sys/emulation/linux/linux_socket.c

   1 /*-
   2  * Copyright (c) 1995 Søren Schmidt
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer
  10  *    in this position and unchanged.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 3. The name of the author may not be used to endorse or promote products
  15  *    derived from this software withough specific prior written permission
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  27  *
  28  * $FreeBSD: src/sys/compat/linux/linux_socket.c,v 1.19.2.8 2001/11/07 20:33:55 marcel Exp $
  29  * $DragonFly: src/sys/emulation/linux/linux_socket.c,v 1.28 2008/07/10 00:19:27 aggelos Exp $
  30  */
  31
  32 #include <sys/param.h>
  33 #include <sys/proc.h>
  34 #include <sys/systm.h>
  35 #include <sys/sysproto.h>
  36 #include <sys/fcntl.h>
  37 #include <sys/file.h>
  38 #include <sys/kern_syscall.h>
  39 #include <sys/socket.h>
  40 #include <sys/socketvar.h>
  41 #include <sys/uio.h>
  42 #include <sys/malloc.h>
  43 #include <sys/mbuf.h>
  44
  45 #include <netinet/in.h>
  46 #include <netinet/in_systm.h>
  47 #include <netinet/ip.h>
  48
  49 #include <arch_linux/linux.h>
  50 #include <arch_linux/linux_proto.h>
  51 #include "linux_socket.h"
  52 #include "linux_util.h"
  53
  54 /*
  55  * Copyin a sockaddr structure provided by a Linux binary.  Linux uses
  56  * the 4.3BSD sockaddr structure which has no sa_len field.  We must
  57  * pass 4.4BSD sockaddr structures when we call native functions in the
  58  * BSD kernel.  This function does the conversion for us.
  59  *
  60  * Also, our socket calls require the sockaddr structure length to agree
  61  * with the address family.  Linux does not, so we must force it.
  62  *
  63  * This function should only need to be called from linux_connect()
  64  * and linux_bind().
  65  */
  66 static int
  67 linux_getsockaddr(struct sockaddr **namp, struct sockaddr *uaddr, size_t len)
  68 {
  69         struct sockaddr *sa;
  70         uint16_t family;        /* XXX: must match Linux sockaddr */
  71         int error;
  72         int sa_len;
  73
  74         *namp = NULL;
  75
  76         if (len > SOCK_MAXADDRLEN)
  77                 return ENAMETOOLONG;
  78         error = copyin(uaddr, &family, sizeof(family));
  79         if (error)
  80                 return (error);
  81
  82         /*
  83          * Force the sa_len field to match the address family.
  84          */
  85         switch (family) {
  86         case AF_INET:
  87                 sa_len = sizeof(struct sockaddr_in);
  88                 break;
  89         case AF_INET6:
  90                 sa_len = sizeof(struct sockaddr_in6);
  91                 break;
  92         default:
  93                 /*
  94                  * This is the default behavior of the old
  95                  * linux_to_bsd_namelen() function.  NOTE!  The
  96                  * minimum length we allocate must cover sa->sa_len and
  97                  * sa->sa_family.
  98                  */
  99                 sa_len = offsetof(struct sockaddr, sa_data[0]);
 100                 if (sa_len < len)
 101                         sa_len = len;
 102                 break;
 103         }
 104
 105         MALLOC(sa, struct sockaddr *, sa_len, M_SONAME, M_WAITOK);
 106         error = copyin(uaddr, sa, sa_len);
 107         if (error) {
 108                 FREE(sa, M_SONAME);
 109         } else {
 110                 /*
 111                  * Convert to the 4.4BSD sockaddr structure.
 112                  */
 113                 sa->sa_family = *(sa_family_t *)sa;
 114                 sa->sa_len = sa_len;
 115                 *namp = sa;
 116         }
 117
 118         return (error);
 119 }
 120
 121 /*
 122  * Transform a 4.4BSD sockaddr structure into a Linux sockaddr structure
 123  * and copy it out to a user address.
