sys/kern/uipc_syscalls.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * sendfile(2) and related extensions:
   6  * Copyright (c) 1998, David Greenman. All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  *      @(#)uipc_syscalls.c     8.4 (Berkeley) 2/21/94
  33  * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
  34  */
  35
  36 #include "opt_ktrace.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/kernel.h>
  41 #include <sys/sysproto.h>
  42 #include <sys/malloc.h>
  43 #include <sys/filedesc.h>
  44 #include <sys/event.h>
  45 #include <sys/proc.h>
  46 #include <sys/fcntl.h>
  47 #include <sys/file.h>
  48 #include <sys/filio.h>
  49 #include <sys/kern_syscall.h>
  50 #include <sys/mbuf.h>
  51 #include <sys/protosw.h>
  52 #include <sys/sfbuf.h>
  53 #include <sys/socket.h>
  54 #include <sys/socketvar.h>
  55 #include <sys/socketops.h>
  56 #include <sys/uio.h>
  57 #include <sys/vnode.h>
  58 #include <sys/lock.h>
  59 #include <sys/mount.h>
  60 #include <sys/jail.h>
  61 #ifdef KTRACE
  62 #include <sys/ktrace.h>
  63 #endif
  64 #include <vm/vm.h>
  65 #include <vm/vm_object.h>
  66 #include <vm/vm_page.h>
  67 #include <vm/vm_pageout.h>
  68 #include <vm/vm_kern.h>
  69 #include <vm/vm_extern.h>
  70 #include <sys/file2.h>
  71 #include <sys/signalvar.h>
  72 #include <sys/serialize.h>
  73
  74 #include <sys/thread2.h>
  75 #include <sys/msgport2.h>
  76 #include <sys/socketvar2.h>
  77 #include <net/netmsg2.h>
  78 #include <vm/vm_page2.h>
  79
  80 extern int use_soaccept_pred_fast;
  81 extern int use_sendfile_async;
  82 extern int use_soconnect_async;
  83
  84 /*
  85  * System call interface to the socket abstraction.
  86  */
  87
  88 extern  struct fileops socketops;
  89
  90 /*
  91  * socket_args(int domain, int type, int protocol)
  92  */
  93 int
  94 kern_socket(int domain, int type, int protocol, int *res)
  95 {
  96         struct thread *td = curthread;
  97         struct filedesc *fdp = td->td_proc->p_fd;
  98         struct socket *so;
  99         struct file *fp;
 100         int fd, error;
 101         u_int fflags = 0;
 102         int oflags = 0;
 103
 104         KKASSERT(td->td_lwp);
 105
 106         if (type & SOCK_NONBLOCK) {
 107                 type &= ~SOCK_NONBLOCK;
 108                 fflags |= FNONBLOCK;
 109         }
 110         if (type & SOCK_CLOEXEC) {
 111                 type &= ~SOCK_CLOEXEC;
 112                 oflags |= O_CLOEXEC;
 113         }
 114
 115         error = falloc(td->td_lwp, &fp, &fd);
 116         if (error)
 117                 return (error);
 118         error = socreate(domain, &so, type, protocol, td);
 119         if (error) {
 120                 fsetfd(fdp, NULL, fd);
 121         } else {
 122                 fp->f_type = DTYPE_SOCKET;
 123                 fp->f_flag = FREAD | FWRITE | fflags;
 124                 fp->f_ops = &socketops;
 125                 fp->f_data = so;
 126                 if (oflags & O_CLOEXEC)
 127                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 128                 *res = fd;
 129                 fsetfd(fdp, fp, fd);
 130         }
 131         fdrop(fp);
 132         return (error);
 133 }
 134
 135 /*
 136  * MPALMOSTSAFE
 137  */
 138 int
 139 sys_socket(struct socket_args *uap)
 140 {
 141         int error;
 142
 143         error = kern_socket(uap->domain, uap->type, uap->protocol,
 144                             &uap->sysmsg_iresult);
 145
 146         return (error);
 147 }
 148
 149 int
 150 kern_bind(int s, struct sockaddr *sa)
 151 {
 152         struct thread *td = curthread;
 153         struct file *fp;
 154         int error;
 155
 156         error = holdsock(td, s, &fp);
 157         if (error)
 158                 return (error);
 159         error = sobind((struct socket *)fp->f_data, sa, td);
 160         dropfp(td, s, fp);
 161
 162         return (error);
 163 }
 164
 165 /*
 166  * bind_args(int s, caddr_t name, int namelen)
 167  *
 168  * MPALMOSTSAFE
 169  */
 170 int
 171 sys_bind(struct bind_args *uap)
 172 {
 173         struct sockaddr *sa;
 174         int error;
 175
 176         error = getsockaddr(&sa, uap->name, uap->namelen);
 177         if (error)
 178                 return (error);
 179         if (!prison_remote_ip(curthread, sa)) {
 180                 kfree(sa, M_SONAME);
 181                 return EAFNOSUPPORT;
 182         }
 183         error = kern_bind(uap->s, sa);
 184         kfree(sa, M_SONAME);
 185
 186         return (error);
 187 }
 188
 189 int
 190 kern_listen(int s, int backlog)
 191 {
 192         struct thread *td = curthread;
 193         struct file *fp;
 194         int error;
 195
 196         error = holdsock(td, s, &fp);
 197         if (error)
 198                 return (error);
 199         error = solisten((struct socket *)fp->f_data, backlog, td);
 200         dropfp(td, s, fp);
 201
 202         return (error);
 203 }
 204
 205 /*
 206  * listen_args(int s, int backlog)
 207  *
 208  * MPALMOSTSAFE
 209  */
 210 int
 211 sys_listen(struct listen_args *uap)
 212 {
 213         int error;
 214
 215         error = kern_listen(uap->s, uap->backlog);
 216         return (error);
 217 }
 218
 219 /*
 220  * Returns the accepted socket as well.
 221  *
 222  * NOTE!  The sockets sitting on so_comp/so_incomp might have 0 refs, the
 223  *        pool token is absolutely required to avoid a sofree() race,
 224  *        as well as to avoid tailq handling races.
 225  */
 226 static boolean_t
 227 soaccept_predicate(struct netmsg_so_notify *msg)
 228 {
 229         struct socket *head = msg->base.nm_so;
 230         struct socket *so;
 231
 232         if (head->so_error != 0) {
 233                 msg->base.lmsg.ms_error = head->so_error;
 234                 return (TRUE);
 235         }
 236         lwkt_getpooltoken(head);
 237         if (!TAILQ_EMPTY(&head->so_comp)) {
 238                 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
 239                 so = TAILQ_FIRST(&head->so_comp);
 240                 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
 241                 TAILQ_REMOVE(&head->so_comp, so, so_list);
 242                 head->so_qlen--;
 243                 soclrstate(so, SS_COMP);
 244
 245                 /*
 246                  * Keep a reference before clearing the so_head
 247                  * to avoid racing socket close in netisr.
 248                  */
 249                 soreference(so);
 250                 so->so_head = NULL;
 251
 252                 lwkt_relpooltoken(head);
 253
 254                 msg->base.lmsg.ms_error = 0;
 255                 msg->base.nm_so = so;
 256                 return (TRUE);
 257         }
 258         lwkt_relpooltoken(head);
 259         if (head->so_state & SS_CANTRCVMORE) {
 260                 msg->base.lmsg.ms_error = ECONNABORTED;
 261                 return (TRUE);
 262         }
 263         if (msg->nm_fflags & FNONBLOCK) {
 264                 msg->base.lmsg.ms_error = EWOULDBLOCK;
 265                 return (TRUE);
 266         }
 267
 268         return (FALSE);
 269 }
 270
 271 /*
 272  * The second argument to kern_accept() is a handle to a struct sockaddr.
