sk)
2084{
2085 return refcount_read(&sk->sk_wmem_alloc) - 1;
2086}
2087
2088/**
2089 * sk_rmem_alloc_get - returns read allocations
2090 * @sk: socket
2091 *
2092 * Return: sk_rmem_alloc
2093 */
2094static inline int sk_rmem_alloc_get(const struct sock *sk)
2095{
2096 return atomic_read(&sk->sk_rmem_alloc);
2097}
2098
2099/**
2100 * sk_has_allocations - check if allocations are outstanding
2101 * @sk: socket
2102 *
2103 * Return: true if socket has write or read allocations
2104 */
2105static inline bool sk_has_allocations(const struct sock *sk)
2106{
2107 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2108}
2109
2110/**
2111 * skwq_has_sleeper - check if there are any waiting processes
2112 * @wq: struct socket_wq
2113 *
2114 * Return: true if socket_wq has waiting processes
2115 *
2116 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2117 * barrier call. They were added due to the race found within the tcp code.
2118 *
2119 * Consider following tcp code paths::
2120 *
2121 * CPU1 CPU2
2122 * sys_select receive packet
2123 * ... ...
2124 * __add_wait_queue update tp->rcv_nxt
2125 * ... ...
2126 * tp->rcv_nxt check sock_def_readable
2127 * ... {
2128 * schedule rcu_read_lock();
2129 * wq = rcu_dereference(sk->sk_wq);
2130 * if (wq && waitqueue_active(&wq->wait))
2131 * wake_up_interruptible(&wq->wait)
2132 * ...
2133 * }
2134 *
2135 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2136 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2137 * could then endup calling schedule and sleep forever if there are no more
2138 * data on the socket.
2139 *
2140 */
2141static inline bool skwq_has_sleeper(struct socket_wq *wq)
2142{
2143 return wq && wq_has_sleeper(&wq->wait);
2144}
2145
2146/**
2147 * sock_poll_wait - place memory barrier behind the poll_wait call.
2148 * @filp: file
2149 * @sock: socket to wait on
2150 * @p: poll_table
2151 *
2152 * See the comments in the wq_has_sleeper function.
2153 */
2154static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2155 poll_table *p)
2156{
2157 if (!poll_does_not_wait(p)) {
2158 poll_wait(filp, &sock->wq.wait, p);
2159 /* We need to be sure we are in sync with the
2160 * socket flags modification.
2161 *
2162 * This memory barrier is paired in the wq_has_sleeper.
2163 */
2164 smp_mb();
2165 }
2166}
2167
2168static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2169{
2170 if (sk->sk_txhash) {
2171 skb->l4_hash = 1;
2172 skb->hash = sk->sk_txhash;
2173 }
2174}
2175
2176void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2177
2178/*
2179 * Queue a received datagram if it will fit. Stream and sequenced
2180 * protocols can't normally use this as they need to fit buffers in
2181 * and play with them.
2182 *
2183 * Inlined as it's very short and called for pretty much every
2184 * packet ever received.
2185 */
2186static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2187{
2188 skb_orphan(skb);
2189 skb->sk = sk;
2190 skb->destructor = sock_rfree;
2191 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2192 sk_mem_charge(sk, skb->truesize);
2193}
2194
2195void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2196 unsigned long expires);
2197
2198void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2199
2200int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2201 struct sk_buff *skb, unsigned int flags,
2202 void (*destructor)(struct sock *sk,
2203 struct sk_buff *skb));
2204int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2205int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2206
2207int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2208struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2209
2210/*
2211 * Recover an error report and clear atomically
2212 */
2213
2214static inline int sock_error(struct sock *sk)
2215{
2216 int err;
2217 if (likely(!sk->sk_err))
2218 return 0;
2219 err = xchg(&sk->sk_err, 0);
2220 return -err;
2221}
2222
2223static inline unsigned long sock_wspace(struct sock *sk)
2224{
2225 int amt = 0;
2226
2227 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2228 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2229 if (amt < 0)
2230 amt = 0;
2231 }
2232 return amt;
2233}
2234
2235/* Note:
2236 * We use sk->sk_wq_raw, from contexts knowing this
2237 * pointer is not NULL and cannot disappear/change.
2238 */
2239static inline void sk_set_bit(int nr, struct sock *sk)
2240{
2241 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2242 !sock_flag(sk, SOCK_FASYNC))
2243 return;
2244
2245 set_bit(nr, &sk->sk_wq_raw->flags);
2246}
2247
2248static inline void sk_clear_bit(int nr, struct sock *sk)
2249{
2250 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2251 !sock_flag(sk, SOCK_FASYNC))
2252 return;
2253
2254 clear_bit(nr, &sk->sk_wq_raw->flags);
2255}
2256
2257static inline void sk_wake_async(const struct sock *sk, int how, int band)
2258{
2259 if (sock_flag(sk, SOCK_FASYNC)) {
2260 rcu_read_lock();
2261 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2262 rcu_read_unlock();
2263 }
2264}
2265
2266/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2267 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2268 * Note: for send buffers, TCP works better if we can build two skbs at
2269 * minimum.
2270 */
2271#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2272
2273#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2274#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2275
2276static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2277{
2278 u32 val;
2279
2280 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2281 return;
2282
2283 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2284
2285 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2286}
2287
2288struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2289 bool force_schedule);
2290
2291/**
2292 * sk_page_frag - return an appropriate page_frag
2293 * @sk: socket
2294 *
2295 * Use the per task page_frag instead of the per socket one for
2296 * optimization when we know that we're in the normal context and owns
2297 * everything that's associated with %current.
2298 *
2299 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2300 * inside other socket operations and end up recursing into sk_page_frag()
2301 * while it's already in use.
2302 *
2303 * Return: a per task page_frag if context allows that,
2304 * otherwise a per socket one.
2305 */
2306static inline struct page_frag *sk_page_frag(struct sock *sk)
2307{
2308 if (gfpflags_normal_context(sk->sk_allocation))
2309 return ¤t->task_frag;
2310
2311 return &sk->sk_frag;
2312}
2313
2314bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2315
2316/*
2317 * Default write policy as shown to user space via poll/select/SIGIO
2318 */
2319static inline bool sock_writeable(const struct sock *sk)
2320{
2321 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2322}
2323
2324static inline gfp_t gfp_any(void)
2325{
2326 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2327}
2328
2329static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2330{
2331 return noblock ? 0 : sk->sk_rcvtimeo;
2332}
2333
2334static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2335{
2336 return noblock ? 0 : sk->sk_sndtimeo;
2337}
2338
2339static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2340{
2341 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2342
2343 return v ?: 1;
2344}
2345
2346/* Alas, with timeout socket operations are not restartable.
2347 * Compare this to poll().
2348 */
2349static inline int sock_intr_errno(long timeo)
2350{
2351 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2352}
2353
2354struct sock_skb_cb {
2355 u32 dropcount;
2356};
2357
2358/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2359 * using skb->cb[] would keep using it directly and utilize its
2360 * alignement guarantee.
2361 */
2362#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2363 sizeof(struct sock_skb_cb)))
2364
2365#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2366 SOCK_SKB_CB_OFFSET))
2367
2368#define sock_skb_cb_check_size(size) \
2369 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2370
2371static inline void
2372sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2373{
2374 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2375 atomic_read(&sk->sk_drops) : 0;
2376}
2377
2378static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2379{
2380 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2381
2382 atomic_add(segs, &sk->sk_drops);
2383}
2384
2385static inline ktime_t sock_read_timestamp(struct sock *sk)
2386{
2387#if BITS_PER_LONG==32
2388 unsigned int seq;
2389 ktime_t kt;
2390
2391 do {
2392 seq = read_seqbegin(&sk->sk_stamp_seq);
2393 kt = sk->sk_stamp;
2394 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2395
2396 return kt;
2397#else
2398 return READ_ONCE(sk->sk_stamp);
2399#endif
2400}
2401
2402static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2403{
2404#if BITS_PER_LONG==32
2405 write_seqlock(&sk->sk_stamp_seq);
2406 sk->sk_stamp = kt;
2407 write_sequnlock(&sk->sk_stamp_seq);
2408#else
2409 WRITE_ONCE(sk->sk_stamp, kt);
2410#endif
2411}
2412
2413void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2414 struct sk_buff *skb);
2415void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2416 struct sk_buff *skb);
2417
2418static inline void
2419sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2420{
2421 ktime_t kt = skb->tstamp;
2422 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2423
2424 /*
2425 * generate control messages if
2426 * - receive time stamping in software requested
2427 * - software time stamp available and wanted
2428 * - hardware time stamps available and wanted
2429 */
2430 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2431 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2432 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2433 (hwtstamps->hwtstamp &&
2434 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2435 __sock_recv_timestamp(msg, sk, skb);
2436 else
2437 sock_write_timestamp(sk, kt);
2438
2439 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2440 __sock_recv_wifi_status(msg, sk, skb);
2441}
2442
2443void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2444 struct sk_buff *skb);
2445
2446#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2447static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2448 struct sk_buff *skb)
2449{
2450#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2451 (1UL << SOCK_RCVTSTAMP))
2452#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2453 SOF_TIMESTAMPING_RAW_HARDWARE)
2454
2455 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2456 __sock_recv_ts_and_drops(msg, sk, skb);
2457 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2458 sock_write_timestamp(sk, skb->tstamp);
2459 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2460 sock_write_timestamp(sk, 0);
2461}
2462
2463void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2464
2465/**
2466 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2467 * @sk: socket sending this packet
2468 * @tsflags: timestamping flags to use
2469 * @tx_flags: completed with instructions for time stamping
2470 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2471 *
2472 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2473 */
2474static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2475 __u8 *tx_flags, __u32 *tskey)
2476{
2477 if (unlikely(tsflags)) {
2478 __sock_tx_timestamp(tsflags, tx_flags);
2479 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2480 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2481 *tskey = sk->sk_tskey++;
2482 }
2483 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2484 *tx_flags |= SKBTX_WIFI_STATUS;
2485}
2486
2487static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2488 __u8 *tx_flags)
2489{
2490 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2491}
2492
2493static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2494{
2495 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2496 &skb_shinfo(skb)->tskey);
2497}
2498
2499DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2500/**
2501 * sk_eat_skb - Release a skb if it is no longer needed
2502 * @sk: socket to eat this skb from
2503 * @skb: socket buffer to eat
2504 *
2505 * This routine must be called with interrupts disabled or with the socket
2506 * locked so that the sk_buff queue operation is ok.
