include/net/sock.h

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Definitions for the AF_INET socket handler.
   7  *
   8  * Version:     @(#)sock.h      1.0.4   05/13/93
   9  *
  10  * Authors:     Ross Biro, <bir7@leland.Stanford.Edu>
  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/config.h>
  44 #include <linux/timer.h>
  45 #include <linux/cache.h>
  46 #include <linux/module.h>
  47 #include <linux/netdevice.h>
  48 #include <linux/skbuff.h>       /* struct sk_buff */
  49 #include <linux/security.h>
  50
  51 #include <linux/filter.h>
  52
  53 #include <asm/atomic.h>
  54 #include <net/dst.h>
  55
  56 /*
  57  * This structure really needs to be cleaned up.
  58  * Most of it is for TCP, and not used by any of
  59  * the other protocols.
  60  */
  61
  62 /* Sock flags */
  63 enum {
  64         SOCK_DEAD,
  65         SOCK_DONE,
  66         SOCK_URGINLINE,
  67         SOCK_KEEPOPEN,
  68         SOCK_LINGER,
  69         SOCK_DESTROY,
  70         SOCK_BROADCAST,
  71 };
  72
  73 /* Define this to get the sk->debug debugging facility. */
  74 #define SOCK_DEBUGGING
  75 #ifdef SOCK_DEBUGGING
  76 #define SOCK_DEBUG(sk, msg...) do { if((sk) && ((sk)->debug)) printk(KERN_DEBUG msg); } while (0)
  77 #else
  78 #define SOCK_DEBUG(sk, msg...) do { } while (0)
  79 #endif
  80
  81 /* This is the per-socket lock.  The spinlock provides a synchronization
  82  * between user contexts and software interrupt processing, whereas the
  83  * mini-semaphore synchronizes multiple users amongst themselves.
  84  */
  85 struct sock_iocb;
  86 typedef struct {
  87         spinlock_t              slock;
  88         struct sock_iocb        *owner;
  89         wait_queue_head_t       wq;
  90 } socket_lock_t;
  91
  92 #define sock_lock_init(__sk) \
  93 do {    spin_lock_init(&((__sk)->lock.slock)); \
  94         (__sk)->lock.owner = NULL; \
  95         init_waitqueue_head(&((__sk)->lock.wq)); \
  96 } while(0)
  97
  98 /**
  99   *     struct sock - network layer representation of sockets
 100   *     @state - Connection state
 101   *     @zapped - ax25 & ipx means !linked
 102   *     @reuse - %SO_REUSEADDR setting
 103   *     @shutdown - mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 104   *     @bound_dev_if - bound device index if != 0
 105   *     @next - main hash linkage for various protocol lookup tables
 106   *     @pprev - main hash linkage for various protocol lookup tables
 107   *     @bind_next - main hash linkage for various protocol lookup tables
 108   *     @bind_pprev - main hash linkage for various protocol lookup tables
 109   *     @refcnt - reference count
 110   *     @family - network address family
 111   *     @use_write_queue - wheter to call sk->write_space(sk) in sock_wfree
 112   *     @userlocks - %SO_SNDBUF and %SO_RCVBUF settings
 113   *     @lock - synchronizer
 114   *     @rcvbuf - size of receive buffer in bytes
 115   *     @sleep - sock wait queue
 116   *     @dst_cache - destination cache
 117   *     @dst_lock - destination cache lock
 118   *     @policy - flow policy
 119   *     @rmem_alloc - receive queue bytes committed
 120   *     @receive_queue - incoming packets
 121   *     @wmem_alloc - transmit queue bytes committed
 122   *     @write_queue - Packet sending queue
 123   *     @omem_alloc - "o" is "option" or "other"
 124   *     @wmem_queued - persistent queue size
 125   *     @forward_alloc - space allocated forward
 126   *     @allocation - allocation mode
 127   *     @sndbuf - size of send buffer in bytes
 128   *     @prev - pointer to previous sock in the list this sock is in
 129   *     @flags - %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, %SO_OOBINLINE settings
 130   *     @no_check - %SO_NO_CHECK setting, wether or not checkup packets
 131   *     @debug - %SO_DEBUG setting
 132   *     @rcvtstamp - %SO_TIMESTAMP setting
 133   *     @no_largesend - whether to sent large segments or not
 134   *     @route_caps - route capabilities (e.g. %NETIF_F_TSO)
 135   *     @lingertime - %SO_LINGER l_linger setting
 136   *     @hashent - hash entry in several tables (e.g. tcp_ehash)