 273  * This allows kern_accept() to return a pointer to an allocated struct
 274  * sockaddr which must be freed later with FREE().  The caller must
 275  * initialize *name to NULL.
 276  */
 277 int
 278 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res,
 279     int sockflags)
 280 {
 281         struct thread *td = curthread;
 282         struct filedesc *fdp = td->td_proc->p_fd;
 283         struct file *lfp = NULL;
 284         struct file *nfp = NULL;
 285         struct sockaddr *sa;
 286         struct socket *head, *so;
 287         struct netmsg_so_notify msg;
 288         int fd;
 289         u_int fflag;            /* type must match fp->f_flag */
 290         int error, tmp;
 291
 292         *res = -1;
 293         if (name && namelen && *namelen < 0)
 294                 return (EINVAL);
 295
 296         error = holdsock(td, s, &lfp);
 297         if (error)
 298                 return (error);
 299
 300         error = falloc(td->td_lwp, &nfp, &fd);
 301         if (error) {            /* Probably ran out of file descriptors. */
 302                 fdrop(lfp);
 303                 return (error);
 304         }
 305         head = (struct socket *)lfp->f_data;
 306         if ((head->so_options & SO_ACCEPTCONN) == 0) {
 307                 error = EINVAL;
 308                 goto done;
 309         }
 310
 311         if (fflags & O_FBLOCKING)
 312                 fflags |= lfp->f_flag & ~FNONBLOCK;
 313         else if (fflags & O_FNONBLOCKING)
 314                 fflags |= lfp->f_flag | FNONBLOCK;
 315         else
 316                 fflags = lfp->f_flag;
 317
 318         if (use_soaccept_pred_fast) {
 319                 boolean_t pred;
 320
 321                 /* Initialize necessary parts for soaccept_predicate() */
 322                 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
 323                 msg.nm_fflags = fflags;
 324
 325                 lwkt_getpooltoken(head);
 326                 pred = soaccept_predicate(&msg);
 327                 lwkt_relpooltoken(head);
 328
 329                 if (pred) {
 330                         error = msg.base.lmsg.ms_error;
 331                         if (error)
 332                                 goto done;
 333                         else
 334                                 goto accepted;
 335                 }
 336         }
 337
 338         /* optimize for uniprocessor case later XXX JH */
 339         netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
 340                               0, netmsg_so_notify, netmsg_so_notify_doabort);
 341         msg.nm_predicate = soaccept_predicate;
 342         msg.nm_fflags = fflags;
 343         msg.nm_etype = NM_REVENT;
 344         error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
 345         if (error)
 346                 goto done;
 347
 348 accepted:
 349         /*
 350          * At this point we have the connection that's ready to be accepted.
 351          *
 352          * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
 353          *       to eat the ref and turn it into a descriptor.
 354          */
 355         so = msg.base.nm_so;
 356
 357         fflag = lfp->f_flag;
 358
 359         /* connection has been removed from the listen queue */
 360         KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
 361
 362         if (sockflags & SOCK_KERN_NOINHERIT) {
 363                 fflag &= ~(FASYNC | FNONBLOCK);
 364                 if (sockflags & SOCK_NONBLOCK)
 365                         fflag |= FNONBLOCK;
 366         } else {
 367                 if (head->so_sigio != NULL)
 368                         fsetown(fgetown(&head->so_sigio), &so->so_sigio);
 369         }
 370
 371         nfp->f_type = DTYPE_SOCKET;
 372         nfp->f_flag = fflag;
 373         nfp->f_ops = &socketops;
 374         nfp->f_data = so;
 375         /* Sync socket async state with file flags */
 376         tmp = fflag & FASYNC;
 377         fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
 378
 379         sa = NULL;
 380         if (so->so_faddr != NULL) {
 381                 sa = so->so_faddr;
 382                 so->so_faddr = NULL;
 383
 384                 soaccept_generic(so);
 385                 error = 0;
 386         } else {
 387                 error = soaccept(so, &sa);
 388         }
 389
 390         /*
 391          * Set the returned name and namelen as applicable.  Set the returned
 392          * namelen to 0 for older code which might ignore the return value
 393          * from accept.
 394          */
 395         if (error == 0) {
 396                 if (sa && name && namelen) {
 397                         if (*namelen > sa->sa_len)
 398                                 *namelen = sa->sa_len;
 399                         *name = sa;
 400                 } else {
 401                         if (sa)
 402                                 kfree(sa, M_SONAME);
 403                 }
 404         }
 405
 406 done:
 407         /*
 408          * If an error occured clear the reserved descriptor, else associate
 409          * nfp with it.
 410          *
 411          * Note that *res is normally ignored if an error is returned but
 412          * a syscall message will still have access to the result code.
 413          */
 414         if (error) {
 415                 fsetfd(fdp, NULL, fd);
 416         } else {
 417                 if (sockflags & SOCK_CLOEXEC)
 418                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 419                 *res = fd;
 420                 fsetfd(fdp, nfp, fd);
 421         }
 422         fdrop(nfp);
 423         dropfp(td, s, lfp);
 424
 425         return (error);
 426 }
 427
 428 /*
 429  * accept(int s, caddr_t name, int *anamelen)
 430  *
 431  * MPALMOSTSAFE
 432  */
 433 int
 434 sys_accept(struct accept_args *uap)
 435 {
 436         struct sockaddr *sa = NULL;
 437         int sa_len;
 438         int error;
 439
 440         if (uap->name) {
 441                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 442                 if (error)
 443                         return (error);
 444
 445                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 446                                     &uap->sysmsg_iresult, 0);
 447
 448                 if (error == 0) {
 449                         prison_local_ip(curthread, sa);
 450                         error = copyout(sa, uap->name, sa_len);
 451                 }
 452                 if (error == 0) {
 453                         error = copyout(&sa_len, uap->anamelen,
 454                                         sizeof(*uap->anamelen));
 455                 }
 456                 if (sa)
 457                         kfree(sa, M_SONAME);
 458         } else {
 459                 error = kern_accept(uap->s, 0, NULL, 0,
 460                                     &uap->sysmsg_iresult, 0);
 461         }
 462         return (error);
 463 }
 464
 465 /*
 466  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
 467  *
 468  * MPALMOSTSAFE
 469  */
 470 int
 471 sys_extaccept(struct extaccept_args *uap)
 472 {
 473         struct sockaddr *sa = NULL;
 474         int sa_len;
 475         int error;
 476         int fflags = uap->flags & O_FMASK;
 477
 478         if (uap->name) {
 479                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 480                 if (error)
 481                         return (error);
 482
 483                 error = kern_accept(uap->s, fflags, &sa, &sa_len,
 484                                     &uap->sysmsg_iresult, 0);
 485
 486                 if (error == 0) {
 487                         prison_local_ip(curthread, sa);
 488                         error = copyout(sa, uap->name, sa_len);
 489                 }
 490                 if (error == 0) {
 491                         error = copyout(&sa_len, uap->anamelen,
 492                             sizeof(*uap->anamelen));
 493                 }
 494                 if (sa)
 495                         kfree(sa, M_SONAME);
 496         } else {
 497                 error = kern_accept(uap->s, fflags, NULL, 0,
 498                                     &uap->sysmsg_iresult, 0);
 499         }
 500         return (error);
 501 }
 502
 503 /*
 504  * accept4(int s, caddr_t name, int *anamelen, int flags)
 505  *
 506  * MPALMOSTSAFE
 507  */
 508 int
 509 sys_accept4(struct accept4_args *uap)
 510 {
 511         struct sockaddr *sa = NULL;
 512         int sa_len;
 513         int error;
 514         int sockflags;
 515
 516         if (uap->flags & ~(SOCK_NONBLOCK | SOCK_CLOEXEC))
 517                 return (EINVAL);
 518         sockflags = uap->flags | SOCK_KERN_NOINHERIT;
 519
 520         if (uap->name) {
 521                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 522                 if (error)
 523                         return (error);
 524
 525                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 526                                     &uap->sysmsg_iresult, sockflags);
 527
 528                 if (error == 0) {
 529                         prison_local_ip(curthread, sa);
 530                         error = copyout(sa, uap->name, sa_len);
 531                 }
 532                 if (error == 0) {
 533                         error = copyout(&sa_len, uap->anamelen,
 534                                         sizeof(*uap->anamelen));
 535                 }
 536                 if (sa)
 537                         kfree(sa, M_SONAME);
 538         } else {
 539                 error = kern_accept(uap->s, 0, NULL, 0,
 540                                     &uap->sysmsg_iresult, sockflags);
 541         }
 542         return (error);
 543 }
 544
 545 /*
 546  * Returns TRUE if predicate satisfied.