2507*/
2508static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2509{
2510 __skb_unlink(skb, &sk->sk_receive_queue);
2511 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2512 !sk->sk_rx_skb_cache) {
2513 sk->sk_rx_skb_cache = skb;
2514 skb_orphan(skb);
2515 return;
2516 }
2517 __kfree_skb(skb);
2518}
2519
2520static inline
2521struct net *sock_net(const struct sock *sk)
2522{
2523 return read_pnet(&sk->sk_net);
2524}
2525
2526static inline
2527void sock_net_set(struct sock *sk, struct net *net)
2528{
2529 write_pnet(&sk->sk_net, net);
2530}
2531
2532static inline bool
2533skb_sk_is_prefetched(struct sk_buff *skb)
2534{
2535#ifdef CONFIG_INET
2536 return skb->destructor == sock_pfree;
2537#else
2538 return false;
2539#endif /* CONFIG_INET */
2540}
2541
2542/* This helper checks if a socket is a full socket,
2543 * ie _not_ a timewait or request socket.
2544 */
2545static inline bool sk_fullsock(const struct sock *sk)
2546{
2547 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2548}
2549
2550static inline bool
2551sk_is_refcounted(struct sock *sk)
2552{
2553 /* Only full sockets have sk->sk_flags. */
2554 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2555}
2556
2557/**
2558 * skb_steal_sock - steal a socket from an sk_buff
2559 * @skb: sk_buff to steal the socket from
2560 * @refcounted: is set to true if the socket is reference-counted
2561 */
2562static inline struct sock *
2563skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2564{
2565 if (skb->sk) {
2566 struct sock *sk = skb->sk;
2567
2568 *refcounted = true;
2569 if (skb_sk_is_prefetched(skb))
2570 *refcounted = sk_is_refcounted(sk);
2571 skb->destructor = NULL;
2572 skb->sk = NULL;
2573 return sk;
2574 }
2575 *refcounted = false;
2576 return NULL;
2577}
2578
2579/* Checks if this SKB belongs to an HW offloaded socket
2580 * and whether any SW fallbacks are required based on dev.
2581 * Check decrypted mark in case skb_orphan() cleared socket.
2582 */
2583static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2584 struct net_device *dev)
2585{
2586#ifdef CONFIG_SOCK_VALIDATE_XMIT
2587 struct sock *sk = skb->sk;
2588
2589 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2590 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2591#ifdef CONFIG_TLS_DEVICE
2592 } else if (unlikely(skb->decrypted)) {
2593 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2594 kfree_skb(skb);
2595 skb = NULL;
2596#endif
2597 }
2598#endif
2599
2600 return skb;
2601}
2602
2603/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2604 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2605 */
2606static inline bool sk_listener(const struct sock *sk)
2607{
2608 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2609}
2610
2611void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2612int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2613 int type);
2614
2615bool sk_ns_capable(const struct sock *sk,
2616 struct user_namespace *user_ns, int cap);
2617bool sk_capable(const struct sock *sk, int cap);
2618bool sk_net_capable(const struct sock *sk, int cap);
2619
2620void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2621
2622/* Take into consideration the size of the struct sk_buff overhead in the
2623 * determination of these values, since that is non-constant across
2624 * platforms. This makes socket queueing behavior and performance
2625 * not depend upon such differences.
2626 */
2627#define _SK_MEM_PACKETS 256
2628#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2629#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2630#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2631
2632extern __u32 sysctl_wmem_max;
2633extern __u32 sysctl_rmem_max;
2634
2635extern int sysctl_tstamp_allow_data;
2636extern int sysctl_optmem_max;
2637
2638extern __u32 sysctl_wmem_default;
2639extern __u32 sysctl_rmem_default;
2640
2641DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2642
2643static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2644{
2645 /* Does this proto have per netns sysctl_wmem ? */
2646 if (proto->sysctl_wmem_offset)
2647 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2648
2649 return *proto->sysctl_wmem;
2650}
2651
2652static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2653{
2654 /* Does this proto have per netns sysctl_rmem ? */
2655 if (proto->sysctl_rmem_offset)
2656 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2657
2658 return *proto->sysctl_rmem;
2659}
2660
2661/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2662 * Some wifi drivers need to tweak it to get more chunks.
2663 * They can use this helper from their ndo_start_xmit()
2664 */
2665static inline void sk_pacing_shift_update(struct sock *sk, int val)
2666{
2667 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2668 return;
2669 WRITE_ONCE(sk->sk_pacing_shift, val);
2670}
2671
2672/* if a socket is bound to a device, check that the given device
2673 * index is either the same or that the socket is bound to an L3
2674 * master device and the given device index is also enslaved to
2675 * that L3 master
2676 */
2677static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2678{
2679 int mdif;
2680
2681 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2682 return true;
2683
2684 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2685 if (mdif && mdif == sk->sk_bound_dev_if)
2686 return true;
2687
2688 return false;
2689}
2690
2691void sock_def_readable(struct sock *sk);
2692
2693int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2694void sock_enable_timestamps(struct sock *sk);
2695void sock_no_linger(struct sock *sk);
2696void sock_set_keepalive(struct sock *sk);
2697void sock_set_priority(struct sock *sk, u32 priority);
2698void sock_set_rcvbuf(struct sock *sk, int val);
2699void sock_set_mark(struct sock *sk, u32 val);
2700void sock_set_reuseaddr(struct sock *sk);
2701void sock_set_reuseport(struct sock *sk);
2702void sock_set_sndtimeo(struct sock *sk, s64 secs);
2703
2704int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2705
2706#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/memcontrol.h>
58#include <linux/res_counter.h>
59#include <linux/static_key.h>
60#include <linux/aio.h>
61#include <linux/sched.h>
62
63#include <linux/filter.h>
64#include <linux/rculist_nulls.h>
65#include <linux/poll.h>
66
67#include <linux/atomic.h>
68#include <net/dst.h>
69#include <net/checksum.h>
70
71struct cgroup;
72struct cgroup_subsys;
73#ifdef CONFIG_NET
74int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
75void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
76#else
77static inline
78int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
79{
80 return 0;
81}
82static inline
83void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
84{
85}
86#endif
87/*
88 * This structure really needs to be cleaned up.
89 * Most of it is for TCP, and not used by any of
90 * the other protocols.
91 */
92
93/* Define this to get the SOCK_DBG debugging facility. */
94#define SOCK_DEBUGGING
95#ifdef SOCK_DEBUGGING
96#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
97 printk(KERN_DEBUG msg); } while (0)
98#else
99/* Validate arguments and do nothing */
100static inline __printf(2, 3)
101void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
102{
103}
104#endif
105
106/* This is the per-socket lock. The spinlock provides a synchronization
107 * between user contexts and software interrupt processing, whereas the
108 * mini-semaphore synchronizes multiple users amongst themselves.
109 */
110typedef struct {
111 spinlock_t slock;
112 int owned;
113 wait_queue_head_t wq;
114 /*
115 * We express the mutex-alike socket_lock semantics
116 * to the lock validator by explicitly managing
117 * the slock as a lock variant (in addition to
118 * the slock itself):
119 */
120#ifdef CONFIG_DEBUG_LOCK_ALLOC
121 struct lockdep_map dep_map;
122#endif
123} socket_lock_t;
124
125struct sock;
126struct proto;
127struct net;
128
129typedef __u32 __bitwise __portpair;
130typedef __u64 __bitwise __addrpair;
131
132/**
133 * struct sock_common - minimal network layer representation of sockets
134 * @skc_daddr: Foreign IPv4 addr
135 * @skc_rcv_saddr: Bound local IPv4 addr
136 * @skc_hash: hash value used with various protocol lookup tables
137 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
138 * @skc_dport: placeholder for inet_dport/tw_dport
139 * @skc_num: placeholder for inet_num/tw_num
140 * @skc_family: network address family
141 * @skc_state: Connection state
142 * @skc_reuse: %SO_REUSEADDR setting
143 * @skc_reuseport: %SO_REUSEPORT setting
144 * @skc_bound_dev_if: bound device index if != 0
145 * @skc_bind_node: bind hash linkage for various protocol lookup tables
146 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
147 * @skc_prot: protocol handlers inside a network family
148 * @skc_net: reference to the network namespace of this socket
149 * @skc_node: main hash linkage for various protocol lookup tables
150 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
151 * @skc_tx_queue_mapping: tx queue number for this connection
152 * @skc_refcnt: reference count
153 *
154 * This is the minimal network layer representation of sockets, the header
155 * for struct sock and struct inet_timewait_sock.