 124  */
 125 static int
 126 linux_copyout_sockaddr(struct sockaddr *sa, struct sockaddr *uaddr, int sa_len)
 127 {
 128         int error;
 129
 130         if (sa_len < (int)sizeof(u_short))
 131                 return (EINVAL);
 132
 133         *(u_short *)sa = sa->sa_family;
 134         error = copyout(sa, uaddr, sa_len);
 135
 136         return (error);
 137 }
 138
 139 static int
 140 linux_to_bsd_domain(int domain)
 141 {
 142
 143         switch (domain) {
 144         case LINUX_AF_UNSPEC:
 145                 return (AF_UNSPEC);
 146         case LINUX_AF_UNIX:
 147                 return (AF_LOCAL);
 148         case LINUX_AF_INET:
 149                 return (AF_INET);
 150         case LINUX_AF_AX25:
 151                 return (AF_CCITT);
 152         case LINUX_AF_IPX:
 153                 return (AF_IPX);
 154         case LINUX_AF_APPLETALK:
 155                 return (AF_APPLETALK);
 156         }
 157         return (-1);
 158 }
 159
 160 static int
 161 linux_to_bsd_sockopt_level(int level)
 162 {
 163
 164         switch (level) {
 165         case LINUX_SOL_SOCKET:
 166                 return (SOL_SOCKET);
 167         }
 168         return (level);
 169 }
 170
 171 static int
 172 linux_to_bsd_ip_sockopt(int opt)
 173 {
 174
 175         switch (opt) {
 176         case LINUX_IP_TOS:
 177                 return (IP_TOS);
 178         case LINUX_IP_TTL:
 179                 return (IP_TTL);
 180         case LINUX_IP_OPTIONS:
 181                 return (IP_OPTIONS);
 182         case LINUX_IP_MULTICAST_IF:
 183                 return (IP_MULTICAST_IF);
 184         case LINUX_IP_MULTICAST_TTL:
 185                 return (IP_MULTICAST_TTL);
 186         case LINUX_IP_MULTICAST_LOOP:
 187                 return (IP_MULTICAST_LOOP);
 188         case LINUX_IP_ADD_MEMBERSHIP:
 189                 return (IP_ADD_MEMBERSHIP);
 190         case LINUX_IP_DROP_MEMBERSHIP:
 191                 return (IP_DROP_MEMBERSHIP);
 192         case LINUX_IP_HDRINCL:
 193                 return (IP_HDRINCL);
 194         }
 195         return (-1);
 196 }
 197
 198 static int
 199 linux_to_bsd_so_sockopt(int opt)
 200 {
 201
 202         switch (opt) {
 203         case LINUX_SO_DEBUG:
 204                 return (SO_DEBUG);
 205         case LINUX_SO_REUSEADDR:
 206                 return (SO_REUSEADDR);
 207         case LINUX_SO_TYPE:
 208                 return (SO_TYPE);
 209         case LINUX_SO_ERROR:
 210                 return (SO_ERROR);
 211         case LINUX_SO_DONTROUTE:
 212                 return (SO_DONTROUTE);
 213         case LINUX_SO_BROADCAST:
 214                 return (SO_BROADCAST);
 215         case LINUX_SO_SNDBUF:
 216                 return (SO_SNDBUF);
 217         case LINUX_SO_RCVBUF:
 218                 return (SO_RCVBUF);
 219         case LINUX_SO_KEEPALIVE:
 220                 return (SO_KEEPALIVE);
 221         case LINUX_SO_OOBINLINE:
 222                 return (SO_OOBINLINE);
 223         case LINUX_SO_LINGER:
 224                 return (SO_LINGER);
 225         }
 226         return (-1);
 227 }
 228
 229 static int
 230 linux_to_bsd_msg_flags(int flags)
 231 {
 232         int ret_flags = 0;
 233
 234         if (flags & LINUX_MSG_OOB)
 235                 ret_flags |= MSG_OOB;
 236         if (flags & LINUX_MSG_PEEK)
 237                 ret_flags |= MSG_PEEK;
 238         if (flags & LINUX_MSG_DONTROUTE)
 239                 ret_flags |= MSG_DONTROUTE;
 240         if (flags & LINUX_MSG_CTRUNC)
 241                 ret_flags |= MSG_CTRUNC;
 242         if (flags & LINUX_MSG_TRUNC)
 243                 ret_flags |= MSG_TRUNC;
 244         if (flags & LINUX_MSG_DONTWAIT)
 245                 ret_flags |= MSG_DONTWAIT;
 246         if (flags & LINUX_MSG_EOR)
 247                 ret_flags |= MSG_EOR;
 248         if (flags & LINUX_MSG_WAITALL)
 249                 ret_flags |= MSG_WAITALL;
 250 #if 0 /* not handled */
 251         if (flags & LINUX_MSG_PROXY)
 252                 ;
 253         if (flags & LINUX_MSG_FIN)
 254                 ;
 255         if (flags & LINUX_MSG_SYN)
 256                 ;
 257         if (flags & LINUX_MSG_CONFIRM)
 258                 ;
 259         if (flags & LINUX_MSG_RST)
 260                 ;
 261         if (flags & LINUX_MSG_ERRQUEUE)
 262                 ;
 263         if (flags & LINUX_MSG_NOSIGNAL)
 264                 ;
 265 #endif
 266         return ret_flags;
 267 }
 268
 269 struct linux_socket_args {
 270         int domain;
 271         int type;
 272         int protocol;
 273 };
 274
 275 static int
 276 linux_socket(struct linux_socket_args *args, int *res)
 277 {
 278         struct linux_socket_args linux_args;
 279         struct sockopt sopt;
 280         int error, domain, optval;
 281
 282         error = copyin(args, &linux_args, sizeof(linux_args));
 283         if (error)
 284                 return (error);
 285
 286         domain = linux_to_bsd_domain(linux_args.domain);
 287         if (domain == -1)
 288                 return (EINVAL);
 289
 290         error = kern_socket(domain, linux_args.type, linux_args.protocol, res);
 291
 292         /* Copy back the return value from socket() */
 293         if (error == 0 && linux_args.type == SOCK_RAW &&
 294             (linux_args.protocol == IPPROTO_RAW || linux_args.protocol == 0) &&
 295             linux_args.domain == AF_INET) {
 296                 /* It's a raw IP socket: set the IP_HDRINCL option. */
 297                 optval = 1;