 547  */
 548 static boolean_t
 549 soconnected_predicate(struct netmsg_so_notify *msg)
 550 {
 551         struct socket *so = msg->base.nm_so;
 552
 553         /* check predicate */
 554         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 555                 msg->base.lmsg.ms_error = so->so_error;
 556                 return (TRUE);
 557         }
 558
 559         return (FALSE);
 560 }
 561
 562 int
 563 kern_connect(int s, int fflags, struct sockaddr *sa)
 564 {
 565         struct thread *td = curthread;
 566         struct file *fp;
 567         struct socket *so;
 568         int error, interrupted = 0;
 569
 570         error = holdsock(td, s, &fp);
 571         if (error)
 572                 return (error);
 573         so = (struct socket *)fp->f_data;
 574
 575         if (fflags & O_FBLOCKING)
 576                 /* fflags &= ~FNONBLOCK; */;
 577         else if (fflags & O_FNONBLOCKING)
 578                 fflags |= FNONBLOCK;
 579         else
 580                 fflags = fp->f_flag;
 581
 582         if (so->so_state & SS_ISCONNECTING) {
 583                 error = EALREADY;
 584                 goto done;
 585         }
 586         error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE);
 587         if (error)
 588                 goto bad;
 589         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 590                 error = EINPROGRESS;
 591                 goto done;
 592         }
 593         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 594                 struct netmsg_so_notify msg;
 595
 596                 netmsg_init_abortable(&msg.base, so,
 597                                       &curthread->td_msgport,
 598                                       0,
 599                                       netmsg_so_notify,
 600                                       netmsg_so_notify_doabort);
 601                 msg.nm_predicate = soconnected_predicate;
 602                 msg.nm_etype = NM_REVENT;
 603                 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
 604                 if (error == EINTR || error == ERESTART)
 605                         interrupted = 1;
 606         }
 607         if (error == 0) {
 608                 error = so->so_error;
 609                 so->so_error = 0;
 610         }
 611 bad:
 612         if (!interrupted)
 613                 soclrstate(so, SS_ISCONNECTING);
 614         if (error == ERESTART)
 615                 error = EINTR;
 616 done:
 617         dropfp(td, s, fp);
 618
 619         return (error);
 620 }
 621
 622 /*
 623  * connect_args(int s, caddr_t name, int namelen)
 624  *
 625  * MPALMOSTSAFE
 626  */
 627 int
 628 sys_connect(struct connect_args *uap)
 629 {
 630         struct sockaddr *sa;
 631         int error;
 632
 633         error = getsockaddr(&sa, uap->name, uap->namelen);
 634         if (error)
 635                 return (error);
 636         if (!prison_remote_ip(curthread, sa)) {
 637                 kfree(sa, M_SONAME);
 638                 return EAFNOSUPPORT;
 639         }
 640         error = kern_connect(uap->s, 0, sa);
 641         kfree(sa, M_SONAME);
 642
 643         return (error);
 644 }
 645
 646 /*
 647  * connect_args(int s, int fflags, caddr_t name, int namelen)
 648  *
 649  * MPALMOSTSAFE
 650  */
 651 int
 652 sys_extconnect(struct extconnect_args *uap)
 653 {
 654         struct sockaddr *sa;
 655         int error;
 656         int fflags = uap->flags & O_FMASK;
 657
 658         error = getsockaddr(&sa, uap->name, uap->namelen);
 659         if (error)
 660                 return (error);
 661         if (!prison_remote_ip(curthread, sa)) {
 662                 kfree(sa, M_SONAME);
 663                 return EAFNOSUPPORT;
 664         }
 665         error = kern_connect(uap->s, fflags, sa);
 666         kfree(sa, M_SONAME);
 667
 668         return (error);
 669 }
 670
 671 int
 672 kern_socketpair(int domain, int type, int protocol, int *sv)
 673 {
 674         struct thread *td = curthread;
 675         struct filedesc *fdp;
 676         struct file *fp1, *fp2;
 677         struct socket *so1, *so2;
 678         int fd1, fd2, error;
 679         u_int fflags = 0;
 680         int oflags = 0;
 681
 682         if (type & SOCK_NONBLOCK) {
 683                 type &= ~SOCK_NONBLOCK;
 684                 fflags |= FNONBLOCK;
 685         }
 686         if (type & SOCK_CLOEXEC) {
 687                 type &= ~SOCK_CLOEXEC;
 688                 oflags |= O_CLOEXEC;
 689         }
 690
 691         fdp = td->td_proc->p_fd;
 692         error = socreate(domain, &so1, type, protocol, td);
 693         if (error)
 694                 return (error);
 695         error = socreate(domain, &so2, type, protocol, td);
 696         if (error)
 697                 goto free1;
 698         error = falloc(td->td_lwp, &fp1, &fd1);
 699         if (error)
 700                 goto free2;
 701         sv[0] = fd1;
 702         fp1->f_data = so1;
 703         error = falloc(td->td_lwp, &fp2, &fd2);
 704         if (error)
 705                 goto free3;
 706         fp2->f_data = so2;
 707         sv[1] = fd2;
 708         error = soconnect2(so1, so2);
 709         if (error)
 710                 goto free4;
 711         if (type == SOCK_DGRAM) {
 712                 /*
 713                  * Datagram socket connection is asymmetric.
