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. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      @(#)uipc_syscalls.c     8.4 (Berkeley) 2/21/94
  37  * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
  38  * $DragonFly: src/sys/kern/uipc_syscalls.c,v 1.59 2005/12/01 18:40:56 dillon Exp $
  39  */
  40
  41 #include "opt_ktrace.h"
  42 #include "opt_sctp.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/sysproto.h>
  48 #include <sys/malloc.h>
  49 #include <sys/filedesc.h>
  50 #include <sys/event.h>
  51 #include <sys/proc.h>
  52 #include <sys/fcntl.h>
  53 #include <sys/file.h>
  54 #include <sys/filio.h>
  55 #include <sys/kern_syscall.h>
  56 #include <sys/mbuf.h>
  57 #include <sys/protosw.h>
  58 #include <sys/sfbuf.h>
  59 #include <sys/socket.h>
  60 #include <sys/socketvar.h>
  61 #include <sys/socketops.h>
  62 #include <sys/uio.h>
  63 #include <sys/vnode.h>
  64 #include <sys/lock.h>
  65 #include <sys/mount.h>
  66 #ifdef KTRACE
  67 #include <sys/ktrace.h>
  68 #endif
  69 #include <vm/vm.h>
  70 #include <vm/vm_object.h>
  71 #include <vm/vm_page.h>
  72 #include <vm/vm_pageout.h>
  73 #include <vm/vm_kern.h>
  74 #include <vm/vm_extern.h>
  75 #include <sys/file2.h>
  76 #include <sys/signalvar.h>
  77 #include <sys/serialize.h>
  78
  79 #include <sys/thread2.h>
  80 #include <sys/msgport2.h>
  81
  82 #ifdef SCTP
  83 #include <netinet/sctp_peeloff.h>
  84 #endif /* SCTP */
  85
  86 struct sfbuf_mref {
  87         struct sf_buf   *sf;
  88         int             mref_count;
  89         struct lwkt_serialize serializer;
  90 };
  91
  92 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
  93
  94 /*
  95  * System call interface to the socket abstraction.
  96  */
  97
  98 extern  struct fileops socketops;
  99
 100 /*
 101  * socket_args(int domain, int type, int protocol)
 102  */
 103 int
 104 kern_socket(int domain, int type, int protocol, int *res)
 105 {
 106         struct thread *td = curthread;
 107         struct proc *p = td->td_proc;
 108         struct filedesc *fdp;
 109         struct socket *so;
 110         struct file *fp;
 111         int fd, error;
 112
 113         KKASSERT(p);
 114         fdp = p->p_fd;
 115
 116         error = falloc(p, &fp, &fd);
 117         if (error)
 118                 return (error);
 119         error = socreate(domain, &so, type, protocol, td);
 120         if (error) {
 121                 if (fdp->fd_files[fd].fp == fp) {
 122                         funsetfd(fdp, fd);
 123                         fdrop(fp, td);
 124                 }
 125         } else {
 126                 fp->f_type = DTYPE_SOCKET;
 127                 fp->f_flag = FREAD | FWRITE;
 128                 fp->f_ops = &socketops;
 129                 fp->f_data = so;
 130                 *res = fd;
 131         }
 132         fdrop(fp, td);
 133         return (error);
 134 }
 135
 136 int
 137 socket(struct socket_args *uap)
 138 {
 139         int error;
 140
 141         error = kern_socket(uap->domain, uap->type, uap->protocol,
 142             &uap->sysmsg_result);
 143
 144         return (error);
 145 }
 146
 147 int
 148 kern_bind(int s, struct sockaddr *sa)
 149 {
 150         struct thread *td = curthread;
 151         struct proc *p = td->td_proc;
 152         struct file *fp;
 153         int error;
 154
 155         KKASSERT(p);
 156         error = holdsock(p->p_fd, s, &fp);
 157         if (error)
 158                 return (error);
 159         error = sobind((struct socket *)fp->f_data, sa, td);
 160         fdrop(fp, td);
 161         return (error);
 162 }
 163
 164 /*
 165  * bind_args(int s, caddr_t name, int namelen)
 166  */
 167 int
 168 bind(struct bind_args *uap)
 169 {
 170         struct sockaddr *sa;
 171         int error;
 172
 173         error = getsockaddr(&sa, uap->name, uap->namelen);
 174         if (error)
 175                 return (error);
 176         error = kern_bind(uap->s, sa);
 177         FREE(sa, M_SONAME);
 178
 179         return (error);
 180 }
 181
 182 int
 183 kern_listen(int s, int backlog)
 184 {
 185         struct thread *td = curthread;
 186         struct proc *p = td->td_proc;
 187         struct file *fp;
 188         int error;
 189
 190         KKASSERT(p);
 191         error = holdsock(p->p_fd, s, &fp);
 192         if (error)
 193                 return (error);
 194         error = solisten((struct socket *)fp->f_data, backlog, td);
 195         fdrop(fp, td);
 196         return(error);
 197 }
 198
 199 /*
 200  * listen_args(int s, int backlog)
 201  */
 202 int
 203 listen(struct listen_args *uap)
 204 {
 205         int error;
 206
 207         error = kern_listen(uap->s, uap->backlog);
 208         return (error);
 209 }
 210
 211 /*
 212  * Returns the accepted socket as well.
 213  */
 214 static boolean_t
 215 soaccept_predicate(struct netmsg *msg0)
 216 {
 217         struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
 218         struct socket *head = msg->nm_so;
 219
 220         if (head->so_error != 0) {
 221                 msg->nm_lmsg.ms_error = head->so_error;
 222                 return (TRUE);
 223         }
 224         if (!TAILQ_EMPTY(&head->so_comp)) {
 225                 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
 226                 msg->nm_so = TAILQ_FIRST(&head->so_comp);
 227                 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
 228                 head->so_qlen--;
 229
 230                 msg->nm_lmsg.ms_error = 0;
 231                 return (TRUE);
 232         }
 233         if (head->so_state & SS_CANTRCVMORE) {
 234                 msg->nm_lmsg.ms_error = ECONNABORTED;
 235                 return (TRUE);
 236         }
 237         if (head->so_state & SS_NBIO) {
 238                 msg->nm_lmsg.ms_error = EWOULDBLOCK;
 239                 return (TRUE);
 240         }
 241
 242         return (FALSE);
 243 }
 244
 245 /*
 246  * The second argument to kern_accept() is a handle to a struct sockaddr.
 247  * This allows kern_accept() to return a pointer to an allocated struct
 248  * sockaddr which must be freed later with FREE().  The caller must
 249  * initialize *name to NULL.
