net/socket.c

   1 /*
   2  * NET          An implementation of the SOCKET network access protocol.
   3  *
   4  * Version:     @(#)socket.c    1.1.93  18/02/95
   5  *
   6  * Authors:     Orest Zborowski, <obz@Kodak.COM>
   7  *              Ross Biro, <bir7@leland.Stanford.Edu>
   8  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
   9  *
  10  * Fixes:
  11  *              Anonymous       :       NOTSOCK/BADF cleanup. Error fix in
  12  *                                      shutdown()
  13  *              Alan Cox        :       verify_area() fixes
  14  *              Alan Cox        :       Removed DDI
  15  *              Jonathan Kamens :       SOCK_DGRAM reconnect bug
  16  *              Alan Cox        :       Moved a load of checks to the very
  17  *                                      top level.
  18  *              Alan Cox        :       Move address structures to/from user
  19  *                                      mode above the protocol layers.
  20  *              Rob Janssen     :       Allow 0 length sends.
  21  *              Alan Cox        :       Asynchronous I/O support (cribbed from the
  22  *                                      tty drivers).
  23  *              Niibe Yutaka    :       Asynchronous I/O for writes (4.4BSD style)
  24  *              Jeff Uphoff     :       Made max number of sockets command-line
  25  *                                      configurable.
  26  *              Matti Aarnio    :       Made the number of sockets dynamic,
  27  *                                      to be allocated when needed, and mr.
  28  *                                      Uphoff's max is used as max to be
  29  *                                      allowed to allocate.
  30  *              Linus           :       Argh. removed all the socket allocation
  31  *                                      altogether: it's in the inode now.
  32  *              Alan Cox        :       Made sock_alloc()/sock_release() public
  33  *                                      for NetROM and future kernel nfsd type
  34  *                                      stuff.
  35  *              Alan Cox        :       sendmsg/recvmsg basics.
  36  *              Tom Dyas        :       Export net symbols.
  37  *              Marcin Dalecki  :       Fixed problems with CONFIG_NET="n".
  38  *              Alan Cox        :       Added thread locking to sys_* calls
  39  *                                      for sockets. May have errors at the
  40  *                                      moment.
  41  *              Kevin Buhr      :       Fixed the dumb errors in the above.
  42  *              Andi Kleen      :       Some small cleanups, optimizations,
  43  *                                      and fixed a copy_from_user() bug.
  44  *              Tigran Aivazian :       sys_send(args) calls sys_sendto(args, NULL, 0)
  45  *
  46  *
  47  *              This program is free software; you can redistribute it and/or
  48  *              modify it under the terms of the GNU General Public License
  49  *              as published by the Free Software Foundation; either version
  50  *              2 of the License, or (at your option) any later version.
  51  *
  52  *
  53  *      This module is effectively the top level interface to the BSD socket
  54  *      paradigm.
  55  *
  56  */
  57
  58 #include <linux/config.h>
  59 #include <linux/mm.h>
  60 #include <linux/smp_lock.h>
  61 #include <linux/socket.h>
  62 #include <linux/file.h>
  63 #include <linux/net.h>
  64 #include <linux/interrupt.h>
  65 #include <linux/netdevice.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/firewall.h>
  68 #include <linux/wanrouter.h>
  69 #include <linux/init.h>
  70 #include <linux/poll.h>
  71
  72 #if defined(CONFIG_KMOD) && defined(CONFIG_NET)
  73 #include <linux/kmod.h>
  74 #endif
  75
  76 #include <asm/uaccess.h>
  77
  78 #include <linux/inet.h>
  79 #include <net/ip.h>
  80 #include <net/sock.h>
  81 #include <net/rarp.h>
  82 #include <net/tcp.h>
  83 #include <net/udp.h>
  84 #include <net/scm.h>
  85
  86 static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
  87 static long long sock_lseek(struct file *file, long long offset, int whence);
  88 static ssize_t sock_read(struct file *file, char *buf,
  89                          size_t size, loff_t *ppos);
  90 static ssize_t sock_write(struct file *file, const char *buf,
  91                           size_t size, loff_t *ppos);
  92
  93 static int sock_close(struct inode *inode, struct file *file);
  94 static unsigned int sock_poll(struct file *file,
  95                               struct poll_table_struct *wait);
  96 static int sock_ioctl(struct inode *inode, struct file *file,
  97                       unsigned int cmd, unsigned long arg);
  98 static int sock_fasync(int fd, struct file *filp, int on);
  99
 100
 101 /*
 102  *      Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
 103  *      in the operation structures but are done directly via the socketcall() multiplexor.
 104  */
 105
 106 static struct file_operations socket_file_ops = {
 107         sock_lseek,
 108         sock_read,
 109         sock_write,
 110         NULL,                   /* readdir */
 111         sock_poll,
 112         sock_ioctl,
 113         NULL,                   /* mmap */
 114         sock_no_open,           /* special open code to disallow open via /proc */
 115         NULL,                   /* flush */
 116         sock_close,
 117         NULL,                   /* no fsync */
 118         sock_fasync
 119 };
 120
 121 /*
 122  *      The protocol list. Each protocol is registered in here.
 123  */
 124
 125 struct net_proto_family *net_families[NPROTO];
 126
 127 /*
 128  *      Statistics counters of the socket lists
 129  */
 130
 131 static int sockets_in_use  = 0;
 132
 133 /*
 134  *      Socket hashing lock.
 135  */
 136 rwlock_t sockhash_lock = RW_LOCK_UNLOCKED;
 137
 138 /*
 139  *      Support routines. Move socket addresses back and forth across the kernel/user
 140  *      divide and look after the messy bits.
 141  */
 142
 143 #define MAX_SOCK_ADDR   128             /* 108 for Unix domain -
 144                                            16 for IP, 16 for IPX,
 145                                            24 for IPv6,
 146                                            about 80 for AX.25
 147                                            must be at least one bigger than
 148                                            the AF_UNIX size (see net/unix/af_unix.c
 149                                            :unix_mkname()).