 714                  */
 715                  error = soconnect2(so2, so1);
 716                  if (error)
 717                         goto free4;
 718         }
 719         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 720         fp1->f_flag = fp2->f_flag = FREAD|FWRITE|fflags;
 721         fp1->f_ops = fp2->f_ops = &socketops;
 722         if (oflags & O_CLOEXEC) {
 723                 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE;
 724                 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE;
 725         }
 726         fsetfd(fdp, fp1, fd1);
 727         fsetfd(fdp, fp2, fd2);
 728         fdrop(fp1);
 729         fdrop(fp2);
 730         return (error);
 731 free4:
 732         fsetfd(fdp, NULL, fd2);
 733         fdrop(fp2);
 734 free3:
 735         fsetfd(fdp, NULL, fd1);
 736         fdrop(fp1);
 737 free2:
 738         (void)soclose(so2, 0);
 739 free1:
 740         (void)soclose(so1, 0);
 741         return (error);
 742 }
 743
 744 /*
 745  * socketpair(int domain, int type, int protocol, int *rsv)
 746  */
 747 int
 748 sys_socketpair(struct socketpair_args *uap)
 749 {
 750         int error, sockv[2];
 751
 752         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 753
 754         if (error == 0) {
 755                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 756
 757                 if (error != 0) {
 758                         kern_close(sockv[0]);
 759                         kern_close(sockv[1]);
 760                 }
 761         }
 762
 763         return (error);
 764 }
 765
 766 int
 767 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 768              struct mbuf *control, int flags, size_t *res)
 769 {
 770         struct thread *td = curthread;
 771         struct lwp *lp = td->td_lwp;
 772         struct proc *p = td->td_proc;
 773         struct file *fp;
 774         size_t len;
 775         int error;
 776         struct socket *so;
 777 #ifdef KTRACE
 778         struct iovec *ktriov = NULL;
 779         struct uio ktruio;
 780 #endif
 781
 782         error = holdsock(td, s, &fp);
 783         if (error)
 784                 return (error);
 785 #ifdef KTRACE
 786         if (KTRPOINT(td, KTR_GENIO)) {
 787                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 788
 789                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 790                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 791                 ktruio = *auio;
 792         }
 793 #endif
 794         len = auio->uio_resid;
 795         so = (struct socket *)fp->f_data;
 796         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 797                 if (fp->f_flag & FNONBLOCK)
 798                         flags |= MSG_FNONBLOCKING;
 799         }
 800         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 801         if (error) {
 802                 if (auio->uio_resid != len && (error == ERESTART ||
 803                     error == EINTR || error == EWOULDBLOCK))
 804                         error = 0;
 805                 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
 806                     !(so->so_options & SO_NOSIGPIPE))
 807                         lwpsignal(p, lp, SIGPIPE);
 808         }
 809 #ifdef KTRACE
 810         if (ktriov != NULL) {
 811                 if (error == 0) {
 812                         ktruio.uio_iov = ktriov;
 813                         ktruio.uio_resid = len - auio->uio_resid;
 814                         ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
 815                 }
 816                 kfree(ktriov, M_TEMP);
 817         }
 818 #endif
 819         if (error == 0)
 820                 *res  = len - auio->uio_resid;
 821         dropfp(td, s, fp);
 822
 823         return (error);
 824 }
 825
 826 /*
 827  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
 828  *
 829  * MPALMOSTSAFE
 830  */
 831 int
 832 sys_sendto(struct sendto_args *uap)
 833 {
 834         struct thread *td = curthread;
 835         struct uio auio;
 836         struct iovec aiov;
 837         struct sockaddr *sa = NULL;
 838         int error;
 839
 840         if (uap->to) {
 841                 error = getsockaddr(&sa, uap->to, uap->tolen);
 842                 if (error)
 843                         return (error);
 844                 if (!prison_remote_ip(curthread, sa)) {
 845                         kfree(sa, M_SONAME);
 846                         return EAFNOSUPPORT;
 847                 }
 848         }
 849         aiov.iov_base = uap->buf;
 850         aiov.iov_len = uap->len;
 851         auio.uio_iov = &aiov;
 852         auio.uio_iovcnt = 1;
 853         auio.uio_offset = 0;
 854         auio.uio_resid = uap->len;
 855         auio.uio_segflg = UIO_USERSPACE;
 856         auio.uio_rw = UIO_WRITE;
 857         auio.uio_td = td;
 858
 859         error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
 860                              &uap->sysmsg_szresult);
 861
 862         if (sa)
 863                 kfree(sa, M_SONAME);
 864         return (error);
 865 }
 866
 867 /*
 868  * sendmsg_args(int s, caddr_t msg, int flags)
 869  *
 870  * MPALMOSTSAFE
 871  */
 872 int
 873 sys_sendmsg(struct sendmsg_args *uap)
 874 {
 875         struct thread *td = curthread;
 876         struct msghdr msg;
 877         struct uio auio;
 878         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 879         struct sockaddr *sa = NULL;
 880         struct mbuf *control = NULL;
 881         int error;
 882
 883         error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
 884         if (error)
 885                 return (error);
 886
 887         /*
 888          * Conditionally copyin msg.msg_name.
 889          */
 890         if (msg.msg_name) {
 891                 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 892                 if (error)
 893                         return (error);
 894                 if (!prison_remote_ip(curthread, sa)) {
 895                         kfree(sa, M_SONAME);
 896                         return EAFNOSUPPORT;
 897                 }
 898         }
 899
 900         /*
 901          * Populate auio.
 902          */
 903         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 904                              &auio.uio_resid);
 905         if (error)
 906                 goto cleanup2;
 907         auio.uio_iov = iov;
 908         auio.uio_iovcnt = msg.msg_iovlen;
 909         auio.uio_offset = 0;
 910         auio.uio_segflg = UIO_USERSPACE;
 911         auio.uio_rw = UIO_WRITE;
 912         auio.uio_td = td;
 913
 914         /*
 915          * Conditionally copyin msg.msg_control.
 916          */
 917         if (msg.msg_control) {
 918                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 919                     msg.msg_controllen > MLEN) {
 920                         error = EINVAL;
 921                         goto cleanup;
 922                 }
 923                 control = m_get(M_WAITOK, MT_CONTROL);
 924                 if (control == NULL) {
 925                         error = ENOBUFS;
 926                         goto cleanup;
 927                 }
 928                 control->m_len = msg.msg_controllen;
 929                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 930                                msg.msg_controllen);
 931                 if (error) {
 932                         m_free(control);
 933                         goto cleanup;
 934                 }
 935         }
 936
 937         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 938                              &uap->sysmsg_szresult);
 939
 940 cleanup:
 941         iovec_free(&iov, aiov);
 942 cleanup2:
 943         if (sa)
 944                 kfree(sa, M_SONAME);
 945         return (error);
 946 }
 947
 948 /*
 949  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 950  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 951  * Don't forget to FREE() and m_free() these if they are returned.