 250  */
 251 int
 252 kern_accept(int s, struct sockaddr **name, int *namelen, int *res)
 253 {
 254         struct thread *td = curthread;
 255         struct proc *p = td->td_proc;
 256         struct filedesc *fdp = p->p_fd;
 257         struct file *lfp = NULL;
 258         struct file *nfp = NULL;
 259         struct sockaddr *sa;
 260         struct socket *head, *so;
 261         struct netmsg_so_notify msg;
 262         lwkt_port_t port;
 263         int fd;
 264         u_int fflag;            /* type must match fp->f_flag */
 265         int error, tmp;
 266
 267         if (name && namelen && *namelen < 0)
 268                 return (EINVAL);
 269
 270         error = holdsock(fdp, s, &lfp);
 271         if (error)
 272                 return (error);
 273
 274         error = falloc(p, &nfp, &fd);
 275         if (error) {            /* Probably ran out of file descriptors. */
 276                 *res = -1;
 277                 fdrop(lfp, td);
 278                 return (error);
 279         }
 280         *res = fd;
 281
 282         head = (struct socket *)lfp->f_data;
 283         if ((head->so_options & SO_ACCEPTCONN) == 0) {
 284                 error = EINVAL;
 285                 goto done;
 286         }
 287
 288         /* optimize for uniprocessor case later XXX JH */
 289         port = head->so_proto->pr_mport(head, NULL, PRU_PRED);
 290         lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport,
 291                      MSGF_PCATCH | MSGF_ABORTABLE,
 292                      lwkt_cmd_func(netmsg_so_notify),
 293                      lwkt_cmd_func(netmsg_so_notify_abort));
 294         msg.nm_predicate = soaccept_predicate;
 295         msg.nm_so = head;
 296         msg.nm_etype = NM_REVENT;
 297         error = lwkt_domsg(port, &msg.nm_lmsg);
 298         if (error)
 299                 goto done;
 300
 301         /*
 302          * At this point we have the connection that's ready to be accepted.
 303          */
 304         so = msg.nm_so;
 305
 306         fflag = lfp->f_flag;
 307
 308         /* connection has been removed from the listen queue */
 309         KNOTE(&head->so_rcv.sb_sel.si_note, 0);
 310
 311         so->so_state &= ~SS_COMP;
 312         so->so_head = NULL;
 313         if (head->so_sigio != NULL)
 314                 fsetown(fgetown(head->so_sigio), &so->so_sigio);
 315
 316         nfp->f_type = DTYPE_SOCKET;
 317         nfp->f_flag = fflag;
 318         nfp->f_ops = &socketops;
 319         nfp->f_data = so;
 320         /* Sync socket nonblocking/async state with file flags */
 321         tmp = fflag & FNONBLOCK;
 322         (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td);
 323         tmp = fflag & FASYNC;
 324         (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td);
 325
 326         sa = NULL;
 327         error = soaccept(so, &sa);
 328
 329         /*
 330          * Set the returned name and namelen as applicable.  Set the returned
 331          * namelen to 0 for older code which might ignore the return value
 332          * from accept.
 333          */
 334         if (error == 0) {
 335                 if (sa && name && namelen) {
 336                         if (*namelen > sa->sa_len)
 337                                 *namelen = sa->sa_len;
 338                         *name = sa;
 339                 } else {
 340                         if (sa)
 341                                 FREE(sa, M_SONAME);
 342                 }
 343         }
 344
 345 done:
 346         /*
 347          * close the new descriptor, assuming someone hasn't ripped it
 348          * out from under us.  Note that *res is normally ignored if an
 349          * error is returned but a syscall message will still have access
 350          * to the result code.
 351          */
 352         if (error) {
 353                 *res = -1;
 354                 if (fdp->fd_files[fd].fp == nfp) {
 355                         funsetfd(fdp, fd);
 356                         fdrop(nfp, td);
 357                 }
 358         }
 359
 360         /*
 361          * Release explicitly held references before returning.
 362          */
 363         if (nfp)
 364                 fdrop(nfp, td);
 365         fdrop(lfp, td);
 366         return (error);
 367 }
 368
 369 /*
 370  * accept_args(int s, caddr_t name, int *anamelen)
 371  */
 372 int
 373 accept(struct accept_args *uap)
 374 {
 375         struct sockaddr *sa = NULL;
 376         int sa_len;
 377         int error;
 378
 379         if (uap->name) {
 380                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 381                 if (error)
 382                         return (error);
 383
 384                 error = kern_accept(uap->s, &sa, &sa_len, &uap->sysmsg_result);
 385
 386                 if (error == 0)
 387                         error = copyout(sa, uap->name, sa_len);
 388                 if (error == 0) {
 389                         error = copyout(&sa_len, uap->anamelen,
 390                             sizeof(*uap->anamelen));
 391                 }
 392                 if (sa)
 393                         FREE(sa, M_SONAME);
 394         } else {
 395                 error = kern_accept(uap->s, NULL, 0, &uap->sysmsg_result);
 396         }
 397         return (error);
 398 }
 399
 400 /*
 401  * Returns TRUE if predicate satisfied.
 402  */
 403 static boolean_t
 404 soconnected_predicate(struct netmsg *msg0)
 405 {
 406         struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
 407         struct socket *so = msg->nm_so;
 408
 409         /* check predicate */
 410         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 411                 msg->nm_lmsg.ms_error = so->so_error;
 412                 return (TRUE);
 413         }
 414
 415         return (FALSE);
 416 }
 417
 418 int
 419 kern_connect(int s, struct sockaddr *sa)
 420 {
 421         struct thread *td = curthread;
 422         struct proc *p = td->td_proc;
 423         struct file *fp;
 424         struct socket *so;
 425         int error;
 426
 427         error = holdsock(p->p_fd, s, &fp);
 428         if (error)
 429                 return (error);
 430         so = (struct socket *)fp->f_data;
 431         if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
 432                 error = EALREADY;
 433                 goto done;
 434         }
 435         error = soconnect(so, sa, td);
 436         if (error)
 437                 goto bad;
 438         if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
 439                 error = EINPROGRESS;
 440                 goto done;
 441         }
 442         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 443                 struct netmsg_so_notify msg;
 444                 lwkt_port_t port;
 445
 446                 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
 447                 lwkt_initmsg(&msg.nm_lmsg,
 448                             &curthread->td_msgport,
 449                             MSGF_PCATCH | MSGF_ABORTABLE,
 450                             lwkt_cmd_func(netmsg_so_notify),
 451                             lwkt_cmd_func(netmsg_so_notify_abort));
 452                 msg.nm_predicate = soconnected_predicate;
 453                 msg.nm_so = so;
 454                 msg.nm_etype = NM_REVENT;
 455                 error = lwkt_domsg(port, &msg.nm_lmsg);
 456         }
 457         if (error == 0) {
 458                 error = so->so_error;
 459                 so->so_error = 0;
 460         }
 461 bad:
 462         so->so_state &= ~SS_ISCONNECTING;
 463         if (error == ERESTART)
 464                 error = EINTR;
 465 done:
 466         fdrop(fp, td);
 467         return (error);
 468 }
 469
 470 /*
 471  * connect_args(int s, caddr_t name, int namelen)
 472  */
 473 int
 474 connect(struct connect_args *uap)
 475 {
 476         struct sockaddr *sa;
 477         int error;
 478
 479         error = getsockaddr(&sa, uap->name, uap->namelen);
 480         if (error)
 481                 return (error);
 482         error = kern_connect(uap->s, sa);
 483         FREE(sa, M_SONAME);
 484
 485         return (error);
 486 }
 487
 488 int
 489 kern_socketpair(int domain, int type, int protocol, int *sv)
 490 {
 491         struct thread *td = curthread;
 492         struct proc *p = td->td_proc;
 493         struct filedesc *fdp;
 494         struct file *fp1, *fp2;
 495         struct socket *so1, *so2;
 496         int fd, error;
 497
 498         KKASSERT(p);
 499         fdp = p->p_fd;
 500         error = socreate(domain, &so1, type, protocol, td);
 501         if (error)
 502                 return (error);
 503         error = socreate(domain, &so2, type, protocol, td);
 504         if (error)
 505                 goto free1;
 506         error = falloc(p, &fp1, &fd);
 507         if (error)
 508                 goto free2;
 509         sv[0] = fd;
 510         fp1->f_data = so1;
 511         error = falloc(p, &fp2, &fd);
 512         if (error)
 513                 goto free3;
 514         fp2->f_data = so2;
 515         sv[1] = fd;
 516         error = soconnect2(so1, so2);
 517         if (error)
 518                 goto free4;
 519         if (type == SOCK_DGRAM) {
 520                 /*
 521                  * Datagram socket connection is asymmetric.