 150                                          */
 151
 152 int move_addr_to_kernel(void *uaddr, int ulen, void *kaddr)
 153 {
 154         if(ulen<0||ulen>MAX_SOCK_ADDR)
 155                 return -EINVAL;
 156         if(ulen==0)
 157                 return 0;
 158         if(copy_from_user(kaddr,uaddr,ulen))
 159                 return -EFAULT;
 160         return 0;
 161 }
 162
 163 int move_addr_to_user(void *kaddr, int klen, void *uaddr, int *ulen)
 164 {
 165         int err;
 166         int len;
 167
 168         if((err=get_user(len, ulen)))
 169                 return err;
 170         if(len>klen)
 171                 len=klen;
 172         if(len<0 || len> MAX_SOCK_ADDR)
 173                 return -EINVAL;
 174         if(len)
 175         {
 176                 if(copy_to_user(uaddr,kaddr,len))
 177                         return -EFAULT;
 178         }
 179         /*
 180          *      "fromlen shall refer to the value before truncation.."
 181          *                      1003.1g
 182          */
 183         return __put_user(klen, ulen);
 184 }
 185
 186 /*
 187  *      Obtains the first available file descriptor and sets it up for use.
 188  */
 189
 190 static int get_fd(struct inode *inode)
 191 {
 192         int fd;
 193
 194         /*
 195          *      Find a file descriptor suitable for return to the user.
 196          */
 197
 198         fd = get_unused_fd();
 199         if (fd >= 0) {
 200                 struct file *file = get_empty_filp();
 201
 202                 if (!file) {
 203                         put_unused_fd(fd);
 204                         return -ENFILE;
 205                 }
 206
 207                 file->f_dentry = d_alloc_root(inode);
 208                 if (!file->f_dentry) {
 209                         put_filp(file);
 210                         put_unused_fd(fd);
 211                         return -ENOMEM;
 212                 }
 213
 214                 /*
 215                  * The socket maintains a reference to the inode, so we
 216                  * have to increment the count.
 217                  */
 218                 inode->i_count++;
 219
 220                 fd_install(fd, file);
 221                 file->f_op = &socket_file_ops;
 222                 file->f_mode = 3;
 223                 file->f_flags = O_RDWR;
 224                 file->f_pos = 0;
 225         }
 226         return fd;
 227 }
 228
 229 extern __inline__ struct socket *socki_lookup(struct inode *inode)
 230 {
 231         return &inode->u.socket_i;
 232 }
 233
 234 /*
 235  *      Go from a file number to its socket slot.
 236  */
 237
 238 extern struct socket *sockfd_lookup(int fd, int *err)
 239 {
 240         struct file *file;
 241         struct inode *inode;
 242         struct socket *sock;
 243
 244         if (!(file = fget(fd)))
 245         {
 246                 *err = -EBADF;
 247                 return NULL;
 248         }
 249
 250         inode = file->f_dentry->d_inode;
 251         if (!inode || !inode->i_sock || !(sock = socki_lookup(inode)))
 252         {
 253                 *err = -ENOTSOCK;
 254                 fput(file);
 255                 return NULL;
 256         }
 257
 258         if (sock->file != file) {
 259                 printk(KERN_ERR "socki_lookup: socket file changed!\n");
 260                 sock->file = file;
 261         }
 262         return sock;
 263 }
 264
 265 extern __inline__ void sockfd_put(struct socket *sock)
 266 {
 267         fput(sock->file);
 268 }
 269
 270 /*
 271  *      Allocate a socket.
 272  */
 273
 274 struct socket *sock_alloc(void)
 275 {
 276         struct inode * inode;
 277         struct socket * sock;
 278
 279         inode = get_empty_inode();
 280         if (!inode)
 281                 return NULL;
 282
 283         sock = socki_lookup(inode);
 284
 285         inode->i_mode = S_IFSOCK|S_IRWXUGO;
 286         inode->i_sock = 1;
 287         inode->i_uid = current->fsuid;
 288         inode->i_gid = current->fsgid;
 289
 290         sock->inode = inode;
 291         init_waitqueue_head(&sock->wait);
 292         sock->fasync_list = NULL;
 293         sock->state = SS_UNCONNECTED;
 294         sock->flags = 0;
 295         sock->ops = NULL;
 296         sock->sk = NULL;
 297         sock->file = NULL;
 298
 299         sockets_in_use++;
 300         return sock;
 301 }
 302
 303 /*
 304  *      In theory you can't get an open on this inode, but /proc provides
 305  *      a back door. Remember to keep it shut otherwise you'll let the
 306  *      creepy crawlies in.
 307  */
 308
 309 static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
 310 {
 311         return -ENXIO;
 312 }
 313
 314 void sock_release(struct socket *sock)
 315 {
 316         if (sock->state != SS_UNCONNECTED)
 317                 sock->state = SS_DISCONNECTING;
 318
 319         if (sock->ops)
 320                 sock->ops->release(sock, NULL);
 321
 322         if (sock->fasync_list)
 323                 printk(KERN_ERR "sock_release: fasync list not empty!\n");
 324
 325         --sockets_in_use;       /* Bookkeeping.. */
 326         sock->file=NULL;
 327         iput(sock->inode);
 328 }
 329
 330 int sock_sendmsg(struct socket *sock, struct msghdr *msg, int size)
 331 {
 332         int err;
 333         struct scm_cookie scm;
 334
 335         err = scm_send(sock, msg, &scm);
 336         if (err >= 0) {
 337                 err = sock->ops->sendmsg(sock, msg, size, &scm);
 338                 scm_destroy(&scm);
 339         }
 340         return err;
 341 }
 342
 343 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int size, int flags)
 344 {
 345         struct scm_cookie scm;
 346
 347         memset(&scm, 0, sizeof(scm));
 348
 349         size = sock->ops->recvmsg(sock, msg, size, flags, &scm);
 350         if (size >= 0)
 351                 scm_recv(sock, msg, &scm, flags);
 352
 353         return size;
 354 }
 355
 356
 357 /*
 358  *      Sockets are not seekable.
 359  */
 360
 361 static long long sock_lseek(struct file *file,long long offset, int whence)
 362 {
 363         return -ESPIPE;
 364 }
 365
 366 /*
 367  *      Read data from a socket. ubuf is a user mode pointer. We make sure the user
 368  *      area ubuf...ubuf+size-1 is writable before asking the protocol.