 952  */
 953 int
 954 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 955              struct mbuf **control, int *flags, size_t *res)
 956 {
 957         struct thread *td = curthread;
 958         struct file *fp;
 959         size_t len;
 960         int error;
 961         int lflags;
 962         struct socket *so;
 963 #ifdef KTRACE
 964         struct iovec *ktriov = NULL;
 965         struct uio ktruio;
 966 #endif
 967
 968         error = holdsock(td, s, &fp);
 969         if (error)
 970                 return (error);
 971 #ifdef KTRACE
 972         if (KTRPOINT(td, KTR_GENIO)) {
 973                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 974
 975                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 976                 bcopy(auio->uio_iov, ktriov, iovlen);
 977                 ktruio = *auio;
 978         }
 979 #endif
 980         len = auio->uio_resid;
 981         so = (struct socket *)fp->f_data;
 982
 983         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 984                 if (fp->f_flag & FNONBLOCK) {
 985                         if (flags) {
 986                                 *flags |= MSG_FNONBLOCKING;
 987                         } else {
 988                                 lflags = MSG_FNONBLOCKING;
 989                                 flags = &lflags;
 990                         }
 991                 }
 992         }
 993
 994         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 995         if (error) {
 996                 if (auio->uio_resid != len && (error == ERESTART ||
 997                     error == EINTR || error == EWOULDBLOCK))
 998                         error = 0;
 999         }
1000 #ifdef KTRACE
1001         if (ktriov != NULL) {
1002                 if (error == 0) {
1003                         ktruio.uio_iov = ktriov;
1004                         ktruio.uio_resid = len - auio->uio_resid;
1005                         ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
1006                 }
1007                 kfree(ktriov, M_TEMP);
1008         }
1009 #endif
1010         if (error == 0)
1011                 *res = len - auio->uio_resid;
1012         dropfp(td, s, fp);
1013
1014         return (error);
1015 }
1016
1017 /*
1018  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
1019  *                      caddr_t from, int *fromlenaddr)
1020  *
1021  * MPALMOSTSAFE
1022  */
1023 int
1024 sys_recvfrom(struct recvfrom_args *uap)
1025 {
1026         struct thread *td = curthread;
1027         struct uio auio;
1028         struct iovec aiov;
1029         struct sockaddr *sa = NULL;
1030         int error, fromlen;
1031
1032         if (uap->from && uap->fromlenaddr) {
1033                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
1034                 if (error)
1035                         return (error);
1036                 if (fromlen < 0)
1037                         return (EINVAL);
1038         } else {
1039                 fromlen = 0;
1040         }
1041         aiov.iov_base = uap->buf;
1042         aiov.iov_len = uap->len;
1043         auio.uio_iov = &aiov;
1044         auio.uio_iovcnt = 1;
1045         auio.uio_offset = 0;
1046         auio.uio_resid = uap->len;
1047         auio.uio_segflg = UIO_USERSPACE;
1048         auio.uio_rw = UIO_READ;
1049         auio.uio_td = td;
1050
1051         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
1052                              &uap->flags, &uap->sysmsg_szresult);
1053
1054         if (error == 0 && uap->from) {
1055                 /* note: sa may still be NULL */
1056                 if (sa) {
1057                         fromlen = MIN(fromlen, sa->sa_len);
1058                         prison_local_ip(curthread, sa);
1059                         error = copyout(sa, uap->from, fromlen);
1060                 } else {
1061                         fromlen = 0;
1062                 }
1063                 if (error == 0) {
1064                         error = copyout(&fromlen, uap->fromlenaddr,
1065                                         sizeof(fromlen));
1066                 }
1067         }
1068         if (sa)
1069                 kfree(sa, M_SONAME);
1070
1071         return (error);
1072 }
1073
1074 /*
1075  * recvmsg_args(int s, struct msghdr *msg, int flags)
1076  *
1077  * MPALMOSTSAFE
1078  */
1079 int
1080 sys_recvmsg(struct recvmsg_args *uap)
1081 {
1082         struct thread *td = curthread;
1083         struct msghdr msg;
1084         struct uio auio;
1085         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1086         struct mbuf *m, *control = NULL;
1087         struct sockaddr *sa = NULL;
1088         caddr_t ctlbuf;
1089         socklen_t *ufromlenp, *ucontrollenp;
1090         int error, fromlen, controllen, len, flags, *uflagsp;
1091
1092         /*
1093          * This copyin handles everything except the iovec.
1094          */
1095         error = copyin(uap->msg, &msg, sizeof(msg));
1096         if (error)
1097                 return (error);
1098
1099         if (msg.msg_name && msg.msg_namelen < 0)
1100                 return (EINVAL);
1101         if (msg.msg_control && msg.msg_controllen < 0)
1102                 return (EINVAL);
1103
1104         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1105                     msg_namelen));
1106         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1107                        msg_controllen));
1108         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1109                                                         msg_flags));
1110
1111         /*
1112          * Populate auio.
1113          */
1114         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1115                              &auio.uio_resid);
1116         if (error)
1117                 return (error);
1118         auio.uio_iov = iov;
1119         auio.uio_iovcnt = msg.msg_iovlen;
1120         auio.uio_offset = 0;
1121         auio.uio_segflg = UIO_USERSPACE;
1122         auio.uio_rw = UIO_READ;
1123         auio.uio_td = td;
1124
1125         flags = uap->flags;
1126
1127         error = kern_recvmsg(uap->s,
1128                              (msg.msg_name ? &sa : NULL), &auio,
1129                              (msg.msg_control ? &control : NULL), &flags,
1130                              &uap->sysmsg_szresult);
1131
1132         /*
1133          * Conditionally copyout the name and populate the namelen field.
1134          */
1135         if (error == 0 && msg.msg_name) {
1136                 /* note: sa may still be NULL */
1137                 if (sa != NULL) {
1138                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
1139                         prison_local_ip(curthread, sa);
1140                         error = copyout(sa, msg.msg_name, fromlen);
1141                 } else {
1142                         fromlen = 0;
1143                 }
1144                 if (error == 0)
1145                         error = copyout(&fromlen, ufromlenp,
1146                             sizeof(*ufromlenp));
1147         }
1148
1149         /*
1150          * Copyout msg.msg_control and msg.msg_controllen.
1151          */
1152         if (error == 0 && msg.msg_control) {
1153                 len = msg.msg_controllen;
1154                 m = control;
1155                 ctlbuf = (caddr_t)msg.msg_control;
1156
1157                 while(m && len > 0) {
1158                         unsigned int tocopy;
1159
1160                         if (len >= m->m_len) {
1161                                 tocopy = m->m_len;
1162                         } else {
1163                                 msg.msg_flags |= MSG_CTRUNC;
1164                                 tocopy = len;
1165                         }
1166
1167                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1168                         if (error)
1169                                 goto cleanup;
1170
1171                         ctlbuf += tocopy;
1172                         len -= tocopy;
1173                         m = m->m_next;
1174                 }
1175                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1176                 error = copyout(&controllen, ucontrollenp,
1177                     sizeof(*ucontrollenp));
1178         }
1179
1180         if (error == 0)
1181                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1182
1183 cleanup:
1184         if (sa)
1185                 kfree(sa, M_SONAME);
1186         iovec_free(&iov, aiov);
1187         if (control)
1188                 m_freem(control);
1189         return (error);
1190 }
1191
1192 /*
1193  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1194  * in kernel pointer instead of a userland pointer.  This allows us
1195  * to manipulate socket options in the emulation code.
1196  */
1197 int
1198 kern_setsockopt(int s, struct sockopt *sopt)
1199 {
1200         struct thread *td = curthread;
1201         struct file *fp;
1202         int error;
1203
1204         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1205                 return (EFAULT);
1206         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1207                 return (EINVAL);
1208         if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1209                 return (EINVAL);
1210
1211         error = holdsock(td, s, &fp);
1212         if (error)
1213                 return (error);
1214
1215         error = sosetopt((struct socket *)fp->f_data, sopt);
1216         dropfp(td, s, fp);
1217
1218         return (error);
1219 }
1220
1221 /*
1222  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1223  *
1224  * MPALMOSTSAFE
1225  */
1226 int
1227 sys_setsockopt(struct setsockopt_args *uap)
1228 {
1229         struct thread *td = curthread;
1230         struct sockopt sopt;
1231         int error;
1232
1233         sopt.sopt_level = uap->level;
1234         sopt.sopt_name = uap->name;
1235         sopt.sopt_valsize = uap->valsize;
1236         sopt.sopt_td = td;
1237         sopt.sopt_val = NULL;
1238
1239         if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1240                 return (EINVAL);
1241         if (uap->val) {
1242                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1243                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1244                 if (error)
1245                         goto out;
1246         }
1247
1248         error = kern_setsockopt(uap->s, &sopt);
1249 out:
1250         if (uap->val)
1251                 kfree(sopt.sopt_val, M_TEMP);
1252         return(error);
1253 }
1254
1255 /*
1256  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1257  * in kernel pointer instead of a userland pointer.  This allows us
1258  * to manipulate socket options in the emulation code.