 137   *     @pair - socket pair (e.g. AF_UNIX/unix_peer)
 138   *     @backlog - always used with the per-socket spinlock held
 139   *     @callback_lock - used with the callbacks in the end of this struct
 140   *     @error_queue - rarely used
 141   *     @prot - protocol handlers inside a network family
 142   *     @err - last error
 143   *     @err_soft - errors that don't cause failure but are the cause of a persistent failure not just 'timed out'
 144   *     @ack_backlog - current listen backlog
 145   *     @max_ack_backlog - listen backlog set in listen()
 146   *     @priority - %SO_PRIORITY setting
 147   *     @type - socket type (%SOCK_STREAM, etc)
 148   *     @localroute - route locally only, %SO_DONTROUTE setting
 149   *     @protocol - which protocol this socket belongs in this network family
 150   *     @peercred - %SO_PEERCRED setting
 151   *     @rcvlowat - %SO_RCVLOWAT setting
 152   *     @rcvtimeo - %SO_RCVTIMEO setting
 153   *     @sndtimeo - %SO_SNDTIMEO setting
 154   *     @filter - socket filtering instructions
 155   *     @protinfo - private area, net family specific, when not using slab
 156   *     @slab - the slabcache this instance was allocated from
 157   *     @timer - sock cleanup timer
 158   *     @stamp - time stamp of last packet received
 159   *     @socket - Identd and reporting IO signals
 160   *     @user_data - RPC layer private data
 161   *     @owner - module that owns this socket
 162   *     @state_change - callback to indicate change in the state of the sock
 163   *     @data_ready - callback to indicate there is data to be processed
 164   *     @write_space - callback to indicate there is bf sending space available
 165   *     @error_report - callback to indicate errors (e.g. %MSG_ERRQUEUE)
 166   *     @backlog_rcv - callback to process the backlog
 167   *     @destruct - called at sock freeing time, i.e. when all refcnt == 0
 168  */
 169 struct sock {
 170         /* Begin of struct sock/struct tcp_tw_bucket shared layout */
 171         volatile unsigned char  state,
 172                                 zapped;
 173         unsigned char           reuse;
 174         unsigned char           shutdown;
 175         int                     bound_dev_if;
 176         struct sock             *next;
 177         struct sock             **pprev;
 178         struct sock             *bind_next;
 179         struct sock             **bind_pprev;
 180         atomic_t                refcnt;
 181         unsigned short          family;
 182         /* End of struct sock/struct tcp_tw_bucket shared layout */
 183         unsigned char           use_write_queue;
 184         unsigned char           userlocks;
 185         socket_lock_t           lock;
 186         int                     rcvbuf;
 187         wait_queue_head_t       *sleep;
 188         struct dst_entry        *dst_cache;
 189         rwlock_t                dst_lock;
 190         struct xfrm_policy      *policy[2];
 191         atomic_t                rmem_alloc;
 192         struct sk_buff_head     receive_queue;
 193         atomic_t                wmem_alloc;
 194         struct sk_buff_head     write_queue;
 195         atomic_t                omem_alloc;
 196         int                     wmem_queued;
 197         int                     forward_alloc;
 198         unsigned int            allocation;
 199         int                     sndbuf;
 200         struct sock             *prev;
 201         unsigned long           flags;
 202         char                    no_check;
 203         unsigned char           debug;
 204         unsigned char           rcvtstamp;
 205         unsigned char           no_largesend;
 206         int                     route_caps;
 207         unsigned long           lingertime;
 208         int                     hashent;
 209         struct sock             *pair;
 210         /*
 211          * The backlog queue is special, it is always used with
 212          * the per-socket spinlock held and requires low latency
 213          * access. Therefore we special case it's implementation.