 369  */
 370
 371 static ssize_t sock_read(struct file *file, char *ubuf,
 372                          size_t size, loff_t *ppos)
 373 {
 374         struct socket *sock;
 375         struct iovec iov;
 376         struct msghdr msg;
 377
 378         if (ppos != &file->f_pos)
 379                 return -ESPIPE;
 380         if (size==0)            /* Match SYS5 behaviour */
 381                 return 0;
 382
 383         sock = socki_lookup(file->f_dentry->d_inode);
 384
 385         msg.msg_name=NULL;
 386         msg.msg_namelen=0;
 387         msg.msg_iov=&iov;
 388         msg.msg_iovlen=1;
 389         msg.msg_control=NULL;
 390         msg.msg_controllen=0;
 391         iov.iov_base=ubuf;
 392         iov.iov_len=size;
 393
 394         return sock_recvmsg(sock, &msg, size,
 395                             !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT);
 396 }
 397
 398
 399 /*
 400  *      Write data to a socket. We verify that the user area ubuf..ubuf+size-1
 401  *      is readable by the user process.
 402  */
 403
 404 static ssize_t sock_write(struct file *file, const char *ubuf,
 405                           size_t size, loff_t *ppos)
 406 {
 407         struct socket *sock;
 408         struct msghdr msg;
 409         struct iovec iov;
 410
 411         if (ppos != &file->f_pos)
 412                 return -ESPIPE;
 413         if(size==0)             /* Match SYS5 behaviour */
 414                 return 0;
 415
 416         sock = socki_lookup(file->f_dentry->d_inode);
 417
 418         msg.msg_name=NULL;
 419         msg.msg_namelen=0;
 420         msg.msg_iov=&iov;
 421         msg.msg_iovlen=1;
 422         msg.msg_control=NULL;
 423         msg.msg_controllen=0;
 424         msg.msg_flags=!(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
 425         iov.iov_base=(void *)ubuf;
 426         iov.iov_len=size;
 427
 428         return sock_sendmsg(sock, &msg, size);
 429 }
 430
 431 int sock_readv_writev(int type, struct inode * inode, struct file * file,
 432                       const struct iovec * iov, long count, long size)
 433 {
 434         struct msghdr msg;
 435         struct socket *sock;
 436
 437         sock = socki_lookup(inode);
 438
 439         msg.msg_name = NULL;
 440         msg.msg_namelen = 0;
 441         msg.msg_control = NULL;
 442         msg.msg_controllen = 0;
 443         msg.msg_iov = (struct iovec *) iov;
 444         msg.msg_iovlen = count;
 445         msg.msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
 446
 447         /* read() does a VERIFY_WRITE */
 448         if (type == VERIFY_WRITE)
 449                 return sock_recvmsg(sock, &msg, size, msg.msg_flags);
 450         return sock_sendmsg(sock, &msg, size);
 451 }
 452
 453
 454 /*
 455  *      With an ioctl arg may well be a user mode pointer, but we don't know what to do
 456  *      with it - that's up to the protocol still.
 457  */
 458
 459 int sock_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
 460            unsigned long arg)
 461 {
 462         struct socket *sock = socki_lookup(inode);
 463         return sock->ops->ioctl(sock, cmd, arg);
 464 }
 465
 466
 467 static unsigned int sock_poll(struct file *file, poll_table * wait)
 468 {
 469         struct socket *sock;
 470
 471         sock = socki_lookup(file->f_dentry->d_inode);
 472
 473         /*
 474          *      We can't return errors to poll, so it's either yes or no.
 475          */
 476
 477         return sock->ops->poll(file, sock, wait);
 478 }
 479
 480
 481 int sock_close(struct inode *inode, struct file *filp)
 482 {
 483         /*
 484          *      It was possible the inode is NULL we were
 485          *      closing an unfinished socket.
 486          */
 487
 488         if (!inode)
 489         {
 490                 printk(KERN_DEBUG "sock_close: NULL inode\n");
 491                 return 0;
 492         }
 493         sock_fasync(-1, filp, 0);
 494         sock_release(socki_lookup(inode));
 495         return 0;
 496 }
 497
 498 /*
 499  *      Update the socket async list
 500  */
 501
 502 static int sock_fasync(int fd, struct file *filp, int on)
 503 {
 504         struct fasync_struct *fa, *fna=NULL, **prev;
 505         struct socket *sock;
 506
 507         if (on)
 508         {
 509                 fna=(struct fasync_struct *)kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
 510                 if(fna==NULL)
 511                         return -ENOMEM;
 512         }
 513
 514         sock = socki_lookup(filp->f_dentry->d_inode);
 515
 516         prev=&(sock->fasync_list);
 517
 518         lock_sock(sock->sk);
 519
 520         for (fa=*prev; fa!=NULL; prev=&fa->fa_next,fa=*prev)
 521                 if (fa->fa_file==filp)
 522                         break;
 523
 524         if(on)
 525         {
 526                 if(fa!=NULL)
 527                 {
 528                         fa->fa_fd=fd;
 529                         kfree_s(fna,sizeof(struct fasync_struct));
 530                         release_sock(sock->sk);
 531                         return 0;
 532                 }
 533                 fna->fa_file=filp;
 534                 fna->fa_fd=fd;
 535                 fna->magic=FASYNC_MAGIC;
 536                 fna->fa_next=sock->fasync_list;
 537                 sock->fasync_list=fna;
 538         }
 539         else
 540         {
 541                 if (fa!=NULL)
 542                 {
 543                         *prev=fa->fa_next;
 544                         kfree_s(fa,sizeof(struct fasync_struct));
 545                 }
 546         }
 547
 548         release_sock(sock->sk);
 549         return 0;
 550 }
 551
 552 int sock_wake_async(struct socket *sock, int how)
 553 {
 554         if (!sock || !sock->fasync_list)
 555                 return -1;
 556         switch (how)
 557         {
 558         case 1:
 559                 if (sock->flags & SO_WAITDATA)
 560                         break;
 561                 goto call_kill;
 562         case 2:
 563                 if (!(sock->flags & SO_NOSPACE))
 564                         break;
 565                 sock->flags &= ~SO_NOSPACE;
 566                 /* fall through */
 567         case 0:
 568         call_kill:
 569                 if(sock->fasync_list != NULL)
 570                         kill_fasync(sock->fasync_list, SIGIO);
 571                 break;
 572         }
 573         return 0;
 574 }
 575
 576
 577 int sock_create(int family, int type, int protocol, struct socket **res)
 578 {
 579         int i;
 580         struct socket *sock;
 581
 582         /*
 583          *      Check protocol is in range
 584          */
 585         if(family<0||family>=NPROTO)
 586                 return -EINVAL;
 587
 588 #if defined(CONFIG_KMOD) && defined(CONFIG_NET)
 589         /* Attempt to load a protocol module if the find failed.