 298                 sopt.sopt_dir = SOPT_SET;
 299                 sopt.sopt_level = IPPROTO_IP;
 300                 sopt.sopt_name = IP_HDRINCL;
 301                 sopt.sopt_val = &optval;
 302                 sopt.sopt_valsize = sizeof(optval);
 303                 sopt.sopt_td = NULL;
 304
 305                 /* We ignore any error returned by setsockopt() */
 306                 kern_setsockopt(*res, &sopt);
 307         }
 308
 309         return (error);
 310 }
 311
 312 struct linux_bind_args {
 313         int s;
 314         struct sockaddr *name;
 315         int namelen;
 316 };
 317
 318 static int
 319 linux_bind(struct linux_bind_args *args, int *res)
 320 {
 321         struct linux_bind_args linux_args;
 322         struct sockaddr *sa;
 323         int error;
 324
 325         error = copyin(args, &linux_args, sizeof(linux_args));
 326         if (error)
 327                 return (error);
 328         error = linux_getsockaddr(&sa, linux_args.name, linux_args.namelen);
 329         if (error)
 330                 return (error);
 331
 332         error = kern_bind(linux_args.s, sa);
 333         FREE(sa, M_SONAME);
 334
 335         return (error);
 336 }
 337
 338 struct linux_connect_args {
 339         int s;
 340         struct sockaddr * name;
 341         int namelen;
 342 };
 343
 344 static int
 345 linux_connect(struct linux_connect_args *args, int *res)
 346 {
 347         struct thread *td = curthread;  /* XXX */
 348         struct proc *p = td->td_proc;
 349         struct linux_connect_args linux_args;
 350         struct sockaddr *sa;
 351         struct socket *so;
 352         struct file *fp;
 353         int error;
 354
 355         KKASSERT(p);
 356
 357         error = copyin(args, &linux_args, sizeof(linux_args));
 358         if (error)
 359                 return (error);
 360         error = linux_getsockaddr(&sa, linux_args.name, linux_args.namelen);
 361         if (error)
 362                 return (error);
 363
 364         error = kern_connect(linux_args.s, 0, sa);
 365         FREE(sa, M_SONAME);
 366
 367         if (error != EISCONN)
 368                 return (error);
 369
 370         /*
 371          * Linux doesn't return EISCONN the first time it occurs,
 372          * when on a non-blocking socket. Instead it returns the
 373          * error getsockopt(SOL_SOCKET, SO_ERROR) would return on BSD.
 374          */
 375         error = holdsock(p->p_fd, linux_args.s, &fp);
 376         if (error)
 377                 return (error);
 378         error = EISCONN;
 379         if (fp->f_flag & FNONBLOCK) {
 380                 so = (struct socket *)fp->f_data;
 381                 if (so->so_emuldata == 0)
 382                         error = so->so_error;
 383                 so->so_emuldata = (void *)1;
 384         }
 385         fdrop(fp);
 386         return (error);
 387 }
 388
 389 struct linux_listen_args {
 390         int s;
 391         int backlog;
 392 };
 393
 394 static int
 395 linux_listen(struct linux_listen_args *args, int *res)
 396 {
 397         struct linux_listen_args linux_args;
 398         int error;
 399
 400         error = copyin(args, &linux_args, sizeof(linux_args));
 401         if (error)
 402                 return (error);
 403
 404         error = kern_listen(linux_args.s, linux_args.backlog);
 405
 406         return(error);
 407 }
 408
 409 struct linux_accept_args {
 410         int s;
 411         struct sockaddr *addr;
 412         int *namelen;
 413 };
 414
 415 static int
 416 linux_accept(struct linux_accept_args *args, int *res)
 417 {
 418         struct linux_accept_args linux_args;
 419         struct sockaddr *sa = NULL;
 420         union fcntl_dat dat = { 0 };
 421         int error, sa_len;
 422
 423         error = copyin(args, &linux_args, sizeof(linux_args));
 424         if (error)
 425                 return (error);
 426
 427         if (linux_args.addr) {
 428                 error = copyin(linux_args.namelen, &sa_len, sizeof(sa_len));
 429                 if (error)
 430                         return (error);
 431
 432                 error = kern_accept(linux_args.s, 0, &sa, &sa_len, res);
 433
 434                 if (error) {
 435                         /*
 436                          * Return a namelen of zero for older code which
 437                          * might ignore the return value from accept().
 438                          */
 439                         sa_len = 0;
 440                         copyout(&sa_len, linux_args.namelen,
 441                             sizeof(*linux_args.namelen));
 442                 } else {
 443                         error = linux_copyout_sockaddr(sa, linux_args.addr,
 444                             sa_len);
 445                         if (error == 0) {
 446                                 error = copyout(&sa_len, linux_args.namelen,
 447                                     sizeof(*linux_args.namelen));
 448                         }
 449                 }
 450                 if (sa)
 451                         FREE(sa, M_SONAME);
 452         } else {
 453                 error = kern_accept(linux_args.s, 0, NULL, 0, res);
 454         }
 455
 456         if (error)
 457                 return (error);
 458
 459         /*
 460          * linux appears not to copy flags from the parent socket to the
 461          * accepted one, so we must clear the flags in the new descriptor.
 462          * Ignore any errors, because we already have an open fd.