 214          */
 215         struct {
 216                 struct sk_buff *head;
 217                 struct sk_buff *tail;
 218         } backlog;
 219         rwlock_t                callback_lock;
 220         struct sk_buff_head     error_queue;
 221         struct proto            *prot;
 222         int                     err,
 223                                 err_soft;
 224         unsigned short          ack_backlog;
 225         unsigned short          max_ack_backlog;
 226         __u32                   priority;
 227         unsigned short          type;
 228         unsigned char           localroute;
 229         unsigned char           protocol;
 230         struct ucred            peercred;
 231         int                     rcvlowat;
 232         long                    rcvtimeo;
 233         long                    sndtimeo;
 234         struct sk_filter        *filter;
 235         void                    *protinfo;
 236         kmem_cache_t            *slab;
 237         struct timer_list       timer;
 238         struct timeval          stamp;
 239         struct socket           *socket;
 240         void                    *user_data;
 241         struct module           *owner;
 242         void                    (*state_change)(struct sock *sk);
 243         void                    (*data_ready)(struct sock *sk, int bytes);
 244         void                    (*write_space)(struct sock *sk);
 245         void                    (*error_report)(struct sock *sk);
 246         int                     (*backlog_rcv) (struct sock *sk,
 247                                                 struct sk_buff *skb);
 248         void                    (*destruct)(struct sock *sk);
 249 };
 250
 251 /* The per-socket spinlock must be held here. */
 252 #define sk_add_backlog(__sk, __skb)                     \
 253 do {    if((__sk)->backlog.tail == NULL) {              \
 254                 (__sk)->backlog.head =                  \
 255                      (__sk)->backlog.tail = (__skb);    \
 256         } else {                                        \
 257                 ((__sk)->backlog.tail)->next = (__skb); \
 258                 (__sk)->backlog.tail = (__skb);         \
 259         }                                               \
 260         (__skb)->next = NULL;                           \
 261 } while(0)
 262
 263 /* IP protocol blocks we attach to sockets.
 264  * socket layer -> transport layer interface
 265  * transport -> network interface is defined by struct inet_proto
 266  */
 267 struct proto {
 268         void                    (*close)(struct sock *sk,
 269                                         long timeout);
 270         int                     (*connect)(struct sock *sk,
 271                                         struct sockaddr *uaddr,
 272                                         int addr_len);
 273         int                     (*disconnect)(struct sock *sk, int flags);
 274
 275         struct sock *           (*accept) (struct sock *sk, int flags, int *err);
 276
 277         int                     (*ioctl)(struct sock *sk, int cmd,
 278                                          unsigned long arg);
 279         int                     (*init)(struct sock *sk);
 280         int                     (*destroy)(struct sock *sk);
 281         void                    (*shutdown)(struct sock *sk, int how);
 282         int                     (*setsockopt)(struct sock *sk, int level,
 283                                         int optname, char *optval, int optlen);
 284         int                     (*getsockopt)(struct sock *sk, int level,
 285                                         int optname, char *optval,
 286                                         int *option);
 287         int                     (*sendmsg)(struct kiocb *iocb, struct sock *sk,
 288                                            struct msghdr *msg, int len);
 289         int                     (*recvmsg)(struct kiocb *iocb, struct sock *sk,
 290                                            struct msghdr *msg,
 291                                         int len, int noblock, int flags,
 292                                         int *addr_len);
 293         int                     (*sendpage)(struct sock *sk, struct page *page,
 294                                         int offset, size_t size, int flags);
 295         int                     (*bind)(struct sock *sk,
 296                                         struct sockaddr *uaddr, int addr_len);
 297
 298         int                     (*backlog_rcv) (struct sock *sk,
 299                                                 struct sk_buff *skb);
 300
 301         /* Keeping track of sk's, looking them up, and port selection methods. */
 302         void                    (*hash)(struct sock *sk);
 303         void                    (*unhash)(struct sock *sk);
 304         int                     (*get_port)(struct sock *sk, unsigned short snum);
 305
 306         char                    name[32];
 307
 308         struct {
 309                 int inuse;
 310                 u8  __pad[SMP_CACHE_BYTES - sizeof(int)];
 311         } stats[NR_CPUS];
 312 };
 313
 314 static __inline__ void sk_set_owner(struct sock *sk, struct module *owner)
 315 {
 316         /*
 317          * One should use sk_set_owner just once, after struct sock creation,
 318          * be it shortly after sk_alloc or after a function that returns a new
 319          * struct sock (and that down the call chain called sk_alloc), e.g. the
 320          * IPv4 and IPv6 modules share tcp_create_openreq_child, so if
 321          * tcp_create_openreq_child called sk_set_owner IPv6 would have to
 322          * change the ownership of this struct sock, with one not needed
 323          * transient sk_set_owner call.