 590          *
 591          * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
 592          * requested real, full-featured networking support upon configuration.
 593          * Otherwise module support will break!
 594          */
 595         if (net_families[family]==NULL)
 596         {
 597                 char module_name[30];
 598                 sprintf(module_name,"net-pf-%d",family);
 599                 request_module(module_name);
 600         }
 601 #endif
 602
 603         if (net_families[family]==NULL)
 604                 return -EINVAL;
 605
 606 /*
 607  *      Check that this is a type that we know how to manipulate and
 608  *      the protocol makes sense here. The family can still reject the
 609  *      protocol later.
 610  */
 611
 612         if ((type != SOCK_STREAM && type != SOCK_DGRAM &&
 613              type != SOCK_SEQPACKET && type != SOCK_RAW && type != SOCK_RDM &&
 614 #ifdef CONFIG_XTP
 615                 type != SOCK_WEB  &&
 616 #endif
 617              type != SOCK_PACKET) || protocol < 0)
 618                         return -EINVAL;
 619
 620 /*
 621  *      Allocate the socket and allow the family to set things up. if
 622  *      the protocol is 0, the family is instructed to select an appropriate
 623  *      default.
 624  */
 625
 626         if (!(sock = sock_alloc()))
 627         {
 628                 printk(KERN_WARNING "socket: no more sockets\n");
 629                 return -ENFILE;         /* Not exactly a match, but its the
 630                                            closest posix thing */
 631         }
 632
 633         sock->type   = type;
 634
 635         if ((i = net_families[family]->create(sock, protocol)) < 0)
 636         {
 637                 sock_release(sock);
 638                 return i;
 639         }
 640
 641         *res = sock;
 642         return 0;
 643 }
 644
 645 asmlinkage int sys_socket(int family, int type, int protocol)
 646 {
 647         int retval;
 648         struct socket *sock;
 649
 650         lock_kernel();
 651
 652         retval = sock_create(family, type, protocol, &sock);
 653         if (retval < 0)
 654                 goto out;
 655
 656         retval = get_fd(sock->inode);
 657         if (retval < 0)
 658                 goto out_release;
 659         sock->file = fcheck(retval);
 660
 661 out:
 662         unlock_kernel();
 663         return retval;
 664
 665 out_release:
 666         sock_release(sock);
 667         goto out;
 668 }
 669
 670 /*
 671  *      Create a pair of connected sockets.
 672  */
 673
 674 asmlinkage int sys_socketpair(int family, int type, int protocol, int usockvec[2])
 675 {
 676         struct socket *sock1, *sock2;
 677         int fd1, fd2, err;
 678
 679         lock_kernel();
 680
 681         /*
 682          * Obtain the first socket and check if the underlying protocol
 683          * supports the socketpair call.
 684          */
 685
 686         err = sys_socket(family, type, protocol);
 687         if (err < 0)
 688                 goto out;
 689         fd1 = err;
 690
 691         /*
 692          * Now grab another socket
 693          */
 694         err = -EINVAL;
 695         fd2 = sys_socket(family, type, protocol);
 696         if (fd2 < 0)
 697                 goto out_close1;
 698
 699         /*
 700          * Get the sockets for the two fd's
 701          */
 702         sock1 = sockfd_lookup(fd1, &err);
 703         if (!sock1)
 704                 goto out_close2;
 705         sock2 = sockfd_lookup(fd2, &err);
 706         if (!sock2)
 707                 goto out_put1;
 708
 709         /* try to connect the two sockets together */
 710         err = sock1->ops->socketpair(sock1, sock2);
 711         if (err < 0)
 712                 goto out_put2;
 713
 714         err = put_user(fd1, &usockvec[0]);
 715         if (err)
 716                 goto out_put2;
 717         err = put_user(fd2, &usockvec[1]);
 718
 719 out_put2:
 720         sockfd_put(sock2);
 721 out_put1:
 722         sockfd_put(sock1);
 723
 724         if (err) {
 725         out_close2:
 726                 sys_close(fd2);
 727         out_close1:
 728                 sys_close(fd1);
 729         }
 730 out:
 731         unlock_kernel();
 732         return err;
 733 }
 734
 735
 736 /*
 737  *      Bind a name to a socket. Nothing much to do here since it's
 738  *      the protocol's responsibility to handle the local address.
 739  *
 740  *      We move the socket address to kernel space before we call
 741  *      the protocol layer (having also checked the address is ok).
 742  */
 743
 744 asmlinkage int sys_bind(int fd, struct sockaddr *umyaddr, int addrlen)
 745 {
 746         struct socket *sock;
 747         char address[MAX_SOCK_ADDR];
 748         int err;
 749
 750         lock_kernel();
 751         if((sock = sockfd_lookup(fd,&err))!=NULL)
 752         {
 753                 if((err=move_addr_to_kernel(umyaddr,addrlen,address))>=0)
 754                         err = sock->ops->bind(sock, (struct sockaddr *)address, addrlen);
 755                 sockfd_put(sock);
 756         }
 757         unlock_kernel();
 758         return err;
 759 }
 760
 761
 762 /*
 763  *      Perform a listen. Basically, we allow the protocol to do anything
 764  *      necessary for a listen, and if that works, we mark the socket as
 765  *      ready for listening.