 522                  */
 523                  error = soconnect2(so2, so1);
 524                  if (error)
 525                         goto free4;
 526         }
 527         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 528         fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
 529         fp1->f_ops = fp2->f_ops = &socketops;
 530         fdrop(fp1, td);
 531         fdrop(fp2, td);
 532         return (error);
 533 free4:
 534         if (fdp->fd_files[sv[1]].fp == fp2) {
 535                 funsetfd(fdp, sv[1]);
 536                 fdrop(fp2, td);
 537         }
 538         fdrop(fp2, td);
 539 free3:
 540         if (fdp->fd_files[sv[0]].fp == fp1) {
 541                 funsetfd(fdp, sv[0]);
 542                 fdrop(fp1, td);
 543         }
 544         fdrop(fp1, td);
 545 free2:
 546         (void)soclose(so2);
 547 free1:
 548         (void)soclose(so1);
 549         return (error);
 550 }
 551
 552 /*
 553  * socketpair(int domain, int type, int protocol, int *rsv)
 554  */
 555 int
 556 socketpair(struct socketpair_args *uap)
 557 {
 558         int error, sockv[2];
 559
 560         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 561
 562         if (error == 0)
 563                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 564         return (error);
 565 }
 566
 567 int
 568 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 569     struct mbuf *control, int flags, int *res)
 570 {
 571         struct thread *td = curthread;
 572         struct proc *p = td->td_proc;
 573         struct file *fp;
 574         int len, error;
 575         struct socket *so;
 576 #ifdef KTRACE
 577         struct iovec *ktriov = NULL;
 578         struct uio ktruio;
 579 #endif
 580
 581         error = holdsock(p->p_fd, s, &fp);
 582         if (error)
 583                 return (error);
 584         if (auio->uio_resid < 0) {
 585                 error = EINVAL;
 586                 goto done;
 587         }
 588 #ifdef KTRACE
 589         if (KTRPOINT(td, KTR_GENIO)) {
 590                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 591
 592                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 593                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 594                 ktruio = *auio;
 595         }
 596 #endif
 597         len = auio->uio_resid;
 598         so = (struct socket *)fp->f_data;
 599         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 600         if (error) {
 601                 if (auio->uio_resid != len && (error == ERESTART ||
 602                     error == EINTR || error == EWOULDBLOCK))
 603                         error = 0;
 604                 if (error == EPIPE)
 605                         psignal(p, SIGPIPE);
 606         }
 607 #ifdef KTRACE
 608         if (ktriov != NULL) {
 609                 if (error == 0) {
 610                         ktruio.uio_iov = ktriov;
 611                         ktruio.uio_resid = len - auio->uio_resid;
 612                         ktrgenio(p->p_tracep, s, UIO_WRITE, &ktruio, error);
 613                 }
 614                 FREE(ktriov, M_TEMP);
 615         }
 616 #endif
 617         if (error == 0)
 618                 *res  = len - auio->uio_resid;
 619 done:
 620         fdrop(fp, td);
 621         return (error);
 622 }
 623
 624 /*
 625  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
 626  */
 627 int
 628 sendto(struct sendto_args *uap)
 629 {
 630         struct thread *td = curthread;
 631         struct uio auio;
 632         struct iovec aiov;
 633         struct sockaddr *sa = NULL;
 634         int error;
 635
 636         if (uap->to) {
 637                 error = getsockaddr(&sa, uap->to, uap->tolen);
 638                 if (error)
 639                         return (error);
 640         }
 641         aiov.iov_base = uap->buf;
 642         aiov.iov_len = uap->len;
 643         auio.uio_iov = &aiov;
 644         auio.uio_iovcnt = 1;
 645         auio.uio_offset = 0;
 646         auio.uio_resid = uap->len;
 647         auio.uio_segflg = UIO_USERSPACE;
 648         auio.uio_rw = UIO_WRITE;
 649         auio.uio_td = td;
 650
 651         error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
 652             &uap->sysmsg_result);
 653
 654         if (sa)
 655                 FREE(sa, M_SONAME);
 656         return (error);
 657 }
 658
 659 /*
 660  * sendmsg_args(int s, caddr_t msg, int flags)
 661  */
 662 int
 663 sendmsg(struct sendmsg_args *uap)
 664 {
 665         struct thread *td = curthread;
 666         struct msghdr msg;
 667         struct uio auio;
 668         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 669         struct sockaddr *sa = NULL;
 670         struct mbuf *control = NULL;
 671         int error;
 672
 673         error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
 674         if (error)
 675                 return (error);
 676
 677         /*
 678          * Conditionally copyin msg.msg_name.
 679          */
 680         if (msg.msg_name) {
 681                 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 682                 if (error)
 683                         return (error);
 684         }
 685
 686         /*
 687          * Populate auio.
 688          */
 689         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 690             &auio.uio_resid);
 691         if (error)
 692                 goto cleanup;
 693         auio.uio_iov = iov;
 694         auio.uio_iovcnt = msg.msg_iovlen;
 695         auio.uio_offset = 0;
 696         auio.uio_segflg = UIO_USERSPACE;
 697         auio.uio_rw = UIO_WRITE;
 698         auio.uio_td = td;
 699
 700         /*
 701          * Conditionally copyin msg.msg_control.