1259  */
1260 int
1261 kern_getsockopt(int s, struct sockopt *sopt)
1262 {
1263         struct thread *td = curthread;
1264         struct file *fp;
1265         int error;
1266
1267         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1268                 return (EFAULT);
1269         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1270                 return (EINVAL);
1271
1272         error = holdsock(td, s, &fp);
1273         if (error)
1274                 return (error);
1275
1276         error = sogetopt((struct socket *)fp->f_data, sopt);
1277         dropfp(td, s, fp);
1278
1279         return (error);
1280 }
1281
1282 /*
1283  * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1284  *
1285  * MPALMOSTSAFE
1286  */
1287 int
1288 sys_getsockopt(struct getsockopt_args *uap)
1289 {
1290         struct thread *td = curthread;
1291         struct sockopt sopt;
1292         int error, valsize, valszmax, mflag = 0;
1293
1294         if (uap->val) {
1295                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1296                 if (error)
1297                         return (error);
1298         } else {
1299                 valsize = 0;
1300         }
1301
1302         sopt.sopt_level = uap->level;
1303         sopt.sopt_name = uap->name;
1304         sopt.sopt_valsize = valsize;
1305         sopt.sopt_td = td;
1306         sopt.sopt_val = NULL;
1307
1308         if (td->td_proc->p_ucred->cr_uid == 0) {
1309                 valszmax = SOMAXOPT_SIZE0;
1310                 mflag = M_NULLOK;
1311         } else {
1312                 valszmax = SOMAXOPT_SIZE;
1313         }
1314         if (sopt.sopt_valsize > valszmax) /* unsigned */
1315                 return (EINVAL);
1316         if (uap->val) {
1317                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP,
1318                     M_WAITOK | mflag);
1319                 if (sopt.sopt_val == NULL)
1320                         return (ENOBUFS);
1321                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1322                 if (error)
1323                         goto out;
1324         }
1325
1326         error = kern_getsockopt(uap->s, &sopt);
1327         if (error)
1328                 goto out;
1329         valsize = sopt.sopt_valsize;
1330         error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1331         if (error)
1332                 goto out;
1333         if (uap->val)
1334                 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1335 out:
1336         if (uap->val)
1337                 kfree(sopt.sopt_val, M_TEMP);
1338         return (error);
1339 }
1340
1341 /*
1342  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1343  * This allows kern_getsockname() to return a pointer to an allocated struct
1344  * sockaddr which must be freed later with FREE().  The caller must
1345  * initialize *name to NULL.
1346  */
1347 int
1348 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1349 {
1350         struct thread *td = curthread;
1351         struct file *fp;
1352         struct socket *so;
1353         struct sockaddr *sa = NULL;
1354         int error;
1355
1356         error = holdsock(td, s, &fp);
1357         if (error)
1358                 return (error);
1359         if (*namelen < 0) {
1360                 fdrop(fp);
1361                 return (EINVAL);
1362         }
1363         so = (struct socket *)fp->f_data;
1364         error = so_pru_sockaddr(so, &sa);
1365         if (error == 0) {
1366                 if (sa == NULL) {
1367                         *namelen = 0;
1368                 } else {
1369                         *namelen = MIN(*namelen, sa->sa_len);
1370                         *name = sa;
1371                 }
1372         }
1373         dropfp(td, s, fp);
1374
1375         return (error);
1376 }
1377
1378 /*
1379  * getsockname_args(int fdes, caddr_t asa, int *alen)
1380  *
1381  * Get socket name.
1382  *
1383  * MPALMOSTSAFE
1384  */
1385 int
1386 sys_getsockname(struct getsockname_args *uap)
1387 {
1388         struct sockaddr *sa = NULL;
1389         struct sockaddr satmp;
1390         int error, sa_len_in, sa_len_out;
1391
1392         error = copyin(uap->alen, &sa_len_in, sizeof(sa_len_in));
1393         if (error)
1394                 return (error);
1395
1396         sa_len_out = sa_len_in;
1397         error = kern_getsockname(uap->fdes, &sa, &sa_len_out);
1398
1399         if (error == 0) {
1400                 if (sa) {
1401                         prison_local_ip(curthread, sa);
1402                         error = copyout(sa, uap->asa, sa_len_out);
1403                 } else {
1404                         /*
1405                          * unnamed uipc sockets don't bother storing
1406                          * sockaddr, simulate an AF_LOCAL sockaddr.
1407                          */
1408                         sa_len_out = sizeof(satmp);
1409                         if (sa_len_out > sa_len_in)
1410                                 sa_len_out = sa_len_in;
1411                         if (sa_len_out < 0)
1412                                 sa_len_out = 0;
1413                         bzero(&satmp, sizeof(satmp));
1414                         satmp.sa_len = sa_len_out;
1415                         satmp.sa_family = AF_LOCAL;
1416                         error = copyout(&satmp, uap->asa, sa_len_out);
1417                 }
1418         }
1419         if (error == 0 && sa_len_out != sa_len_in)
1420                 error = copyout(&sa_len_out, uap->alen, sizeof(*uap->alen));
1421         if (sa)
1422                 kfree(sa, M_SONAME);
1423         return (error);
1424 }
1425
1426 /*
1427  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1428  * This allows kern_getpeername() to return a pointer to an allocated struct
1429  * sockaddr which must be freed later with FREE().  The caller must
1430  * initialize *name to NULL.
1431  */
1432 int
1433 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1434 {
1435         struct thread *td = curthread;
1436         struct file *fp;
1437         struct socket *so;
1438         struct sockaddr *sa = NULL;
1439         int error;
1440
1441         error = holdsock(td, s, &fp);
1442         if (error)
1443                 return (error);
1444         if (*namelen < 0) {
1445                 fdrop(fp);
1446                 return (EINVAL);
1447         }
1448         so = (struct socket *)fp->f_data;
1449         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1450                 fdrop(fp);
1451                 return (ENOTCONN);
1452         }
1453         error = so_pru_peeraddr(so, &sa);
1454         if (error == 0) {
1455                 if (sa == NULL) {
1456                         *namelen = 0;
1457                 } else {
1458                         *namelen = MIN(*namelen, sa->sa_len);
1459                         *name = sa;
1460                 }
1461         }
1462         dropfp(td, s, fp);
1463
1464         return (error);
1465 }
1466
1467 /*
1468  * getpeername_args(int fdes, caddr_t asa, int *alen)
1469  *
1470  * Get name of peer for connected socket.
1471  *
1472  * MPALMOSTSAFE
1473  */
1474 int
1475 sys_getpeername(struct getpeername_args *uap)
1476 {
1477         struct sockaddr *sa = NULL;
1478         int error, sa_len;
1479
1480         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1481         if (error)
1482                 return (error);
1483
1484         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1485
1486         if (error == 0) {
1487                 prison_local_ip(curthread, sa);
1488                 error = copyout(sa, uap->asa, sa_len);
1489         }
1490         if (error == 0)
1491                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1492         if (sa)
1493                 kfree(sa, M_SONAME);
1494         return (error);
1495 }
1496
1497 int
1498 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1499 {
1500         struct sockaddr *sa;
1501         int error;
1502
1503         *namp = NULL;
1504         if (len > SOCK_MAXADDRLEN)
1505                 return ENAMETOOLONG;
1506         if (len < offsetof(struct sockaddr, sa_data[0]))
1507                 return EDOM;
1508         sa = kmalloc(len, M_SONAME, M_WAITOK);
1509         error = copyin(uaddr, sa, len);
1510         if (error) {
1511                 kfree(sa, M_SONAME);
1512         } else {
1513                 sa->sa_len = len;
1514                 *namp = sa;
1515         }
1516         return error;
1517 }
1518
1519 /*
1520  * Detach a mapped page and release resources back to the system.