 324          */
 325         if (unlikely(sk->owner != NULL))
 326                 BUG();
 327         sk->owner = owner;
 328         __module_get(owner);
 329 }
 330
 331 /* Called with local bh disabled */
 332 static __inline__ void sock_prot_inc_use(struct proto *prot)
 333 {
 334         prot->stats[smp_processor_id()].inuse++;
 335 }
 336
 337 static __inline__ void sock_prot_dec_use(struct proto *prot)
 338 {
 339         prot->stats[smp_processor_id()].inuse--;
 340 }
 341
 342 /* About 10 seconds */
 343 #define SOCK_DESTROY_TIME (10*HZ)
 344
 345 /* Sockets 0-1023 can't be bound to unless you are superuser */
 346 #define PROT_SOCK       1024
 347
 348 #define SHUTDOWN_MASK   3
 349 #define RCV_SHUTDOWN    1
 350 #define SEND_SHUTDOWN   2
 351
 352 #define SOCK_SNDBUF_LOCK        1
 353 #define SOCK_RCVBUF_LOCK        2
 354 #define SOCK_BINDADDR_LOCK      4
 355 #define SOCK_BINDPORT_LOCK      8
 356
 357 /* sock_iocb: used to kick off async processing of socket ios */
 358 struct sock_iocb {
 359         struct list_head        list;
 360
 361         int                     flags;
 362         int                     size;
 363         struct socket           *sock;
 364         struct sock             *sk;
 365         struct scm_cookie       *scm;
 366         struct msghdr           *msg, async_msg;
 367         struct iovec            async_iov;
 368 };
 369
 370 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
 371 {
 372         BUG_ON(sizeof(struct sock_iocb) > KIOCB_PRIVATE_SIZE);
 373         return (struct sock_iocb *)iocb->private;
 374 }
 375
 376 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
 377 {
 378         return container_of((void *)si, struct kiocb, private);
 379 }
 380
 381 struct socket_alloc {
 382         struct socket socket;
 383         struct inode vfs_inode;
 384 };
 385
 386 static inline struct socket *SOCKET_I(struct inode *inode)
 387 {
 388         return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
 389 }
 390
 391 static inline struct inode *SOCK_INODE(struct socket *socket)
 392 {
 393         return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
 394 }
 395
 396 /* Used by processes to "lock" a socket state, so that
 397  * interrupts and bottom half handlers won't change it
 398  * from under us. It essentially blocks any incoming
 399  * packets, so that we won't get any new data or any
 400  * packets that change the state of the socket.
 401  *
 402  * While locked, BH processing will add new packets to
 403  * the backlog queue.  This queue is processed by the
 404  * owner of the socket lock right before it is released.
 405  *
 406  * Since ~2.3.5 it is also exclusive sleep lock serializing
 407  * accesses from user process context.
 408  */
 409 extern void __lock_sock(struct sock *sk);
 410 extern void __release_sock(struct sock *sk);
 411 #define sock_owned_by_user(sk)  (NULL != (sk)->lock.owner)
 412 #define lock_sock(__sk) \
 413 do {    might_sleep(); \
 414         spin_lock_bh(&((__sk)->lock.slock)); \
 415         if ((__sk)->lock.owner != NULL) \
 416                 __lock_sock(__sk); \
 417         (__sk)->lock.owner = (void *)1; \
 418         spin_unlock_bh(&((__sk)->lock.slock)); \
 419 } while(0)
 420
 421 #define release_sock(__sk) \
 422 do {    spin_lock_bh(&((__sk)->lock.slock)); \
 423         if ((__sk)->backlog.tail != NULL) \
 424                 __release_sock(__sk); \
 425         (__sk)->lock.owner = NULL; \
 426         if (waitqueue_active(&((__sk)->lock.wq))) wake_up(&((__sk)->lock.wq)); \
 427         spin_unlock_bh(&((__sk)->lock.slock)); \
 428 } while(0)
 429
 430 /* BH context may only use the following locking interface. */
 431 #define bh_lock_sock(__sk)      spin_lock(&((__sk)->lock.slock))
 432 #define bh_unlock_sock(__sk)    spin_unlock(&((__sk)->lock.slock))
 433
 434 extern struct sock *            sk_alloc(int family, int priority, int zero_it,
 435                                          kmem_cache_t *slab);
 436 extern void                     sk_free(struct sock *sk);
 437
 438 extern struct sk_buff           *sock_wmalloc(struct sock *sk,
 439                                               unsigned long size, int force,
 440                                               int priority);
 441 extern struct sk_buff           *sock_rmalloc(struct sock *sk,
 442                                               unsigned long size, int force,
 443                                               int priority);
 444 extern void                     sock_wfree(struct sk_buff *skb);
 445 extern void                     sock_rfree(struct sk_buff *skb);
 446
 447 extern int                      sock_setsockopt(struct socket *sock, int level,
 448                                                 int op, char __user *optval,
 449                                                 int optlen);
 450
 451 extern int                      sock_getsockopt(struct socket *sock, int level,
 452                                                 int op, char __user *optval,
 453                                                 int __user *optlen);
 454 extern struct sk_buff           *sock_alloc_send_skb(struct sock *sk,
 455                                                      unsigned long size,
 456                                                      int noblock,
 457                                                      int *errcode);
 458 extern struct sk_buff           *sock_alloc_send_pskb(struct sock *sk,
 459                                                       unsigned long header_len,
 460                                                       unsigned long data_len,
 461                                                       int noblock,
 462                                                       int *errcode);
 463 extern void *sock_kmalloc(struct sock *sk, int size, int priority);
 464 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
 465 extern void sk_send_sigurg(struct sock *sk);
 466
 467 /*
 468  * Functions to fill in entries in struct proto_ops when a protocol
 469  * does not implement a particular function.
