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 #include <linux/android_kabi.h>
72 #include <linux/android_vendor.h>
73
74 /*
75 * This structure really needs to be cleaned up.
76 * Most of it is for TCP, and not used by any of
77 * the other protocols.
78 */
79
80 /* Define this to get the SOCK_DBG debugging facility. */
81 #define SOCK_DEBUGGING
82 #ifdef SOCK_DEBUGGING
83 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
84 printk(KERN_DEBUG msg); } while (0)
85 #else
86 /* Validate arguments and do nothing */
87 static inline __printf(2, 3)
SOCK_DEBUG(const struct sock * sk,const char * msg,...)88 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
89 {
90 }
91 #endif
92
93 /* This is the per-socket lock. The spinlock provides a synchronization
94 * between user contexts and software interrupt processing, whereas the
95 * mini-semaphore synchronizes multiple users amongst themselves.
96 */
97 typedef struct {
98 spinlock_t slock;
99 int owned;
100 wait_queue_head_t wq;
101 /*
102 * We express the mutex-alike socket_lock semantics
103 * to the lock validator by explicitly managing
104 * the slock as a lock variant (in addition to
105 * the slock itself):
106 */
107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
108 struct lockdep_map dep_map;
109 #endif
110 } socket_lock_t;
111
112 struct sock;
113 struct proto;
114 struct net;
115
116 typedef __u32 __bitwise __portpair;
117 typedef __u64 __bitwise __addrpair;
118
119 /**
120 * struct sock_common - minimal network layer representation of sockets
121 * @skc_daddr: Foreign IPv4 addr
122 * @skc_rcv_saddr: Bound local IPv4 addr
123 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
124 * @skc_hash: hash value used with various protocol lookup tables
125 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
126 * @skc_dport: placeholder for inet_dport/tw_dport
127 * @skc_num: placeholder for inet_num/tw_num
128 * @skc_portpair: __u32 union of @skc_dport & @skc_num
129 * @skc_family: network address family
130 * @skc_state: Connection state
131 * @skc_reuse: %SO_REUSEADDR setting
132 * @skc_reuseport: %SO_REUSEPORT setting
133 * @skc_ipv6only: socket is IPV6 only
134 * @skc_net_refcnt: socket is using net ref counting
135 * @skc_bound_dev_if: bound device index if != 0
136 * @skc_bind_node: bind hash linkage for various protocol lookup tables
137 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
138 * @skc_prot: protocol handlers inside a network family
139 * @skc_net: reference to the network namespace of this socket
140 * @skc_v6_daddr: IPV6 destination address
141 * @skc_v6_rcv_saddr: IPV6 source address
142 * @skc_cookie: socket's cookie value
143 * @skc_node: main hash linkage for various protocol lookup tables
144 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
145 * @skc_tx_queue_mapping: tx queue number for this connection
146 * @skc_rx_queue_mapping: rx queue number for this connection
147 * @skc_flags: place holder for sk_flags
148 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
149 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
150 * @skc_listener: connection request listener socket (aka rsk_listener)
151 * [union with @skc_flags]
152 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
153 * [union with @skc_flags]
154 * @skc_incoming_cpu: record/match cpu processing incoming packets
155 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
156 * [union with @skc_incoming_cpu]
157 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
158 * [union with @skc_incoming_cpu]
159 * @skc_refcnt: reference count
160 *
161 * This is the minimal network layer representation of sockets, the header
162 * for struct sock and struct inet_timewait_sock.
163 */
164 struct sock_common {
165 union {
166 __addrpair skc_addrpair;
167 struct {
168 __be32 skc_daddr;
169 __be32 skc_rcv_saddr;
170 };
171 };
172 union {
173 unsigned int skc_hash;
174 __u16 skc_u16hashes[2];
175 };
176 /* skc_dport && skc_num must be grouped as well */
177 union {
178 __portpair skc_portpair;
179 struct {
180 __be16 skc_dport;
181 __u16 skc_num;
182 };
183 };
184
185 unsigned short skc_family;
186 volatile unsigned char skc_state;
187 unsigned char skc_reuse:4;
188 unsigned char skc_reuseport:1;
189 unsigned char skc_ipv6only:1;
190 unsigned char skc_net_refcnt:1;
191 int skc_bound_dev_if;
192 union {
193 struct hlist_node skc_bind_node;
194 struct hlist_node skc_portaddr_node;
195 };
196 struct proto *skc_prot;
197 possible_net_t skc_net;
198
199 #if IS_ENABLED(CONFIG_IPV6)
200 struct in6_addr skc_v6_daddr;
201 struct in6_addr skc_v6_rcv_saddr;
202 #endif
203
204 atomic64_t skc_cookie;
205
206 /* following fields are padding to force
207 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
208 * assuming IPV6 is enabled. We use this padding differently
209 * for different kind of 'sockets'
210 */
211 union {
212 unsigned long skc_flags;
213 struct sock *skc_listener; /* request_sock */
214 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
215 };
216 /*
217 * fields between dontcopy_begin/dontcopy_end
218 * are not copied in sock_copy()
219 */
220 /* private: */
221 int skc_dontcopy_begin[0];
222 /* public: */
223 union {
224 struct hlist_node skc_node;
225 struct hlist_nulls_node skc_nulls_node;
226 };
227 unsigned short skc_tx_queue_mapping;
228 #ifdef CONFIG_XPS
229 unsigned short skc_rx_queue_mapping;
230 #endif
231 union {
232 int skc_incoming_cpu;
233 u32 skc_rcv_wnd;
234 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
235 };
236
237 refcount_t skc_refcnt;
238 /* private: */
239 int skc_dontcopy_end[0];
240 union {
241 u32 skc_rxhash;
242 u32 skc_window_clamp;
243 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
244 };
245 /* public: */
246 };
247
248 struct bpf_local_storage;
249
250 /**
251 * struct sock - network layer representation of sockets
252 * @__sk_common: shared layout with inet_timewait_sock
253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
255 * @sk_lock: synchronizer
256 * @sk_kern_sock: True if sock is using kernel lock classes
257 * @sk_rcvbuf: size of receive buffer in bytes
258 * @sk_wq: sock wait queue and async head
259 * @sk_rx_dst: receive input route used by early demux
260 * @sk_dst_cache: destination cache
261 * @sk_dst_pending_confirm: need to confirm neighbour
262 * @sk_policy: flow policy
263 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
264 * @sk_receive_queue: incoming packets
265 * @sk_wmem_alloc: transmit queue bytes committed
266 * @sk_tsq_flags: TCP Small Queues flags
267 * @sk_write_queue: Packet sending queue
268 * @sk_omem_alloc: "o" is "option" or "other"
269 * @sk_wmem_queued: persistent queue size
270 * @sk_forward_alloc: space allocated forward
271 * @sk_napi_id: id of the last napi context to receive data for sk
272 * @sk_ll_usec: usecs to busypoll when there is no data
273 * @sk_allocation: allocation mode
274 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
275 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
276 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
277 * @sk_sndbuf: size of send buffer in bytes
278 * @__sk_flags_offset: empty field used to determine location of bitfield
279 * @sk_padding: unused element for alignment
280 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
281 * @sk_no_check_rx: allow zero checksum in RX packets
282 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
283 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
284 * @sk_route_forced_caps: static, forced route capabilities
285 * (set in tcp_init_sock())
286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
287 * @sk_gso_max_size: Maximum GSO segment size to build
288 * @sk_gso_max_segs: Maximum number of GSO segments
289 * @sk_pacing_shift: scaling factor for TCP Small Queues
290 * @sk_lingertime: %SO_LINGER l_linger setting
291 * @sk_backlog: always used with the per-socket spinlock held
292 * @sk_callback_lock: used with the callbacks in the end of this struct
293 * @sk_error_queue: rarely used
294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
295 * IPV6_ADDRFORM for instance)
296 * @sk_err: last error
297 * @sk_err_soft: errors that don't cause failure but are the cause of a
298 * persistent failure not just 'timed out'
299 * @sk_drops: raw/udp drops counter
300 * @sk_ack_backlog: current listen backlog
301 * @sk_max_ack_backlog: listen backlog set in listen()
302 * @sk_uid: user id of owner
303 * @sk_priority: %SO_PRIORITY setting
304 * @sk_type: socket type (%SOCK_STREAM, etc)
305 * @sk_protocol: which protocol this socket belongs in this network family
306 * @sk_peer_pid: &struct pid for this socket's peer
307 * @sk_peer_cred: %SO_PEERCRED setting
308 * @sk_rcvlowat: %SO_RCVLOWAT setting
309 * @sk_rcvtimeo: %SO_RCVTIMEO setting
310 * @sk_sndtimeo: %SO_SNDTIMEO setting
311 * @sk_txhash: computed flow hash for use on transmit
312 * @sk_filter: socket filtering instructions
313 * @sk_timer: sock cleanup timer
314 * @sk_stamp: time stamp of last packet received
315 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
316 * @sk_tsflags: SO_TIMESTAMPING socket options
317 * @sk_tskey: counter to disambiguate concurrent tstamp requests
318 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
319 * @sk_socket: Identd and reporting IO signals
320 * @sk_user_data: RPC layer private data
321 * @sk_frag: cached page frag
322 * @sk_peek_off: current peek_offset value
323 * @sk_send_head: front of stuff to transmit
324 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
325 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
326 * @sk_security: used by security modules
327 * @sk_mark: generic packet mark
328 * @sk_cgrp_data: cgroup data for this cgroup
329 * @sk_memcg: this socket's memory cgroup association
330 * @sk_write_pending: a write to stream socket waits to start
331 * @sk_state_change: callback to indicate change in the state of the sock
332 * @sk_data_ready: callback to indicate there is data to be processed
333 * @sk_write_space: callback to indicate there is bf sending space available
334 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
335 * @sk_backlog_rcv: callback to process the backlog
336 * @sk_validate_xmit_skb: ptr to an optional validate function
337 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
338 * @sk_reuseport_cb: reuseport group container
339 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
340 * @sk_rcu: used during RCU grace period
341 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
342 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
343 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
344 * @sk_txtime_unused: unused txtime flags
345 */
346 struct sock {
347 /*
348 * Now struct inet_timewait_sock also uses sock_common, so please just
349 * don't add nothing before this first member (__sk_common) --acme
350 */
351 struct sock_common __sk_common;
352 #define sk_node __sk_common.skc_node
353 #define sk_nulls_node __sk_common.skc_nulls_node
354 #define sk_refcnt __sk_common.skc_refcnt
355 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
356 #ifdef CONFIG_XPS
357 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
358 #endif
359
360 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
361 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
362 #define sk_hash __sk_common.skc_hash
363 #define sk_portpair __sk_common.skc_portpair
364 #define sk_num __sk_common.skc_num
365 #define sk_dport __sk_common.skc_dport
366 #define sk_addrpair __sk_common.skc_addrpair
367 #define sk_daddr __sk_common.skc_daddr
368 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
369 #define sk_family __sk_common.skc_family
370 #define sk_state __sk_common.skc_state
371 #define sk_reuse __sk_common.skc_reuse
372 #define sk_reuseport __sk_common.skc_reuseport
373 #define sk_ipv6only __sk_common.skc_ipv6only
374 #define sk_net_refcnt __sk_common.skc_net_refcnt
375 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
376 #define sk_bind_node __sk_common.skc_bind_node
377 #define sk_prot __sk_common.skc_prot
378 #define sk_net __sk_common.skc_net
379 #define sk_v6_daddr __sk_common.skc_v6_daddr
380 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
381 #define sk_cookie __sk_common.skc_cookie
382 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
383 #define sk_flags __sk_common.skc_flags
384 #define sk_rxhash __sk_common.skc_rxhash
385
386 socket_lock_t sk_lock;
387 atomic_t sk_drops;
388 int sk_rcvlowat;
389 struct sk_buff_head sk_error_queue;
390 struct sk_buff *sk_rx_skb_cache;
391 struct sk_buff_head sk_receive_queue;
392 /*
393 * The backlog queue is special, it is always used with
394 * the per-socket spinlock held and requires low latency
395 * access. Therefore we special case it's implementation.