 766  */
 767
 768 asmlinkage int sys_listen(int fd, int backlog)
 769 {
 770         struct socket *sock;
 771         int err;
 772
 773         lock_kernel();
 774         if((sock = sockfd_lookup(fd, &err))!=NULL)
 775         {
 776                 err=sock->ops->listen(sock, backlog);
 777                 sockfd_put(sock);
 778         }
 779         unlock_kernel();
 780         return err;
 781 }
 782
 783
 784 /*
 785  *      For accept, we attempt to create a new socket, set up the link
 786  *      with the client, wake up the client, then return the new
 787  *      connected fd. We collect the address of the connector in kernel
 788  *      space and move it to user at the very end. This is unclean because
 789  *      we open the socket then return an error.
 790  *
 791  *      1003.1g adds the ability to recvmsg() to query connection pending
 792  *      status to recvmsg. We need to add that support in a way thats
 793  *      clean when we restucture accept also.
 794  */
 795
 796 asmlinkage int sys_accept(int fd, struct sockaddr *upeer_sockaddr, int *upeer_addrlen)
 797 {
 798         struct inode *inode;
 799         struct socket *sock, *newsock;
 800         int err, len;
 801         char address[MAX_SOCK_ADDR];
 802
 803         lock_kernel();
 804         sock = sockfd_lookup(fd, &err);
 805         if (!sock)
 806                 goto out;
 807
 808 restart:
 809         err = -EMFILE;
 810         if (!(newsock = sock_alloc()))
 811                 goto out_put;
 812
 813         inode = newsock->inode;
 814         newsock->type = sock->type;
 815
 816         err = sock->ops->dup(newsock, sock);
 817         if (err < 0)
 818                 goto out_release;
 819
 820         err = newsock->ops->accept(sock, newsock, sock->file->f_flags);
 821         if (err < 0)
 822                 goto out_release;
 823         newsock = socki_lookup(inode);
 824
 825         if ((err = get_fd(inode)) < 0)
 826                 goto out_release;
 827         newsock->file = fcheck(err);
 828
 829         if (upeer_sockaddr)
 830         {
 831                 /* Handle the race where the accept works and we
 832                    then getname after it has closed again */
 833                 if(newsock->ops->getname(newsock, (struct sockaddr *)address, &len, 1)<0)
 834                 {
 835                         sys_close(err);
 836                         goto restart;
 837                 }
 838                 /* N.B. Should check for errors here */
 839                 move_addr_to_user(address, len, upeer_sockaddr, upeer_addrlen);
 840         }
 841
 842 out_put:
 843         sockfd_put(sock);
 844 out:
 845         unlock_kernel();
 846         return err;
 847
 848 out_release:
 849         sock_release(newsock);
 850         goto out_put;
 851 }
 852
 853
 854 /*
 855  *      Attempt to connect to a socket with the server address.  The address
 856  *      is in user space so we verify it is OK and move it to kernel space.
 857  *
 858  *      For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
 859  *      break bindings
 860  *
 861  *      NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
 862  *      other SEQPACKET protocols that take time to connect() as it doesn't
 863  *      include the -EINPROGRESS status for such sockets.
 864  */
 865
 866 asmlinkage int sys_connect(int fd, struct sockaddr *uservaddr, int addrlen)
 867 {
 868         struct socket *sock;
 869         char address[MAX_SOCK_ADDR];
 870         int err;
 871
 872         lock_kernel();
 873         sock = sockfd_lookup(fd, &err);
 874         if (!sock)
 875                 goto out;
 876         err = move_addr_to_kernel(uservaddr, addrlen, address);
 877         if (err < 0)
 878                 goto out_put;
 879         err = sock->ops->connect(sock, (struct sockaddr *) address, addrlen,
 880                                  sock->file->f_flags);
 881 out_put:
 882         sockfd_put(sock);
 883 out:
 884         unlock_kernel();
 885         return err;
 886 }
 887
 888 /*
 889  *      Get the local address ('name') of a socket object. Move the obtained
 890  *      name to user space.
 891  */
 892
 893 asmlinkage int sys_getsockname(int fd, struct sockaddr *usockaddr, int *usockaddr_len)
 894 {
 895         struct socket *sock;
 896         char address[MAX_SOCK_ADDR];
 897         int len, err;
 898
 899         lock_kernel();
 900         sock = sockfd_lookup(fd, &err);
 901         if (!sock)
 902                 goto out;
 903         err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 0);
 904         if (err)
 905                 goto out_put;
 906         err = move_addr_to_user(address, len, usockaddr, usockaddr_len);
 907
 908 out_put:
 909         sockfd_put(sock);
 910 out:
 911         unlock_kernel();
 912         return err;
 913 }
 914
 915 /*
 916  *      Get the remote address ('name') of a socket object. Move the obtained
 917  *      name to user space.
 918  */
 919
 920 asmlinkage int sys_getpeername(int fd, struct sockaddr *usockaddr, int *usockaddr_len)
 921 {
 922         struct socket *sock;
 923         char address[MAX_SOCK_ADDR];
 924         int len, err;
 925
 926         lock_kernel();
 927         if ((sock = sockfd_lookup(fd, &err))!=NULL)
 928         {
 929                 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 1);
 930                 if (!err)
 931                         err=move_addr_to_user(address,len, usockaddr, usockaddr_len);
 932                 sockfd_put(sock);
 933         }
 934         unlock_kernel();
 935         return err;
 936 }
 937
 938 /*
 939  *      Send a datagram to a given address. We move the address into kernel
 940  *      space and check the user space data area is readable before invoking
 941  *      the protocol.