 470  */
 471 extern int                      sock_no_release(struct socket *);
 472 extern int                      sock_no_bind(struct socket *,
 473                                              struct sockaddr *, int);
 474 extern int                      sock_no_connect(struct socket *,
 475                                                 struct sockaddr *, int, int);
 476 extern int                      sock_no_socketpair(struct socket *,
 477                                                    struct socket *);
 478 extern int                      sock_no_accept(struct socket *,
 479                                                struct socket *, int);
 480 extern int                      sock_no_getname(struct socket *,
 481                                                 struct sockaddr *, int *, int);
 482 extern unsigned int             sock_no_poll(struct file *, struct socket *,
 483                                              struct poll_table_struct *);
 484 extern int                      sock_no_ioctl(struct socket *, unsigned int,
 485                                               unsigned long);
 486 extern int                      sock_no_listen(struct socket *, int);
 487 extern int                      sock_no_shutdown(struct socket *, int);
 488 extern int                      sock_no_getsockopt(struct socket *, int , int,
 489                                                    char *, int *);
 490 extern int                      sock_no_setsockopt(struct socket *, int, int,
 491                                                    char *, int);
 492 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
 493                                                 struct msghdr *, int);
 494 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
 495                                                 struct msghdr *, int, int);
 496 extern int                      sock_no_mmap(struct file *file,
 497                                              struct socket *sock,
 498                                              struct vm_area_struct *vma);
 499 extern ssize_t                  sock_no_sendpage(struct socket *sock,
 500                                                 struct page *page,
 501                                                 int offset, size_t size,
 502                                                 int flags);
 503
 504 /*
 505  *      Default socket callbacks and setup code
 506  */
 507
 508 extern void sock_def_destruct(struct sock *);
 509
 510 /* Initialise core socket variables */
 511 extern void sock_init_data(struct socket *sock, struct sock *sk);
 512
 513 /**
 514  *      sk_filter - run a packet through a socket filter
 515  *      @sk: sock associated with &sk_buff
 516  *      @skb: buffer to filter
 517  *      @needlock: set to 1 if the sock is not locked by caller.
 518  *
 519  * Run the filter code and then cut skb->data to correct size returned by
 520  * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
 521  * than pkt_len we keep whole skb->data. This is the socket level
 522  * wrapper to sk_run_filter. It returns 0 if the packet should
 523  * be accepted or -EPERM if the packet should be tossed.
 524  *
 525  */
 526
 527 static inline int sk_filter(struct sock *sk, struct sk_buff *skb, int needlock)
 528 {
 529         int err;
 530
 531         err = security_sock_rcv_skb(sk, skb);
 532         if (err)
 533                 return err;
 534
 535         if (sk->filter) {
 536                 struct sk_filter *filter;
 537
 538                 if (needlock)
 539                         bh_lock_sock(sk);
 540
 541                 filter = sk->filter;
 542                 if (filter) {
 543                         int pkt_len = sk_run_filter(skb, filter->insns,
 544                                                     filter->len);
 545                         if (!pkt_len)
 546                                 err = -EPERM;
 547                         else
 548                                 skb_trim(skb, pkt_len);
 549                 }
 550
 551                 if (needlock)
 552                         bh_unlock_sock(sk);
 553         }
 554         return err;
 555 }
 556
 557 /**
 558  *      sk_filter_release: Release a socket filter
 559  *      @sk: socket
 560  *      @fp: filter to remove
 561  *
 562  *      Remove a filter from a socket and release its resources.