396 * Note : rmem_alloc is in this structure to fill a hole
397 * on 64bit arches, not because its logically part of
398 * backlog.
399 */
400 struct {
401 atomic_t rmem_alloc;
402 int len;
403 struct sk_buff *head;
404 struct sk_buff *tail;
405 } sk_backlog;
406 #define sk_rmem_alloc sk_backlog.rmem_alloc
407
408 int sk_forward_alloc;
409 #ifdef CONFIG_NET_RX_BUSY_POLL
410 unsigned int sk_ll_usec;
411 /* ===== mostly read cache line ===== */
412 unsigned int sk_napi_id;
413 #endif
414 int sk_rcvbuf;
415
416 struct sk_filter __rcu *sk_filter;
417 union {
418 struct socket_wq __rcu *sk_wq;
419 /* private: */
420 struct socket_wq *sk_wq_raw;
421 /* public: */
422 };
423 #ifdef CONFIG_XFRM
424 struct xfrm_policy __rcu *sk_policy[2];
425 #endif
426 struct dst_entry __rcu *sk_rx_dst;
427 struct dst_entry __rcu *sk_dst_cache;
428 atomic_t sk_omem_alloc;
429 int sk_sndbuf;
430
431 /* ===== cache line for TX ===== */
432 int sk_wmem_queued;
433 refcount_t sk_wmem_alloc;
434 unsigned long sk_tsq_flags;
435 union {
436 struct sk_buff *sk_send_head;
437 struct rb_root tcp_rtx_queue;
438 };
439 struct sk_buff *sk_tx_skb_cache;
440 struct sk_buff_head sk_write_queue;
441 __s32 sk_peek_off;
442 int sk_write_pending;
443 __u32 sk_dst_pending_confirm;
444 u32 sk_pacing_status; /* see enum sk_pacing */
445 long sk_sndtimeo;
446 struct timer_list sk_timer;
447 __u32 sk_priority;
448 __u32 sk_mark;
449 unsigned long sk_pacing_rate; /* bytes per second */
450 unsigned long sk_max_pacing_rate;
451 struct page_frag sk_frag;
452 netdev_features_t sk_route_caps;
453 netdev_features_t sk_route_nocaps;
454 netdev_features_t sk_route_forced_caps;
455 int sk_gso_type;
456 unsigned int sk_gso_max_size;
457 gfp_t sk_allocation;
458 __u32 sk_txhash;
459
460 /*
461 * Because of non atomicity rules, all
462 * changes are protected by socket lock.
463 */
464 u8 sk_padding : 1,
465 sk_kern_sock : 1,
466 sk_no_check_tx : 1,
467 sk_no_check_rx : 1,
468 sk_userlocks : 4;
469 u8 sk_pacing_shift;
470 u16 sk_type;
471 u16 sk_protocol;
472 u16 sk_gso_max_segs;
473 unsigned long sk_lingertime;
474 struct proto *sk_prot_creator;
475 rwlock_t sk_callback_lock;
476 int sk_err,
477 sk_err_soft;
478 u32 sk_ack_backlog;
479 u32 sk_max_ack_backlog;
480 kuid_t sk_uid;
481 #if IS_ENABLED(CONFIG_DEBUG_SPINLOCK) || IS_ENABLED(CONFIG_DEBUG_LOCK_ALLOC)
482 spinlock_t sk_peer_lock;
483 #else
484 /* sk_peer_lock is in the ANDROID_KABI_RESERVE(1) field below */
485 #endif
486 struct pid *sk_peer_pid;
487 const struct cred *sk_peer_cred;
488
489 long sk_rcvtimeo;
490 ktime_t sk_stamp;
491 #if BITS_PER_LONG==32
492 seqlock_t sk_stamp_seq;
493 #endif
494 u16 sk_tsflags;
495 u8 sk_shutdown;
496 u32 sk_tskey;
497 atomic_t sk_zckey;
498
499 u8 sk_clockid;
500 u8 sk_txtime_deadline_mode : 1,
501 sk_txtime_report_errors : 1,
502 sk_txtime_unused : 6;
503
504 struct socket *sk_socket;
505 void *sk_user_data;
506 #ifdef CONFIG_SECURITY
507 void *sk_security;
508 #endif
509 struct sock_cgroup_data sk_cgrp_data;
510 struct mem_cgroup *sk_memcg;
511 void (*sk_state_change)(struct sock *sk);
512 void (*sk_data_ready)(struct sock *sk);
513 void (*sk_write_space)(struct sock *sk);
514 void (*sk_error_report)(struct sock *sk);
515 int (*sk_backlog_rcv)(struct sock *sk,
516 struct sk_buff *skb);
517 #ifdef CONFIG_SOCK_VALIDATE_XMIT
518 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
519 struct net_device *dev,
520 struct sk_buff *skb);
521 #endif
522 void (*sk_destruct)(struct sock *sk);
523 struct sock_reuseport __rcu *sk_reuseport_cb;
524 #ifdef CONFIG_BPF_SYSCALL
525 struct bpf_local_storage __rcu *sk_bpf_storage;
526 #endif
527 struct rcu_head sk_rcu;
528
529 #if IS_ENABLED(CONFIG_DEBUG_SPINLOCK) || IS_ENABLED(CONFIG_DEBUG_LOCK_ALLOC)
530 ANDROID_KABI_RESERVE(1);
531 #else
532 ANDROID_KABI_USE(1, spinlock_t sk_peer_lock);
533 #endif
534 ANDROID_KABI_RESERVE(2);
535 ANDROID_KABI_RESERVE(3);
536 ANDROID_KABI_RESERVE(4);
537 ANDROID_KABI_RESERVE(5);
538 ANDROID_KABI_RESERVE(6);
539 ANDROID_KABI_RESERVE(7);
540 ANDROID_KABI_RESERVE(8);
541
542 ANDROID_OEM_DATA(1);
543 };
544
545 enum sk_pacing {
546 SK_PACING_NONE = 0,
547 SK_PACING_NEEDED = 1,
548 SK_PACING_FQ = 2,
549 };
550
551 /* flag bits in sk_user_data
552 *
553 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
554 * not be suitable for copying when cloning the socket. For instance,
555 * it can point to a reference counted object. sk_user_data bottom
556 * bit is set if pointer must not be copied.
557 *
558 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
559 * managed/owned by a BPF reuseport array. This bit should be set
560 * when sk_user_data's sk is added to the bpf's reuseport_array.
561 *
562 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
563 * sk_user_data points to psock type. This bit should be set
564 * when sk_user_data is assigned to a psock object.
565 */
566 #define SK_USER_DATA_NOCOPY 1UL
567 #define SK_USER_DATA_BPF 2UL
568 #define SK_USER_DATA_PSOCK 4UL
569 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
570 SK_USER_DATA_PSOCK)
571
572 /**
573 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
574 * @sk: socket
575 */
sk_user_data_is_nocopy(const struct sock * sk)576 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
577 {
578 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
579 }
580
581 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
582
583 /**
584 * __rcu_dereference_sk_user_data_with_flags - return the pointer
585 * only if argument flags all has been set in sk_user_data. Otherwise
586 * return NULL
587 *
588 * @sk: socket
589 * @flags: flag bits
590 */
591 static inline void *
__rcu_dereference_sk_user_data_with_flags(const struct sock * sk,uintptr_t flags)592 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
593 uintptr_t flags)
594 {
595 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
596
597 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
598
599 if ((sk_user_data & flags) == flags)
600 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
601 return NULL;
602 }
603
604 #define rcu_dereference_sk_user_data(sk) \
605 __rcu_dereference_sk_user_data_with_flags(sk, 0)
606 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
607 ({ \
608 uintptr_t __tmp1 = (uintptr_t)(ptr), \
609 __tmp2 = (uintptr_t)(flags); \
610 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
611 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
612 rcu_assign_pointer(__sk_user_data((sk)), \
613 __tmp1 | __tmp2); \
614 })
615 #define rcu_assign_sk_user_data(sk, ptr) \
616 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
617
618 /*
619 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
620 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
621 * on a socket means that the socket will reuse everybody else's port
622 * without looking at the other's sk_reuse value.
623 */
624
625 #define SK_NO_REUSE 0
626 #define SK_CAN_REUSE 1
627 #define SK_FORCE_REUSE 2
628
629 int sk_set_peek_off(struct sock *sk, int val);
630
sk_peek_offset(struct sock * sk,int flags)631 static inline int sk_peek_offset(struct sock *sk, int flags)
632 {
633 if (unlikely(flags & MSG_PEEK)) {
634 return READ_ONCE(sk->sk_peek_off);
635 }
636
637 return 0;
638 }
639
sk_peek_offset_bwd(struct sock * sk,int val)640 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
641 {
642 s32 off = READ_ONCE(sk->sk_peek_off);
643
644 if (unlikely(off >= 0)) {
645 off = max_t(s32, off - val, 0);
646 WRITE_ONCE(sk->sk_peek_off, off);
647 }
648 }
649
sk_peek_offset_fwd(struct sock * sk,int val)650 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
651 {
652 sk_peek_offset_bwd(sk, -val);
653 }
654
655 /*
656 * Hashed lists helper routines
657 */
sk_entry(const struct hlist_node * node)658 static inline struct sock *sk_entry(const struct hlist_node *node)
659 {
660 return hlist_entry(node, struct sock, sk_node);
661 }
662
__sk_head(const struct hlist_head * head)663 static inline struct sock *__sk_head(const struct hlist_head *head)
664 {
665 return hlist_entry(head->first, struct sock, sk_node);
666 }
667
sk_head(const struct hlist_head * head)668 static inline struct sock *sk_head(const struct hlist_head *head)
669 {
670 return hlist_empty(head) ? NULL : __sk_head(head);
671 }
672
__sk_nulls_head(const struct hlist_nulls_head * head)673 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
674 {
675 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
676 }
677
sk_nulls_head(const struct hlist_nulls_head * head)678 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
679 {
680 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
681 }
682
sk_next(const struct sock * sk)683 static inline struct sock *sk_next(const struct sock *sk)
684 {
685 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
686 }
687
sk_nulls_next(const struct sock * sk)688 static inline struct sock *sk_nulls_next(const struct sock *sk)
689 {
690 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
691 hlist_nulls_entry(sk->sk_nulls_node.next,
692 struct sock, sk_nulls_node) :
693 NULL;
694 }
695
sk_unhashed(const struct sock * sk)696 static inline bool sk_unhashed(const struct sock *sk)
697 {
698 return hlist_unhashed(&sk->sk_node);
699 }
700
sk_hashed(const struct sock * sk)701 static inline bool sk_hashed(const struct sock *sk)
702 {
703 return !sk_unhashed(sk);
704 }
705
sk_node_init(struct hlist_node * node)706 static inline void sk_node_init(struct hlist_node *node)
707 {
708 node->pprev = NULL;
709 }
710
sk_nulls_node_init(struct hlist_nulls_node * node)711 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
712 {
713 node->pprev = NULL;
714 }
715
__sk_del_node(struct sock * sk)716 static inline void __sk_del_node(struct sock *sk)
717 {
718 __hlist_del(&sk->sk_node);
719 }
720
721 /* NB: equivalent to hlist_del_init_rcu */
__sk_del_node_init(struct sock * sk)722 static inline bool __sk_del_node_init(struct sock *sk)
723 {
724 if (sk_hashed(sk)) {
725 __sk_del_node(sk);
726 sk_node_init(&sk->sk_node);
727 return true;
728 }
729 return false;
730 }
731
732 /* Grab socket reference count. This operation is valid only
733 when sk is ALREADY grabbed f.e. it is found in hash table
734 or a list and the lookup is made under lock preventing hash table
735 modifications.