156 */
157struct sock_common {
158 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
159 * address on 64bit arches : cf INET_MATCH()
160 */
161 union {
162 __addrpair skc_addrpair;
163 struct {
164 __be32 skc_daddr;
165 __be32 skc_rcv_saddr;
166 };
167 };
168 union {
169 unsigned int skc_hash;
170 __u16 skc_u16hashes[2];
171 };
172 /* skc_dport && skc_num must be grouped as well */
173 union {
174 __portpair skc_portpair;
175 struct {
176 __be16 skc_dport;
177 __u16 skc_num;
178 };
179 };
180
181 unsigned short skc_family;
182 volatile unsigned char skc_state;
183 unsigned char skc_reuse:4;
184 unsigned char skc_reuseport:4;
185 int skc_bound_dev_if;
186 union {
187 struct hlist_node skc_bind_node;
188 struct hlist_nulls_node skc_portaddr_node;
189 };
190 struct proto *skc_prot;
191#ifdef CONFIG_NET_NS
192 struct net *skc_net;
193#endif
194
195#if IS_ENABLED(CONFIG_IPV6)
196 struct in6_addr skc_v6_daddr;
197 struct in6_addr skc_v6_rcv_saddr;
198#endif
199
200 /*
201 * fields between dontcopy_begin/dontcopy_end
202 * are not copied in sock_copy()
203 */
204 /* private: */
205 int skc_dontcopy_begin[0];
206 /* public: */
207 union {
208 struct hlist_node skc_node;
209 struct hlist_nulls_node skc_nulls_node;
210 };
211 int skc_tx_queue_mapping;
212 atomic_t skc_refcnt;
213 /* private: */
214 int skc_dontcopy_end[0];
215 /* public: */
216};
217
218struct cg_proto;
219/**
220 * struct sock - network layer representation of sockets
221 * @__sk_common: shared layout with inet_timewait_sock
222 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
223 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
224 * @sk_lock: synchronizer
225 * @sk_rcvbuf: size of receive buffer in bytes
226 * @sk_wq: sock wait queue and async head
227 * @sk_rx_dst: receive input route used by early demux
228 * @sk_dst_cache: destination cache
229 * @sk_dst_lock: destination cache lock
230 * @sk_policy: flow policy
231 * @sk_receive_queue: incoming packets
232 * @sk_wmem_alloc: transmit queue bytes committed
233 * @sk_write_queue: Packet sending queue
234 * @sk_async_wait_queue: DMA copied packets
235 * @sk_omem_alloc: "o" is "option" or "other"
236 * @sk_wmem_queued: persistent queue size
237 * @sk_forward_alloc: space allocated forward
238 * @sk_napi_id: id of the last napi context to receive data for sk
239 * @sk_ll_usec: usecs to busypoll when there is no data
240 * @sk_allocation: allocation mode
241 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
242 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
243 * @sk_sndbuf: size of send buffer in bytes
244 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
245 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
246 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets
247 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
248 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
249 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
250 * @sk_gso_max_size: Maximum GSO segment size to build
251 * @sk_gso_max_segs: Maximum number of GSO segments
252 * @sk_lingertime: %SO_LINGER l_linger setting
253 * @sk_backlog: always used with the per-socket spinlock held
254 * @sk_callback_lock: used with the callbacks in the end of this struct
255 * @sk_error_queue: rarely used
256 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
257 * IPV6_ADDRFORM for instance)
258 * @sk_err: last error
259 * @sk_err_soft: errors that don't cause failure but are the cause of a
260 * persistent failure not just 'timed out'
261 * @sk_drops: raw/udp drops counter
262 * @sk_ack_backlog: current listen backlog
263 * @sk_max_ack_backlog: listen backlog set in listen()
264 * @sk_priority: %SO_PRIORITY setting
265 * @sk_cgrp_prioidx: socket group's priority map index
266 * @sk_type: socket type (%SOCK_STREAM, etc)
267 * @sk_protocol: which protocol this socket belongs in this network family
268 * @sk_peer_pid: &struct pid for this socket's peer
269 * @sk_peer_cred: %SO_PEERCRED setting
270 * @sk_rcvlowat: %SO_RCVLOWAT setting
271 * @sk_rcvtimeo: %SO_RCVTIMEO setting
272 * @sk_sndtimeo: %SO_SNDTIMEO setting
273 * @sk_rxhash: flow hash received from netif layer
274 * @sk_filter: socket filtering instructions
275 * @sk_protinfo: private area, net family specific, when not using slab
276 * @sk_timer: sock cleanup timer
277 * @sk_stamp: time stamp of last packet received
278 * @sk_socket: Identd and reporting IO signals
279 * @sk_user_data: RPC layer private data
280 * @sk_frag: cached page frag
281 * @sk_peek_off: current peek_offset value
282 * @sk_send_head: front of stuff to transmit
283 * @sk_security: used by security modules
284 * @sk_mark: generic packet mark
285 * @sk_classid: this socket's cgroup classid
286 * @sk_cgrp: this socket's cgroup-specific proto data
287 * @sk_write_pending: a write to stream socket waits to start
288 * @sk_state_change: callback to indicate change in the state of the sock
289 * @sk_data_ready: callback to indicate there is data to be processed
290 * @sk_write_space: callback to indicate there is bf sending space available
291 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
292 * @sk_backlog_rcv: callback to process the backlog
293 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
294 */
295struct sock {
296 /*
297 * Now struct inet_timewait_sock also uses sock_common, so please just
298 * don't add nothing before this first member (__sk_common) --acme
299 */
300 struct sock_common __sk_common;
301#define sk_node __sk_common.skc_node
302#define sk_nulls_node __sk_common.skc_nulls_node
303#define sk_refcnt __sk_common.skc_refcnt
304#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
305
306#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
307#define sk_dontcopy_end __sk_common.skc_dontcopy_end
308#define sk_hash __sk_common.skc_hash
309#define sk_portpair __sk_common.skc_portpair
310#define sk_num __sk_common.skc_num
311#define sk_dport __sk_common.skc_dport
312#define sk_addrpair __sk_common.skc_addrpair
313#define sk_daddr __sk_common.skc_daddr
314#define sk_rcv_saddr __sk_common.skc_rcv_saddr
315#define sk_family __sk_common.skc_family
316#define sk_state __sk_common.skc_state
317#define sk_reuse __sk_common.skc_reuse
318#define sk_reuseport __sk_common.skc_reuseport
319#define sk_bound_dev_if __sk_common.skc_bound_dev_if
320#define sk_bind_node __sk_common.skc_bind_node
321#define sk_prot __sk_common.skc_prot
322#define sk_net __sk_common.skc_net
323#define sk_v6_daddr __sk_common.skc_v6_daddr
324#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
325
326 socket_lock_t sk_lock;
327 struct sk_buff_head sk_receive_queue;
328 /*
329 * The backlog queue is special, it is always used with
330 * the per-socket spinlock held and requires low latency
331 * access. Therefore we special case it's implementation.
332 * Note : rmem_alloc is in this structure to fill a hole
333 * on 64bit arches, not because its logically part of
334 * backlog.
335 */
336 struct {
337 atomic_t rmem_alloc;
338 int len;
339 struct sk_buff *head;
340 struct sk_buff *tail;
341 } sk_backlog;
342#define sk_rmem_alloc sk_backlog.rmem_alloc
343 int sk_forward_alloc;
344#ifdef CONFIG_RPS
345 __u32 sk_rxhash;
346#endif
347#ifdef CONFIG_NET_RX_BUSY_POLL
348 unsigned int sk_napi_id;
349 unsigned int sk_ll_usec;
350#endif
351 atomic_t sk_drops;
352 int sk_rcvbuf;
353
354 struct sk_filter __rcu *sk_filter;
355 struct socket_wq __rcu *sk_wq;
356
357#ifdef CONFIG_NET_DMA
358 struct sk_buff_head sk_async_wait_queue;
359#endif
360
361#ifdef CONFIG_XFRM
362 struct xfrm_policy *sk_policy[2];
363#endif
364 unsigned long sk_flags;
365 struct dst_entry *sk_rx_dst;
366 struct dst_entry __rcu *sk_dst_cache;
367 spinlock_t sk_dst_lock;
368 atomic_t sk_wmem_alloc;
369 atomic_t sk_omem_alloc;
370 int sk_sndbuf;
371 struct sk_buff_head sk_write_queue;
372 kmemcheck_bitfield_begin(flags);
373 unsigned int sk_shutdown : 2,
374 sk_no_check : 2,
375 sk_userlocks : 4,
376 sk_protocol : 8,
377 sk_type : 16;
378 kmemcheck_bitfield_end(flags);
379 int sk_wmem_queued;
380 gfp_t sk_allocation;
381 u32 sk_pacing_rate; /* bytes per second */
382 u32 sk_max_pacing_rate;
383 netdev_features_t sk_route_caps;
384 netdev_features_t sk_route_nocaps;
385 int sk_gso_type;
386 unsigned int sk_gso_max_size;
387 u16 sk_gso_max_segs;
388 int sk_rcvlowat;
389 unsigned long sk_lingertime;
390 struct sk_buff_head sk_error_queue;
391 struct proto *sk_prot_creator;
392 rwlock_t sk_callback_lock;
393 int sk_err,
394 sk_err_soft;
395 unsigned short sk_ack_backlog;
396 unsigned short sk_max_ack_backlog;
397 __u32 sk_priority;
398#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
399 __u32 sk_cgrp_prioidx;
400#endif
401 struct pid *sk_peer_pid;
402 const struct cred *sk_peer_cred;
403 long sk_rcvtimeo;
404 long sk_sndtimeo;
405 void *sk_protinfo;
406 struct timer_list sk_timer;
407 ktime_t sk_stamp;
408 struct socket *sk_socket;
409 void *sk_user_data;
410 struct page_frag sk_frag;
411 struct sk_buff *sk_send_head;
412 __s32 sk_peek_off;
413 int sk_write_pending;
414#ifdef CONFIG_SECURITY
415 void *sk_security;
416#endif
417 __u32 sk_mark;
418 u32 sk_classid;
419 struct cg_proto *sk_cgrp;
420 void (*sk_state_change)(struct sock *sk);
421 void (*sk_data_ready)(struct sock *sk);
422 void (*sk_write_space)(struct sock *sk);
423 void (*sk_error_report)(struct sock *sk);
424 int (*sk_backlog_rcv)(struct sock *sk,
425 struct sk_buff *skb);
426 void (*sk_destruct)(struct sock *sk);
427};
428
429#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
430
431#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
432#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
433
434/*
435 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
436 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
437 * on a socket means that the socket will reuse everybody else's port
438 * without looking at the other's sk_reuse value.
439 */
440
441#define SK_NO_REUSE 0
442#define SK_CAN_REUSE 1
443#define SK_FORCE_REUSE 2
444
445static inline int sk_peek_offset(struct sock *sk, int flags)
446{
447 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
448 return sk->sk_peek_off;
449 else
450 return 0;
451}
452
453static inline void sk_peek_offset_bwd(struct sock *sk, int val)
454{
455 if (sk->sk_peek_off >= 0) {
456 if (sk->sk_peek_off >= val)
457 sk->sk_peek_off -= val;
458 else
459 sk->sk_peek_off = 0;
460 }
461}
462
463static inline void sk_peek_offset_fwd(struct sock *sk, int val)
464{
465 if (sk->sk_peek_off >= 0)
466 sk->sk_peek_off += val;
467}
468
469/*
470 * Hashed lists helper routines
471 */
472static inline struct sock *sk_entry(const struct hlist_node *node)
473{
474 return hlist_entry(node, struct sock, sk_node);
475}
476
477static inline struct sock *__sk_head(const struct hlist_head *head)
478{
479 return hlist_entry(head->first, struct sock, sk_node);
480}
481
482static inline struct sock *sk_head(const struct hlist_head *head)
483{
484 return hlist_empty(head) ? NULL : __sk_head(head);
485}
486
487static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
488{
489 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
490}
491
492static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
493{
494 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
495}
496
497static inline struct sock *sk_next(const struct sock *sk)
498{
499 return sk->sk_node.next ?