 463          */
 464         kern_fcntl(*res, F_SETFL, &dat, curproc->p_ucred);
 465         return (0);
 466 }
 467
 468 struct linux_getsockname_args {
 469         int s;
 470         struct sockaddr *addr;
 471         int *namelen;
 472 };
 473
 474 static int
 475 linux_getsockname(struct linux_getsockname_args *args, int *res)
 476 {
 477         struct linux_getsockname_args linux_args;
 478         struct sockaddr *sa = NULL;
 479         int error, sa_len;
 480
 481
 482         error = copyin(args, &linux_args, sizeof(linux_args));
 483         if (error)
 484                 return (error);
 485         error = copyin(linux_args.namelen, &sa_len, sizeof(sa_len));
 486         if (error)
 487                 return (error);
 488
 489         error = kern_getsockname(linux_args.s, &sa, &sa_len);
 490
 491         if (error == 0)
 492                 error = linux_copyout_sockaddr(sa, linux_args.addr, sa_len);
 493         if (error == 0)
 494                 error = copyout(&sa_len, linux_args.namelen,
 495                     sizeof(*linux_args.namelen));
 496         if (sa)
 497                 FREE(sa, M_SONAME);
 498         return(error);
 499 }
 500
 501 struct linux_getpeername_args {
 502         int s;
 503         struct sockaddr *addr;
 504         int *namelen;
 505 };
 506
 507 static int
 508 linux_getpeername(struct linux_getpeername_args *args, int *res)
 509 {
 510         struct linux_getpeername_args linux_args;
 511         struct sockaddr *sa = NULL;
 512         int error, sa_len;
 513
 514         error = copyin(args, &linux_args, sizeof(linux_args));
 515         if (error)
 516                 return (error);
 517         error = copyin(linux_args.namelen, &sa_len, sizeof(sa_len));
 518         if (error)
 519                 return (error);
 520
 521         error = kern_getpeername(linux_args.s, &sa, &sa_len);
 522
 523         if (error == 0)
 524                 error = linux_copyout_sockaddr(sa, linux_args.addr, sa_len);
 525         if (error == 0)
 526                 error = copyout(&sa_len, linux_args.namelen,
 527                     sizeof(*linux_args.namelen));
 528         if (sa)
 529                 FREE(sa, M_SONAME);
 530         return(error);
 531 }
 532
 533 struct linux_socketpair_args {
 534         int domain;
 535         int type;
 536         int protocol;
 537         int *rsv;
 538 };
 539
 540 static int
 541 linux_socketpair(struct linux_socketpair_args *args, int *res)
 542 {
 543         struct linux_socketpair_args linux_args;
 544         int error, domain, sockv[2];
 545
 546         error = copyin(args, &linux_args, sizeof(linux_args));
 547         if (error)
 548                 return (error);
 549
 550         domain = linux_to_bsd_domain(linux_args.domain);
 551         if (domain == -1)
 552                 return (EINVAL);
 553         error = kern_socketpair(domain, linux_args.type, linux_args.protocol,
 554             sockv);
 555
 556         if (error == 0)
 557                 error = copyout(sockv, linux_args.rsv, sizeof(sockv));
 558         return(error);
 559 }
 560
 561 struct linux_send_args {
 562         int s;
 563         void *msg;
 564         int len;
 565         int flags;
 566 };
 567
 568 static int
 569 linux_send(struct linux_send_args *args, int *res)
 570 {
 571         struct linux_send_args linux_args;
 572         struct thread *td = curthread;
 573         struct uio auio;
 574         struct iovec aiov;
 575         int error;
 576
 577         error = copyin(args, &linux_args, sizeof(linux_args));
 578         if (error)
 579                 return (error);
 580
 581         aiov.iov_base = linux_args.msg;
 582         aiov.iov_len = linux_args.len;
 583         auio.uio_iov = &aiov;
 584         auio.uio_iovcnt = 1;
 585         auio.uio_offset = 0;
 586         auio.uio_resid = linux_args.len;
 587         auio.uio_segflg = UIO_USERSPACE;
 588         auio.uio_rw = UIO_WRITE;
 589         auio.uio_td = td;
 590
 591         error = kern_sendmsg(linux_args.s, NULL, &auio, NULL,
 592             linux_args.flags, res);
 593
 594         return(error);
 595 }
 596
 597 struct linux_recv_args {
 598         int s;
 599         void *msg;
 600         int len;
 601         int flags;
 602 };
 603
 604 static int
 605 linux_recv(struct linux_recv_args *args, int *res)
 606 {
 607         struct linux_recv_args linux_args;
 608         struct thread *td = curthread;
 609         struct uio auio;
 610         struct iovec aiov;
 611         int error;
 612
 613         error = copyin(args, &linux_args, sizeof(linux_args));
 614         if (error)
 615                 return (error);
 616
 617         aiov.iov_base = linux_args.msg;
 618         aiov.iov_len = linux_args.len;
 619         auio.uio_iov = &aiov;
 620         auio.uio_iovcnt = 1;
 621         auio.uio_offset = 0;
 622         auio.uio_resid = linux_args.len;
 623         auio.uio_segflg = UIO_USERSPACE;
 624         auio.uio_rw = UIO_READ;
 625         auio.uio_td = td;
 626
 627         error = kern_recvmsg(linux_args.s, NULL, &auio, NULL,
 628             &linux_args.flags, res);
 629
 630         return(error);
 631 }
 632
 633 struct linux_sendto_args {
 634         int s;
 635         void *msg;
 636         int len;
 637         int flags;
 638         struct sockaddr *to;
 639         int tolen;
 640 };
 641
 642 static int
 643 linux_sendto(struct linux_sendto_args *args, int *res)
 644 {
 645         struct linux_sendto_args linux_args;
 646         struct thread *td = curthread;
 647         struct uio auio;
 648         struct iovec aiov;
 649         struct sockopt sopt;
 650         struct sockaddr *sa = NULL;
 651         caddr_t msg = NULL;
 652         int error, optval;
 653
 654         error = copyin(args, &linux_args, sizeof(linux_args));
 655         if (error)
 656                 return (error);
 657
 658         if (linux_args.to) {
 659                 error = linux_getsockaddr(&sa, linux_args.to,
 660                     linux_args.tolen);
 661                 if (error)
 662                         return (error);
 663         }
 664
 665         /*
 666          * Check to see if the IP_HDRINCL option is set.