736 */
737
sock_hold(struct sock * sk)738 static __always_inline void sock_hold(struct sock *sk)
739 {
740 refcount_inc(&sk->sk_refcnt);
741 }
742
743 /* Ungrab socket in the context, which assumes that socket refcnt
744 cannot hit zero, f.e. it is true in context of any socketcall.
745 */
__sock_put(struct sock * sk)746 static __always_inline void __sock_put(struct sock *sk)
747 {
748 refcount_dec(&sk->sk_refcnt);
749 }
750
sk_del_node_init(struct sock * sk)751 static inline bool sk_del_node_init(struct sock *sk)
752 {
753 bool rc = __sk_del_node_init(sk);
754
755 if (rc) {
756 /* paranoid for a while -acme */
757 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
758 __sock_put(sk);
759 }
760 return rc;
761 }
762 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
763
__sk_nulls_del_node_init_rcu(struct sock * sk)764 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
765 {
766 if (sk_hashed(sk)) {
767 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
768 return true;
769 }
770 return false;
771 }
772
sk_nulls_del_node_init_rcu(struct sock * sk)773 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
774 {
775 bool rc = __sk_nulls_del_node_init_rcu(sk);
776
777 if (rc) {
778 /* paranoid for a while -acme */
779 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
780 __sock_put(sk);
781 }
782 return rc;
783 }
784
__sk_add_node(struct sock * sk,struct hlist_head * list)785 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
786 {
787 hlist_add_head(&sk->sk_node, list);
788 }
789
sk_add_node(struct sock * sk,struct hlist_head * list)790 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
791 {
792 sock_hold(sk);
793 __sk_add_node(sk, list);
794 }
795
sk_add_node_rcu(struct sock * sk,struct hlist_head * list)796 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
797 {
798 sock_hold(sk);
799 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
800 sk->sk_family == AF_INET6)
801 hlist_add_tail_rcu(&sk->sk_node, list);
802 else
803 hlist_add_head_rcu(&sk->sk_node, list);
804 }
805
sk_add_node_tail_rcu(struct sock * sk,struct hlist_head * list)806 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
807 {
808 sock_hold(sk);
809 hlist_add_tail_rcu(&sk->sk_node, list);
810 }
811
__sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)812 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
813 {
814 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
815 }
816
__sk_nulls_add_node_tail_rcu(struct sock * sk,struct hlist_nulls_head * list)817 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
818 {
819 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
820 }
821
sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)822 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
823 {
824 sock_hold(sk);
825 __sk_nulls_add_node_rcu(sk, list);
826 }
827
__sk_del_bind_node(struct sock * sk)828 static inline void __sk_del_bind_node(struct sock *sk)
829 {
830 __hlist_del(&sk->sk_bind_node);
831 }
832
sk_add_bind_node(struct sock * sk,struct hlist_head * list)833 static inline void sk_add_bind_node(struct sock *sk,
834 struct hlist_head *list)
835 {
836 hlist_add_head(&sk->sk_bind_node, list);
837 }
838
839 #define sk_for_each(__sk, list) \
840 hlist_for_each_entry(__sk, list, sk_node)
841 #define sk_for_each_rcu(__sk, list) \
842 hlist_for_each_entry_rcu(__sk, list, sk_node)
843 #define sk_nulls_for_each(__sk, node, list) \
844 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
845 #define sk_nulls_for_each_rcu(__sk, node, list) \
846 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
847 #define sk_for_each_from(__sk) \
848 hlist_for_each_entry_from(__sk, sk_node)
849 #define sk_nulls_for_each_from(__sk, node) \
850 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
851 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
852 #define sk_for_each_safe(__sk, tmp, list) \
853 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
854 #define sk_for_each_bound(__sk, list) \
855 hlist_for_each_entry(__sk, list, sk_bind_node)
856
857 /**
858 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
859 * @tpos: the type * to use as a loop cursor.
860 * @pos: the &struct hlist_node to use as a loop cursor.
861 * @head: the head for your list.
862 * @offset: offset of hlist_node within the struct.
863 *
864 */
865 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
866 for (pos = rcu_dereference(hlist_first_rcu(head)); \
867 pos != NULL && \
868 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
869 pos = rcu_dereference(hlist_next_rcu(pos)))
870
sk_user_ns(struct sock * sk)871 static inline struct user_namespace *sk_user_ns(struct sock *sk)
872 {
873 /* Careful only use this in a context where these parameters
874 * can not change and must all be valid, such as recvmsg from
875 * userspace.
876 */
877 return sk->sk_socket->file->f_cred->user_ns;
878 }
879
880 /* Sock flags */
881 enum sock_flags {
882 SOCK_DEAD,
883 SOCK_DONE,
884 SOCK_URGINLINE,
885 SOCK_KEEPOPEN,
886 SOCK_LINGER,
887 SOCK_DESTROY,
888 SOCK_BROADCAST,
889 SOCK_TIMESTAMP,
890 SOCK_ZAPPED,
891 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
892 SOCK_DBG, /* %SO_DEBUG setting */
893 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
894 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
895 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
896 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
897 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
898 SOCK_FASYNC, /* fasync() active */
899 SOCK_RXQ_OVFL,
900 SOCK_ZEROCOPY, /* buffers from userspace */
901 SOCK_WIFI_STATUS, /* push wifi status to userspace */
902 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
903 * Will use last 4 bytes of packet sent from
904 * user-space instead.
905 */
906 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
907 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
908 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
909 SOCK_TXTIME,
910 SOCK_XDP, /* XDP is attached */
911 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
912 };
913
914 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
915
sock_copy_flags(struct sock * nsk,struct sock * osk)916 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
917 {
918 nsk->sk_flags = osk->sk_flags;
919 }
920
sock_set_flag(struct sock * sk,enum sock_flags flag)921 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
922 {
923 __set_bit(flag, &sk->sk_flags);
924 }
925
sock_reset_flag(struct sock * sk,enum sock_flags flag)926 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
927 {
928 __clear_bit(flag, &sk->sk_flags);
929 }
930
sock_valbool_flag(struct sock * sk,enum sock_flags bit,int valbool)931 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
932 int valbool)
933 {
934 if (valbool)
935 sock_set_flag(sk, bit);
936 else
937 sock_reset_flag(sk, bit);
938 }
939
sock_flag(const struct sock * sk,enum sock_flags flag)940 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
941 {
942 return test_bit(flag, &sk->sk_flags);
943 }
944
945 #ifdef CONFIG_NET
946 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
sk_memalloc_socks(void)947 static inline int sk_memalloc_socks(void)
948 {
949 return static_branch_unlikely(&memalloc_socks_key);
950 }
951
952 void __receive_sock(struct file *file);
953 #else
954
sk_memalloc_socks(void)955 static inline int sk_memalloc_socks(void)
956 {
957 return 0;
958 }
959
__receive_sock(struct file * file)960 static inline void __receive_sock(struct file *file)
961 { }
962 #endif
963
sk_gfp_mask(const struct sock * sk,gfp_t gfp_mask)964 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
965 {
966 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
967 }
968
sk_acceptq_removed(struct sock * sk)969 static inline void sk_acceptq_removed(struct sock *sk)
970 {
971 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
972 }
973
sk_acceptq_added(struct sock * sk)974 static inline void sk_acceptq_added(struct sock *sk)
975 {
976 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
977 }
978
sk_acceptq_is_full(const struct sock * sk)979 static inline bool sk_acceptq_is_full(const struct sock *sk)
980 {
981 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
982 }
983
984 /*
985 * Compute minimal free write space needed to queue new packets.
986 */
sk_stream_min_wspace(const struct sock * sk)987 static inline int sk_stream_min_wspace(const struct sock *sk)
988 {
989 return READ_ONCE(sk->sk_wmem_queued) >> 1;
990 }
991
sk_stream_wspace(const struct sock * sk)992 static inline int sk_stream_wspace(const struct sock *sk)
993 {
994 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
995 }
996
sk_wmem_queued_add(struct sock * sk,int val)997 static inline void sk_wmem_queued_add(struct sock *sk, int val)
998 {
999 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1000 }
1001
1002 void sk_stream_write_space(struct sock *sk);
1003
1004 /* OOB backlog add */
__sk_add_backlog(struct sock * sk,struct sk_buff * skb)1005 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1006 {
1007 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1008 skb_dst_force(skb);
1009
1010 if (!sk->sk_backlog.tail)
1011 WRITE_ONCE(sk->sk_backlog.head, skb);
1012 else
1013 sk->sk_backlog.tail->next = skb;
1014
1015 WRITE_ONCE(sk->sk_backlog.tail, skb);
1016 skb->next = NULL;
1017 }
1018
1019 /*
1020 * Take into account size of receive queue and backlog queue
1021 * Do not take into account this skb truesize,
1022 * to allow even a single big packet to come.