500 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
501}
502
503static inline struct sock *sk_nulls_next(const struct sock *sk)
504{
505 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
506 hlist_nulls_entry(sk->sk_nulls_node.next,
507 struct sock, sk_nulls_node) :
508 NULL;
509}
510
511static inline bool sk_unhashed(const struct sock *sk)
512{
513 return hlist_unhashed(&sk->sk_node);
514}
515
516static inline bool sk_hashed(const struct sock *sk)
517{
518 return !sk_unhashed(sk);
519}
520
521static inline void sk_node_init(struct hlist_node *node)
522{
523 node->pprev = NULL;
524}
525
526static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
527{
528 node->pprev = NULL;
529}
530
531static inline void __sk_del_node(struct sock *sk)
532{
533 __hlist_del(&sk->sk_node);
534}
535
536/* NB: equivalent to hlist_del_init_rcu */
537static inline bool __sk_del_node_init(struct sock *sk)
538{
539 if (sk_hashed(sk)) {
540 __sk_del_node(sk);
541 sk_node_init(&sk->sk_node);
542 return true;
543 }
544 return false;
545}
546
547/* Grab socket reference count. This operation is valid only
548 when sk is ALREADY grabbed f.e. it is found in hash table
549 or a list and the lookup is made under lock preventing hash table
550 modifications.
551 */
552
553static inline void sock_hold(struct sock *sk)
554{
555 atomic_inc(&sk->sk_refcnt);
556}
557
558/* Ungrab socket in the context, which assumes that socket refcnt
559 cannot hit zero, f.e. it is true in context of any socketcall.
560 */
561static inline void __sock_put(struct sock *sk)
562{
563 atomic_dec(&sk->sk_refcnt);
564}
565
566static inline bool sk_del_node_init(struct sock *sk)
567{
568 bool rc = __sk_del_node_init(sk);
569
570 if (rc) {
571 /* paranoid for a while -acme */
572 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
573 __sock_put(sk);
574 }
575 return rc;
576}
577#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
578
579static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
580{
581 if (sk_hashed(sk)) {
582 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
583 return true;
584 }
585 return false;
586}
587
588static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
589{
590 bool rc = __sk_nulls_del_node_init_rcu(sk);
591
592 if (rc) {
593 /* paranoid for a while -acme */
594 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
595 __sock_put(sk);
596 }
597 return rc;
598}
599
600static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
601{
602 hlist_add_head(&sk->sk_node, list);
603}
604
605static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
606{
607 sock_hold(sk);
608 __sk_add_node(sk, list);
609}
610
611static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
612{
613 sock_hold(sk);
614 hlist_add_head_rcu(&sk->sk_node, list);
615}
616
617static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
618{
619 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
620}
621
622static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
623{
624 sock_hold(sk);
625 __sk_nulls_add_node_rcu(sk, list);
626}
627
628static inline void __sk_del_bind_node(struct sock *sk)
629{
630 __hlist_del(&sk->sk_bind_node);
631}
632
633static inline void sk_add_bind_node(struct sock *sk,
634 struct hlist_head *list)
635{
636 hlist_add_head(&sk->sk_bind_node, list);
637}
638
639#define sk_for_each(__sk, list) \
640 hlist_for_each_entry(__sk, list, sk_node)
641#define sk_for_each_rcu(__sk, list) \
642 hlist_for_each_entry_rcu(__sk, list, sk_node)
643#define sk_nulls_for_each(__sk, node, list) \
644 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
645#define sk_nulls_for_each_rcu(__sk, node, list) \
646 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
647#define sk_for_each_from(__sk) \
648 hlist_for_each_entry_from(__sk, sk_node)
649#define sk_nulls_for_each_from(__sk, node) \
650 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
651 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
652#define sk_for_each_safe(__sk, tmp, list) \
653 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
654#define sk_for_each_bound(__sk, list) \
655 hlist_for_each_entry(__sk, list, sk_bind_node)
656
657static inline struct user_namespace *sk_user_ns(struct sock *sk)
658{
659 /* Careful only use this in a context where these parameters
660 * can not change and must all be valid, such as recvmsg from
661 * userspace.
662 */
663 return sk->sk_socket->file->f_cred->user_ns;
664}
665
666/* Sock flags */
667enum sock_flags {
668 SOCK_DEAD,
669 SOCK_DONE,
670 SOCK_URGINLINE,
671 SOCK_KEEPOPEN,
672 SOCK_LINGER,
673 SOCK_DESTROY,
674 SOCK_BROADCAST,
675 SOCK_TIMESTAMP,
676 SOCK_ZAPPED,
677 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
678 SOCK_DBG, /* %SO_DEBUG setting */
679 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
680 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
681 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
682 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
683 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
684 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
685 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
686 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
687 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
688 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
689 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
690 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
691 SOCK_FASYNC, /* fasync() active */
692 SOCK_RXQ_OVFL,
693 SOCK_ZEROCOPY, /* buffers from userspace */
694 SOCK_WIFI_STATUS, /* push wifi status to userspace */
695 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
696 * Will use last 4 bytes of packet sent from
697 * user-space instead.
698 */
699 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
700 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
701};
702
703static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
704{
705 nsk->sk_flags = osk->sk_flags;
706}
707
708static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
709{
710 __set_bit(flag, &sk->sk_flags);
711}
712
713static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
714{
715 __clear_bit(flag, &sk->sk_flags);
716}
717
718static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
719{
720 return test_bit(flag, &sk->sk_flags);
721}
722
723#ifdef CONFIG_NET
724extern struct static_key memalloc_socks;
725static inline int sk_memalloc_socks(void)
726{
727 return static_key_false(&memalloc_socks);
728}
729#else
730
731static inline int sk_memalloc_socks(void)
732{
733 return 0;
734}
735
736#endif
737
738static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
739{
740 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
741}
742
743static inline void sk_acceptq_removed(struct sock *sk)
744{
745 sk->sk_ack_backlog--;
746}
747
748static inline void sk_acceptq_added(struct sock *sk)
749{
750 sk->sk_ack_backlog++;
751}
752
753static inline bool sk_acceptq_is_full(const struct sock *sk)
754{
755 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
756}
757
758/*
759 * Compute minimal free write space needed to queue new packets.
760 */
761static inline int sk_stream_min_wspace(const struct sock *sk)
762{
763 return sk->sk_wmem_queued >> 1;
764}
765
766static inline int sk_stream_wspace(const struct sock *sk)
767{
768 return sk->sk_sndbuf - sk->sk_wmem_queued;
769}
770
771void sk_stream_write_space(struct sock *sk);
772
773/* OOB backlog add */
774static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
775{
776 /* dont let skb dst not refcounted, we are going to leave rcu lock */
777 skb_dst_force(skb);
778
779 if (!sk->sk_backlog.tail)
780 sk->sk_backlog.head = skb;
781 else
782 sk->sk_backlog.tail->next = skb;
783
784 sk->sk_backlog.tail = skb;
785 skb->next = NULL;
786}
787
788/*
789 * Take into account size of receive queue and backlog queue
790 * Do not take into account this skb truesize,
791 * to allow even a single big packet to come.
792 */
793static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
794 unsigned int limit)
795{
796 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
797
798 return qsize > limit;
799}
800
801/* The per-socket spinlock must be held here. */
802static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
803 unsigned int limit)
804{
805 if (sk_rcvqueues_full(sk, skb, limit))
806 return -ENOBUFS;
807
808 __sk_add_backlog(sk, skb);
809 sk->sk_backlog.len += skb->truesize;
810 return 0;
811}
812
813int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
814
815static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
816{
817 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
818 return __sk_backlog_rcv(sk, skb);
819
820 return sk->sk_backlog_rcv(sk, skb);
821}
822
823static inline void sock_rps_record_flow_hash(__u32 hash)
824{
825#ifdef CONFIG_RPS
826 struct rps_sock_flow_table *sock_flow_table;
827
828 rcu_read_lock();
829 sock_flow_table = rcu_dereference(rps_sock_flow_table);
830 rps_record_sock_flow(sock_flow_table, hash);
831 rcu_read_unlock();
832#endif
833}
834
835static inline void sock_rps_reset_flow_hash(__u32 hash)
836{
837#ifdef CONFIG_RPS
838 struct rps_sock_flow_table *sock_flow_table;
839
840 rcu_read_lock();
841 sock_flow_table = rcu_dereference(rps_sock_flow_table);
842 rps_reset_sock_flow(sock_flow_table, hash);
843 rcu_read_unlock();
844#endif
845}
846
847static inline void sock_rps_record_flow(const struct sock *sk)
848{
849#ifdef CONFIG_RPS
850 sock_rps_record_flow_hash(sk->sk_rxhash);
851#endif
852}
853
854static inline void sock_rps_reset_flow(const struct sock *sk)
855{
856#ifdef CONFIG_RPS
857 sock_rps_reset_flow_hash(sk->sk_rxhash);
858#endif
859}
860
861static inline void sock_rps_save_rxhash(struct sock *sk,
862 const struct sk_buff *skb)
863{
864#ifdef CONFIG_RPS
865 if (unlikely(sk->sk_rxhash != skb->hash)) {
866 sock_rps_reset_flow(sk);
867 sk->sk_rxhash = skb->hash;
868 }
869#endif
870}
871
872static inline void sock_rps_reset_rxhash(struct sock *sk)
873{
874#ifdef CONFIG_RPS
875 sock_rps_reset_flow(sk);
876 sk->sk_rxhash = 0;
877#endif
878}
879
880#define sk_wait_event(__sk, __timeo, __condition) \
881 ({ int __rc; \
882 release_sock(__sk); \
883 __rc = __condition; \
884 if (!__rc) { \
885 *(__timeo) = schedule_timeout(*(__timeo)); \
886 } \
887 lock_sock(__sk); \
888 __rc = __condition; \
889 __rc; \
890 })
891
892int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
893int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
894void sk_stream_wait_close(struct sock *sk, long timeo_p);
895int sk_stream_error(struct sock *sk, int flags, int err);
896void sk_stream_kill_queues(struct sock *sk);
897void sk_set_memalloc(struct sock *sk);
898void sk_clear_memalloc(struct sock *sk);
899
900int sk_wait_data(struct sock *sk, long *timeo);
901
902struct request_sock_ops;
903struct timewait_sock_ops;
904struct inet_hashinfo;
905struct raw_hashinfo;
906struct module;
907
908/*
909 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
910 * un-modified. Special care is taken when initializing object to zero.