 667          */
 668         sopt.sopt_dir = SOPT_GET;
 669         sopt.sopt_level = IPPROTO_IP;
 670         sopt.sopt_name = IP_HDRINCL;
 671         sopt.sopt_val = &optval;
 672         sopt.sopt_valsize = sizeof(optval);
 673         sopt.sopt_td = NULL;
 674
 675         if (kern_getsockopt(linux_args.s, &sopt) != 0)
 676                 optval = 0;
 677
 678         if (optval == 0) {
 679                 /*
 680                  * IP_HDRINCL is not set.  Package the message as usual.
 681                  */
 682                 aiov.iov_base = linux_args.msg;
 683                 aiov.iov_len = linux_args.len;
 684                 auio.uio_iov = &aiov;
 685                 auio.uio_iovcnt = 1;
 686                 auio.uio_offset = 0;
 687                 auio.uio_resid = linux_args.len;
 688                 auio.uio_segflg = UIO_USERSPACE;
 689                 auio.uio_rw = UIO_WRITE;
 690                 auio.uio_td = td;
 691         } else {
 692                 /*
 693                  * IP_HDRINCL is set.  We must convert the beginning of
 694                  * the packet header so we can feed it to the BSD kernel.
 695                  */
 696
 697                 /*
 698                  * Check that the packet header is long enough to contain
 699                  * the fields of interest.  This relies on the fact that
 700                  * the fragment offset field comes after the length field.
 701                  */
 702                 if (linux_args.len < offsetof(struct ip, ip_off))
 703                         return (EINVAL);
 704
 705                 MALLOC(msg, caddr_t, linux_args.len, M_LINUX, M_WAITOK);
 706                 error = copyin(linux_args.msg, msg, linux_args.len);
 707                 if (error)
 708                         goto cleanup;
 709
 710                 /* Fix the ip_len and ip_off fields.  */
 711                 ((struct ip *)msg)->ip_len = linux_args.len;
 712                 ((struct ip *)msg)->ip_off = ntohs(((struct ip *)msg)->ip_off);
 713
 714                 aiov.iov_base = msg;
 715                 aiov.iov_len = linux_args.len;
 716                 auio.uio_iov = &aiov;
 717                 auio.uio_iovcnt = 1;
 718                 auio.uio_offset = 0;
 719                 auio.uio_resid = linux_args.len;
 720                 auio.uio_segflg = UIO_SYSSPACE;
 721                 auio.uio_rw = UIO_WRITE;
 722                 auio.uio_td = td;
 723         }
 724
 725         error = kern_sendmsg(linux_args.s, sa, &auio, NULL, linux_args.flags,
 726             res);
 727
 728 cleanup:
 729         if (sa)
 730                 FREE(sa, M_SONAME);
 731         if (msg)
 732                 FREE(msg, M_LINUX);
 733         return(error);
 734 }
 735
 736 struct linux_recvfrom_args {
 737         int s;
 738         void *buf;
 739         int len;
 740         int flags;
 741         struct sockaddr *from;
 742         int *fromlen;
 743 };
 744
 745 static int
 746 linux_recvfrom(struct linux_recvfrom_args *args, int *res)
 747 {
 748         struct linux_recvfrom_args linux_args;
 749         struct thread *td = curthread;
 750         struct uio auio;
 751         struct iovec aiov;
 752         struct sockaddr *sa = NULL;
 753         int error, fromlen, flags;
 754
 755         error = copyin(args, &linux_args, sizeof(linux_args));
 756         if (error)
 757                 return (error);
 758
 759         if (linux_args.from && linux_args.fromlen) {
 760                 error = copyin(linux_args.fromlen, &fromlen, sizeof(fromlen));
 761                 if (error)
 762                         return (error);
 763                 if (fromlen < 0)
 764                         return (EINVAL);
 765         } else {
 766                 fromlen = 0;
 767         }
 768         aiov.iov_base = linux_args.buf;
 769         aiov.iov_len = linux_args.len;
 770         auio.uio_iov = &aiov;
 771         auio.uio_iovcnt = 1;
 772         auio.uio_offset = 0;
 773         auio.uio_resid = linux_args.len;
 774         auio.uio_segflg = UIO_USERSPACE;
 775         auio.uio_rw = UIO_READ;
 776         auio.uio_td = td;
 777
 778         flags = linux_to_bsd_msg_flags(linux_args.flags);
 779
 780         error = kern_recvmsg(linux_args.s, linux_args.from ? &sa : NULL, &auio,
 781             NULL, &flags, res);
 782
 783         if (error == 0 && linux_args.from) {
 784                 if (sa != NULL) {
 785                         fromlen = MIN(fromlen, sa->sa_len);
 786                         error = linux_copyout_sockaddr(sa, linux_args.from,
 787                                                         fromlen);
 788                 } else
 789                         fromlen = 0;
 790                 if (error == 0)
 791                         copyout(&fromlen, linux_args.fromlen,
 792                             sizeof(fromlen));
 793         }
 794         if (sa)
 795                 FREE(sa, M_SONAME);
 796
 797         return(error);
 798 }
 799
 800 struct linux_sendmsg_args {
 801         int s;
 802         struct msghdr *msg;
 803         int flags;
 804 };
 805
 806 static int
 807 linux_sendmsg(struct linux_sendmsg_args *args, int *res)
 808 {
 809         struct linux_sendmsg_args linux_args;
 810         struct thread *td = curthread;
 811         struct msghdr msg;
 812         struct uio auio;
 813         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 814         struct sockaddr *sa = NULL;
 815         struct mbuf *control = NULL;
 816         int error;
 817
 818         error = copyin(args, &linux_args, sizeof(linux_args));
 819         if (error)
 820                 return (error);
 821
 822         error = copyin(linux_args.msg, &msg, sizeof(msg));
 823         if (error)
 824                 return (error);
 825
 826         /*
 827          * Conditionally copyin msg.msg_name.