1023 */
sk_rcvqueues_full(const struct sock * sk,unsigned int limit)1024 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1025 {
1026 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1027
1028 return qsize > limit;
1029 }
1030
1031 /* The per-socket spinlock must be held here. */
sk_add_backlog(struct sock * sk,struct sk_buff * skb,unsigned int limit)1032 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1033 unsigned int limit)
1034 {
1035 if (sk_rcvqueues_full(sk, limit))
1036 return -ENOBUFS;
1037
1038 /*
1039 * If the skb was allocated from pfmemalloc reserves, only
1040 * allow SOCK_MEMALLOC sockets to use it as this socket is
1041 * helping free memory
1042 */
1043 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1044 return -ENOMEM;
1045
1046 __sk_add_backlog(sk, skb);
1047 sk->sk_backlog.len += skb->truesize;
1048 return 0;
1049 }
1050
1051 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1052
sk_backlog_rcv(struct sock * sk,struct sk_buff * skb)1053 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1054 {
1055 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1056 return __sk_backlog_rcv(sk, skb);
1057
1058 return sk->sk_backlog_rcv(sk, skb);
1059 }
1060
sk_incoming_cpu_update(struct sock * sk)1061 static inline void sk_incoming_cpu_update(struct sock *sk)
1062 {
1063 int cpu = raw_smp_processor_id();
1064
1065 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1066 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1067 }
1068
sock_rps_record_flow_hash(__u32 hash)1069 static inline void sock_rps_record_flow_hash(__u32 hash)
1070 {
1071 #ifdef CONFIG_RPS
1072 struct rps_sock_flow_table *sock_flow_table;
1073
1074 rcu_read_lock();
1075 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1076 rps_record_sock_flow(sock_flow_table, hash);
1077 rcu_read_unlock();
1078 #endif
1079 }
1080
sock_rps_record_flow(const struct sock * sk)1081 static inline void sock_rps_record_flow(const struct sock *sk)
1082 {
1083 #ifdef CONFIG_RPS
1084 if (static_branch_unlikely(&rfs_needed)) {
1085 /* Reading sk->sk_rxhash might incur an expensive cache line
1086 * miss.
1087 *
1088 * TCP_ESTABLISHED does cover almost all states where RFS
1089 * might be useful, and is cheaper [1] than testing :
1090 * IPv4: inet_sk(sk)->inet_daddr
1091 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1092 * OR an additional socket flag
1093 * [1] : sk_state and sk_prot are in the same cache line.
1094 */
1095 if (sk->sk_state == TCP_ESTABLISHED)
1096 sock_rps_record_flow_hash(sk->sk_rxhash);
1097 }
1098 #endif
1099 }
1100
sock_rps_save_rxhash(struct sock * sk,const struct sk_buff * skb)1101 static inline void sock_rps_save_rxhash(struct sock *sk,
1102 const struct sk_buff *skb)
1103 {
1104 #ifdef CONFIG_RPS
1105 if (unlikely(sk->sk_rxhash != skb->hash))
1106 sk->sk_rxhash = skb->hash;
1107 #endif
1108 }
1109
sock_rps_reset_rxhash(struct sock * sk)1110 static inline void sock_rps_reset_rxhash(struct sock *sk)
1111 {
1112 #ifdef CONFIG_RPS
1113 sk->sk_rxhash = 0;
1114 #endif
1115 }
1116
1117 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1118 ({ int __rc; \
1119 release_sock(__sk); \
1120 __rc = __condition; \
1121 if (!__rc) { \
1122 *(__timeo) = wait_woken(__wait, \
1123 TASK_INTERRUPTIBLE, \
1124 *(__timeo)); \
1125 } \
1126 sched_annotate_sleep(); \
1127 lock_sock(__sk); \
1128 __rc = __condition; \
1129 __rc; \
1130 })
1131
1132 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1133 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1134 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1135 int sk_stream_error(struct sock *sk, int flags, int err);
1136 void sk_stream_kill_queues(struct sock *sk);
1137 void sk_set_memalloc(struct sock *sk);
1138 void sk_clear_memalloc(struct sock *sk);
1139
1140 void __sk_flush_backlog(struct sock *sk);
1141
sk_flush_backlog(struct sock * sk)1142 static inline bool sk_flush_backlog(struct sock *sk)
1143 {
1144 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1145 __sk_flush_backlog(sk);
1146 return true;
1147 }
1148 return false;
1149 }
1150
1151 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1152
1153 struct request_sock_ops;
1154 struct timewait_sock_ops;
1155 struct inet_hashinfo;
1156 struct raw_hashinfo;
1157 struct smc_hashinfo;
1158 struct module;
1159
1160 /*
1161 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1162 * un-modified. Special care is taken when initializing object to zero.
1163 */
sk_prot_clear_nulls(struct sock * sk,int size)1164 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1165 {
1166 if (offsetof(struct sock, sk_node.next) != 0)
1167 memset(sk, 0, offsetof(struct sock, sk_node.next));
1168 memset(&sk->sk_node.pprev, 0,
1169 size - offsetof(struct sock, sk_node.pprev));
1170 }
1171
1172 /* Networking protocol blocks we attach to sockets.
1173 * socket layer -> transport layer interface
1174 */
1175 struct proto {
1176 void (*close)(struct sock *sk,
1177 long timeout);
1178 int (*pre_connect)(struct sock *sk,
1179 struct sockaddr *uaddr,
1180 int addr_len);
1181 int (*connect)(struct sock *sk,
1182 struct sockaddr *uaddr,
1183 int addr_len);
1184 int (*disconnect)(struct sock *sk, int flags);
1185
1186 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1187 bool kern);
1188
1189 int (*ioctl)(struct sock *sk, int cmd,
1190 unsigned long arg);
1191 int (*init)(struct sock *sk);
1192 void (*destroy)(struct sock *sk);
1193 void (*shutdown)(struct sock *sk, int how);
1194 int (*setsockopt)(struct sock *sk, int level,
1195 int optname, sockptr_t optval,
1196 unsigned int optlen);
1197 int (*getsockopt)(struct sock *sk, int level,
1198 int optname, char __user *optval,
1199 int __user *option);
1200 void (*keepalive)(struct sock *sk, int valbool);
1201 #ifdef CONFIG_COMPAT
1202 int (*compat_ioctl)(struct sock *sk,
1203 unsigned int cmd, unsigned long arg);
1204 #endif
1205 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1206 size_t len);
1207 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1208 size_t len, int noblock, int flags,
1209 int *addr_len);
1210 int (*sendpage)(struct sock *sk, struct page *page,
1211 int offset, size_t size, int flags);
1212 int (*bind)(struct sock *sk,
1213 struct sockaddr *addr, int addr_len);
1214 int (*bind_add)(struct sock *sk,
1215 struct sockaddr *addr, int addr_len);
1216
1217 int (*backlog_rcv) (struct sock *sk,
1218 struct sk_buff *skb);
1219
1220 void (*release_cb)(struct sock *sk);
1221
1222 /* Keeping track of sk's, looking them up, and port selection methods. */
1223 int (*hash)(struct sock *sk);
1224 void (*unhash)(struct sock *sk);
1225 void (*rehash)(struct sock *sk);
1226 int (*get_port)(struct sock *sk, unsigned short snum);
1227
1228 /* Keeping track of sockets in use */
1229 #ifdef CONFIG_PROC_FS
1230 unsigned int inuse_idx;
1231 #endif
1232
1233 bool (*stream_memory_free)(const struct sock *sk, int wake);
1234 bool (*stream_memory_read)(const struct sock *sk);
1235 /* Memory pressure */
1236 void (*enter_memory_pressure)(struct sock *sk);
1237 void (*leave_memory_pressure)(struct sock *sk);
1238 atomic_long_t *memory_allocated; /* Current allocated memory. */
1239 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1240 /*
1241 * Pressure flag: try to collapse.
1242 * Technical note: it is used by multiple contexts non atomically.
1243 * All the __sk_mem_schedule() is of this nature: accounting
1244 * is strict, actions are advisory and have some latency.
1245 */
1246 unsigned long *memory_pressure;
1247 long *sysctl_mem;
1248
1249 int *sysctl_wmem;
1250 int *sysctl_rmem;
1251 u32 sysctl_wmem_offset;
1252 u32 sysctl_rmem_offset;
1253
1254 int max_header;
1255 bool no_autobind;
1256
1257 struct kmem_cache *slab;
1258 unsigned int obj_size;
1259 slab_flags_t slab_flags;
1260 unsigned int useroffset; /* Usercopy region offset */
1261 unsigned int usersize; /* Usercopy region size */
1262
1263 struct percpu_counter *orphan_count;
1264
1265 struct request_sock_ops *rsk_prot;
1266 struct timewait_sock_ops *twsk_prot;
1267
1268 union {
1269 struct inet_hashinfo *hashinfo;
1270 struct udp_table *udp_table;
1271 struct raw_hashinfo *raw_hash;
1272 struct smc_hashinfo *smc_hash;
1273 } h;
1274
1275 struct module *owner;
1276
1277 char name[32];
1278
1279 struct list_head node;
1280 #ifdef SOCK_REFCNT_DEBUG
1281 atomic_t socks;
1282 #endif
1283 int (*diag_destroy)(struct sock *sk, int err);
1284 } __randomize_layout;
1285
1286 int proto_register(struct proto *prot, int alloc_slab);
1287 void proto_unregister(struct proto *prot);
1288 int sock_load_diag_module(int family, int protocol);
1289
1290 #ifdef SOCK_REFCNT_DEBUG
sk_refcnt_debug_inc(struct sock * sk)1291 static inline void sk_refcnt_debug_inc(struct sock *sk)
1292 {
1293 atomic_inc(&sk->sk_prot->socks);
1294 }
1295
sk_refcnt_debug_dec(struct sock * sk)1296 static inline void sk_refcnt_debug_dec(struct sock *sk)
1297 {
1298 atomic_dec(&sk->sk_prot->socks);
1299 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1300 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1301 }
1302
sk_refcnt_debug_release(const struct sock * sk)1303 static inline void sk_refcnt_debug_release(const struct sock *sk)
1304 {
1305 if (refcount_read(&sk->sk_refcnt) != 1)
1306 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1307 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1308 }
1309 #else /* SOCK_REFCNT_DEBUG */
1310 #define sk_refcnt_debug_inc(sk) do { } while (0)
1311 #define sk_refcnt_debug_dec(sk) do { } while (0)
1312 #define sk_refcnt_debug_release(sk) do { } while (0)
1313 #endif /* SOCK_REFCNT_DEBUG */
1314
__sk_stream_memory_free(const struct sock * sk,int wake)1315 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1316 {
1317 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1318 return false;
1319
1320 return sk->sk_prot->stream_memory_free ?