911 */
912static inline void sk_prot_clear_nulls(struct sock *sk, int size)
913{
914 if (offsetof(struct sock, sk_node.next) != 0)
915 memset(sk, 0, offsetof(struct sock, sk_node.next));
916 memset(&sk->sk_node.pprev, 0,
917 size - offsetof(struct sock, sk_node.pprev));
918}
919
920/* Networking protocol blocks we attach to sockets.
921 * socket layer -> transport layer interface
922 * transport -> network interface is defined by struct inet_proto
923 */
924struct proto {
925 void (*close)(struct sock *sk,
926 long timeout);
927 int (*connect)(struct sock *sk,
928 struct sockaddr *uaddr,
929 int addr_len);
930 int (*disconnect)(struct sock *sk, int flags);
931
932 struct sock * (*accept)(struct sock *sk, int flags, int *err);
933
934 int (*ioctl)(struct sock *sk, int cmd,
935 unsigned long arg);
936 int (*init)(struct sock *sk);
937 void (*destroy)(struct sock *sk);
938 void (*shutdown)(struct sock *sk, int how);
939 int (*setsockopt)(struct sock *sk, int level,
940 int optname, char __user *optval,
941 unsigned int optlen);
942 int (*getsockopt)(struct sock *sk, int level,
943 int optname, char __user *optval,
944 int __user *option);
945#ifdef CONFIG_COMPAT
946 int (*compat_setsockopt)(struct sock *sk,
947 int level,
948 int optname, char __user *optval,
949 unsigned int optlen);
950 int (*compat_getsockopt)(struct sock *sk,
951 int level,
952 int optname, char __user *optval,
953 int __user *option);
954 int (*compat_ioctl)(struct sock *sk,
955 unsigned int cmd, unsigned long arg);
956#endif
957 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
958 struct msghdr *msg, size_t len);
959 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
960 struct msghdr *msg,
961 size_t len, int noblock, int flags,
962 int *addr_len);
963 int (*sendpage)(struct sock *sk, struct page *page,
964 int offset, size_t size, int flags);
965 int (*bind)(struct sock *sk,
966 struct sockaddr *uaddr, int addr_len);
967
968 int (*backlog_rcv) (struct sock *sk,
969 struct sk_buff *skb);
970
971 void (*release_cb)(struct sock *sk);
972 void (*mtu_reduced)(struct sock *sk);
973
974 /* Keeping track of sk's, looking them up, and port selection methods. */
975 void (*hash)(struct sock *sk);
976 void (*unhash)(struct sock *sk);
977 void (*rehash)(struct sock *sk);
978 int (*get_port)(struct sock *sk, unsigned short snum);
979 void (*clear_sk)(struct sock *sk, int size);
980
981 /* Keeping track of sockets in use */
982#ifdef CONFIG_PROC_FS
983 unsigned int inuse_idx;
984#endif
985
986 bool (*stream_memory_free)(const struct sock *sk);
987 /* Memory pressure */
988 void (*enter_memory_pressure)(struct sock *sk);
989 atomic_long_t *memory_allocated; /* Current allocated memory. */
990 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
991 /*
992 * Pressure flag: try to collapse.
993 * Technical note: it is used by multiple contexts non atomically.
994 * All the __sk_mem_schedule() is of this nature: accounting
995 * is strict, actions are advisory and have some latency.
996 */
997 int *memory_pressure;
998 long *sysctl_mem;
999 int *sysctl_wmem;
1000 int *sysctl_rmem;
1001 int max_header;
1002 bool no_autobind;
1003
1004 struct kmem_cache *slab;
1005 unsigned int obj_size;
1006 int slab_flags;
1007
1008 struct percpu_counter *orphan_count;
1009
1010 struct request_sock_ops *rsk_prot;
1011 struct timewait_sock_ops *twsk_prot;
1012
1013 union {
1014 struct inet_hashinfo *hashinfo;
1015 struct udp_table *udp_table;
1016 struct raw_hashinfo *raw_hash;
1017 } h;
1018
1019 struct module *owner;
1020
1021 char name[32];
1022
1023 struct list_head node;
1024#ifdef SOCK_REFCNT_DEBUG
1025 atomic_t socks;
1026#endif
1027#ifdef CONFIG_MEMCG_KMEM
1028 /*
1029 * cgroup specific init/deinit functions. Called once for all
1030 * protocols that implement it, from cgroups populate function.
1031 * This function has to setup any files the protocol want to
1032 * appear in the kmem cgroup filesystem.
1033 */
1034 int (*init_cgroup)(struct mem_cgroup *memcg,
1035 struct cgroup_subsys *ss);
1036 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1037 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1038#endif
1039};
1040
1041/*
1042 * Bits in struct cg_proto.flags
1043 */
1044enum cg_proto_flags {
1045 /* Currently active and new sockets should be assigned to cgroups */
1046 MEMCG_SOCK_ACTIVE,
1047 /* It was ever activated; we must disarm static keys on destruction */
1048 MEMCG_SOCK_ACTIVATED,
1049};
1050
1051struct cg_proto {
1052 struct res_counter memory_allocated; /* Current allocated memory. */
1053 struct percpu_counter sockets_allocated; /* Current number of sockets. */
1054 int memory_pressure;
1055 long sysctl_mem[3];
1056 unsigned long flags;
1057 /*
1058 * memcg field is used to find which memcg we belong directly
1059 * Each memcg struct can hold more than one cg_proto, so container_of
1060 * won't really cut.
1061 *
1062 * The elegant solution would be having an inverse function to
1063 * proto_cgroup in struct proto, but that means polluting the structure
1064 * for everybody, instead of just for memcg users.
1065 */
1066 struct mem_cgroup *memcg;
1067};
1068
1069int proto_register(struct proto *prot, int alloc_slab);
1070void proto_unregister(struct proto *prot);
1071
1072static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1073{
1074 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1075}
1076
1077static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1078{
1079 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1080}
1081
1082#ifdef SOCK_REFCNT_DEBUG
1083static inline void sk_refcnt_debug_inc(struct sock *sk)
1084{
1085 atomic_inc(&sk->sk_prot->socks);
1086}
1087
1088static inline void sk_refcnt_debug_dec(struct sock *sk)
1089{
1090 atomic_dec(&sk->sk_prot->socks);
1091 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1092 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1093}
1094
1095static inline void sk_refcnt_debug_release(const struct sock *sk)
1096{
1097 if (atomic_read(&sk->sk_refcnt) != 1)
1098 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1099 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1100}
1101#else /* SOCK_REFCNT_DEBUG */
1102#define sk_refcnt_debug_inc(sk) do { } while (0)
1103#define sk_refcnt_debug_dec(sk) do { } while (0)
1104#define sk_refcnt_debug_release(sk) do { } while (0)
1105#endif /* SOCK_REFCNT_DEBUG */
1106
1107#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1108extern struct static_key memcg_socket_limit_enabled;
1109static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1110 struct cg_proto *cg_proto)
1111{
1112 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1113}
1114#define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1115#else
1116#define mem_cgroup_sockets_enabled 0
1117static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1118 struct cg_proto *cg_proto)
1119{
1120 return NULL;
1121}
1122#endif
1123
1124static inline bool sk_stream_memory_free(const struct sock *sk)
1125{
1126 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1127 return false;
1128
1129 return sk->sk_prot->stream_memory_free ?