 702          */
 703         if (msg.msg_control) {
 704                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 705                     msg.msg_controllen > MLEN) {
 706                         error = EINVAL;
 707                         goto cleanup;
 708                 }
 709                 control = m_get(MB_WAIT, MT_CONTROL);
 710                 if (control == NULL) {
 711                         error = ENOBUFS;
 712                         goto cleanup;
 713                 }
 714                 control->m_len = msg.msg_controllen;
 715                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 716                     msg.msg_controllen);
 717                 if (error) {
 718                         m_free(control);
 719                         goto cleanup;
 720                 }
 721         }
 722
 723         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 724             &uap->sysmsg_result);
 725
 726 cleanup:
 727         if (sa)
 728                 FREE(sa, M_SONAME);
 729         iovec_free(&iov, aiov);
 730         return (error);
 731 }
 732
 733 /*
 734  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 735  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 736  * Don't forget to FREE() and m_free() these if they are returned.
 737  */
 738 int
 739 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 740     struct mbuf **control, int *flags, int *res)
 741 {
 742         struct thread *td = curthread;
 743         struct proc *p = td->td_proc;
 744         struct file *fp;
 745         int len, error;
 746         struct socket *so;
 747 #ifdef KTRACE
 748         struct iovec *ktriov = NULL;
 749         struct uio ktruio;
 750 #endif
 751
 752         error = holdsock(p->p_fd, s, &fp);
 753         if (error)
 754                 return (error);
 755         if (auio->uio_resid < 0) {
 756                 error = EINVAL;
 757                 goto done;
 758         }
 759 #ifdef KTRACE
 760         if (KTRPOINT(td, KTR_GENIO)) {
 761                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 762
 763                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 764                 bcopy(auio->uio_iov, ktriov, iovlen);
 765                 ktruio = *auio;
 766         }
 767 #endif
 768         len = auio->uio_resid;
 769         so = (struct socket *)fp->f_data;
 770         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 771         if (error) {
 772                 if (auio->uio_resid != len && (error == ERESTART ||
 773                     error == EINTR || error == EWOULDBLOCK))
 774                         error = 0;
 775         }
 776 #ifdef KTRACE
 777         if (ktriov != NULL) {
 778                 if (error == 0) {
 779                         ktruio.uio_iov = ktriov;
 780                         ktruio.uio_resid = len - auio->uio_resid;
 781                         ktrgenio(p->p_tracep, s, UIO_READ, &ktruio, error);
 782                 }
 783                 FREE(ktriov, M_TEMP);
 784         }
 785 #endif
 786         if (error == 0)
 787                 *res = len - auio->uio_resid;
 788 done:
 789         fdrop(fp, td);
 790         return (error);
 791 }
 792
 793 /*
 794  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
 795  *                      caddr_t from, int *fromlenaddr)
 796  */
 797 int
 798 recvfrom(struct recvfrom_args *uap)
 799 {
 800         struct thread *td = curthread;
 801         struct uio auio;
 802         struct iovec aiov;
 803         struct sockaddr *sa = NULL;
 804         int error, fromlen;
 805
 806         if (uap->from && uap->fromlenaddr) {
 807                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
 808                 if (error)
 809                         return (error);
 810                 if (fromlen < 0)
 811                         return (EINVAL);
 812         } else {
 813                 fromlen = 0;
 814         }
 815         aiov.iov_base = uap->buf;
 816         aiov.iov_len = uap->len;
 817         auio.uio_iov = &aiov;
 818         auio.uio_iovcnt = 1;
 819         auio.uio_offset = 0;
 820         auio.uio_resid = uap->len;
 821         auio.uio_segflg = UIO_USERSPACE;
 822         auio.uio_rw = UIO_READ;
 823         auio.uio_td = td;
 824
 825         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
 826             &uap->flags, &uap->sysmsg_result);
 827
 828         if (error == 0 && uap->from) {
 829                 /* note: sa may still be NULL */
 830                 if (sa) {
 831                         fromlen = MIN(fromlen, sa->sa_len);
 832                         error = copyout(sa, uap->from, fromlen);
 833                 } else {
 834                         fromlen = 0;
 835                 }
 836                 if (error == 0) {
 837                         error = copyout(&fromlen, uap->fromlenaddr,
 838                                         sizeof(fromlen));
 839                 }
 840         }
 841         if (sa)
 842                 FREE(sa, M_SONAME);
 843
 844         return (error);
 845 }
 846
 847 /*
 848  * recvmsg_args(int s, struct msghdr *msg, int flags)
 849  */
 850 int
 851 recvmsg(struct recvmsg_args *uap)
 852 {
 853         struct thread *td = curthread;
 854         struct msghdr msg;
 855         struct uio auio;
 856         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 857         struct mbuf *m, *control = NULL;
 858         struct sockaddr *sa = NULL;
 859         caddr_t ctlbuf;
 860         socklen_t *ufromlenp, *ucontrollenp;
 861         int error, fromlen, controllen, len, flags, *uflagsp;
 862
 863         /*
 864          * This copyin handles everything except the iovec.
 865          */
 866         error = copyin(uap->msg, &msg, sizeof(msg));
 867         if (error)
 868                 return (error);
 869
 870         if (msg.msg_name && msg.msg_namelen < 0)
 871                 return (EINVAL);
 872         if (msg.msg_control && msg.msg_controllen < 0)
 873                 return (EINVAL);
 874
 875         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 876             msg_namelen));
 877         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 878             msg_controllen));
 879         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
 880             msg_flags));
 881
 882         /*
 883          * Populate auio.
 884          */
 885         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 886             &auio.uio_resid);
 887         if (error)
 888                 return (error);
 889         auio.uio_iov = iov;
 890         auio.uio_iovcnt = msg.msg_iovlen;
 891         auio.uio_offset = 0;
 892         auio.uio_segflg = UIO_USERSPACE;
 893         auio.uio_rw = UIO_READ;
 894         auio.uio_td = td;
 895
 896         flags = uap->flags;
 897
 898         error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
 899             msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
 900
 901         /*
 902          * Conditionally copyout the name and populate the namelen field.
 903          */
 904         if (error == 0 && msg.msg_name) {
 905                 fromlen = MIN(msg.msg_namelen, sa->sa_len);
 906                 error = copyout(sa, msg.msg_name, fromlen);
 907                 if (error == 0)
 908                         error = copyout(&fromlen, ufromlenp,
 909                             sizeof(*ufromlenp));
 910         }
 911
 912         /*
 913          * Copyout msg.msg_control and msg.msg_controllen.
