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