1521  * We must release our wiring and if the object is ripped out
1522  * from under the vm_page we become responsible for freeing the
1523  * page.
1524  *
1525  * MPSAFE
1526  */
1527 static void
1528 sf_buf_mfree(void *arg)
1529 {
1530         struct sf_buf *sf = arg;
1531         vm_page_t m;
1532
1533         m = sf_buf_page(sf);
1534         if (sf_buf_free(sf)) {
1535                 /* sf invalid now */
1536                 vm_page_sbusy_drop(m);
1537 #if 0
1538                 if (m->object == NULL &&
1539                     m->wire_count == 0 &&
1540                     (m->flags & PG_NEED_COMMIT) == 0) {
1541                         vm_page_free(m);
1542                 } else {
1543                         vm_page_wakeup(m);
1544                 }
1545 #endif
1546         }
1547 }
1548
1549 /*
1550  * sendfile(2).
1551  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1552  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1553  *
1554  * Send a file specified by 'fd' and starting at 'offset' to a socket
1555  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1556  * nbytes == 0. Optionally add a header and/or trailer to the socket
1557  * output. If specified, write the total number of bytes sent into *sbytes.
1558  *
1559  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1560  * the headers to count against the remaining bytes to be sent from
1561  * the file descriptor.  We may wish to implement a compatibility syscall
1562  * in the future.
1563  *
1564  * MPALMOSTSAFE
1565  */
1566 int
1567 sys_sendfile(struct sendfile_args *uap)
1568 {
1569         struct thread *td = curthread;
1570         struct file *fp;
1571         struct vnode *vp = NULL;
1572         struct sf_hdtr hdtr;
1573         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1574         struct uio auio;
1575         struct mbuf *mheader = NULL;
1576         size_t hbytes = 0;
1577         size_t tbytes;
1578         off_t hdtr_size = 0;
1579         off_t sbytes;
1580         int error;
1581
1582         /*
1583          * Do argument checking. Must be a regular file in, stream
1584          * type and connected socket out, positive offset.
1585          */
1586         fp = holdfp(td, uap->fd, FREAD);
1587         if (fp == NULL) {
1588                 return (EBADF);
1589         }
1590         if (fp->f_type != DTYPE_VNODE) {
1591                 fdrop(fp);
1592                 return (EINVAL);
1593         }
1594         vp = (struct vnode *)fp->f_data;
1595         vref(vp);
1596         dropfp(td, uap->fd, fp);
1597
1598         /*
1599          * If specified, get the pointer to the sf_hdtr struct for
1600          * any headers/trailers.
1601          */
1602         if (uap->hdtr) {
1603                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1604                 if (error)
1605                         goto done;
1606                 /*
1607                  * Send any headers.
1608                  */
1609                 if (hdtr.headers) {
1610                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1611                                              hdtr.hdr_cnt, &hbytes);
1612                         if (error)
1613                                 goto done;
1614                         auio.uio_iov = iov;
1615                         auio.uio_iovcnt = hdtr.hdr_cnt;
1616                         auio.uio_offset = 0;
1617                         auio.uio_segflg = UIO_USERSPACE;
1618                         auio.uio_rw = UIO_WRITE;
1619                         auio.uio_td = td;
1620                         auio.uio_resid = hbytes;
1621
1622                         mheader = m_uiomove(&auio);
1623
1624                         iovec_free(&iov, aiov);
1625                         if (mheader == NULL)
1626                                 goto done;
1627                 }
1628         }
1629
1630         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1631                               &sbytes, uap->flags);
1632         if (error)
1633                 goto done;
1634
1635         /*
1636          * Send trailers. Wimp out and use writev(2).
1637          */
1638         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1639                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1640                                      hdtr.trl_cnt, &auio.uio_resid);
1641                 if (error)
1642                         goto done;
1643                 auio.uio_iov = iov;
1644                 auio.uio_iovcnt = hdtr.trl_cnt;
1645                 auio.uio_offset = 0;
1646                 auio.uio_segflg = UIO_USERSPACE;
1647                 auio.uio_rw = UIO_WRITE;
1648                 auio.uio_td = td;
1649
1650                 tbytes = 0;     /* avoid gcc warnings */
1651                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1652
1653                 iovec_free(&iov, aiov);
1654                 if (error)
1655                         goto done;
1656                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1657         }
1658
1659 done:
1660         if (vp)
1661                 vrele(vp);
1662         if (uap->sbytes != NULL) {
1663                 sbytes += hdtr_size;
1664                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1665         }
1666         return (error);
1667 }
1668
1669 int
1670 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1671               struct mbuf *mheader, off_t *sbytes, int flags)
1672 {
1673         struct thread *td = curthread;
1674         struct vm_object *obj;
1675         struct socket *so;
1676         struct file *fp;
1677         struct mbuf *m, *mp;
1678         struct sf_buf *sf;
1679         struct vm_page *pg;
1680         off_t off, xfsize, xbytes;
1681         off_t hbytes = 0;
1682         int error = 0;
1683
1684         if (vp->v_type != VREG) {
1685                 error = EINVAL;
1686                 goto done0;
1687         }
1688         if ((obj = vp->v_object) == NULL) {
1689                 error = EINVAL;
1690                 goto done0;
1691         }
1692         error = holdsock(td, sfd, &fp);
1693         if (error)
1694                 goto done0;
1695         so = (struct socket *)fp->f_data;
1696         if (so->so_type != SOCK_STREAM) {
1697                 error = EINVAL;
1698                 goto done1;
1699         }
1700         if ((so->so_state & SS_ISCONNECTED) == 0) {
1701                 error = ENOTCONN;
1702                 goto done1;
1703         }
1704         if (offset < 0) {
1705                 error = EINVAL;
1706                 goto done1;
1707         }
1708
1709         /*
1710          * preallocation is required for asynchronous passing of mbufs,
1711          * otherwise we can wind up building up an infinite number of
1712          * mbufs during the asynchronous latency.
1713          */
1714         if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) {
1715                 error = EINVAL;
1716                 goto done1;
1717         }
1718
1719         *sbytes = 0;
1720         xbytes = 0;
1721
1722         /*
1723          * Protect against multiple writers to the socket.
1724          * We need at least a shared lock on the VM object
1725          */
1726         ssb_lock(&so->so_snd, M_WAITOK);
1727         vm_object_hold_shared(obj);
1728
1729         /*
1730          * Loop through the pages in the file, starting with the requested
1731          * offset. Get a file page (do I/O if necessary), map the file page
1732          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1733          * it on the socket.
1734          */
1735         for (off = offset; ;
1736              off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) {
1737                 vm_pindex_t pindex;
1738                 vm_offset_t pgoff;
1739                 long space;
1740                 int loops;
1741
1742                 pindex = OFF_TO_IDX(off);
1743                 loops = 0;
1744
1745 retry_lookup:
1746                 /*
1747                  * Calculate the amount to transfer. Not to exceed a page,
1748                  * the EOF, or the passed in nbytes.