 563  */
 564
 565 static inline void sk_filter_release(struct sock *sk, struct sk_filter *fp)
 566 {
 567         unsigned int size = sk_filter_len(fp);
 568
 569         atomic_sub(size, &sk->omem_alloc);
 570
 571         if (atomic_dec_and_test(&fp->refcnt))
 572                 kfree(fp);
 573 }
 574
 575 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
 576 {
 577         atomic_inc(&fp->refcnt);
 578         atomic_add(sk_filter_len(fp), &sk->omem_alloc);
 579 }
 580
 581 /*
 582  * Socket reference counting postulates.
 583  *
 584  * * Each user of socket SHOULD hold a reference count.
 585  * * Each access point to socket (an hash table bucket, reference from a list,
 586  *   running timer, skb in flight MUST hold a reference count.
 587  * * When reference count hits 0, it means it will never increase back.
 588  * * When reference count hits 0, it means that no references from
 589  *   outside exist to this socket and current process on current CPU
 590  *   is last user and may/should destroy this socket.
 591  * * sk_free is called from any context: process, BH, IRQ. When
 592  *   it is called, socket has no references from outside -> sk_free
 593  *   may release descendant resources allocated by the socket, but
 594  *   to the time when it is called, socket is NOT referenced by any
 595  *   hash tables, lists etc.
 596  * * Packets, delivered from outside (from network or from another process)
 597  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
 598  *   when they sit in queue. Otherwise, packets will leak to hole, when
 599  *   socket is looked up by one cpu and unhasing is made by another CPU.
 600  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
 601  *   (leak to backlog). Packet socket does all the processing inside
 602  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
 603  *   use separate SMP lock, so that they are prone too.
 604  */
 605
 606 /* Grab socket reference count. This operation is valid only
 607    when sk is ALREADY grabbed f.e. it is found in hash table
 608    or a list and the lookup is made under lock preventing hash table
 609    modifications.
 610  */
 611
 612 static inline void sock_hold(struct sock *sk)
 613 {
 614         atomic_inc(&sk->refcnt);
 615 }
 616
 617 /* Ungrab socket in the context, which assumes that socket refcnt
 618    cannot hit zero, f.e. it is true in context of any socketcall.
 619  */
 620 static inline void __sock_put(struct sock *sk)
 621 {
 622         atomic_dec(&sk->refcnt);
 623 }
 624
 625 /* Ungrab socket and destroy it, if it was the last reference. */
 626 static inline void sock_put(struct sock *sk)
 627 {
 628         if (atomic_dec_and_test(&sk->refcnt))
 629                 sk_free(sk);
 630 }
 631
 632 /* Detach socket from process context.
 633  * Announce socket dead, detach it from wait queue and inode.
 634  * Note that parent inode held reference count on this struct sock,
 635  * we do not release it in this function, because protocol
 636  * probably wants some additional cleanups or even continuing
 637  * to work with this socket (TCP).