 914          */
 915         if (error == 0 && msg.msg_control) {
 916                 len = msg.msg_controllen;
 917                 m = control;
 918                 ctlbuf = (caddr_t)msg.msg_control;
 919
 920                 while(m && len > 0) {
 921                         unsigned int tocopy;
 922
 923                         if (len >= m->m_len) {
 924                                 tocopy = m->m_len;
 925                         } else {
 926                                 msg.msg_flags |= MSG_CTRUNC;
 927                                 tocopy = len;
 928                         }
 929
 930                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
 931                         if (error)
 932                                 goto cleanup;
 933
 934                         ctlbuf += tocopy;
 935                         len -= tocopy;
 936                         m = m->m_next;
 937                 }
 938                 controllen = ctlbuf - (caddr_t)msg.msg_control;
 939                 error = copyout(&controllen, ucontrollenp,
 940                     sizeof(*ucontrollenp));
 941         }
 942
 943         if (error == 0)
 944                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
 945
 946 cleanup:
 947         if (sa)
 948                 FREE(sa, M_SONAME);
 949         iovec_free(&iov, aiov);
 950         if (control)
 951                 m_freem(control);
 952         return (error);
 953 }
 954
 955 /*
 956  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
 957  * in kernel pointer instead of a userland pointer.  This allows us
 958  * to manipulate socket options in the emulation code.
 959  */
 960 int
 961 kern_setsockopt(int s, struct sockopt *sopt)
 962 {
 963         struct thread *td = curthread;
 964         struct proc *p = td->td_proc;
 965         struct file *fp;
 966         int error;
 967
 968         if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
 969                 return (EFAULT);
 970         if (sopt->sopt_valsize < 0)
 971                 return (EINVAL);
 972
 973         error = holdsock(p->p_fd, s, &fp);
 974         if (error)
 975                 return (error);
 976
 977         error = sosetopt((struct socket *)fp->f_data, sopt);
 978         fdrop(fp, td);
 979         return (error);
 980 }
 981
 982 /*
 983  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
 984  */
 985 int
 986 setsockopt(struct setsockopt_args *uap)
 987 {
 988         struct thread *td = curthread;
 989         struct sockopt sopt;
 990         int error;
 991
 992         sopt.sopt_level = uap->level;
 993         sopt.sopt_name = uap->name;
 994         sopt.sopt_val = uap->val;
 995         sopt.sopt_valsize = uap->valsize;
 996         sopt.sopt_td = td;
 997
 998         error = kern_setsockopt(uap->s, &sopt);
 999         return(error);
1000 }
1001
1002 /*
1003  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1004  * in kernel pointer instead of a userland pointer.  This allows us
1005  * to manipulate socket options in the emulation code.
1006  */
1007 int
1008 kern_getsockopt(int s, struct sockopt *sopt)
1009 {
1010         struct thread *td = curthread;
1011         struct proc *p = td->td_proc;
1012         struct file *fp;
1013         int error;
1014
1015         if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1016                 return (EFAULT);
1017         if (sopt->sopt_valsize < 0)
1018                 return (EINVAL);
1019
1020         error = holdsock(p->p_fd, s, &fp);
1021         if (error)
1022                 return (error);
1023
1024         error = sogetopt((struct socket *)fp->f_data, sopt);
1025         fdrop(fp, td);
1026         return (error);
1027 }
1028
1029 /*
1030  * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1031  */
1032 int
1033 getsockopt(struct getsockopt_args *uap)
1034 {
1035         struct thread *td = curthread;
1036         struct  sockopt sopt;
1037         int     error, valsize;
1038
1039         if (uap->val) {
1040                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1041                 if (error)
1042                         return (error);
1043                 if (valsize < 0)
1044                         return (EINVAL);
1045         } else {
1046                 valsize = 0;
1047         }
1048
1049         sopt.sopt_level = uap->level;
1050         sopt.sopt_name = uap->name;
1051         sopt.sopt_val = uap->val;
1052         sopt.sopt_valsize = valsize;
1053         sopt.sopt_td = td;
1054
1055         error = kern_getsockopt(uap->s, &sopt);
1056         if (error == 0) {
1057                 valsize = sopt.sopt_valsize;
1058                 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1059         }
1060         return (error);
1061 }
1062
1063 /*
1064  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1065  * This allows kern_getsockname() to return a pointer to an allocated struct
1066  * sockaddr which must be freed later with FREE().  The caller must
1067  * initialize *name to NULL.
1068  */
1069 int
1070 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1071 {
1072         struct thread *td = curthread;
1073         struct proc *p = td->td_proc;
1074         struct file *fp;
1075         struct socket *so;
1076         struct sockaddr *sa = NULL;
1077         int error;
1078
1079         error = holdsock(p->p_fd, s, &fp);
1080         if (error)
1081                 return (error);
1082         if (*namelen < 0) {
1083                 fdrop(fp, td);
1084                 return (EINVAL);
1085         }
1086         so = (struct socket *)fp->f_data;
1087         error = so_pru_sockaddr(so, &sa);
1088         if (error == 0) {
1089                 if (sa == 0) {
1090                         *namelen = 0;
1091                 } else {
1092                         *namelen = MIN(*namelen, sa->sa_len);
1093                         *name = sa;
1094                 }
1095         }
1096
1097         fdrop(fp, td);
1098         return (error);
1099 }
1100
1101 /*
1102  * getsockname_args(int fdes, caddr_t asa, int *alen)
1103  *
1104  * Get socket name.
1105  */
1106 int
1107 getsockname(struct getsockname_args *uap)
1108 {
1109         struct sockaddr *sa = NULL;
1110         int error, sa_len;
1111
1112         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1113         if (error)
1114                 return (error);
1115
1116         error = kern_getsockname(uap->fdes, &sa, &sa_len);
1117
1118         if (error == 0)
1119                 error = copyout(sa, uap->asa, sa_len);
1120         if (error == 0)
1121                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1122         if (sa)
1123                 FREE(sa, M_SONAME);
1124         return (error);
1125 }
1126
1127 /*
1128  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1129  * This allows kern_getpeername() to return a pointer to an allocated struct
1130  * sockaddr which must be freed later with FREE().  The caller must
1131  * initialize *name to NULL.
1132  */
1133 int
1134 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1135 {
1136         struct thread *td = curthread;
1137         struct proc *p = td->td_proc;
1138         struct file *fp;
1139         struct socket *so;
1140         struct sockaddr *sa = NULL;
1141         int error;
1142
1143         error = holdsock(p->p_fd, s, &fp);
1144         if (error)
1145                 return (error);
1146         if (*namelen < 0) {
1147                 fdrop(fp, td);
1148                 return (EINVAL);
1149         }
1150         so = (struct socket *)fp->f_data;
1151         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1152                 fdrop(fp, td);
1153                 return (ENOTCONN);
1154         }
1155         error = so_pru_peeraddr(so, &sa);
1156         if (error == 0) {
1157                 if (sa == 0) {
1158                         *namelen = 0;
1159                 } else {
1160                         *namelen = MIN(*namelen, sa->sa_len);
1161                         *name = sa;
1162                 }
1163         }
1164
1165         fdrop(fp, td);
1166         return (error);
1167 }
1168
1169 /*
1170  * getpeername_args(int fdes, caddr_t asa, int *alen)
1171  *
1172  * Get name of peer for connected socket.