1130 sk->sk_prot->stream_memory_free(sk) : true;
1131}
1132
1133static inline bool sk_stream_is_writeable(const struct sock *sk)
1134{
1135 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1136 sk_stream_memory_free(sk);
1137}
1138
1139
1140static inline bool sk_has_memory_pressure(const struct sock *sk)
1141{
1142 return sk->sk_prot->memory_pressure != NULL;
1143}
1144
1145static inline bool sk_under_memory_pressure(const struct sock *sk)
1146{
1147 if (!sk->sk_prot->memory_pressure)
1148 return false;
1149
1150 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1151 return !!sk->sk_cgrp->memory_pressure;
1152
1153 return !!*sk->sk_prot->memory_pressure;
1154}
1155
1156static inline void sk_leave_memory_pressure(struct sock *sk)
1157{
1158 int *memory_pressure = sk->sk_prot->memory_pressure;
1159
1160 if (!memory_pressure)
1161 return;
1162
1163 if (*memory_pressure)
1164 *memory_pressure = 0;
1165
1166 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1167 struct cg_proto *cg_proto = sk->sk_cgrp;
1168 struct proto *prot = sk->sk_prot;
1169
1170 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1171 cg_proto->memory_pressure = 0;
1172 }
1173
1174}
1175
1176static inline void sk_enter_memory_pressure(struct sock *sk)
1177{
1178 if (!sk->sk_prot->enter_memory_pressure)
1179 return;
1180
1181 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1182 struct cg_proto *cg_proto = sk->sk_cgrp;
1183 struct proto *prot = sk->sk_prot;
1184
1185 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1186 cg_proto->memory_pressure = 1;
1187 }
1188
1189 sk->sk_prot->enter_memory_pressure(sk);
1190}
1191
1192static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1193{
1194 long *prot = sk->sk_prot->sysctl_mem;
1195 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1196 prot = sk->sk_cgrp->sysctl_mem;
1197 return prot[index];
1198}
1199
1200static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1201 unsigned long amt,
1202 int *parent_status)
1203{
1204 struct res_counter *fail;
1205 int ret;
1206
1207 ret = res_counter_charge_nofail(&prot->memory_allocated,
1208 amt << PAGE_SHIFT, &fail);
1209 if (ret < 0)
1210 *parent_status = OVER_LIMIT;
1211}
1212
1213static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1214 unsigned long amt)
1215{
1216 res_counter_uncharge(&prot->memory_allocated, amt << PAGE_SHIFT);
1217}
1218
1219static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1220{
1221 u64 ret;
1222 ret = res_counter_read_u64(&prot->memory_allocated, RES_USAGE);
1223 return ret >> PAGE_SHIFT;
1224}
1225
1226static inline long
1227sk_memory_allocated(const struct sock *sk)
1228{
1229 struct proto *prot = sk->sk_prot;
1230 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1231 return memcg_memory_allocated_read(sk->sk_cgrp);
1232
1233 return atomic_long_read(prot->memory_allocated);
1234}
1235
1236static inline long
1237sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1238{
1239 struct proto *prot = sk->sk_prot;
1240
1241 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1242 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1243 /* update the root cgroup regardless */
1244 atomic_long_add_return(amt, prot->memory_allocated);
1245 return memcg_memory_allocated_read(sk->sk_cgrp);
1246 }
1247
1248 return atomic_long_add_return(amt, prot->memory_allocated);
1249}
1250
1251static inline void
1252sk_memory_allocated_sub(struct sock *sk, int amt)
1253{
1254 struct proto *prot = sk->sk_prot;
1255
1256 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1257 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1258
1259 atomic_long_sub(amt, prot->memory_allocated);
1260}
1261
1262static inline void sk_sockets_allocated_dec(struct sock *sk)
1263{
1264 struct proto *prot = sk->sk_prot;
1265
1266 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1267 struct cg_proto *cg_proto = sk->sk_cgrp;
1268
1269 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1270 percpu_counter_dec(&cg_proto->sockets_allocated);
1271 }
1272
1273 percpu_counter_dec(prot->sockets_allocated);
1274}
1275
1276static inline void sk_sockets_allocated_inc(struct sock *sk)
1277{
1278 struct proto *prot = sk->sk_prot;
1279
1280 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1281 struct cg_proto *cg_proto = sk->sk_cgrp;
1282
1283 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1284 percpu_counter_inc(&cg_proto->sockets_allocated);
1285 }
1286
1287 percpu_counter_inc(prot->sockets_allocated);
1288}
1289
1290static inline int
1291sk_sockets_allocated_read_positive(struct sock *sk)
1292{
1293 struct proto *prot = sk->sk_prot;
1294
1295 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1296 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1297
1298 return percpu_counter_read_positive(prot->sockets_allocated);
1299}
1300
1301static inline int
1302proto_sockets_allocated_sum_positive(struct proto *prot)
1303{
1304 return percpu_counter_sum_positive(prot->sockets_allocated);
1305}
1306
1307static inline long
1308proto_memory_allocated(struct proto *prot)
1309{
1310 return atomic_long_read(prot->memory_allocated);
1311}
1312
1313static inline bool
1314proto_memory_pressure(struct proto *prot)
1315{
1316 if (!prot->memory_pressure)
1317 return false;
1318 return !!*prot->memory_pressure;
1319}
1320
1321
1322#ifdef CONFIG_PROC_FS
1323/* Called with local bh disabled */
1324void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1325int sock_prot_inuse_get(struct net *net, struct proto *proto);
1326#else
1327static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1328 int inc)
1329{
1330}
1331#endif
1332
1333
1334/* With per-bucket locks this operation is not-atomic, so that
1335 * this version is not worse.
1336 */
1337static inline void __sk_prot_rehash(struct sock *sk)
1338{
1339 sk->sk_prot->unhash(sk);
1340 sk->sk_prot->hash(sk);
1341}
1342
1343void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1344
1345/* About 10 seconds */
1346#define SOCK_DESTROY_TIME (10*HZ)
1347
1348/* Sockets 0-1023 can't be bound to unless you are superuser */
1349#define PROT_SOCK 1024
1350
1351#define SHUTDOWN_MASK 3
1352#define RCV_SHUTDOWN 1
1353#define SEND_SHUTDOWN 2
1354
1355#define SOCK_SNDBUF_LOCK 1
1356#define SOCK_RCVBUF_LOCK 2
1357#define SOCK_BINDADDR_LOCK 4
1358#define SOCK_BINDPORT_LOCK 8
1359
1360/* sock_iocb: used to kick off async processing of socket ios */
1361struct sock_iocb {
1362 struct list_head list;
1363
1364 int flags;
1365 int size;
1366 struct socket *sock;
1367 struct sock *sk;
1368 struct scm_cookie *scm;
1369 struct msghdr *msg, async_msg;
1370 struct kiocb *kiocb;
1371};
1372
1373static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1374{
1375 return (struct sock_iocb *)iocb->private;
1376}
1377
1378static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1379{
1380 return si->kiocb;
1381}
1382
1383struct socket_alloc {
1384 struct socket socket;
1385 struct inode vfs_inode;
1386};
1387
1388static inline struct socket *SOCKET_I(struct inode *inode)
1389{
1390 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1391}
1392
1393static inline struct inode *SOCK_INODE(struct socket *socket)
1394{
1395 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1396}
1397
1398/*
1399 * Functions for memory accounting
1400 */
1401int __sk_mem_schedule(struct sock *sk, int size, int kind);
1402void __sk_mem_reclaim(struct sock *sk);
1403
1404#define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1405#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1406#define SK_MEM_SEND 0
1407#define SK_MEM_RECV 1
1408
1409static inline int sk_mem_pages(int amt)
1410{
1411 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1412}
1413
1414static inline bool sk_has_account(struct sock *sk)
1415{
1416 /* return true if protocol supports memory accounting */
1417 return !!sk->sk_prot->memory_allocated;
1418}
1419
1420static inline bool sk_wmem_schedule(struct sock *sk, int size)
1421{
1422 if (!sk_has_account(sk))
1423 return true;
1424 return size <= sk->sk_forward_alloc ||
1425 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1426}
1427
1428static inline bool
1429sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1430{
1431 if (!sk_has_account(sk))
1432 return true;
1433 return size<= sk->sk_forward_alloc ||
1434 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1435 skb_pfmemalloc(skb);
1436}
1437
1438static inline void sk_mem_reclaim(struct sock *sk)
1439{
1440 if (!sk_has_account(sk))
1441 return;
1442 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1443 __sk_mem_reclaim(sk);
1444}
1445
1446static inline void sk_mem_reclaim_partial(struct sock *sk)
1447{
1448 if (!sk_has_account(sk))
1449 return;
1450 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1451 __sk_mem_reclaim(sk);
1452}
1453
1454static inline void sk_mem_charge(struct sock *sk, int size)
1455{
1456 if (!sk_has_account(sk))
1457 return;
1458 sk->sk_forward_alloc -= size;
1459}
1460
1461static inline void sk_mem_uncharge(struct sock *sk, int size)
1462{
1463 if (!sk_has_account(sk))
1464 return;
1465 sk->sk_forward_alloc += size;
1466}
1467
1468static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1469{
1470 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1471 sk->sk_wmem_queued -= skb->truesize;
1472 sk_mem_uncharge(sk, skb->truesize);
1473 __kfree_skb(skb);
1474}
1475
1476/* Used by processes to "lock" a socket state, so that
1477 * interrupts and bottom half handlers won't change it
1478 * from under us. It essentially blocks any incoming
1479 * packets, so that we won't get any new data or any
1480 * packets that change the state of the socket.
1481 *
1482 * While locked, BH processing will add new packets to
1483 * the backlog queue. This queue is processed by the
1484 * owner of the socket lock right before it is released.
1485 *
1486 * Since ~2.3.5 it is also exclusive sleep lock serializing
1487 * accesses from user process context.
1488 */
1489#define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1490
1491static inline void sock_release_ownership(struct sock *sk)
1492{
1493 sk->sk_lock.owned = 0;
1494}
1495
1496/*
1497 * Macro so as to not evaluate some arguments when
1498 * lockdep is not enabled.
1499 *
1500 * Mark both the sk_lock and the sk_lock.slock as a
1501 * per-address-family lock class.
1502 */
1503#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1504do { \
1505 sk->sk_lock.owned = 0; \
1506 init_waitqueue_head(&sk->sk_lock.wq); \
1507 spin_lock_init(&(sk)->sk_lock.slock); \
1508 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1509 sizeof((sk)->sk_lock)); \
1510 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1511 (skey), (sname)); \
1512 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1513} while (0)
1514
1515void lock_sock_nested(struct sock *sk, int subclass);
1516
1517static inline void lock_sock(struct sock *sk)
1518{
1519 lock_sock_nested(sk, 0);
1520}
1521
1522void release_sock(struct sock *sk);
1523
1524/* BH context may only use the following locking interface. */
1525#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1526#define bh_lock_sock_nested(__sk) \
1527 spin_lock_nested(&((__sk)->sk_lock.slock), \
1528 SINGLE_DEPTH_NESTING)
1529#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1530
1531bool lock_sock_fast(struct sock *sk);
1532/**
1533 * unlock_sock_fast - complement of lock_sock_fast
1534 * @sk: socket
1535 * @slow: slow mode
1536 *
1537 * fast unlock socket for user context.
1538 * If slow mode is on, we call regular release_sock()
1539 */
1540static inline void unlock_sock_fast(struct sock *sk, bool slow)
1541{
1542 if (slow)
1543 release_sock(sk);
1544 else
1545 spin_unlock_bh(&sk->sk_lock.slock);
1546}
1547
1548
1549struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1550 struct proto *prot);
1551void sk_free(struct sock *sk);
1552void sk_release_kernel(struct sock *sk);
1553struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1554
1555struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1556 gfp_t priority);
1557void sock_wfree(struct sk_buff *skb);
1558void skb_orphan_partial(struct sk_buff *skb);
1559void sock_rfree(struct sk_buff *skb);
1560void sock_edemux(struct sk_buff *skb);
1561
1562int sock_setsockopt(struct socket *sock, int level, int op,
1563 char __user *optval, unsigned int optlen);
1564
1565int sock_getsockopt(struct socket *sock, int level, int op,
1566 char __user *optval, int __user *optlen);
1567struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1568 int noblock, int *errcode);
1569struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1570 unsigned long data_len, int noblock,
1571 int *errcode, int max_page_order);
1572void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1573void sock_kfree_s(struct sock *sk, void *mem, int size);
1574void sk_send_sigurg(struct sock *sk);
1575
1576/*
1577 * Functions to fill in entries in struct proto_ops when a protocol
1578 * does not implement a particular function.