 638  */
 639 static inline void sock_orphan(struct sock *sk)
 640 {
 641         write_lock_bh(&sk->callback_lock);
 642         __set_bit(SOCK_DEAD, &sk->flags);
 643         sk->socket = NULL;
 644         sk->sleep = NULL;
 645         write_unlock_bh(&sk->callback_lock);
 646 }
 647
 648 static inline void sock_graft(struct sock *sk, struct socket *parent)
 649 {
 650         write_lock_bh(&sk->callback_lock);
 651         sk->sleep = &parent->wait;
 652         parent->sk = sk;
 653         sk->socket = parent;
 654         write_unlock_bh(&sk->callback_lock);
 655 }
 656
 657 static inline int sock_i_uid(struct sock *sk)
 658 {
 659         int uid;
 660
 661         read_lock(&sk->callback_lock);
 662         uid = sk->socket ? SOCK_INODE(sk->socket)->i_uid : 0;
 663         read_unlock(&sk->callback_lock);
 664         return uid;
 665 }
 666
 667 static inline unsigned long sock_i_ino(struct sock *sk)
 668 {
 669         unsigned long ino;
 670
 671         read_lock(&sk->callback_lock);
 672         ino = sk->socket ? SOCK_INODE(sk->socket)->i_ino : 0;
 673         read_unlock(&sk->callback_lock);
 674         return ino;
 675 }
 676
 677 static inline struct dst_entry *
 678 __sk_dst_get(struct sock *sk)
 679 {
 680         return sk->dst_cache;
 681 }
 682
 683 static inline struct dst_entry *
 684 sk_dst_get(struct sock *sk)
 685 {
 686         struct dst_entry *dst;
 687
 688         read_lock(&sk->dst_lock);
 689         dst = sk->dst_cache;
 690         if (dst)
 691                 dst_hold(dst);
 692         read_unlock(&sk->dst_lock);
 693         return dst;
 694 }
 695
 696 static inline void
 697 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
 698 {
 699         struct dst_entry *old_dst;
 700
 701         old_dst = sk->dst_cache;
 702         sk->dst_cache = dst;
 703         dst_release(old_dst);
 704 }
 705
 706 static inline void
 707 sk_dst_set(struct sock *sk, struct dst_entry *dst)
 708 {
 709         write_lock(&sk->dst_lock);
 710         __sk_dst_set(sk, dst);
 711         write_unlock(&sk->dst_lock);
 712 }
 713
 714 static inline void
 715 __sk_dst_reset(struct sock *sk)
 716 {
 717         struct dst_entry *old_dst;
 718
 719         old_dst = sk->dst_cache;
 720         sk->dst_cache = NULL;
 721         dst_release(old_dst);
 722 }
 723
 724 static inline void
 725 sk_dst_reset(struct sock *sk)
 726 {
 727         write_lock(&sk->dst_lock);
 728         __sk_dst_reset(sk);
 729         write_unlock(&sk->dst_lock);
 730 }
 731
 732 static inline struct dst_entry *
 733 __sk_dst_check(struct sock *sk, u32 cookie)
 734 {
 735         struct dst_entry *dst = sk->dst_cache;
 736
 737         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 738                 sk->dst_cache = NULL;
 739                 return NULL;
 740         }
 741
 742         return dst;
 743 }
 744
 745 static inline struct dst_entry *
 746 sk_dst_check(struct sock *sk, u32 cookie)
 747 {
 748         struct dst_entry *dst = sk_dst_get(sk);
 749
 750         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 751                 sk_dst_reset(sk);
 752                 return NULL;
 753         }
 754
 755         return dst;
 756 }
 757
 758
 759 /*
 760  *      Queue a received datagram if it will fit. Stream and sequenced
 761  *      protocols can't normally use this as they need to fit buffers in
 762  *      and play with them.
 763  *
 764  *      Inlined as it's very short and called for pretty much every
 765  *      packet ever received.
 766  */
 767
 768 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
 769 {
 770         sock_hold(sk);
 771         skb->sk = sk;
 772         skb->destructor = sock_wfree;
 773         atomic_add(skb->truesize, &sk->wmem_alloc);
 774 }
 775
 776 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
 777 {
 778         skb->sk = sk;
 779         skb->destructor = sock_rfree;
 780         atomic_add(skb->truesize, &sk->rmem_alloc);
 781 }
 782
 783 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 784 {
 785         int err = 0;
 786
 787         /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
 788            number of warnings when compiling with -W --ANK
 789          */
 790         if (atomic_read(&sk->rmem_alloc) + skb->truesize >= (unsigned)sk->rcvbuf) {
 791                 err = -ENOMEM;
 792                 goto out;
 793         }
 794
 795         /* It would be deadlock, if sock_queue_rcv_skb is used
 796            with socket lock! We assume that users of this
 797            function are lock free.