1321 sk->sk_prot->stream_memory_free(sk, wake) : true;
1322 }
1323
sk_stream_memory_free(const struct sock * sk)1324 static inline bool sk_stream_memory_free(const struct sock *sk)
1325 {
1326 return __sk_stream_memory_free(sk, 0);
1327 }
1328
__sk_stream_is_writeable(const struct sock * sk,int wake)1329 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1330 {
1331 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1332 __sk_stream_memory_free(sk, wake);
1333 }
1334
sk_stream_is_writeable(const struct sock * sk)1335 static inline bool sk_stream_is_writeable(const struct sock *sk)
1336 {
1337 return __sk_stream_is_writeable(sk, 0);
1338 }
1339
sk_under_cgroup_hierarchy(struct sock * sk,struct cgroup * ancestor)1340 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1341 struct cgroup *ancestor)
1342 {
1343 #ifdef CONFIG_SOCK_CGROUP_DATA
1344 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1345 ancestor);
1346 #else
1347 return -ENOTSUPP;
1348 #endif
1349 }
1350
sk_has_memory_pressure(const struct sock * sk)1351 static inline bool sk_has_memory_pressure(const struct sock *sk)
1352 {
1353 return sk->sk_prot->memory_pressure != NULL;
1354 }
1355
sk_under_memory_pressure(const struct sock * sk)1356 static inline bool sk_under_memory_pressure(const struct sock *sk)
1357 {
1358 if (!sk->sk_prot->memory_pressure)
1359 return false;
1360
1361 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1362 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1363 return true;
1364
1365 return !!*sk->sk_prot->memory_pressure;
1366 }
1367
1368 static inline long
sk_memory_allocated(const struct sock * sk)1369 sk_memory_allocated(const struct sock *sk)
1370 {
1371 return atomic_long_read(sk->sk_prot->memory_allocated);
1372 }
1373
1374 static inline long
sk_memory_allocated_add(struct sock * sk,int amt)1375 sk_memory_allocated_add(struct sock *sk, int amt)
1376 {
1377 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1378 }
1379
1380 static inline void
sk_memory_allocated_sub(struct sock * sk,int amt)1381 sk_memory_allocated_sub(struct sock *sk, int amt)
1382 {
1383 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1384 }
1385
sk_sockets_allocated_dec(struct sock * sk)1386 static inline void sk_sockets_allocated_dec(struct sock *sk)
1387 {
1388 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1389 }
1390
sk_sockets_allocated_inc(struct sock * sk)1391 static inline void sk_sockets_allocated_inc(struct sock *sk)
1392 {
1393 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1394 }
1395
1396 static inline u64
sk_sockets_allocated_read_positive(struct sock * sk)1397 sk_sockets_allocated_read_positive(struct sock *sk)
1398 {
1399 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1400 }
1401
1402 static inline int
proto_sockets_allocated_sum_positive(struct proto * prot)1403 proto_sockets_allocated_sum_positive(struct proto *prot)
1404 {
1405 return percpu_counter_sum_positive(prot->sockets_allocated);
1406 }
1407
1408 static inline long
proto_memory_allocated(struct proto * prot)1409 proto_memory_allocated(struct proto *prot)
1410 {
1411 return atomic_long_read(prot->memory_allocated);
1412 }
1413
1414 static inline bool
proto_memory_pressure(struct proto * prot)1415 proto_memory_pressure(struct proto *prot)
1416 {
1417 if (!prot->memory_pressure)
1418 return false;
1419 return !!*prot->memory_pressure;
1420 }
1421
1422
1423 #ifdef CONFIG_PROC_FS
1424 /* Called with local bh disabled */
1425 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1426 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1427 int sock_inuse_get(struct net *net);
1428 #else
sock_prot_inuse_add(struct net * net,struct proto * prot,int inc)1429 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1430 int inc)
1431 {
1432 }
1433 #endif
1434
1435
1436 /* With per-bucket locks this operation is not-atomic, so that
1437 * this version is not worse.
1438 */
__sk_prot_rehash(struct sock * sk)1439 static inline int __sk_prot_rehash(struct sock *sk)
1440 {
1441 sk->sk_prot->unhash(sk);
1442 return sk->sk_prot->hash(sk);
1443 }
1444
1445 /* About 10 seconds */
1446 #define SOCK_DESTROY_TIME (10*HZ)
1447
1448 /* Sockets 0-1023 can't be bound to unless you are superuser */
1449 #define PROT_SOCK 1024
1450
1451 #define SHUTDOWN_MASK 3
1452 #define RCV_SHUTDOWN 1
1453 #define SEND_SHUTDOWN 2
1454
1455 #define SOCK_SNDBUF_LOCK 1
1456 #define SOCK_RCVBUF_LOCK 2
1457 #define SOCK_BINDADDR_LOCK 4
1458 #define SOCK_BINDPORT_LOCK 8
1459
1460 struct socket_alloc {
1461 struct socket socket;
1462 struct inode vfs_inode;
1463 };
1464
SOCKET_I(struct inode * inode)1465 static inline struct socket *SOCKET_I(struct inode *inode)
1466 {
1467 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1468 }
1469
SOCK_INODE(struct socket * socket)1470 static inline struct inode *SOCK_INODE(struct socket *socket)
1471 {
1472 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1473 }
1474
1475 /*
1476 * Functions for memory accounting
1477 */
1478 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1479 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1480 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1481 void __sk_mem_reclaim(struct sock *sk, int amount);
1482
1483 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1484 * do not necessarily have 16x time more memory than 4KB ones.
1485 */
1486 #define SK_MEM_QUANTUM 4096
1487 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1488 #define SK_MEM_SEND 0
1489 #define SK_MEM_RECV 1
1490
1491 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
sk_prot_mem_limits(const struct sock * sk,int index)1492 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1493 {
1494 long val = READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1495
1496 #if PAGE_SIZE > SK_MEM_QUANTUM
1497 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1498 #elif PAGE_SIZE < SK_MEM_QUANTUM
1499 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1500 #endif
1501 return val;
1502 }
1503
sk_mem_pages(int amt)1504 static inline int sk_mem_pages(int amt)
1505 {
1506 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1507 }
1508
sk_has_account(struct sock * sk)1509 static inline bool sk_has_account(struct sock *sk)
1510 {
1511 /* return true if protocol supports memory accounting */
1512 return !!sk->sk_prot->memory_allocated;
1513 }
1514
sk_wmem_schedule(struct sock * sk,int size)1515 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1516 {
1517 int delta;
1518
1519 if (!sk_has_account(sk))
1520 return true;
1521 delta = size - sk->sk_forward_alloc;
1522 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1523 }
1524
1525 static inline bool
sk_rmem_schedule(struct sock * sk,struct sk_buff * skb,int size)1526 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1527 {
1528 int delta;
1529
1530 if (!sk_has_account(sk))
1531 return true;
1532 delta = size - sk->sk_forward_alloc;
1533 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1534 skb_pfmemalloc(skb);
1535 }
1536
sk_mem_reclaim(struct sock * sk)1537 static inline void sk_mem_reclaim(struct sock *sk)
1538 {
1539 if (!sk_has_account(sk))
1540 return;
1541 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1542 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1543 }
1544
sk_mem_reclaim_partial(struct sock * sk)1545 static inline void sk_mem_reclaim_partial(struct sock *sk)
1546 {
1547 if (!sk_has_account(sk))
1548 return;
1549 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1550 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1551 }
1552
sk_mem_charge(struct sock * sk,int size)1553 static inline void sk_mem_charge(struct sock *sk, int size)
1554 {
1555 if (!sk_has_account(sk))
1556 return;
1557 sk->sk_forward_alloc -= size;
1558 }
1559
sk_mem_uncharge(struct sock * sk,int size)1560 static inline void sk_mem_uncharge(struct sock *sk, int size)
1561 {
1562 if (!sk_has_account(sk))
1563 return;
1564 sk->sk_forward_alloc += size;
1565
1566 /* Avoid a possible overflow.
1567 * TCP send queues can make this happen, if sk_mem_reclaim()
1568 * is not called and more than 2 GBytes are released at once.
1569 *
1570 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1571 * no need to hold that much forward allocation anyway.
1572 */
1573 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1574 __sk_mem_reclaim(sk, 1 << 20);
1575 }
1576
1577 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
sk_wmem_free_skb(struct sock * sk,struct sk_buff * skb)1578 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1579 {
1580 sk_wmem_queued_add(sk, -skb->truesize);
1581 sk_mem_uncharge(sk, skb->truesize);
1582 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1583 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1584 skb_ext_reset(skb);
1585 skb_zcopy_clear(skb, true);
1586 sk->sk_tx_skb_cache = skb;
1587 return;
1588 }
1589 __kfree_skb(skb);
1590 }
1591
sock_release_ownership(struct sock * sk)1592 static inline void sock_release_ownership(struct sock *sk)
1593 {
1594 if (sk->sk_lock.owned) {
1595 sk->sk_lock.owned = 0;
1596
1597 /* The sk_lock has mutex_unlock() semantics: */
1598 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1599 }
1600 }
1601
1602 /*
1603 * Macro so as to not evaluate some arguments when
1604 * lockdep is not enabled.
1605 *
1606 * Mark both the sk_lock and the sk_lock.slock as a
1607 * per-address-family lock class.
1608 */
1609 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1610 do { \
1611 sk->sk_lock.owned = 0; \
1612 init_waitqueue_head(&sk->sk_lock.wq); \
1613 spin_lock_init(&(sk)->sk_lock.slock); \
1614 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1615 sizeof((sk)->sk_lock)); \
1616 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1617 (skey), (sname)); \
1618 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1619 } while (0)
1620
1621 #ifdef CONFIG_LOCKDEP
lockdep_sock_is_held(const struct sock * sk)1622 static inline bool lockdep_sock_is_held(const struct sock *sk)
1623 {
1624 return lockdep_is_held(&sk->sk_lock) ||
1625 lockdep_is_held(&sk->sk_lock.slock);
1626 }
1627 #endif
1628
1629 void lock_sock_nested(struct sock *sk, int subclass);
1630
lock_sock(struct sock * sk)1631 static inline void lock_sock(struct sock *sk)
1632 {
1633 lock_sock_nested(sk, 0);
1634 }
1635
1636 void __release_sock(struct sock *sk);
1637 void release_sock(struct sock *sk);
1638
1639 /* BH context may only use the following locking interface. */
1640 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1641 #define bh_lock_sock_nested(__sk) \
1642 spin_lock_nested(&((__sk)->sk_lock.slock), \
1643 SINGLE_DEPTH_NESTING)
1644 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1645
1646 bool lock_sock_fast(struct sock *sk);
1647 /**
1648 * unlock_sock_fast - complement of lock_sock_fast
1649 * @sk: socket
1650 * @slow: slow mode
1651 *
1652 * fast unlock socket for user context.
1653 * If slow mode is on, we call regular release_sock()
1654 */
unlock_sock_fast(struct sock * sk,bool slow)1655 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1656 {
1657 if (slow)
1658 release_sock(sk);
1659 else
1660 spin_unlock_bh(&sk->sk_lock.slock);
1661 }
1662
1663 /* Used by processes to "lock" a socket state, so that
1664 * interrupts and bottom half handlers won't change it
1665 * from under us. It essentially blocks any incoming
1666 * packets, so that we won't get any new data or any
1667 * packets that change the state of the socket.
1668 *
1669 * While locked, BH processing will add new packets to
1670 * the backlog queue. This queue is processed by the
1671 * owner of the socket lock right before it is released.
1672 *
1673 * Since ~2.3.5 it is also exclusive sleep lock serializing
1674 * accesses from user process context.