1579 */
1580int sock_no_bind(struct socket *, struct sockaddr *, int);
1581int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1582int sock_no_socketpair(struct socket *, struct socket *);
1583int sock_no_accept(struct socket *, struct socket *, int);
1584int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1585unsigned int sock_no_poll(struct file *, struct socket *,
1586 struct poll_table_struct *);
1587int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1588int sock_no_listen(struct socket *, int);
1589int sock_no_shutdown(struct socket *, int);
1590int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1591int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1592int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t);
1593int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t,
1594 int);
1595int sock_no_mmap(struct file *file, struct socket *sock,
1596 struct vm_area_struct *vma);
1597ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1598 size_t size, int flags);
1599
1600/*
1601 * Functions to fill in entries in struct proto_ops when a protocol
1602 * uses the inet style.
1603 */
1604int sock_common_getsockopt(struct socket *sock, int level, int optname,
1605 char __user *optval, int __user *optlen);
1606int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1607 struct msghdr *msg, size_t size, int flags);
1608int sock_common_setsockopt(struct socket *sock, int level, int optname,
1609 char __user *optval, unsigned int optlen);
1610int compat_sock_common_getsockopt(struct socket *sock, int level,
1611 int optname, char __user *optval, int __user *optlen);
1612int compat_sock_common_setsockopt(struct socket *sock, int level,
1613 int optname, char __user *optval, unsigned int optlen);
1614
1615void sk_common_release(struct sock *sk);
1616
1617/*
1618 * Default socket callbacks and setup code
1619 */
1620
1621/* Initialise core socket variables */
1622void sock_init_data(struct socket *sock, struct sock *sk);
1623
1624/*
1625 * Socket reference counting postulates.
1626 *
1627 * * Each user of socket SHOULD hold a reference count.
1628 * * Each access point to socket (an hash table bucket, reference from a list,
1629 * running timer, skb in flight MUST hold a reference count.
1630 * * When reference count hits 0, it means it will never increase back.
1631 * * When reference count hits 0, it means that no references from
1632 * outside exist to this socket and current process on current CPU
1633 * is last user and may/should destroy this socket.
1634 * * sk_free is called from any context: process, BH, IRQ. When
1635 * it is called, socket has no references from outside -> sk_free
1636 * may release descendant resources allocated by the socket, but
1637 * to the time when it is called, socket is NOT referenced by any
1638 * hash tables, lists etc.
1639 * * Packets, delivered from outside (from network or from another process)
1640 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1641 * when they sit in queue. Otherwise, packets will leak to hole, when
1642 * socket is looked up by one cpu and unhasing is made by another CPU.
1643 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1644 * (leak to backlog). Packet socket does all the processing inside
1645 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1646 * use separate SMP lock, so that they are prone too.
1647 */
1648
1649/* Ungrab socket and destroy it, if it was the last reference. */
1650static inline void sock_put(struct sock *sk)
1651{
1652 if (atomic_dec_and_test(&sk->sk_refcnt))
1653 sk_free(sk);
1654}
1655/* Generic version of sock_put(), dealing with all sockets
1656 * (TCP_TIMEWAIT, ESTABLISHED...)
1657 */
1658void sock_gen_put(struct sock *sk);
1659
1660int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1661
1662static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1663{
1664 sk->sk_tx_queue_mapping = tx_queue;
1665}
1666
1667static inline void sk_tx_queue_clear(struct sock *sk)
1668{
1669 sk->sk_tx_queue_mapping = -1;
1670}
1671
1672static inline int sk_tx_queue_get(const struct sock *sk)
1673{
1674 return sk ? sk->sk_tx_queue_mapping : -1;
1675}
1676
1677static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1678{
1679 sk_tx_queue_clear(sk);
1680 sk->sk_socket = sock;
1681}
1682
1683static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1684{
1685 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1686 return &rcu_dereference_raw(sk->sk_wq)->wait;
1687}
1688/* Detach socket from process context.
1689 * Announce socket dead, detach it from wait queue and inode.
1690 * Note that parent inode held reference count on this struct sock,
1691 * we do not release it in this function, because protocol
1692 * probably wants some additional cleanups or even continuing
1693 * to work with this socket (TCP).
1694 */
1695static inline void sock_orphan(struct sock *sk)
1696{
1697 write_lock_bh(&sk->sk_callback_lock);
1698 sock_set_flag(sk, SOCK_DEAD);
1699 sk_set_socket(sk, NULL);
1700 sk->sk_wq = NULL;
1701 write_unlock_bh(&sk->sk_callback_lock);
1702}
1703
1704static inline void sock_graft(struct sock *sk, struct socket *parent)
1705{
1706 write_lock_bh(&sk->sk_callback_lock);
1707 sk->sk_wq = parent->wq;
1708 parent->sk = sk;
1709 sk_set_socket(sk, parent);
1710 security_sock_graft(sk, parent);
1711 write_unlock_bh(&sk->sk_callback_lock);
1712}
1713
1714kuid_t sock_i_uid(struct sock *sk);
1715unsigned long sock_i_ino(struct sock *sk);
1716
1717static inline struct dst_entry *
1718__sk_dst_get(struct sock *sk)
1719{
1720 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1721 lockdep_is_held(&sk->sk_lock.slock));
1722}
1723
1724static inline struct dst_entry *
1725sk_dst_get(struct sock *sk)
1726{
1727 struct dst_entry *dst;
1728
1729 rcu_read_lock();
1730 dst = rcu_dereference(sk->sk_dst_cache);
1731 if (dst)
1732 dst_hold(dst);
1733 rcu_read_unlock();
1734 return dst;
1735}
1736
1737static inline void dst_negative_advice(struct sock *sk)
1738{
1739 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1740
1741 if (dst && dst->ops->negative_advice) {
1742 ndst = dst->ops->negative_advice(dst);
1743
1744 if (ndst != dst) {
1745 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1746 sk_tx_queue_clear(sk);
1747 }
1748 }
1749}
1750
1751static inline void
1752__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1753{
1754 struct dst_entry *old_dst;
1755
1756 sk_tx_queue_clear(sk);
1757 /*
1758 * This can be called while sk is owned by the caller only,
1759 * with no state that can be checked in a rcu_dereference_check() cond
1760 */
1761 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1762 rcu_assign_pointer(sk->sk_dst_cache, dst);
1763 dst_release(old_dst);
1764}
1765
1766static inline void
1767sk_dst_set(struct sock *sk, struct dst_entry *dst)
1768{
1769 spin_lock(&sk->sk_dst_lock);
1770 __sk_dst_set(sk, dst);
1771 spin_unlock(&sk->sk_dst_lock);
1772}
1773
1774static inline void
1775__sk_dst_reset(struct sock *sk)
1776{
1777 __sk_dst_set(sk, NULL);
1778}
1779
1780static inline void
1781sk_dst_reset(struct sock *sk)
1782{
1783 spin_lock(&sk->sk_dst_lock);
1784 __sk_dst_reset(sk);
1785 spin_unlock(&sk->sk_dst_lock);
1786}
1787
1788struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1789
1790struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1791
1792static inline bool sk_can_gso(const struct sock *sk)
1793{
1794 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1795}
1796
1797void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1798
1799static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1800{
1801 sk->sk_route_nocaps |= flags;
1802 sk->sk_route_caps &= ~flags;
1803}
1804
1805static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1806 char __user *from, char *to,
1807 int copy, int offset)
1808{
1809 if (skb->ip_summed == CHECKSUM_NONE) {
1810 int err = 0;
1811 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1812 if (err)
1813 return err;
1814 skb->csum = csum_block_add(skb->csum, csum, offset);
1815 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1816 if (!access_ok(VERIFY_READ, from, copy) ||
1817 __copy_from_user_nocache(to, from, copy))
1818 return -EFAULT;
1819 } else if (copy_from_user(to, from, copy))
1820 return -EFAULT;
1821
1822 return 0;
1823}
1824
1825static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1826 char __user *from, int copy)
1827{
1828 int err, offset = skb->len;
1829
1830 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1831 copy, offset);
1832 if (err)
1833 __skb_trim(skb, offset);
1834
1835 return err;
1836}
1837
1838static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1839 struct sk_buff *skb,
1840 struct page *page,
1841 int off, int copy)
1842{
1843 int err;
1844
1845 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1846 copy, skb->len);
1847 if (err)
1848 return err;
1849
1850 skb->len += copy;
1851 skb->data_len += copy;
1852 skb->truesize += copy;
1853 sk->sk_wmem_queued += copy;
1854 sk_mem_charge(sk, copy);
1855 return 0;
1856}
1857
1858static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1859 struct sk_buff *skb, struct page *page,
1860 int off, int copy)
1861{
1862 if (skb->ip_summed == CHECKSUM_NONE) {
1863 int err = 0;
1864 __wsum csum = csum_and_copy_from_user(from,
1865 page_address(page) + off,
1866 copy, 0, &err);
1867 if (err)
1868 return err;
1869 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1870 } else if (copy_from_user(page_address(page) + off, from, copy))
1871 return -EFAULT;
1872
1873 skb->len += copy;
1874 skb->data_len += copy;
1875 skb->truesize += copy;
1876 sk->sk_wmem_queued += copy;
1877 sk_mem_charge(sk, copy);
1878 return 0;
1879}
1880
1881/**
1882 * sk_wmem_alloc_get - returns write allocations
1883 * @sk: socket
1884 *
1885 * Returns sk_wmem_alloc minus initial offset of one
1886 */
1887static inline int sk_wmem_alloc_get(const struct sock *sk)
1888{
1889 return atomic_read(&sk->sk_wmem_alloc) - 1;
1890}
1891
1892/**
1893 * sk_rmem_alloc_get - returns read allocations
1894 * @sk: socket
1895 *
1896 * Returns sk_rmem_alloc
1897 */
1898static inline int sk_rmem_alloc_get(const struct sock *sk)
1899{
1900 return atomic_read(&sk->sk_rmem_alloc);
1901}
1902
1903/**
1904 * sk_has_allocations - check if allocations are outstanding
1905 * @sk: socket
1906 *
1907 * Returns true if socket has write or read allocations
1908 */
1909static inline bool sk_has_allocations(const struct sock *sk)
1910{
1911 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1912}
1913
1914/**
1915 * wq_has_sleeper - check if there are any waiting processes
1916 * @wq: struct socket_wq
1917 *
1918 * Returns true if socket_wq has waiting processes
1919 *
1920 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1921 * barrier call. They were added due to the race found within the tcp code.
