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