 828          */
 829         if (msg.msg_name) {
 830                 error = linux_getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 831                 if (error)
 832                         return (error);
 833         }
 834
 835         /*
 836          * Populate auio.
 837          */
 838         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 839                              &auio.uio_resid);
 840         if (error)
 841                 goto cleanup2;
 842         auio.uio_iov = iov;
 843         auio.uio_iovcnt = msg.msg_iovlen;
 844         auio.uio_offset = 0;
 845         auio.uio_segflg = UIO_USERSPACE;
 846         auio.uio_rw = UIO_WRITE;
 847         auio.uio_td = td;
 848
 849         /*
 850          * Conditionally copyin msg.msg_control.
 851          */
 852         if (msg.msg_control) {
 853                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 854                     msg.msg_controllen > MLEN) {
 855                         error = EINVAL;
 856                         goto cleanup;
 857                 }
 858                 control = m_get(MB_WAIT, MT_CONTROL);
 859                 if (control == NULL) {
 860                         error = ENOBUFS;
 861                         goto cleanup;
 862                 }
 863                 control->m_len = msg.msg_controllen;
 864                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 865                     msg.msg_controllen);
 866                 if (error) {
 867                         m_free(control);
 868                         goto cleanup;
 869                 }
 870                 /*
 871                  * Linux and BSD both support SCM_RIGHTS.  If a linux binary
 872                  * wants anything else with an option level of SOL_SOCKET,
 873                  * we don't support it.
 874                  */
 875                 if (mtod(control, struct cmsghdr *)->cmsg_level ==
 876                     SOL_SOCKET &&
 877                     mtod(control, struct cmsghdr *)->cmsg_type !=
 878                     SCM_RIGHTS) {
 879                         m_free(control);
 880                         error = EINVAL;
 881                         goto cleanup;
 882                 }
 883         }
 884
 885         error = kern_sendmsg(linux_args.s, sa, &auio, control,
 886             linux_args.flags, res);
 887
 888 cleanup:
 889         iovec_free(&iov, aiov);
 890 cleanup2:
 891         if (sa)
 892                 FREE(sa, M_SONAME);
 893         return (error);
 894 }
 895
 896 struct linux_recvmsg_args {
 897         int s;
 898         struct msghdr *msg;
 899         int flags;
 900 };
 901
 902 static int
 903 linux_recvmsg(struct linux_recvmsg_args *args, int *res)
 904 {
 905         struct linux_recvmsg_args linux_args;
 906         struct thread *td = curthread;
 907         struct msghdr msg;
 908         struct uio auio;
 909         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 910         struct mbuf *m, *control = NULL;
 911         struct sockaddr *sa = NULL;
 912         caddr_t ctlbuf;
 913         socklen_t *ufromlenp, *ucontrollenp;
 914         int error, fromlen, controllen, len, flags, *uflagsp;
 915
 916         error = copyin(args, &linux_args, sizeof(linux_args));
 917         if (error)
 918                 return (error);
 919
 920         error = copyin(linux_args.msg, &msg, sizeof(struct msghdr));
 921         if (error)
 922                 return (error);
 923
 924         if (msg.msg_name && msg.msg_namelen < 0)
 925                 return (EINVAL);
 926         if (msg.msg_control && msg.msg_controllen < 0)
 927                 return (EINVAL);
 928
 929         ufromlenp = (socklen_t *)((caddr_t)linux_args.msg +
 930             offsetof(struct msghdr, msg_namelen));
 931         ucontrollenp = (socklen_t *)((caddr_t)linux_args.msg +
 932             offsetof(struct msghdr, msg_controllen));
 933         uflagsp = (int *)((caddr_t)linux_args.msg +
 934             offsetof(struct msghdr, msg_flags));
 935
 936         /*
 937          * Populate auio.