1675 */
1676
sock_owned_by_me(const struct sock * sk)1677 static inline void sock_owned_by_me(const struct sock *sk)
1678 {
1679 #ifdef CONFIG_LOCKDEP
1680 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1681 #endif
1682 }
1683
sock_owned_by_user(const struct sock * sk)1684 static inline bool sock_owned_by_user(const struct sock *sk)
1685 {
1686 sock_owned_by_me(sk);
1687 return sk->sk_lock.owned;
1688 }
1689
sock_owned_by_user_nocheck(const struct sock * sk)1690 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1691 {
1692 return sk->sk_lock.owned;
1693 }
1694
1695 /* no reclassification while locks are held */
sock_allow_reclassification(const struct sock * csk)1696 static inline bool sock_allow_reclassification(const struct sock *csk)
1697 {
1698 struct sock *sk = (struct sock *)csk;
1699
1700 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1701 }
1702
1703 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1704 struct proto *prot, int kern);
1705 void sk_free(struct sock *sk);
1706 void sk_destruct(struct sock *sk);
1707 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1708 void sk_free_unlock_clone(struct sock *sk);
1709
1710 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1711 gfp_t priority);
1712 void __sock_wfree(struct sk_buff *skb);
1713 void sock_wfree(struct sk_buff *skb);
1714 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1715 gfp_t priority);
1716 void skb_orphan_partial(struct sk_buff *skb);
1717 void sock_rfree(struct sk_buff *skb);
1718 void sock_efree(struct sk_buff *skb);
1719 #ifdef CONFIG_INET
1720 void sock_edemux(struct sk_buff *skb);
1721 void sock_pfree(struct sk_buff *skb);
1722 #else
1723 #define sock_edemux sock_efree
1724 #endif
1725
1726 int sock_setsockopt(struct socket *sock, int level, int op,
1727 sockptr_t optval, unsigned int optlen);
1728
1729 int sock_getsockopt(struct socket *sock, int level, int op,
1730 char __user *optval, int __user *optlen);
1731 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1732 bool timeval, bool time32);
1733 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1734 int noblock, int *errcode);
1735 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1736 unsigned long data_len, int noblock,
1737 int *errcode, int max_page_order);
1738 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1739 void sock_kfree_s(struct sock *sk, void *mem, int size);
1740 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1741 void sk_send_sigurg(struct sock *sk);
1742
1743 struct sockcm_cookie {
1744 u64 transmit_time;
1745 u32 mark;
1746 u16 tsflags;
1747 };
1748
sockcm_init(struct sockcm_cookie * sockc,const struct sock * sk)1749 static inline void sockcm_init(struct sockcm_cookie *sockc,
1750 const struct sock *sk)
1751 {
1752 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1753 }
1754
1755 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1756 struct sockcm_cookie *sockc);
1757 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1758 struct sockcm_cookie *sockc);
1759
1760 /*
1761 * Functions to fill in entries in struct proto_ops when a protocol
1762 * does not implement a particular function.
1763 */
1764 int sock_no_bind(struct socket *, struct sockaddr *, int);
1765 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1766 int sock_no_socketpair(struct socket *, struct socket *);
1767 int sock_no_accept(struct socket *, struct socket *, int, bool);
1768 int sock_no_getname(struct socket *, struct sockaddr *, int);
1769 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1770 int sock_no_listen(struct socket *, int);
1771 int sock_no_shutdown(struct socket *, int);
1772 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1773 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1774 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1775 int sock_no_mmap(struct file *file, struct socket *sock,
1776 struct vm_area_struct *vma);
1777 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1778 size_t size, int flags);
1779 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1780 int offset, size_t size, int flags);
1781
1782 /*
1783 * Functions to fill in entries in struct proto_ops when a protocol
1784 * uses the inet style.
1785 */
1786 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1787 char __user *optval, int __user *optlen);
1788 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1789 int flags);
1790 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1791 sockptr_t optval, unsigned int optlen);
1792
1793 void sk_common_release(struct sock *sk);
1794
1795 /*
1796 * Default socket callbacks and setup code
1797 */
1798
1799 /* Initialise core socket variables */
1800 void sock_init_data(struct socket *sock, struct sock *sk);
1801
1802 /*
1803 * Socket reference counting postulates.
1804 *
1805 * * Each user of socket SHOULD hold a reference count.
1806 * * Each access point to socket (an hash table bucket, reference from a list,
1807 * running timer, skb in flight MUST hold a reference count.
1808 * * When reference count hits 0, it means it will never increase back.
1809 * * When reference count hits 0, it means that no references from
1810 * outside exist to this socket and current process on current CPU
1811 * is last user and may/should destroy this socket.
1812 * * sk_free is called from any context: process, BH, IRQ. When
1813 * it is called, socket has no references from outside -> sk_free
1814 * may release descendant resources allocated by the socket, but
1815 * to the time when it is called, socket is NOT referenced by any
1816 * hash tables, lists etc.
1817 * * Packets, delivered from outside (from network or from another process)
1818 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1819 * when they sit in queue. Otherwise, packets will leak to hole, when
1820 * socket is looked up by one cpu and unhasing is made by another CPU.
1821 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1822 * (leak to backlog). Packet socket does all the processing inside
1823 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1824 * use separate SMP lock, so that they are prone too.
1825 */
1826
1827 /* Ungrab socket and destroy it, if it was the last reference. */
sock_put(struct sock * sk)1828 static inline void sock_put(struct sock *sk)
1829 {
1830 if (refcount_dec_and_test(&sk->sk_refcnt))
1831 sk_free(sk);
1832 }
1833 /* Generic version of sock_put(), dealing with all sockets
1834 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1835 */
1836 void sock_gen_put(struct sock *sk);
1837
1838 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1839 unsigned int trim_cap, bool refcounted);
sk_receive_skb(struct sock * sk,struct sk_buff * skb,const int nested)1840 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1841 const int nested)
1842 {
1843 return __sk_receive_skb(sk, skb, nested, 1, true);
1844 }
1845
sk_tx_queue_set(struct sock * sk,int tx_queue)1846 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1847 {
1848 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1849 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1850 return;
1851 sk->sk_tx_queue_mapping = tx_queue;
1852 }
1853
1854 #define NO_QUEUE_MAPPING USHRT_MAX
1855
sk_tx_queue_clear(struct sock * sk)1856 static inline void sk_tx_queue_clear(struct sock *sk)
1857 {
1858 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1859 }
1860
sk_tx_queue_get(const struct sock * sk)1861 static inline int sk_tx_queue_get(const struct sock *sk)
1862 {
1863 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1864 return sk->sk_tx_queue_mapping;
1865
1866 return -1;
1867 }
1868
sk_rx_queue_set(struct sock * sk,const struct sk_buff * skb)1869 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1870 {
1871 #ifdef CONFIG_XPS
1872 if (skb_rx_queue_recorded(skb)) {
1873 u16 rx_queue = skb_get_rx_queue(skb);
1874
1875 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1876 return;
1877
1878 sk->sk_rx_queue_mapping = rx_queue;
1879 }
1880 #endif
1881 }
1882
sk_rx_queue_clear(struct sock * sk)1883 static inline void sk_rx_queue_clear(struct sock *sk)
1884 {
1885 #ifdef CONFIG_XPS
1886 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1887 #endif
1888 }
1889
1890 #ifdef CONFIG_XPS
sk_rx_queue_get(const struct sock * sk)1891 static inline int sk_rx_queue_get(const struct sock *sk)
1892 {
1893 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1894 return sk->sk_rx_queue_mapping;
1895
1896 return -1;
1897 }
1898 #endif
1899
sk_set_socket(struct sock * sk,struct socket * sock)1900 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1901 {
1902 sk->sk_socket = sock;
1903 }
1904
sk_sleep(struct sock * sk)1905 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1906 {
1907 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1908 return &rcu_dereference_raw(sk->sk_wq)->wait;
1909 }
1910 /* Detach socket from process context.
1911 * Announce socket dead, detach it from wait queue and inode.
1912 * Note that parent inode held reference count on this struct sock,
1913 * we do not release it in this function, because protocol
1914 * probably wants some additional cleanups or even continuing
1915 * to work with this socket (TCP).
1916 */
sock_orphan(struct sock * sk)1917 static inline void sock_orphan(struct sock *sk)
1918 {
1919 write_lock_bh(&sk->sk_callback_lock);
1920 sock_set_flag(sk, SOCK_DEAD);
1921 sk_set_socket(sk, NULL);
1922 sk->sk_wq = NULL;
1923 write_unlock_bh(&sk->sk_callback_lock);
1924 }
1925
sock_graft(struct sock * sk,struct socket * parent)1926 static inline void sock_graft(struct sock *sk, struct socket *parent)
1927 {
1928 WARN_ON(parent->sk);
1929 write_lock_bh(&sk->sk_callback_lock);
1930 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1931 parent->sk = sk;
1932 sk_set_socket(sk, parent);
1933 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1934 security_sock_graft(sk, parent);
1935 write_unlock_bh(&sk->sk_callback_lock);
1936 }
1937
1938 kuid_t sock_i_uid(struct sock *sk);
1939 unsigned long sock_i_ino(struct sock *sk);
1940
sock_net_uid(const struct net * net,const struct sock * sk)1941 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1942 {
1943 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1944 }
1945
net_tx_rndhash(void)1946 static inline u32 net_tx_rndhash(void)
1947 {
1948 u32 v = prandom_u32();
1949
1950 return v ?: 1;
1951 }
1952
sk_set_txhash(struct sock * sk)1953 static inline void sk_set_txhash(struct sock *sk)
1954 {
1955 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
1956 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
1957 }
1958
sk_rethink_txhash(struct sock * sk)1959 static inline bool sk_rethink_txhash(struct sock *sk)
1960 {
1961 if (sk->sk_txhash) {
1962 sk_set_txhash(sk);
1963 return true;
1964 }
1965 return false;
1966 }
1967
1968 static inline struct dst_entry *
__sk_dst_get(struct sock * sk)1969 __sk_dst_get(struct sock *sk)
1970 {
1971 return rcu_dereference_check(sk->sk_dst_cache,
1972 lockdep_sock_is_held(sk));
1973 }
1974
1975 static inline struct dst_entry *
sk_dst_get(struct sock * sk)1976 sk_dst_get(struct sock *sk)
1977 {
1978 struct dst_entry *dst;
1979
1980 rcu_read_lock();
1981 dst = rcu_dereference(sk->sk_dst_cache);
1982 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1983 dst = NULL;
1984 rcu_read_unlock();
1985 return dst;
1986 }
1987
__dst_negative_advice(struct sock * sk)1988 static inline void __dst_negative_advice(struct sock *sk)
1989 {
1990 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1991
1992 if (dst && dst->ops->negative_advice) {
1993 ndst = dst->ops->negative_advice(dst);
1994
1995 if (ndst != dst) {
1996 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1997 sk_tx_queue_clear(sk);
1998 sk->sk_dst_pending_confirm = 0;
1999 }
2000 }
2001 }
2002
dst_negative_advice(struct sock * sk)2003 static inline void dst_negative_advice(struct sock *sk)
2004 {
2005 sk_rethink_txhash(sk);
2006 __dst_negative_advice(sk);
2007 }
2008
2009 static inline void
__sk_dst_set(struct sock * sk,struct dst_entry * dst)2010 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2011 {
2012 struct dst_entry *old_dst;
2013
2014 sk_tx_queue_clear(sk);
2015 sk->sk_dst_pending_confirm = 0;
2016 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2017 lockdep_sock_is_held(sk));
2018 rcu_assign_pointer(sk->sk_dst_cache, dst);
2019 dst_release(old_dst);
2020 }
2021
2022 static inline void
sk_dst_set(struct sock * sk,struct dst_entry * dst)2023 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2024 {
2025 struct dst_entry *old_dst;
2026
2027 sk_tx_queue_clear(sk);
2028 sk->sk_dst_pending_confirm = 0;