1922 *
1923 * Consider following tcp code paths:
1924 *
1925 * CPU1 CPU2
1926 *
1927 * sys_select receive packet
1928 * ... ...
1929 * __add_wait_queue update tp->rcv_nxt
1930 * ... ...
1931 * tp->rcv_nxt check sock_def_readable
1932 * ... {
1933 * schedule rcu_read_lock();
1934 * wq = rcu_dereference(sk->sk_wq);
1935 * if (wq && waitqueue_active(&wq->wait))
1936 * wake_up_interruptible(&wq->wait)
1937 * ...
1938 * }
1939 *
1940 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1941 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1942 * could then endup calling schedule and sleep forever if there are no more
1943 * data on the socket.
1944 *
1945 */
1946static inline bool wq_has_sleeper(struct socket_wq *wq)
1947{
1948 /* We need to be sure we are in sync with the
1949 * add_wait_queue modifications to the wait queue.
1950 *
1951 * This memory barrier is paired in the sock_poll_wait.
1952 */
1953 smp_mb();
1954 return wq && waitqueue_active(&wq->wait);
1955}
1956
1957/**
1958 * sock_poll_wait - place memory barrier behind the poll_wait call.
1959 * @filp: file
1960 * @wait_address: socket wait queue
1961 * @p: poll_table
1962 *
1963 * See the comments in the wq_has_sleeper function.
1964 */
1965static inline void sock_poll_wait(struct file *filp,
1966 wait_queue_head_t *wait_address, poll_table *p)
1967{
1968 if (!poll_does_not_wait(p) && wait_address) {
1969 poll_wait(filp, wait_address, p);
1970 /* We need to be sure we are in sync with the
1971 * socket flags modification.
1972 *
1973 * This memory barrier is paired in the wq_has_sleeper.
1974 */
1975 smp_mb();
1976 }
1977}
1978
1979/*
1980 * Queue a received datagram if it will fit. Stream and sequenced
1981 * protocols can't normally use this as they need to fit buffers in
1982 * and play with them.
1983 *
1984 * Inlined as it's very short and called for pretty much every
1985 * packet ever received.
1986 */
1987
1988static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1989{
1990 skb_orphan(skb);
1991 skb->sk = sk;
1992 skb->destructor = sock_wfree;
1993 /*
1994 * We used to take a refcount on sk, but following operation
1995 * is enough to guarantee sk_free() wont free this sock until
1996 * all in-flight packets are completed
1997 */
1998 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1999}
2000
2001static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2002{
2003 skb_orphan(skb);
2004 skb->sk = sk;
2005 skb->destructor = sock_rfree;
2006 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2007 sk_mem_charge(sk, skb->truesize);
2008}
2009
2010void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2011 unsigned long expires);
2012
2013void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2014
2015int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2016
2017int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2018
2019/*
2020 * Recover an error report and clear atomically
2021 */
2022
2023static inline int sock_error(struct sock *sk)
2024{
2025 int err;
2026 if (likely(!sk->sk_err))
2027 return 0;
2028 err = xchg(&sk->sk_err, 0);
2029 return -err;
2030}
2031
2032static inline unsigned long sock_wspace(struct sock *sk)
2033{
2034 int amt = 0;
2035
2036 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2037 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2038 if (amt < 0)
2039 amt = 0;
2040 }
2041 return amt;
2042}
2043
2044static inline void sk_wake_async(struct sock *sk, int how, int band)
2045{
2046 if (sock_flag(sk, SOCK_FASYNC))
2047 sock_wake_async(sk->sk_socket, how, band);
2048}
2049
2050/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2051 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2052 * Note: for send buffers, TCP works better if we can build two skbs at
2053 * minimum.
2054 */
2055#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2056
2057#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2058#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2059
2060static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2061{
2062 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2063 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2064 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2065 }
2066}
2067
2068struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2069
2070/**
2071 * sk_page_frag - return an appropriate page_frag
2072 * @sk: socket
2073 *
2074 * If socket allocation mode allows current thread to sleep, it means its
2075 * safe to use the per task page_frag instead of the per socket one.
2076 */
2077static inline struct page_frag *sk_page_frag(struct sock *sk)
2078{
2079 if (sk->sk_allocation & __GFP_WAIT)
2080 return ¤t->task_frag;
2081
2082 return &sk->sk_frag;
2083}
2084
2085bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2086
2087/*
2088 * Default write policy as shown to user space via poll/select/SIGIO
2089 */
2090static inline bool sock_writeable(const struct sock *sk)
2091{
2092 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2093}
2094
2095static inline gfp_t gfp_any(void)
2096{
2097 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2098}
2099
2100static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2101{
2102 return noblock ? 0 : sk->sk_rcvtimeo;
2103}
2104
2105static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2106{
2107 return noblock ? 0 : sk->sk_sndtimeo;
2108}
2109
2110static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2111{
2112 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2113}
2114
2115/* Alas, with timeout socket operations are not restartable.
2116 * Compare this to poll().
2117 */
2118static inline int sock_intr_errno(long timeo)
2119{
2120 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2121}
2122
2123void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2124 struct sk_buff *skb);
2125void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2126 struct sk_buff *skb);
2127
2128static inline void
2129sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2130{
2131 ktime_t kt = skb->tstamp;
2132 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2133
2134 /*
2135 * generate control messages if
2136 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2137 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2138 * - software time stamp available and wanted
2139 * (SOCK_TIMESTAMPING_SOFTWARE)
2140 * - hardware time stamps available and wanted
2141 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2142 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2143 */
2144 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2145 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2146 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2147 (hwtstamps->hwtstamp.tv64 &&
2148 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2149 (hwtstamps->syststamp.tv64 &&
2150 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2151 __sock_recv_timestamp(msg, sk, skb);
2152 else
2153 sk->sk_stamp = kt;
2154
2155 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2156 __sock_recv_wifi_status(msg, sk, skb);
2157}
2158
2159void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2160 struct sk_buff *skb);
2161
2162static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2163 struct sk_buff *skb)
2164{
2165#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2166 (1UL << SOCK_RCVTSTAMP) | \
2167 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2168 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2169 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2170
2171 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2172 __sock_recv_ts_and_drops(msg, sk, skb);
2173 else
2174 sk->sk_stamp = skb->tstamp;
2175}
2176
2177/**
2178 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2179 * @sk: socket sending this packet
2180 * @tx_flags: filled with instructions for time stamping
2181 *
2182 * Currently only depends on SOCK_TIMESTAMPING* flags.
2183 */
2184void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2185
2186/**
2187 * sk_eat_skb - Release a skb if it is no longer needed
2188 * @sk: socket to eat this skb from
2189 * @skb: socket buffer to eat
2190 * @copied_early: flag indicating whether DMA operations copied this data early
2191 *
2192 * This routine must be called with interrupts disabled or with the socket
2193 * locked so that the sk_buff queue operation is ok.
2194*/
2195#ifdef CONFIG_NET_DMA
2196static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2197{
2198 __skb_unlink(skb, &sk->sk_receive_queue);
2199 if (!copied_early)
2200 __kfree_skb(skb);
2201 else
2202 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2203}
2204#else
2205static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2206{
2207 __skb_unlink(skb, &sk->sk_receive_queue);
2208 __kfree_skb(skb);
2209}
2210#endif
2211
2212static inline
2213struct net *sock_net(const struct sock *sk)
2214{
2215 return read_pnet(&sk->sk_net);
2216}
2217
2218static inline
2219void sock_net_set(struct sock *sk, struct net *net)
2220{
2221 write_pnet(&sk->sk_net, net);
2222}
2223
2224/*
2225 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2226 * They should not hold a reference to a namespace in order to allow
2227 * to stop it.
2228 * Sockets after sk_change_net should be released using sk_release_kernel
2229 */
2230static inline void sk_change_net(struct sock *sk, struct net *net)
2231{
2232 struct net *current_net = sock_net(sk);
2233
2234 if (!net_eq(current_net, net)) {
2235 put_net(current_net);
2236 sock_net_set(sk, hold_net(net));
2237 }
2238}
2239
2240static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2241{
2242 if (skb->sk) {
2243 struct sock *sk = skb->sk;
2244
2245 skb->destructor = NULL;
2246 skb->sk = NULL;
2247 return sk;
2248 }
2249 return NULL;
2250}
2251
2252void sock_enable_timestamp(struct sock *sk, int flag);
2253int sock_get_timestamp(struct sock *, struct timeval __user *);
2254int sock_get_timestampns(struct sock *, struct timespec __user *);
2255int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2256 int type);
2257
2258bool sk_ns_capable(const struct sock *sk,
2259 struct user_namespace *user_ns, int cap);
2260bool sk_capable(const struct sock *sk, int cap);
2261bool sk_net_capable(const struct sock *sk, int cap);
2262
2263/*
2264 * Enable debug/info messages
2265 */
2266extern int net_msg_warn;
2267#define NETDEBUG(fmt, args...) \
2268 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2269
2270#define LIMIT_NETDEBUG(fmt, args...) \
2271 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2272
2273extern __u32 sysctl_wmem_max;
2274extern __u32 sysctl_rmem_max;
2275
2276extern int sysctl_optmem_max;
2277
2278extern __u32 sysctl_wmem_default;
2279extern __u32 sysctl_rmem_default;
2280
2281#endif /* _SOCK_H */