 942  */
 943
 944 asmlinkage int sys_sendto(int fd, void * buff, size_t len, unsigned flags,
 945            struct sockaddr *addr, int addr_len)
 946 {
 947         struct socket *sock;
 948         char address[MAX_SOCK_ADDR];
 949         int err;
 950         struct msghdr msg;
 951         struct iovec iov;
 952
 953         sock = sockfd_lookup(fd, &err);
 954         if (!sock)
 955                 goto out;
 956         iov.iov_base=buff;
 957         iov.iov_len=len;
 958         msg.msg_name=NULL;
 959         msg.msg_iov=&iov;
 960         msg.msg_iovlen=1;
 961         msg.msg_control=NULL;
 962         msg.msg_controllen=0;
 963         msg.msg_namelen=addr_len;
 964         if(addr)
 965         {
 966                 err = move_addr_to_kernel(addr, addr_len, address);
 967                 if (err < 0)
 968                         goto out_put;
 969                 msg.msg_name=address;
 970         }
 971         if (sock->file->f_flags & O_NONBLOCK)
 972                 flags |= MSG_DONTWAIT;
 973         msg.msg_flags = flags;
 974         err = sock_sendmsg(sock, &msg, len);
 975
 976 out_put:
 977         sockfd_put(sock);
 978 out:
 979         return err;
 980 }
 981
 982 /*
 983  *      Send a datagram down a socket.
 984  */
 985
 986 asmlinkage int sys_send(int fd, void * buff, size_t len, unsigned flags)
 987 {
 988         return sys_sendto(fd, buff, len, flags, NULL, 0);
 989 }
 990
 991 /*
 992  *      Receive a frame from the socket and optionally record the address of the
 993  *      sender. We verify the buffers are writable and if needed move the
 994  *      sender address from kernel to user space.
 995  */
 996
 997 asmlinkage int sys_recvfrom(int fd, void * ubuf, size_t size, unsigned flags,
 998              struct sockaddr *addr, int *addr_len)
 999 {
1000         struct socket *sock;
1001         struct iovec iov;
1002         struct msghdr msg;
1003         char address[MAX_SOCK_ADDR];
1004         int err,err2;
1005
1006         sock = sockfd_lookup(fd, &err);
1007         if (!sock)
1008                 goto out;
1009
1010         msg.msg_control=NULL;
1011         msg.msg_controllen=0;
1012         msg.msg_iovlen=1;
1013         msg.msg_iov=&iov;
1014         iov.iov_len=size;
1015         iov.iov_base=ubuf;
1016         msg.msg_name=address;
1017         msg.msg_namelen=MAX_SOCK_ADDR;
1018         if (sock->file->f_flags & O_NONBLOCK)
1019                 flags |= MSG_DONTWAIT;
1020         err=sock_recvmsg(sock, &msg, size, flags);
1021
1022         if(err >= 0 && addr != NULL)
1023         {
1024                 err2=move_addr_to_user(address, msg.msg_namelen, addr, addr_len);
1025                 if(err2<0)
1026                         err=err2;
1027         }
1028         sockfd_put(sock);
1029 out:
1030         return err;
1031 }
1032
1033 /*
1034  *      Receive a datagram from a socket.
1035  */
1036
1037 asmlinkage int sys_recv(int fd, void * ubuf, size_t size, unsigned flags)
1038 {
1039         return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1040 }
1041
1042 /*
1043  *      Set a socket option. Because we don't know the option lengths we have
1044  *      to pass the user mode parameter for the protocols to sort out.
1045  */
1046
1047 asmlinkage int sys_setsockopt(int fd, int level, int optname, char *optval, int optlen)
1048 {
1049         int err;
1050         struct socket *sock;
1051
1052         lock_kernel();
1053         if ((sock = sockfd_lookup(fd, &err))!=NULL)
1054         {
1055                 if (level == SOL_SOCKET)
1056                         err=sock_setsockopt(sock,level,optname,optval,optlen);
1057                 else
1058                         err=sock->ops->setsockopt(sock, level, optname, optval, optlen);
1059                 sockfd_put(sock);
1060         }
1061         unlock_kernel();
1062         return err;
1063 }
1064
1065 /*
1066  *      Get a socket option. Because we don't know the option lengths we have
1067  *      to pass a user mode parameter for the protocols to sort out.
1068  */
1069
1070 asmlinkage int sys_getsockopt(int fd, int level, int optname, char *optval, int *optlen)
1071 {
1072         int err;
1073         struct socket *sock;
1074
1075         lock_kernel();
1076         if ((sock = sockfd_lookup(fd, &err))!=NULL)
1077         {
1078                 if (level == SOL_SOCKET)
1079                         err=sock_getsockopt(sock,level,optname,optval,optlen);
1080                 else
1081                         err=sock->ops->getsockopt(sock, level, optname, optval, optlen);
1082                 sockfd_put(sock);
1083         }
1084         unlock_kernel();
1085         return err;
1086 }
1087
1088
1089 /*
1090  *      Shutdown a socket.
1091  */
1092
1093 asmlinkage int sys_shutdown(int fd, int how)
1094 {
1095         int err;
1096         struct socket *sock;
1097
1098         lock_kernel();
1099         if ((sock = sockfd_lookup(fd, &err))!=NULL)
1100         {
1101                 err=sock->ops->shutdown(sock, how);
1102                 sockfd_put(sock);
1103         }
1104         unlock_kernel();
1105         return err;
1106 }
1107
1108 /*
1109  *      BSD sendmsg interface
1110  */
1111
1112 asmlinkage int sys_sendmsg(int fd, struct msghdr *msg, unsigned flags)
1113 {
1114         struct socket *sock;
1115         char address[MAX_SOCK_ADDR];
1116         struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1117         unsigned char ctl[sizeof(struct cmsghdr) + 20]; /* 20 is size of ipv6_pktinfo */
1118         unsigned char *ctl_buf = ctl;
1119         struct msghdr msg_sys;
1120         int err, ctl_len, iov_size, total_len;
1121
1122         err = -EFAULT;
1123         if (copy_from_user(&msg_sys,msg,sizeof(struct msghdr)))
1124                 goto out;
1125
1126         sock = sockfd_lookup(fd, &err);
1127         if (!sock)
1128                 goto out;
1129
1130         /* do not move before msg_sys is valid */
1131         err = -EINVAL;
1132         if (msg_sys.msg_iovlen > UIO_MAXIOV)
1133                 goto out_put;
1134
1135         /* Check whether to allocate the iovec area*/
1136         err = -ENOMEM;
1137         iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1138         if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1139                 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1140                 if (!iov)
1141                         goto out_put;
1142         }
1143
1144         /* This will also move the address data into kernel space */
1145         err = verify_iovec(&msg_sys, iov, address, VERIFY_READ);
1146         if (err < 0)
1147                 goto out_freeiov;
1148         total_len = err;
1149
1150         err = -ENOBUFS;
1151
1152         /* msg_controllen must fit to int */
1153         if (msg_sys.msg_controllen > INT_MAX)
1154                 goto out_freeiov;
1155         ctl_len = msg_sys.msg_controllen;
1156         if (ctl_len)
1157         {
1158                 if (ctl_len > sizeof(ctl))
1159                 {
1160                         /* Suggested by the Advanced Sockets API for IPv6 draft:
1161                          * Limit the msg_controllen size by the SO_SNDBUF size.