 938          */
 939         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 940                              &auio.uio_resid);
 941         if (error)
 942                 return (error);
 943         auio.uio_iov = iov;
 944         auio.uio_iovcnt = msg.msg_iovlen;
 945         auio.uio_offset = 0;
 946         auio.uio_segflg = UIO_USERSPACE;
 947         auio.uio_rw = UIO_READ;
 948         auio.uio_td = td;
 949
 950         flags = linux_to_bsd_msg_flags(linux_args.flags);
 951
 952         error = kern_recvmsg(linux_args.s, msg.msg_name ? &sa : NULL, &auio,
 953             msg.msg_control ? &control : NULL, &flags, res);
 954
 955         /*
 956          * Copyout msg.msg_name and msg.msg_namelen.
 957          */
 958         if (error == 0 && msg.msg_name) {
 959                 if (sa != NULL) {
 960                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
 961                         error = linux_copyout_sockaddr(sa, msg.msg_name,
 962                                                         fromlen);
 963                 } else
 964                         fromlen = 0;
 965                 if (error == 0)
 966                         error = copyout(&fromlen, ufromlenp,
 967                             sizeof(*ufromlenp));
 968         }
 969
 970         /*
 971          * Copyout msg.msg_control and msg.msg_controllen.
 972          */
 973         if (error == 0 && msg.msg_control) {
 974                 /*
 975                  * Linux and BSD both support SCM_RIGHTS.  If a linux binary
 976                  * wants anything else with an option level of SOL_SOCKET,
 977                  * we don't support it.
 978                  */
 979                 if (mtod((struct mbuf *)msg.msg_control,
 980                     struct cmsghdr *)->cmsg_level == SOL_SOCKET &&
 981                     mtod((struct mbuf *)msg.msg_control,
 982                     struct cmsghdr *)->cmsg_type != SCM_RIGHTS) {
 983                         error = EINVAL;
 984                         goto cleanup;
 985                 }
 986
 987                 len = msg.msg_controllen;
 988                 m = control;
 989                 ctlbuf = (caddr_t)msg.msg_control;
 990
 991                 while (m && len > 0) {
 992                         unsigned int tocopy;
 993
 994                         if (len >= m->m_len) {
 995                                 tocopy = m->m_len;
 996                         } else {
 997                                 msg.msg_flags |= MSG_CTRUNC;
 998                                 tocopy = len;
 999                         }
1000
1001                         error = copyout(mtod(m, caddr_t), ctlbuf,
1002                             tocopy);
1003                         if (error)
1004                                 goto cleanup;
1005
1006                         ctlbuf += tocopy;
1007                         len -= tocopy;
1008                         m = m->m_next;
1009                 }
1010                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1011                 error = copyout(&controllen, ucontrollenp,
1012                     sizeof(*ucontrollenp));
1013         }
1014
1015         if (error == 0)
1016                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1017
1018 cleanup:
1019         if (sa)
1020                 FREE(sa, M_SONAME);
1021         iovec_free(&iov, aiov);
1022         if (control)
1023                 m_freem(control);
1024         return (error);
1025 }
1026
1027 struct linux_shutdown_args {
1028         int s;
1029         int how;
1030 };
1031
1032 static int
1033 linux_shutdown(struct linux_shutdown_args *args, int *res)
1034 {
1035         struct linux_shutdown_args linux_args;
1036         int error;
1037
1038         error = copyin(args, &linux_args, sizeof(linux_args));
1039         if (error)
1040                 return (error);
1041
1042         error = kern_shutdown(linux_args.s, linux_args.how);
1043
1044         return (error);
1045 }
1046
1047 struct linux_setsockopt_args {
1048         int s;
1049         int level;
1050         int optname;
1051         void *optval;
1052         int optlen;
1053 };
1054
1055 static int
1056 linux_setsockopt(struct linux_setsockopt_args *args, int *res)
1057 {
1058         struct linux_setsockopt_args linux_args;
1059         struct thread *td = curthread;
1060         struct sockopt sopt;
1061         int error, name, level;
1062
1063         error = copyin(args, &linux_args, sizeof(linux_args));
1064         if (error)
1065                 return (error);
1066
1067         level = linux_to_bsd_sockopt_level(linux_args.level);
1068         switch (level) {
1069         case SOL_SOCKET:
1070                 name = linux_to_bsd_so_sockopt(linux_args.optname);
1071                 break;
1072         case IPPROTO_IP:
1073                 name = linux_to_bsd_ip_sockopt(linux_args.optname);
1074                 break;
1075         case IPPROTO_TCP:
1076                 /* Linux TCP option values match BSD's */
1077                 name = linux_args.optname;
1078                 break;
1079         default:
1080                 name = -1;
1081                 break;
1082         }
1083         if (name == -1)
1084                 return (EINVAL);
1085
1086         sopt.sopt_dir = SOPT_SET;
1087         sopt.sopt_level = level;
1088         sopt.sopt_name = name;
1089         sopt.sopt_valsize = linux_args.optlen;
1090         sopt.sopt_td = td;
1091
1092         if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1093                 return (EINVAL);
1094