 798         */
 799         err = sk_filter(sk, skb, 1);
 800         if (err)
 801                 goto out;
 802
 803         skb->dev = NULL;
 804         skb_set_owner_r(skb, sk);
 805         skb_queue_tail(&sk->receive_queue, skb);
 806         if (!test_bit(SOCK_DEAD, &sk->flags))
 807                 sk->data_ready(sk,skb->len);
 808 out:
 809         return err;
 810 }
 811
 812 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
 813 {
 814         /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
 815            number of warnings when compiling with -W --ANK
 816          */
 817         if (atomic_read(&sk->rmem_alloc) + skb->truesize >= (unsigned)sk->rcvbuf)
 818                 return -ENOMEM;
 819         skb_set_owner_r(skb, sk);
 820         skb_queue_tail(&sk->error_queue,skb);
 821         if (!test_bit(SOCK_DEAD, &sk->flags))
 822                 sk->data_ready(sk,skb->len);
 823         return 0;
 824 }
 825
 826 /*
 827  *      Recover an error report and clear atomically
 828  */
 829
 830 static inline int sock_error(struct sock *sk)
 831 {
 832         int err=xchg(&sk->err,0);
 833         return -err;
 834 }
 835
 836 static inline unsigned long sock_wspace(struct sock *sk)
 837 {
 838         int amt = 0;
 839
 840         if (!(sk->shutdown & SEND_SHUTDOWN)) {
 841                 amt = sk->sndbuf - atomic_read(&sk->wmem_alloc);
 842                 if (amt < 0)
 843                         amt = 0;
 844         }
 845         return amt;
 846 }
 847
 848 static inline void sk_wake_async(struct sock *sk, int how, int band)
 849 {
 850         if (sk->socket && sk->socket->fasync_list)
 851                 sock_wake_async(sk->socket, how, band);
 852 }
 853
 854 #define SOCK_MIN_SNDBUF 2048
 855 #define SOCK_MIN_RCVBUF 256
 856
 857 /*
 858  *      Default write policy as shown to user space via poll/select/SIGIO
 859  */
 860 static inline int sock_writeable(struct sock *sk)
 861 {
 862         return atomic_read(&sk->wmem_alloc) < (sk->sndbuf / 2);
 863 }
 864
 865 static inline int gfp_any(void)
 866 {
 867         return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
 868 }
 869
 870 static inline long sock_rcvtimeo(struct sock *sk, int noblock)
 871 {
 872         return noblock ? 0 : sk->rcvtimeo;
 873 }
 874
 875 static inline long sock_sndtimeo(struct sock *sk, int noblock)
 876 {
 877         return noblock ? 0 : sk->sndtimeo;
 878 }
 879
 880 static inline int sock_rcvlowat(struct sock *sk, int waitall, int len)
 881 {
 882         return (waitall ? len : min_t(int, sk->rcvlowat, len)) ? : 1;
 883 }
 884
 885 /* Alas, with timeout socket operations are not restartable.
 886  * Compare this to poll().
 887  */
 888 static inline int sock_intr_errno(long timeo)
 889 {
 890         return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
 891 }
 892
 893 static __inline__ void
 894 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
 895 {
 896         if (sk->rcvtstamp)
 897                 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP, sizeof(skb->stamp), &skb->stamp);
 898         else
 899                 sk->stamp = skb->stamp;
 900 }
 901
 902 /*
 903  *      Enable debug/info messages
 904  */
 905
 906 #if 0
 907 #define NETDEBUG(x)     do { } while (0)
 908 #else
 909 #define NETDEBUG(x)     do { x; } while (0)
 910 #endif
 911
 912 /*
 913  * Macros for sleeping on a socket. Use them like this:
 914  *
 915  * SOCK_SLEEP_PRE(sk)
 916  * if (condition)
 917  *      schedule();
 918  * SOCK_SLEEP_POST(sk)
 919  *
 920  * N.B. These are now obsolete and were, afaik, only ever used in DECnet
 921  * and when the last use of them in DECnet has gone, I'm intending to
 922  * remove them.
 923  */
 924
 925 #define SOCK_SLEEP_PRE(sk)      { struct task_struct *tsk = current; \
 926                                 DECLARE_WAITQUEUE(wait, tsk); \
 927                                 tsk->state = TASK_INTERRUPTIBLE; \
 928                                 add_wait_queue((sk)->sleep, &wait); \
 929                                 release_sock(sk);
 930
 931 #define SOCK_SLEEP_POST(sk)     tsk->state = TASK_RUNNING; \
 932                                 remove_wait_queue((sk)->sleep, &wait); \
 933                                 lock_sock(sk); \
 934                                 }
 935
 936 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
 937 {
 938         if (valbool)
 939                 __set_bit(bit, &sk->flags);
 940         else
 941                 __clear_bit(bit, &sk->flags);
 942 }
 943
 944 extern __u32 sysctl_wmem_max;
 945 extern __u32 sysctl_rmem_max;
 946
 947 int siocdevprivate_ioctl(unsigned int fd, unsigned int cmd, unsigned long arg);
 948
 949 #endif  /* _SOCK_H */