1173  */
1174 int
1175 getpeername(struct getpeername_args *uap)
1176 {
1177         struct sockaddr *sa = NULL;
1178         int error, sa_len;
1179
1180         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1181         if (error)
1182                 return (error);
1183
1184         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1185
1186         if (error == 0)
1187                 error = copyout(sa, uap->asa, sa_len);
1188         if (error == 0)
1189                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1190         if (sa)
1191                 FREE(sa, M_SONAME);
1192         return (error);
1193 }
1194
1195 int
1196 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1197 {
1198         struct sockaddr *sa;
1199         int error;
1200
1201         *namp = NULL;
1202         if (len > SOCK_MAXADDRLEN)
1203                 return ENAMETOOLONG;
1204         if (len < offsetof(struct sockaddr, sa_data[0]))
1205                 return EDOM;
1206         MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1207         error = copyin(uaddr, sa, len);
1208         if (error) {
1209                 FREE(sa, M_SONAME);
1210         } else {
1211 #if BYTE_ORDER != BIG_ENDIAN
1212                 /*
1213                  * The bind(), connect(), and sendto() syscalls were not
1214                  * versioned for COMPAT_43.  Thus, this check must stay.
1215                  */
1216                 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1217                         sa->sa_family = sa->sa_len;
1218 #endif
1219                 sa->sa_len = len;
1220                 *namp = sa;
1221         }
1222         return error;
1223 }
1224
1225 /*
1226  * holdsock() - load the struct file pointer associated
1227  * with a socket into *fpp.  If an error occurs, non-zero
1228  * will be returned and *fpp will be set to NULL.
1229  */
1230 int
1231 holdsock(struct filedesc *fdp, int fdes, struct file **fpp)
1232 {
1233         struct file *fp;
1234         int error = 0;
1235
1236         *fpp = NULL;
1237         if ((unsigned)fdes >= fdp->fd_nfiles)
1238                 return EBADF;
1239         if ((fp = fdp->fd_files[fdes].fp) == NULL)
1240                 return EBADF;
1241         if (fp->f_type != DTYPE_SOCKET)
1242                 return ENOTSOCK;
1243         fhold(fp);
1244         *fpp = fp;
1245         return (error);
1246 }
1247
1248 /*
1249  * Detach a mapped page and release resources back to the system.
1250  * We must release our wiring and if the object is ripped out
1251  * from under the vm_page we become responsible for freeing the
1252  * page.  These routines must be MPSAFE.
1253  *
1254  * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1255  *
1256  * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1257  */
1258 static void
1259 sf_buf_mref(void *arg)
1260 {
1261         struct sfbuf_mref *sfm = arg;
1262
1263         /*
1264          * We must already hold a ref so there is no race to 0, just
1265          * atomically increment the count.
1266          */
1267         atomic_add_int(&sfm->mref_count, 1);
1268 }
1269
1270 static void
1271 sf_buf_mfree(void *arg)
1272 {
1273         struct sfbuf_mref *sfm = arg;
1274         vm_page_t m;
1275
1276         KKASSERT(sfm->mref_count > 0);
1277         if (sfm->mref_count == 1) {
1278                 /*
1279                  * We are the only holder so no further locking is required,
1280                  * the sfbuf can simply be freed.
1281                  */
1282                 sfm->mref_count = 0;
1283                 goto freeit;
1284         } else {
1285                 /*
1286                  * There may be other holders, we must obtain the serializer
1287                  * to protect against a sf_buf_mfree() race to 0.  An atomic
1288                  * operation is still required for races against
1289                  * sf_buf_mref().
1290                  *
1291                  * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1292                  */
1293                 lwkt_serialize_enter(&sfm->serializer);
1294                 atomic_subtract_int(&sfm->mref_count, 1);
1295                 if (sfm->mref_count == 0) {
1296                         lwkt_serialize_exit(&sfm->serializer);
1297 freeit:
1298                         get_mplock();
1299                         crit_enter();
1300                         m = sf_buf_page(sfm->sf);
1301                         sf_buf_free(sfm->sf);
1302                         vm_page_unwire(m, 0);
1303                         if (m->wire_count == 0 && m->object == NULL)
1304                                 vm_page_try_to_free(m);
1305                         crit_exit();
1306                         rel_mplock();
1307                         free(sfm, M_SENDFILE);
1308                 } else {
1309                         lwkt_serialize_exit(&sfm->serializer);
1310                 }
1311         }
1312 }
1313
1314 /*
1315  * sendfile(2).
1316  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1317  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1318  *
1319  * Send a file specified by 'fd' and starting at 'offset' to a socket
1320  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1321  * nbytes == 0. Optionally add a header and/or trailer to the socket
1322  * output. If specified, write the total number of bytes sent into *sbytes.
1323  *
1324  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1325  * the headers to count against the remaining bytes to be sent from
1326  * the file descriptor.  We may wish to implement a compatibility syscall
1327  * in the future.
1328  */
1329 int
1330 sendfile(struct sendfile_args *uap)
1331 {
1332         struct thread *td = curthread;
1333         struct proc *p = td->td_proc;
1334         struct file *fp;
1335         struct filedesc *fdp;
1336         struct vnode *vp = NULL;
1337         struct sf_hdtr hdtr;
1338         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1339         struct uio auio;
1340         struct mbuf *mheader = NULL;
1341         off_t hdtr_size = 0, sbytes;
1342         int error, hbytes = 0, tbytes;
1343
1344         KKASSERT(p);
1345         fdp = p->p_fd;
1346
1347         /*
1348          * Do argument checking. Must be a regular file in, stream
1349          * type and connected socket out, positive offset.
1350          */
1351         fp = holdfp(fdp, uap->fd, FREAD);
1352         if (fp == NULL) {
1353                 return (EBADF);
1354         }
1355         if (fp->f_type != DTYPE_VNODE) {
1356                 fdrop(fp, td);
1357                 return (EINVAL);
1358         }
1359         vp = (struct vnode *)fp->f_data;
1360         vref(vp);
1361         fdrop(fp, td);
1362
1363         /*
1364          * If specified, get the pointer to the sf_hdtr struct for
1365          * any headers/trailers.
1366          */
1367         if (uap->hdtr) {
1368                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1369                 if (error)
1370                         goto done;
1371                 /*
1372                  * Send any headers.
1373                  */
1374                 if (hdtr.headers) {
1375                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1376                             hdtr.hdr_cnt, &hbytes);
1377                         if (error)
1378                                 goto done;
1379                         auio.uio_iov = iov;
1380                         auio.uio_iovcnt = hdtr.hdr_cnt;
1381                         auio.uio_offset = 0;
1382                         auio.uio_segflg = UIO_USERSPACE;
1383                         auio.uio_rw = UIO_WRITE;
1384                         auio.uio_td = td;
1385                         auio.uio_resid = hbytes;
1386
1387                         mheader = m_uiomove(&auio);
1388
1389                         iovec_free(&iov, aiov);
1390                         if (mheader == NULL)
1391                                 goto done;
1392                 }
1393         }
1394
1395         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1396             &sbytes, uap->flags);
1397         if (error)
1398                 goto done;
1399
1400         /*
1401          * Send trailers. Wimp out and use writev(2).