1095         if (linux_args.optval) {
1096                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1097                 error = copyin(linux_args.optval, sopt.sopt_val, sopt.sopt_valsize);
1098                 if (error)
1099                         goto out;
1100         } else {
1101                 sopt.sopt_val = NULL;
1102         }
1103         error = kern_setsockopt(linux_args.s, &sopt);
1104         if (error)
1105                 goto out;
1106         if (linux_args.optval)
1107                 error = copyout(sopt.sopt_val, linux_args.optval,
1108                                 sopt.sopt_valsize);
1109 out:
1110         if (linux_args.optval)
1111                 kfree(sopt.sopt_val, M_TEMP);
1112         return(error);
1113 }
1114
1115 struct linux_getsockopt_args {
1116         int s;
1117         int level;
1118         int optname;
1119         void *optval;
1120         int *optlen;
1121 };
1122
1123 static int
1124 linux_getsockopt(struct linux_getsockopt_args *args, int *res)
1125 {
1126         struct linux_getsockopt_args linux_args;
1127         struct thread *td = curthread;
1128         struct sockopt sopt;
1129         int error, name, valsize, level;
1130
1131         error = copyin(args, &linux_args, sizeof(linux_args));
1132         if (error)
1133                 return (error);
1134
1135         if (linux_args.optval) {
1136                 error = copyin(linux_args.optlen, &valsize, sizeof(valsize));
1137                 if (error)
1138                         return (error);
1139                 if (valsize < 0 || valsize > SOMAXOPT_SIZE)
1140                         return (EINVAL);
1141         } else {
1142                 valsize = 0;
1143         }
1144
1145         level = linux_to_bsd_sockopt_level(linux_args.level);
1146         switch (level) {
1147         case SOL_SOCKET:
1148                 name = linux_to_bsd_so_sockopt(linux_args.optname);
1149                 break;
1150         case IPPROTO_IP:
1151                 name = linux_to_bsd_ip_sockopt(linux_args.optname);
1152                 break;
1153         case IPPROTO_TCP:
1154                 /* Linux TCP option values match BSD's */
1155                 name = linux_args.optname;
1156                 break;
1157         default:
1158                 name = -1;
1159                 break;
1160         }
1161         if (name == -1)
1162                 return (EINVAL);
1163
1164         sopt.sopt_dir = SOPT_GET;
1165         sopt.sopt_level = level;
1166         sopt.sopt_name = name;
1167         sopt.sopt_valsize = valsize;
1168         sopt.sopt_td = td;
1169
1170         if (linux_args.optval) {
1171                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1172                 error = copyin(linux_args.optval, sopt.sopt_val, sopt.sopt_valsize);
1173                 if (error)
1174                         goto out;
1175         } else {
1176                 sopt.sopt_val = NULL;
1177         }
1178         error = kern_getsockopt(linux_args.s, &sopt);
1179         if (error)
1180                 goto out;
1181         valsize = sopt.sopt_valsize;
1182         error = copyout(&valsize, linux_args.optlen, sizeof(valsize));
1183         if (error)
1184                 goto out;
1185         if (linux_args.optval)
1186                 error = copyout(sopt.sopt_val, linux_args.optval, sopt.sopt_valsize);
1187 out:
1188         if (linux_args.optval)
1189                 kfree(sopt.sopt_val, M_TEMP);
1190         return(error);
1191 }
1192
1193 int
1194 sys_linux_socketcall(struct linux_socketcall_args *args)
1195 {
1196         void *arg = (void *)args->args;
1197
1198         switch (args->what) {
1199         case LINUX_SOCKET:
1200                 return (linux_socket(arg, &args->sysmsg_result));
1201         case LINUX_BIND:
1202                 return (linux_bind(arg, &args->sysmsg_result));
1203         case LINUX_CONNECT:
1204                 return (linux_connect(arg, &args->sysmsg_result));
1205         case LINUX_LISTEN:
1206                 return (linux_listen(arg, &args->sysmsg_result));
1207         case LINUX_ACCEPT:
1208                 return (linux_accept(arg, &args->sysmsg_result));
1209         case LINUX_GETSOCKNAME:
1210                 return (linux_getsockname(arg, &args->sysmsg_result));
1211         case LINUX_GETPEERNAME:
1212                 return (linux_getpeername(arg, &args->sysmsg_result));
1213         case LINUX_SOCKETPAIR:
1214                 return (linux_socketpair(arg, &args->sysmsg_result));
1215         case LINUX_SEND:
1216                 return (linux_send(arg, &args->sysmsg_result));
1217         case LINUX_RECV:
1218                 return (linux_recv(arg, &args->sysmsg_result));
1219         case LINUX_SENDTO:
1220                 return (linux_sendto(arg, &args->sysmsg_result));
1221         case LINUX_RECVFROM:
1222                 return (linux_recvfrom(arg, &args->sysmsg_result));
1223         case LINUX_SHUTDOWN:
1224                 return (linux_shutdown(arg, &args->sysmsg_result));
1225         case LINUX_SETSOCKOPT:
1226                 return (linux_setsockopt(arg, &args->sysmsg_result));
1227         case LINUX_GETSOCKOPT:
1228                 return (linux_getsockopt(arg, &args->sysmsg_result));
1229         case LINUX_SENDMSG:
1230                 return (linux_sendmsg(arg, &args->sysmsg_result));
1231         case LINUX_RECVMSG:
1232                 return (linux_recvmsg(arg, &args->sysmsg_result));
1233         }
1234
1235         uprintf("LINUX: 'socket' typ=%d not implemented\n", args->what);
1236         return (ENOSYS);
1237 }