1749                  */
1750                 xfsize = vp->v_filesize - off;
1751                 if (xfsize > PAGE_SIZE)
1752                         xfsize = PAGE_SIZE;
1753                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1754                 if (PAGE_SIZE - pgoff < xfsize)
1755                         xfsize = PAGE_SIZE - pgoff;
1756                 if (nbytes && xfsize > (nbytes - xbytes))
1757                         xfsize = nbytes - xbytes;
1758                 if (xfsize <= 0)
1759                         break;
1760                 /*
1761                  * Optimize the non-blocking case by looking at the socket space
1762                  * before going to the extra work of constituting the sf_buf.
1763                  */
1764                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1765                         space = ssb_space_prealloc(&so->so_snd);
1766                 else
1767                         space = ssb_space(&so->so_snd);
1768
1769                 if ((fp->f_flag & FNONBLOCK) && space <= 0) {
1770                         if (so->so_state & SS_CANTSENDMORE)
1771                                 error = EPIPE;
1772                         else
1773                                 error = EAGAIN;
1774                         goto done;
1775                 }
1776
1777                 /*
1778                  * Attempt to look up the page.
1779                  *
1780                  * Try to find the data using a shared vm_object token and
1781                  * vm_page_lookup_sbusy_try() first.
1782                  *
1783                  * If data is missing, use a UIO_NOCOPY VOP_READ to load
1784                  * the missing data and loop back up.  We avoid all sorts
1785                  * of problems by not trying to hold onto the page during
1786                  * the I/O.
1787                  *
1788                  * NOTE: The soft-busy will temporary block filesystem
1789                  *       truncation operations when a file is removed
1790                  *       while the sendfile is running.
1791                  */
1792                 pg = vm_page_lookup_sbusy_try(obj, pindex, pgoff, xfsize);
1793                 if (pg == NULL) {
1794                         struct uio auio;
1795                         struct iovec aiov;
1796                         int bsize;
1797
1798                         if (++loops > 100000) {
1799                                 kprintf("sendfile: VOP operation failed "
1800                                         "to retain page\n");
1801                                 error = EIO;
1802                                 goto done;
1803                         }
1804
1805                         vm_object_drop(obj);
1806                         bsize = vp->v_mount->mnt_stat.f_iosize;
1807                         auio.uio_iov = &aiov;
1808                         auio.uio_iovcnt = 1;
1809                         aiov.iov_base = 0;
1810                         aiov.iov_len = MAXBSIZE;
1811                         auio.uio_resid = MAXBSIZE;
1812                         auio.uio_offset = trunc_page(off);
1813                         auio.uio_segflg = UIO_NOCOPY;
1814                         auio.uio_rw = UIO_READ;
1815                         auio.uio_td = td;
1816
1817                         vn_lock(vp, LK_SHARED | LK_RETRY);
1818                         error = VOP_READ_FP(vp, &auio,
1819                                          IO_VMIO | ((MAXBSIZE / bsize) << 16),
1820                                          td->td_ucred, fp);
1821                         vn_unlock(vp);
1822                         vm_object_hold_shared(obj);
1823
1824                         if (error)
1825                                 goto done;
1826                         goto retry_lookup;
1827                 }
1828
1829                 /*
1830                  * Get a sendfile buf. We usually wait as long as necessary,
1831                  * but this wait can be interrupted.
1832                  */
1833                 if ((sf = sf_buf_alloc(pg)) == NULL) {
1834                         vm_page_sbusy_drop(pg);
1835                         /* vm_page_try_to_free(pg); */
1836                         error = EINTR;
1837                         goto done;
1838                 }
1839
1840                 /*
1841                  * Get an mbuf header and set it up as having external storage.
1842                  */
1843                 MGETHDR(m, M_WAITOK, MT_DATA);
1844                 if (m == NULL) {
1845                         error = ENOBUFS;
1846                         vm_page_sbusy_drop(pg);
1847                         /* vm_page_try_to_free(pg); */
1848                         sf_buf_free(sf);
1849                         goto done;
1850                 }
1851
1852                 m->m_ext.ext_free = sf_buf_mfree;
1853                 m->m_ext.ext_ref = sf_buf_ref;
1854                 m->m_ext.ext_arg = sf;
1855                 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1856                 m->m_ext.ext_size = PAGE_SIZE;
1857                 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1858                 m->m_flags |= M_EXT;
1859                 m->m_pkthdr.len = m->m_len = xfsize;
1860                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1861
1862                 if (mheader != NULL) {
1863                         hbytes = mheader->m_pkthdr.len;
1864                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1865                         m_cat(mheader, m);
1866                         m = mheader;
1867                         mheader = NULL;
1868                 } else {
1869                         hbytes = 0;
1870                 }
1871
1872                 /*
1873                  * Add the buffer to the socket buffer chain.
1874                  */
1875                 crit_enter();
1876 retry_space:
1877                 /*
1878                  * Make sure that the socket is still able to take more data.
1879                  * CANTSENDMORE being true usually means that the connection
1880                  * was closed. so_error is true when an error was sensed after
1881                  * a previous send.
1882                  * The state is checked after the page mapping and buffer
1883                  * allocation above since those operations may block and make
1884                  * any socket checks stale. From this point forward, nothing
1885                  * blocks before the pru_send (or more accurately, any blocking
1886                  * results in a loop back to here to re-check).
1887                  */
1888                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1889                         if (so->so_state & SS_CANTSENDMORE) {
1890                                 error = EPIPE;
1891                         } else {
1892                                 error = so->so_error;
1893                                 so->so_error = 0;
1894                         }
1895                         m_freem(m);
1896                         crit_exit();
1897                         goto done;
1898                 }
1899                 /*
1900                  * Wait for socket space to become available. We do this just
1901                  * after checking the connection state above in order to avoid
1902                  * a race condition with ssb_wait().
1903                  */
1904                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1905                         space = ssb_space_prealloc(&so->so_snd);
1906                 else
1907                         space = ssb_space(&so->so_snd);
1908
1909                 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1910                         if (fp->f_flag & FNONBLOCK) {
1911                                 m_freem(m);
1912                                 crit_exit();
1913                                 error = EAGAIN;
1914                                 goto done;
1915                         }
1916                         error = ssb_wait(&so->so_snd);
1917                         /*
1918                          * An error from ssb_wait usually indicates that we've
1919                          * been interrupted by a signal. If we've sent anything
1920                          * then return bytes sent, otherwise return the error.
1921                          */
1922                         if (error) {
1923                                 m_freem(m);
1924                                 crit_exit();
1925                                 goto done;
1926                         }
1927                         goto retry_space;
1928                 }
1929
1930                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1931                         for (mp = m; mp != NULL; mp = mp->m_next)
1932                                 ssb_preallocstream(&so->so_snd, mp);
1933                 }
1934                 if (use_sendfile_async)
1935                         error = so_pru_senda(so, 0, m, NULL, NULL, td);
1936                 else
1937                         error = so_pru_send(so, 0, m, NULL, NULL, td);
1938
1939                 crit_exit();
1940                 if (error)
1941                         goto done;
1942         }
1943         if (mheader != NULL) {
1944                 *sbytes += mheader->m_pkthdr.len;
1945
1946                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1947                         for (mp = mheader; mp != NULL; mp = mp->m_next)
1948                                 ssb_preallocstream(&so->so_snd, mp);
1949                 }
1950                 if (use_sendfile_async)
1951                         error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1952                 else
1953                         error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1954
1955                 mheader = NULL;
1956         }
1957 done:
1958         vm_object_drop(obj);
1959         ssb_unlock(&so->so_snd);
1960 done1:
1961         dropfp(td, sfd, fp);
1962 done0:
1963         if (mheader != NULL)
1964                 m_freem(mheader);
1965         return (error);
1966 }