2029 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2030 dst_release(old_dst);
2031 }
2032
2033 static inline void
__sk_dst_reset(struct sock * sk)2034 __sk_dst_reset(struct sock *sk)
2035 {
2036 __sk_dst_set(sk, NULL);
2037 }
2038
2039 static inline void
sk_dst_reset(struct sock * sk)2040 sk_dst_reset(struct sock *sk)
2041 {
2042 sk_dst_set(sk, NULL);
2043 }
2044
2045 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2046
2047 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2048
sk_dst_confirm(struct sock * sk)2049 static inline void sk_dst_confirm(struct sock *sk)
2050 {
2051 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2052 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2053 }
2054
sock_confirm_neigh(struct sk_buff * skb,struct neighbour * n)2055 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2056 {
2057 if (skb_get_dst_pending_confirm(skb)) {
2058 struct sock *sk = skb->sk;
2059 unsigned long now = jiffies;
2060
2061 /* avoid dirtying neighbour */
2062 if (READ_ONCE(n->confirmed) != now)
2063 WRITE_ONCE(n->confirmed, now);
2064 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2065 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2066 }
2067 }
2068
2069 bool sk_mc_loop(struct sock *sk);
2070
sk_can_gso(const struct sock * sk)2071 static inline bool sk_can_gso(const struct sock *sk)
2072 {
2073 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2074 }
2075
2076 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2077
sk_nocaps_add(struct sock * sk,netdev_features_t flags)2078 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2079 {
2080 sk->sk_route_nocaps |= flags;
2081 sk->sk_route_caps &= ~flags;
2082 }
2083
skb_do_copy_data_nocache(struct sock * sk,struct sk_buff * skb,struct iov_iter * from,char * to,int copy,int offset)2084 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2085 struct iov_iter *from, char *to,
2086 int copy, int offset)
2087 {
2088 if (skb->ip_summed == CHECKSUM_NONE) {
2089 __wsum csum = 0;
2090 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2091 return -EFAULT;
2092 skb->csum = csum_block_add(skb->csum, csum, offset);
2093 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2094 if (!copy_from_iter_full_nocache(to, copy, from))
2095 return -EFAULT;
2096 } else if (!copy_from_iter_full(to, copy, from))
2097 return -EFAULT;
2098
2099 return 0;
2100 }
2101
skb_add_data_nocache(struct sock * sk,struct sk_buff * skb,struct iov_iter * from,int copy)2102 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2103 struct iov_iter *from, int copy)
2104 {
2105 int err, offset = skb->len;
2106
2107 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2108 copy, offset);
2109 if (err)
2110 __skb_trim(skb, offset);
2111
2112 return err;
2113 }
2114
skb_copy_to_page_nocache(struct sock * sk,struct iov_iter * from,struct sk_buff * skb,struct page * page,int off,int copy)2115 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2116 struct sk_buff *skb,
2117 struct page *page,
2118 int off, int copy)
2119 {
2120 int err;
2121
2122 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2123 copy, skb->len);
2124 if (err)
2125 return err;
2126
2127 skb->len += copy;
2128 skb->data_len += copy;
2129 skb->truesize += copy;
2130 sk_wmem_queued_add(sk, copy);
2131 sk_mem_charge(sk, copy);
2132 return 0;
2133 }
2134
2135 /**
2136 * sk_wmem_alloc_get - returns write allocations
2137 * @sk: socket
2138 *
2139 * Return: sk_wmem_alloc minus initial offset of one
2140 */
sk_wmem_alloc_get(const struct sock * sk)2141 static inline int sk_wmem_alloc_get(const struct sock *sk)
2142 {
2143 return refcount_read(&sk->sk_wmem_alloc) - 1;
2144 }
2145
2146 /**
2147 * sk_rmem_alloc_get - returns read allocations
2148 * @sk: socket
2149 *
2150 * Return: sk_rmem_alloc
2151 */
sk_rmem_alloc_get(const struct sock * sk)2152 static inline int sk_rmem_alloc_get(const struct sock *sk)
2153 {
2154 return atomic_read(&sk->sk_rmem_alloc);
2155 }
2156
2157 /**
2158 * sk_has_allocations - check if allocations are outstanding
2159 * @sk: socket
2160 *
2161 * Return: true if socket has write or read allocations
2162 */
sk_has_allocations(const struct sock * sk)2163 static inline bool sk_has_allocations(const struct sock *sk)
2164 {
2165 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2166 }
2167
2168 /**
2169 * skwq_has_sleeper - check if there are any waiting processes
2170 * @wq: struct socket_wq
2171 *
2172 * Return: true if socket_wq has waiting processes
2173 *
2174 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2175 * barrier call. They were added due to the race found within the tcp code.
2176 *
2177 * Consider following tcp code paths::
2178 *
2179 * CPU1 CPU2
2180 * sys_select receive packet
2181 * ... ...
2182 * __add_wait_queue update tp->rcv_nxt
2183 * ... ...
2184 * tp->rcv_nxt check sock_def_readable
2185 * ... {
2186 * schedule rcu_read_lock();
2187 * wq = rcu_dereference(sk->sk_wq);
2188 * if (wq && waitqueue_active(&wq->wait))
2189 * wake_up_interruptible(&wq->wait)
2190 * ...
2191 * }
2192 *
2193 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2194 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2195 * could then endup calling schedule and sleep forever if there are no more
2196 * data on the socket.
2197 *
2198 */
skwq_has_sleeper(struct socket_wq * wq)2199 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2200 {
2201 return wq && wq_has_sleeper(&wq->wait);
2202 }
2203
2204 /**
2205 * sock_poll_wait - place memory barrier behind the poll_wait call.
2206 * @filp: file
2207 * @sock: socket to wait on
2208 * @p: poll_table
2209 *
2210 * See the comments in the wq_has_sleeper function.
2211 */
sock_poll_wait(struct file * filp,struct socket * sock,poll_table * p)2212 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2213 poll_table *p)
2214 {
2215 if (!poll_does_not_wait(p)) {
2216 poll_wait(filp, &sock->wq.wait, p);
2217 /* We need to be sure we are in sync with the
2218 * socket flags modification.
2219 *
2220 * This memory barrier is paired in the wq_has_sleeper.
2221 */
2222 smp_mb();
2223 }
2224 }
2225
skb_set_hash_from_sk(struct sk_buff * skb,struct sock * sk)2226 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2227 {
2228 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2229 u32 txhash = READ_ONCE(sk->sk_txhash);
2230
2231 if (txhash) {
2232 skb->l4_hash = 1;
2233 skb->hash = txhash;
2234 }
2235 }
2236
2237 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2238
2239 /*
2240 * Queue a received datagram if it will fit. Stream and sequenced
2241 * protocols can't normally use this as they need to fit buffers in
2242 * and play with them.
2243 *
2244 * Inlined as it's very short and called for pretty much every
2245 * packet ever received.
2246 */
skb_set_owner_r(struct sk_buff * skb,struct sock * sk)2247 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2248 {
2249 skb_orphan(skb);
2250 skb->sk = sk;
2251 skb->destructor = sock_rfree;
2252 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2253 sk_mem_charge(sk, skb->truesize);
2254 }
2255
skb_set_owner_sk_safe(struct sk_buff * skb,struct sock * sk)2256 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2257 {
2258 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2259 skb_orphan(skb);
2260 skb->destructor = sock_efree;
2261 skb->sk = sk;
2262 return true;
2263 }
2264 return false;
2265 }
2266
2267 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2268 unsigned long expires);
2269
2270 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2271
2272 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2273
2274 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2275 struct sk_buff *skb, unsigned int flags,
2276 void (*destructor)(struct sock *sk,
2277 struct sk_buff *skb));
2278 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2279 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2280
2281 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2282 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2283
2284 /*
2285 * Recover an error report and clear atomically
2286 */
2287
sock_error(struct sock * sk)2288 static inline int sock_error(struct sock *sk)
2289 {
2290 int err;
2291
2292 /* Avoid an atomic operation for the common case.
2293 * This is racy since another cpu/thread can change sk_err under us.
2294 */
2295 if (likely(data_race(!sk->sk_err)))
2296 return 0;
2297
2298 err = xchg(&sk->sk_err, 0);
2299 return -err;
2300 }
2301
sock_wspace(struct sock * sk)2302 static inline unsigned long sock_wspace(struct sock *sk)
2303 {
2304 int amt = 0;
2305
2306 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2307 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2308 if (amt < 0)
2309 amt = 0;
2310 }
2311 return amt;
2312 }
2313
2314 /* Note:
2315 * We use sk->sk_wq_raw, from contexts knowing this
2316 * pointer is not NULL and cannot disappear/change.
2317 */
sk_set_bit(int nr,struct sock * sk)2318 static inline void sk_set_bit(int nr, struct sock *sk)
2319 {
2320 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2321 !sock_flag(sk, SOCK_FASYNC))
2322 return;
2323
2324 set_bit(nr, &sk->sk_wq_raw->flags);
2325 }
2326
sk_clear_bit(int nr,struct sock * sk)2327 static inline void sk_clear_bit(int nr, struct sock *sk)
2328 {
2329 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2330 !sock_flag(sk, SOCK_FASYNC))
2331 return;
2332
2333 clear_bit(nr, &sk->sk_wq_raw->flags);
2334 }
2335
sk_wake_async(const struct sock * sk,int how,int band)2336 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2337 {
2338 if (sock_flag(sk, SOCK_FASYNC)) {
2339 rcu_read_lock();
2340 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2341 rcu_read_unlock();
2342 }
2343 }
2344
2345 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2346 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2347 * Note: for send buffers, TCP works better if we can build two skbs at
2348 * minimum.
2349 */
2350 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2351
2352 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2353 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2354
sk_stream_moderate_sndbuf(struct sock * sk)2355 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2356 {
2357 u32 val;
2358
2359 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2360 return;
2361
2362 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2363
2364 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2365 }
2366
2367 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2368 bool force_schedule);
2369
2370 /**
2371 * sk_page_frag - return an appropriate page_frag
2372 * @sk: socket
2373 *
2374 * Use the per task page_frag instead of the per socket one for
2375 * optimization when we know that we're in process context and own
2376 * everything that's associated with %current.
2377 *
2378 * Both direct reclaim and page faults can nest inside other
2379 * socket operations and end up recursing into sk_page_frag()
2380 * while it's already in use: explicitly avoid task page_frag
2381 * usage if the caller is potentially doing any of them.
2382 * This assumes that page fault handlers use the GFP_NOFS flags.
2383 *
2384 * Return: a per task page_frag if context allows that,
2385 * otherwise a per socket one.
2386 */
sk_page_frag(struct sock * sk)2387 static inline struct page_frag *sk_page_frag(struct sock *sk)
2388 {
2389 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) ==
2390 (__GFP_DIRECT_RECLAIM | __GFP_FS))
2391 return ¤t->task_frag;
2392
2393 return &sk->sk_frag;
2394 }
2395
2396 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2397
2398 /*
2399 * Default write policy as shown to user space via poll/select/SIGIO
2400 */
sock_writeable(const struct sock * sk)2401 static inline bool sock_writeable(const struct sock *sk)
2402 {
2403 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2404 }
2405
gfp_any(void)2406 static inline gfp_t gfp_any(void)
2407 {
2408 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2409 }
2410
sock_rcvtimeo(const struct sock * sk,bool noblock)2411 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2412 {
2413 return noblock ? 0 : sk->sk_rcvtimeo;
2414 }
2415
sock_sndtimeo(const struct sock * sk,bool noblock)2416 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2417 {
2418 return noblock ? 0 : sk->sk_sndtimeo;
2419 }
2420
sock_rcvlowat(const struct sock * sk,int waitall,int len)2421 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2422 {
2423 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2424
2425 return v ?: 1;
2426 }
2427
2428 /* Alas, with timeout socket operations are not restartable.