1162                          */
1163                         /* Note - when this code becomes multithreaded on
1164                          * SMP machines you have a race to fix here.
1165                          */
1166                         err = -ENOBUFS;
1167                         ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1168                         if (ctl_buf == NULL)
1169                                 goto out_freeiov;
1170                 }
1171                 err = -EFAULT;
1172                 if (copy_from_user(ctl_buf, msg_sys.msg_control, ctl_len))
1173                         goto out_freectl;
1174                 msg_sys.msg_control = ctl_buf;
1175         }
1176         msg_sys.msg_flags = flags;
1177
1178         if (sock->file->f_flags & O_NONBLOCK)
1179                 msg_sys.msg_flags |= MSG_DONTWAIT;
1180         err = sock_sendmsg(sock, &msg_sys, total_len);
1181
1182 out_freectl:
1183         if (ctl_buf != ctl)
1184                 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1185 out_freeiov:
1186         if (iov != iovstack)
1187                 sock_kfree_s(sock->sk, iov, iov_size);
1188 out_put:
1189         sockfd_put(sock);
1190 out:
1191         return err;
1192 }
1193
1194 /*
1195  *      BSD recvmsg interface
1196  */
1197
1198 asmlinkage int sys_recvmsg(int fd, struct msghdr *msg, unsigned int flags)
1199 {
1200         struct socket *sock;
1201         struct iovec iovstack[UIO_FASTIOV];
1202         struct iovec *iov=iovstack;
1203         struct msghdr msg_sys;
1204         unsigned long cmsg_ptr;
1205         int err, iov_size, total_len, len;
1206
1207         /* kernel mode address */
1208         char addr[MAX_SOCK_ADDR];
1209
1210         /* user mode address pointers */
1211         struct sockaddr *uaddr;
1212         int *uaddr_len;
1213
1214         err=-EFAULT;
1215         if (copy_from_user(&msg_sys,msg,sizeof(struct msghdr)))
1216                 goto out;
1217
1218         sock = sockfd_lookup(fd, &err);
1219         if (!sock)
1220                 goto out;
1221
1222         err = -EINVAL;
1223         if (msg_sys.msg_iovlen > UIO_MAXIOV)
1224                 goto out_put;
1225
1226         /* Check whether to allocate the iovec area*/
1227         err = -ENOMEM;
1228         iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1229         if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1230                 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1231                 if (!iov)
1232                         goto out_put;
1233         }
1234
1235         /*
1236          *      Save the user-mode address (verify_iovec will change the
1237          *      kernel msghdr to use the kernel address space)
1238          */
1239
1240         uaddr = msg_sys.msg_name;
1241         uaddr_len = &msg->msg_namelen;
1242         err = verify_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1243         if (err < 0)
1244                 goto out_freeiov;
1245         total_len=err;
1246
1247         cmsg_ptr = (unsigned long)msg_sys.msg_control;
1248         msg_sys.msg_flags = 0;
1249
1250         if (sock->file->f_flags & O_NONBLOCK)
1251                 flags |= MSG_DONTWAIT;
1252         err = sock_recvmsg(sock, &msg_sys, total_len, flags);
1253         if (err < 0)
1254                 goto out_freeiov;
1255         len = err;
1256
1257         if (uaddr != NULL) {
1258                 err = move_addr_to_user(addr, msg_sys.msg_namelen, uaddr, uaddr_len);
1259                 if (err < 0)
1260                         goto out_freeiov;
1261         }
1262         err = __put_user(msg_sys.msg_flags, &msg->msg_flags);
1263         if (err)
1264                 goto out_freeiov;
1265         err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1266                                                          &msg->msg_controllen);
1267         if (err)
1268                 goto out_freeiov;
1269         err = len;
1270
1271 out_freeiov:
1272         if (iov != iovstack)
1273                 sock_kfree_s(sock->sk, iov, iov_size);
1274 out_put:
1275         sockfd_put(sock);
1276 out:
1277         return err;
1278 }
1279
1280
1281 /*
1282  *      Perform a file control on a socket file descriptor.
1283  *
1284  *      Doesn't aquire a fd lock, because no network fcntl
1285  *      function sleeps currently.
1286  */
1287
1288 int sock_fcntl(struct file *filp, unsigned int cmd, unsigned long arg)
1289 {
1290         struct socket *sock;
1291
1292         sock = socki_lookup (filp->f_dentry->d_inode);
1293         if (sock && sock->ops)
1294                 return sock->ops->fcntl(sock, cmd, arg);
1295         return(-EINVAL);
1296 }
1297
1298 /* Argument list sizes for sys_socketcall */
1299 #define AL(x) ((x) * sizeof(unsigned long))
1300 static unsigned char nargs[18]={AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
1301                                 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
1302                                 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3)};
1303 #undef AL
1304
1305 /*
1306  *      System call vectors.
1307  *
1308  *      Argument checking cleaned up. Saved 20% in size.
1309  *  This function doesn't need to set the kernel lock because
1310  *  it is set by the callees.