1402          */
1403         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1404                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1405                     hdtr.trl_cnt, &auio.uio_resid);
1406                 if (error)
1407                         goto done;
1408                 auio.uio_iov = iov;
1409                 auio.uio_iovcnt = hdtr.trl_cnt;
1410                 auio.uio_offset = 0;
1411                 auio.uio_segflg = UIO_USERSPACE;
1412                 auio.uio_rw = UIO_WRITE;
1413                 auio.uio_td = td;
1414
1415                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1416
1417                 iovec_free(&iov, aiov);
1418                 if (error)
1419                         goto done;
1420                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1421         }
1422
1423 done:
1424         if (uap->sbytes != NULL) {
1425                 sbytes += hdtr_size;
1426                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1427         }
1428         if (vp)
1429                 vrele(vp);
1430         return (error);
1431 }
1432
1433 int
1434 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1435     struct mbuf *mheader, off_t *sbytes, int flags)
1436 {
1437         struct thread *td = curthread;
1438         struct proc *p = td->td_proc;
1439         struct vm_object *obj;
1440         struct socket *so;
1441         struct file *fp;
1442         struct mbuf *m;
1443         struct sf_buf *sf;
1444         struct sfbuf_mref *sfm;
1445         struct vm_page *pg;
1446         off_t off, xfsize;
1447         off_t hbytes = 0;
1448         int error = 0;
1449
1450         if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1451                 error = EINVAL;
1452                 goto done0;
1453         }
1454         error = holdsock(p->p_fd, sfd, &fp);
1455         if (error)
1456                 goto done0;
1457         so = (struct socket *)fp->f_data;
1458         if (so->so_type != SOCK_STREAM) {
1459                 error = EINVAL;
1460                 goto done;
1461         }
1462         if ((so->so_state & SS_ISCONNECTED) == 0) {
1463                 error = ENOTCONN;
1464                 goto done;
1465         }
1466         if (offset < 0) {
1467                 error = EINVAL;
1468                 goto done;
1469         }
1470
1471         *sbytes = 0;
1472         /*
1473          * Protect against multiple writers to the socket.
1474          */
1475         (void) sblock(&so->so_snd, M_WAITOK);
1476
1477         /*
1478          * Loop through the pages in the file, starting with the requested
1479          * offset. Get a file page (do I/O if necessary), map the file page
1480          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1481          * it on the socket.
1482          */
1483         for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1484                 vm_pindex_t pindex;
1485                 vm_offset_t pgoff;
1486
1487                 pindex = OFF_TO_IDX(off);
1488 retry_lookup:
1489                 /*
1490                  * Calculate the amount to transfer. Not to exceed a page,
1491                  * the EOF, or the passed in nbytes.
1492                  */
1493                 xfsize = obj->un_pager.vnp.vnp_size - off;
1494                 if (xfsize > PAGE_SIZE)
1495                         xfsize = PAGE_SIZE;
1496                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1497                 if (PAGE_SIZE - pgoff < xfsize)
1498                         xfsize = PAGE_SIZE - pgoff;
1499                 if (nbytes && xfsize > (nbytes - *sbytes))
1500                         xfsize = nbytes - *sbytes;
1501                 if (xfsize <= 0)
1502                         break;
1503                 /*
1504                  * Optimize the non-blocking case by looking at the socket space
1505                  * before going to the extra work of constituting the sf_buf.
1506                  */
1507                 if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1508                         if (so->so_state & SS_CANTSENDMORE)
1509                                 error = EPIPE;
1510                         else
1511                                 error = EAGAIN;
1512                         sbunlock(&so->so_snd);
1513                         goto done;
1514                 }
1515                 /*
1516                  * Attempt to look up the page.
1517                  *
1518                  *      Allocate if not found, wait and loop if busy, then
1519                  *      wire the page.  critical section protection is
1520                  *      required to maintain the object association (an
1521                  *      interrupt can free the page) through to the
1522                  *      vm_page_wire() call.
1523                  */
1524                 crit_enter();
1525                 pg = vm_page_lookup(obj, pindex);
1526                 if (pg == NULL) {
1527                         pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1528                         if (pg == NULL) {
1529                                 vm_wait();
1530                                 crit_exit();
1531                                 goto retry_lookup;
1532                         }
1533                         vm_page_wakeup(pg);
1534                 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1535                         crit_exit();
1536                         goto retry_lookup;
1537                 }
1538                 vm_page_wire(pg);
1539                 crit_exit();
1540
1541                 /*
1542                  * If page is not valid for what we need, initiate I/O
1543                  */
1544
1545                 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1546                         struct uio auio;
1547                         struct iovec aiov;
1548                         int bsize;
1549
1550                         /*
1551                          * Ensure that our page is still around when the I/O
1552                          * completes.
1553                          */
1554                         vm_page_io_start(pg);
1555
1556                         /*
1557                          * Get the page from backing store.
1558                          */
1559                         bsize = vp->v_mount->mnt_stat.f_iosize;
1560                         auio.uio_iov = &aiov;
1561                         auio.uio_iovcnt = 1;
1562                         aiov.iov_base = 0;
1563                         aiov.iov_len = MAXBSIZE;
1564                         auio.uio_resid = MAXBSIZE;
1565                         auio.uio_offset = trunc_page(off);
1566                         auio.uio_segflg = UIO_NOCOPY;
1567                         auio.uio_rw = UIO_READ;
1568                         auio.uio_td = td;
1569                         vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1570                         error = VOP_READ(vp, &auio,
1571                                     IO_VMIO | ((MAXBSIZE / bsize) << 16),
1572                                     p->p_ucred);
1573                         VOP_UNLOCK(vp, 0, td);
1574                         vm_page_flag_clear(pg, PG_ZERO);
1575                         vm_page_io_finish(pg);
1576                         if (error) {
1577                                 crit_enter();
1578                                 vm_page_unwire(pg, 0);
1579                                 vm_page_try_to_free(pg);
1580                                 crit_exit();
1581                                 sbunlock(&so->so_snd);
1582                                 goto done;
1583                         }
1584                 }
1585
1586
1587                 /*
1588                  * Get a sendfile buf. We usually wait as long as necessary,
1589                  * but this wait can be interrupted.
1590                  */
1591                 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1592                         crit_enter();
1593                         vm_page_unwire(pg, 0);
1594                         vm_page_try_to_free(pg);
1595                         crit_exit();
1596                         sbunlock(&so->so_snd);
1597                         error = EINTR;
1598                         goto done;
1599                 }
1600
1601                 /*
1602                  * Get an mbuf header and set it up as having external storage.