2429 * Compare this to poll().
2430 */
sock_intr_errno(long timeo)2431 static inline int sock_intr_errno(long timeo)
2432 {
2433 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2434 }
2435
2436 struct sock_skb_cb {
2437 u32 dropcount;
2438 };
2439
2440 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2441 * using skb->cb[] would keep using it directly and utilize its
2442 * alignement guarantee.
2443 */
2444 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2445 sizeof(struct sock_skb_cb)))
2446
2447 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2448 SOCK_SKB_CB_OFFSET))
2449
2450 #define sock_skb_cb_check_size(size) \
2451 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2452
2453 static inline void
sock_skb_set_dropcount(const struct sock * sk,struct sk_buff * skb)2454 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2455 {
2456 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2457 atomic_read(&sk->sk_drops) : 0;
2458 }
2459
sk_drops_add(struct sock * sk,const struct sk_buff * skb)2460 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2461 {
2462 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2463
2464 atomic_add(segs, &sk->sk_drops);
2465 }
2466
sock_read_timestamp(struct sock * sk)2467 static inline ktime_t sock_read_timestamp(struct sock *sk)
2468 {
2469 #if BITS_PER_LONG==32
2470 unsigned int seq;
2471 ktime_t kt;
2472
2473 do {
2474 seq = read_seqbegin(&sk->sk_stamp_seq);
2475 kt = sk->sk_stamp;
2476 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2477
2478 return kt;
2479 #else
2480 return READ_ONCE(sk->sk_stamp);
2481 #endif
2482 }
2483
sock_write_timestamp(struct sock * sk,ktime_t kt)2484 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2485 {
2486 #if BITS_PER_LONG==32
2487 write_seqlock(&sk->sk_stamp_seq);
2488 sk->sk_stamp = kt;
2489 write_sequnlock(&sk->sk_stamp_seq);
2490 #else
2491 WRITE_ONCE(sk->sk_stamp, kt);
2492 #endif
2493 }
2494
2495 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2496 struct sk_buff *skb);
2497 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2498 struct sk_buff *skb);
2499
2500 static inline void
sock_recv_timestamp(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2501 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2502 {
2503 ktime_t kt = skb->tstamp;
2504 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2505
2506 /*
2507 * generate control messages if
2508 * - receive time stamping in software requested
2509 * - software time stamp available and wanted
2510 * - hardware time stamps available and wanted
2511 */
2512 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2513 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2514 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2515 (hwtstamps->hwtstamp &&
2516 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2517 __sock_recv_timestamp(msg, sk, skb);
2518 else
2519 sock_write_timestamp(sk, kt);
2520
2521 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2522 __sock_recv_wifi_status(msg, sk, skb);
2523 }
2524
2525 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2526 struct sk_buff *skb);
2527
2528 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
sock_recv_ts_and_drops(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2529 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2530 struct sk_buff *skb)
2531 {
2532 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2533 (1UL << SOCK_RCVTSTAMP))
2534 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2535 SOF_TIMESTAMPING_RAW_HARDWARE)
2536
2537 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2538 __sock_recv_ts_and_drops(msg, sk, skb);
2539 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2540 sock_write_timestamp(sk, skb->tstamp);
2541 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2542 sock_write_timestamp(sk, 0);
2543 }
2544
2545 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2546
2547 /**
2548 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2549 * @sk: socket sending this packet
2550 * @tsflags: timestamping flags to use
2551 * @tx_flags: completed with instructions for time stamping
2552 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2553 *
2554 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2555 */
_sock_tx_timestamp(struct sock * sk,__u16 tsflags,__u8 * tx_flags,__u32 * tskey)2556 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2557 __u8 *tx_flags, __u32 *tskey)
2558 {
2559 if (unlikely(tsflags)) {
2560 __sock_tx_timestamp(tsflags, tx_flags);
2561 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2562 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2563 *tskey = sk->sk_tskey++;
2564 }
2565 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2566 *tx_flags |= SKBTX_WIFI_STATUS;
2567 }
2568
sock_tx_timestamp(struct sock * sk,__u16 tsflags,__u8 * tx_flags)2569 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2570 __u8 *tx_flags)
2571 {
2572 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2573 }
2574
skb_setup_tx_timestamp(struct sk_buff * skb,__u16 tsflags)2575 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2576 {
2577 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2578 &skb_shinfo(skb)->tskey);
2579 }
2580
2581 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2582 /**
2583 * sk_eat_skb - Release a skb if it is no longer needed
2584 * @sk: socket to eat this skb from
2585 * @skb: socket buffer to eat
2586 *
2587 * This routine must be called with interrupts disabled or with the socket
2588 * locked so that the sk_buff queue operation is ok.
2589 */
sk_eat_skb(struct sock * sk,struct sk_buff * skb)2590 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2591 {
2592 __skb_unlink(skb, &sk->sk_receive_queue);
2593 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2594 !sk->sk_rx_skb_cache) {
2595 sk->sk_rx_skb_cache = skb;
2596 skb_orphan(skb);
2597 return;
2598 }
2599 __kfree_skb(skb);
2600 }
2601
2602 static inline
sock_net(const struct sock * sk)2603 struct net *sock_net(const struct sock *sk)
2604 {
2605 return read_pnet(&sk->sk_net);
2606 }
2607
2608 static inline
sock_net_set(struct sock * sk,struct net * net)2609 void sock_net_set(struct sock *sk, struct net *net)
2610 {
2611 write_pnet(&sk->sk_net, net);
2612 }
2613
2614 static inline bool
skb_sk_is_prefetched(struct sk_buff * skb)2615 skb_sk_is_prefetched(struct sk_buff *skb)
2616 {
2617 #ifdef CONFIG_INET
2618 return skb->destructor == sock_pfree;
2619 #else
2620 return false;
2621 #endif /* CONFIG_INET */
2622 }
2623
2624 /* This helper checks if a socket is a full socket,
2625 * ie _not_ a timewait or request socket.
2626 */
sk_fullsock(const struct sock * sk)2627 static inline bool sk_fullsock(const struct sock *sk)
2628 {
2629 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2630 }
2631
2632 static inline bool
sk_is_refcounted(struct sock * sk)2633 sk_is_refcounted(struct sock *sk)
2634 {
2635 /* Only full sockets have sk->sk_flags. */
2636 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2637 }
2638
2639 /**
2640 * skb_steal_sock - steal a socket from an sk_buff
2641 * @skb: sk_buff to steal the socket from
2642 * @refcounted: is set to true if the socket is reference-counted
2643 */
2644 static inline struct sock *
skb_steal_sock(struct sk_buff * skb,bool * refcounted)2645 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2646 {
2647 if (skb->sk) {
2648 struct sock *sk = skb->sk;
2649
2650 *refcounted = true;
2651 if (skb_sk_is_prefetched(skb))
2652 *refcounted = sk_is_refcounted(sk);
2653 skb->destructor = NULL;
2654 skb->sk = NULL;
2655 return sk;
2656 }
2657 *refcounted = false;
2658 return NULL;
2659 }
2660
2661 /* Checks if this SKB belongs to an HW offloaded socket
2662 * and whether any SW fallbacks are required based on dev.
2663 * Check decrypted mark in case skb_orphan() cleared socket.
2664 */
sk_validate_xmit_skb(struct sk_buff * skb,struct net_device * dev)2665 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2666 struct net_device *dev)
2667 {
2668 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2669 struct sock *sk = skb->sk;
2670
2671 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2672 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2673 #ifdef CONFIG_TLS_DEVICE
2674 } else if (unlikely(skb->decrypted)) {
2675 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2676 kfree_skb(skb);
2677 skb = NULL;
2678 #endif
2679 }
2680 #endif
2681
2682 return skb;
2683 }
2684
2685 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2686 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2687 */
sk_listener(const struct sock * sk)2688 static inline bool sk_listener(const struct sock *sk)
2689 {
2690 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2691 }
2692
2693 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2694 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2695 int type);
2696
2697 bool sk_ns_capable(const struct sock *sk,
2698 struct user_namespace *user_ns, int cap);
2699 bool sk_capable(const struct sock *sk, int cap);
2700 bool sk_net_capable(const struct sock *sk, int cap);
2701
2702 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2703
2704 /* Take into consideration the size of the struct sk_buff overhead in the
2705 * determination of these values, since that is non-constant across
2706 * platforms. This makes socket queueing behavior and performance
2707 * not depend upon such differences.
2708 */
2709 #define _SK_MEM_PACKETS 256
2710 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2711 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2712 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2713
2714 extern __u32 sysctl_wmem_max;
2715 extern __u32 sysctl_rmem_max;
2716
2717 extern int sysctl_tstamp_allow_data;
2718 extern int sysctl_optmem_max;
2719
2720 extern __u32 sysctl_wmem_default;
2721 extern __u32 sysctl_rmem_default;
2722
2723 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2724 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2725
sk_get_wmem0(const struct sock * sk,const struct proto * proto)2726 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2727 {
2728 /* Does this proto have per netns sysctl_wmem ? */
2729 if (proto->sysctl_wmem_offset)
2730 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2731
2732 return READ_ONCE(*proto->sysctl_wmem);
2733 }
2734
sk_get_rmem0(const struct sock * sk,const struct proto * proto)2735 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2736 {
2737 /* Does this proto have per netns sysctl_rmem ? */
2738 if (proto->sysctl_rmem_offset)
2739 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2740
2741 return READ_ONCE(*proto->sysctl_rmem);
2742 }
2743
2744 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2745 * Some wifi drivers need to tweak it to get more chunks.
2746 * They can use this helper from their ndo_start_xmit()
2747 */
sk_pacing_shift_update(struct sock * sk,int val)2748 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2749 {
2750 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2751 return;
2752 WRITE_ONCE(sk->sk_pacing_shift, val);
2753 }
2754
2755 /* if a socket is bound to a device, check that the given device
2756 * index is either the same or that the socket is bound to an L3
2757 * master device and the given device index is also enslaved to
2758 * that L3 master
2759 */
sk_dev_equal_l3scope(struct sock * sk,int dif)2760 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2761 {
2762 int mdif;
2763
2764 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2765 return true;
2766
2767 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2768 if (mdif && mdif == sk->sk_bound_dev_if)
2769 return true;
2770
2771 return false;
2772 }
2773
2774 void sock_def_readable(struct sock *sk);
2775
2776 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2777 void sock_enable_timestamps(struct sock *sk);
2778 void sock_no_linger(struct sock *sk);
2779 void sock_set_keepalive(struct sock *sk);
2780 void sock_set_priority(struct sock *sk, u32 priority);
2781 void sock_set_rcvbuf(struct sock *sk, int val);
2782 void sock_set_mark(struct sock *sk, u32 val);
2783 void sock_set_reuseaddr(struct sock *sk);
2784 void sock_set_reuseport(struct sock *sk);
2785 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2786
2787 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2788
2789 #endif /* _SOCK_H */
2790