1311  */
1312
1313 asmlinkage int sys_socketcall(int call, unsigned long *args)
1314 {
1315         unsigned long a[6];
1316         unsigned long a0,a1;
1317         int err;
1318
1319         if(call<1||call>SYS_RECVMSG)
1320                 return -EINVAL;
1321
1322         /* copy_from_user should be SMP safe. */
1323         if (copy_from_user(a, args, nargs[call]))
1324                 return -EFAULT;
1325
1326         a0=a[0];
1327         a1=a[1];
1328
1329         switch(call)
1330         {
1331                 case SYS_SOCKET:
1332                         err = sys_socket(a0,a1,a[2]);
1333                         break;
1334                 case SYS_BIND:
1335                         err = sys_bind(a0,(struct sockaddr *)a1, a[2]);
1336                         break;
1337                 case SYS_CONNECT:
1338                         err = sys_connect(a0, (struct sockaddr *)a1, a[2]);
1339                         break;
1340                 case SYS_LISTEN:
1341                         err = sys_listen(a0,a1);
1342                         break;
1343                 case SYS_ACCEPT:
1344                         err = sys_accept(a0,(struct sockaddr *)a1, (int *)a[2]);
1345                         break;
1346                 case SYS_GETSOCKNAME:
1347                         err = sys_getsockname(a0,(struct sockaddr *)a1, (int *)a[2]);
1348                         break;
1349                 case SYS_GETPEERNAME:
1350                         err = sys_getpeername(a0, (struct sockaddr *)a1, (int *)a[2]);
1351                         break;
1352                 case SYS_SOCKETPAIR:
1353                         err = sys_socketpair(a0,a1, a[2], (int *)a[3]);
1354                         break;
1355                 case SYS_SEND:
1356                         err = sys_send(a0, (void *)a1, a[2], a[3]);
1357                         break;
1358                 case SYS_SENDTO:
1359                         err = sys_sendto(a0,(void *)a1, a[2], a[3],
1360                                          (struct sockaddr *)a[4], a[5]);
1361                         break;
1362                 case SYS_RECV:
1363                         err = sys_recv(a0, (void *)a1, a[2], a[3]);
1364                         break;
1365                 case SYS_RECVFROM:
1366                         err = sys_recvfrom(a0, (void *)a1, a[2], a[3],
1367                                            (struct sockaddr *)a[4], (int *)a[5]);
1368                         break;
1369                 case SYS_SHUTDOWN:
1370                         err = sys_shutdown(a0,a1);
1371                         break;
1372                 case SYS_SETSOCKOPT:
1373                         err = sys_setsockopt(a0, a1, a[2], (char *)a[3], a[4]);
1374                         break;
1375                 case SYS_GETSOCKOPT:
1376                         err = sys_getsockopt(a0, a1, a[2], (char *)a[3], (int *)a[4]);
1377                         break;
1378                 case SYS_SENDMSG:
1379                         err = sys_sendmsg(a0, (struct msghdr *) a1, a[2]);
1380                         break;
1381                 case SYS_RECVMSG:
1382                         err = sys_recvmsg(a0, (struct msghdr *) a1, a[2]);
1383                         break;
1384                 default:
1385                         err = -EINVAL;
1386                         break;
1387         }
1388         return err;
1389 }
1390
1391 /*
1392  *      This function is called by a protocol handler that wants to
1393  *      advertise its address family, and have it linked into the
1394  *      SOCKET module.
1395  */
1396
1397 int sock_register(struct net_proto_family *ops)
1398 {
1399         if (ops->family >= NPROTO) {
1400                 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
1401                 return -ENOBUFS;
1402         }
1403         net_families[ops->family]=ops;
1404         return 0;
1405 }
1406
1407 /*
1408  *      This function is called by a protocol handler that wants to
1409  *      remove its address family, and have it unlinked from the
1410  *      SOCKET module.
1411  */
1412
1413 int sock_unregister(int family)
1414 {
1415         if (family < 0 || family >= NPROTO)
1416                 return -1;
1417
1418         net_families[family]=NULL;
1419         return 0;
1420 }
1421
1422 void __init proto_init(void)
1423 {
1424         extern struct net_proto protocols[];    /* Network protocols */
1425         struct net_proto *pro;
1426
1427         /* Kick all configured protocols. */
1428         pro = protocols;
1429         while (pro->name != NULL)
1430         {
1431                 (*pro->init_func)(pro);
1432                 pro++;
1433         }
1434         /* We're all done... */
1435 }
1436
1437 extern void sk_init(void);
1438 #ifdef CONFIG_WAN_ROUTER
1439 extern void wanrouter_init(void);
1440 #endif
1441
1442 void __init sock_init(void)
1443 {
1444         int i;
1445
1446         printk(KERN_INFO "Linux NET4.0 for Linux 2.3\n");
1447         printk(KERN_INFO "Based upon Swansea University Computer Society NET3.039\n");
1448
1449         /*
1450          *      Initialize all address (protocol) families.
1451          */
1452
1453         for (i = 0; i < NPROTO; i++)
1454                 net_families[i] = NULL;
1455
1456         /*
1457          *      Initialize sock SLAB cache.
1458          */
1459
1460         sk_init();
1461
1462 #ifdef SLAB_SKB
1463         /*
1464          *      Initialize skbuff SLAB cache
1465          */
1466         skb_init();
1467 #endif
1468
1469
1470         /*
1471          *      Wan router layer.
1472          */
1473
1474 #ifdef CONFIG_WAN_ROUTER
1475         wanrouter_init();
1476 #endif
1477
1478         /*
1479          *      Attach the firewall module if configured
1480          */
1481
1482 #ifdef CONFIG_FIREWALL
1483         fwchain_init();
1484 #endif
1485
1486         /*
1487          *      Initialize the protocols module.
1488          */
1489
1490         proto_init();
1491
1492         /*
1493          *      The netlink device handler may be needed early.
1494          */
1495
1496 #ifdef  CONFIG_RTNETLINK
1497         rtnetlink_init();
1498 #endif
1499 #ifdef CONFIG_NETLINK_DEV
1500         init_netlink();
1501 #endif
1502 }
1503
1504 int socket_get_info(char *buffer, char **start, off_t offset, int length)
1505 {
1506         int len = sprintf(buffer, "sockets: used %d\n", sockets_in_use);
1507         if (offset >= len)
1508         {
1509                 *start = buffer;
1510                 return 0;
1511         }
1512         *start = buffer + offset;
1513         len -= offset;
1514         if (len > length)
1515                 len = length;
1516         return len;
1517 }