1603                  */
1604                 MGETHDR(m, MB_WAIT, MT_DATA);
1605                 if (m == NULL) {
1606                         error = ENOBUFS;
1607                         sf_buf_free(sf);
1608                         sbunlock(&so->so_snd);
1609                         goto done;
1610                 }
1611
1612                 /*
1613                  * sfm is a temporary hack, use a per-cpu cache for this.
1614                  */
1615                 sfm = malloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1616                 sfm->sf = sf;
1617                 sfm->mref_count = 1;
1618                 lwkt_serialize_init(&sfm->serializer);
1619
1620                 m->m_ext.ext_free = sf_buf_mfree;
1621                 m->m_ext.ext_ref = sf_buf_mref;
1622                 m->m_ext.ext_arg = sfm;
1623                 m->m_ext.ext_buf = (void *)sf->kva;
1624                 m->m_ext.ext_size = PAGE_SIZE;
1625                 m->m_data = (char *) sf->kva + pgoff;
1626                 m->m_flags |= M_EXT;
1627                 m->m_pkthdr.len = m->m_len = xfsize;
1628                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1629
1630                 if (mheader != NULL) {
1631                         hbytes = mheader->m_pkthdr.len;
1632                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1633                         m_cat(mheader, m);
1634                         m = mheader;
1635                         mheader = NULL;
1636                 } else
1637                         hbytes = 0;
1638
1639                 /*
1640                  * Add the buffer to the socket buffer chain.
1641                  */
1642                 crit_enter();
1643 retry_space:
1644                 /*
1645                  * Make sure that the socket is still able to take more data.
1646                  * CANTSENDMORE being true usually means that the connection
1647                  * was closed. so_error is true when an error was sensed after
1648                  * a previous send.
1649                  * The state is checked after the page mapping and buffer
1650                  * allocation above since those operations may block and make
1651                  * any socket checks stale. From this point forward, nothing
1652                  * blocks before the pru_send (or more accurately, any blocking
1653                  * results in a loop back to here to re-check).
1654                  */
1655                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1656                         if (so->so_state & SS_CANTSENDMORE) {
1657                                 error = EPIPE;
1658                         } else {
1659                                 error = so->so_error;
1660                                 so->so_error = 0;
1661                         }
1662                         m_freem(m);
1663                         sbunlock(&so->so_snd);
1664                         crit_exit();
1665                         goto done;
1666                 }
1667                 /*
1668                  * Wait for socket space to become available. We do this just
1669                  * after checking the connection state above in order to avoid
1670                  * a race condition with sbwait().
1671                  */
1672                 if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1673                         if (so->so_state & SS_NBIO) {
1674                                 m_freem(m);
1675                                 sbunlock(&so->so_snd);
1676                                 crit_exit();
1677                                 error = EAGAIN;
1678                                 goto done;
1679                         }
1680                         error = sbwait(&so->so_snd);
1681                         /*
1682                          * An error from sbwait usually indicates that we've
1683                          * been interrupted by a signal. If we've sent anything
1684                          * then return bytes sent, otherwise return the error.
1685                          */
1686                         if (error) {
1687                                 m_freem(m);
1688                                 sbunlock(&so->so_snd);
1689                                 crit_exit();
1690                                 goto done;
1691                         }
1692                         goto retry_space;
1693                 }
1694                 error = so_pru_send(so, 0, m, NULL, NULL, td);
1695                 crit_exit();
1696                 if (error) {
1697                         sbunlock(&so->so_snd);
1698                         goto done;
1699                 }
1700         }
1701         if (mheader != NULL) {
1702                 *sbytes += mheader->m_pkthdr.len;
1703                 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1704                 mheader = NULL;
1705         }
1706         sbunlock(&so->so_snd);
1707
1708 done:
1709         fdrop(fp, td);
1710 done0:
1711         if (mheader != NULL)
1712                 m_freem(mheader);
1713         return (error);
1714 }
1715
1716 int
1717 sctp_peeloff(struct sctp_peeloff_args *uap)
1718 {
1719 #ifdef SCTP
1720         struct thread *td = curthread;
1721         struct proc *p = td->td_proc;
1722         struct filedesc *fdp = p->p_fd;
1723         struct file *lfp = NULL;
1724         struct file *nfp = NULL;
1725         int error;
1726         struct socket *head, *so;
1727         caddr_t assoc_id;
1728         int fd;
1729         short fflag;            /* type must match fp->f_flag */
1730
1731         assoc_id = uap->name;
1732         error = holdsock(fdp, uap->sd, &lfp);
1733         if (error) {
1734                 return (error);
1735         }
1736         crit_enter();
1737         head = (struct socket *)lfp->f_data;
1738         error = sctp_can_peel_off(head, assoc_id);
1739         if (error) {
1740                 crit_exit();
1741                 goto done;
1742         }
1743         /*
1744          * At this point we know we do have a assoc to pull
1745          * we proceed to get the fd setup. This may block
1746          * but that is ok.
1747          */
1748
1749         fflag = lfp->f_flag;
1750         error = falloc(p, &nfp, &fd);
1751         if (error) {
1752                 /*
1753                  * Probably ran out of file descriptors. Put the
1754                  * unaccepted connection back onto the queue and
1755                  * do another wakeup so some other process might
1756                  * have a chance at it.
1757                  */
1758                 crit_exit();
1759                 goto done;
1760         }
1761         fhold(nfp);
1762         uap->sysmsg_result = fd;
1763
1764         so = sctp_get_peeloff(head, assoc_id, &error);
1765         if (so == NULL) {
1766                 /*
1767                  * Either someone else peeled it off OR
1768                  * we can't get a socket.
1769                  */
1770                 goto noconnection;
1771         }
1772         so->so_state &= ~SS_COMP;
1773         so->so_state &= ~SS_NOFDREF;
1774         so->so_head = NULL;
1775         if (head->so_sigio != NULL)
1776                 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1777
1778         nfp->f_type = DTYPE_SOCKET;
1779         nfp->f_flag = fflag;
1780         nfp->f_ops = &socketops;
1781         nfp->f_data = so;
1782
1783 noconnection:
1784         /*
1785          * close the new descriptor, assuming someone hasn't ripped it
1786          * out from under us.
1787          */
1788         if (error) {
1789                 if (fdp->fd_files[fd].fp == nfp) {
1790                         funsetfd(fdp, fd);
1791                         fdrop(nfp, td);
1792                 }
1793         }
1794         crit_exit();
1795         /*
1796          * Release explicitly held references before returning.
1797          */
1798 done:
1799         if (nfp != NULL)
1800                 fdrop(nfp, td);
1801         fdrop(lfp, td);
1802         return (error);
1803 #else /* SCTP */
1804         return(EOPNOTSUPP);
1805 #endif /* SCTP */
1806 }