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 1.81 2007/05/23 08:57:05 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 #include <net/netmsg2.h>
  82
  83 #ifdef SCTP
  84 #include <netinet/sctp_peeloff.h>
  85 #endif /* SCTP */
  86
  87 struct sfbuf_mref {
  88         struct sf_buf   *sf;
  89         int             mref_count;
  90         struct lwkt_serialize serializer;
  91 };
  92
  93 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
  94
  95 /*
  96  * System call interface to the socket abstraction.
  97  */
  98
  99 extern  struct fileops socketops;
 100
 101 /*
 102  * socket_args(int domain, int type, int protocol)
 103  */
 104 int
 105 kern_socket(int domain, int type, int protocol, int *res)
 106 {
 107         struct thread *td = curthread;
 108         struct proc *p = td->td_proc;
 109         struct socket *so;
 110         struct file *fp;
 111         int fd, error;
 112
 113         KKASSERT(p);
 114
 115         error = falloc(p, &fp, &fd);
 116         if (error)
 117                 return (error);
 118         error = socreate(domain, &so, type, protocol, td);
 119         if (error) {
 120                 fsetfd(p, NULL, fd);
 121         } else {
 122                 fp->f_type = DTYPE_SOCKET;
 123                 fp->f_flag = FREAD | FWRITE;
 124                 fp->f_ops = &socketops;
 125                 fp->f_data = so;
 126                 *res = fd;
 127                 fsetfd(p, fp, fd);
 128         }
 129         fdrop(fp);
 130         return (error);
 131 }
 132
 133 int
 134 sys_socket(struct socket_args *uap)
 135 {
 136         int error;
 137
 138         error = kern_socket(uap->domain, uap->type, uap->protocol,
 139             &uap->sysmsg_result);
 140
 141         return (error);
 142 }
 143
 144 int
 145 kern_bind(int s, struct sockaddr *sa)
 146 {
 147         struct thread *td = curthread;
 148         struct proc *p = td->td_proc;
 149         struct file *fp;
 150         int error;
 151
 152         KKASSERT(p);
 153         error = holdsock(p->p_fd, s, &fp);
 154         if (error)
 155                 return (error);
 156         error = sobind((struct socket *)fp->f_data, sa, td);
 157         fdrop(fp);
 158         return (error);
 159 }
 160
 161 /*
 162  * bind_args(int s, caddr_t name, int namelen)
 163  */
 164 int
 165 sys_bind(struct bind_args *uap)
 166 {
 167         struct sockaddr *sa;
 168         int error;
 169
 170         error = getsockaddr(&sa, uap->name, uap->namelen);
 171         if (error)
 172                 return (error);
 173         error = kern_bind(uap->s, sa);
 174         FREE(sa, M_SONAME);
 175
 176         return (error);
 177 }
 178
 179 int
 180 kern_listen(int s, int backlog)
 181 {
 182         struct thread *td = curthread;
 183         struct proc *p = td->td_proc;
 184         struct file *fp;
 185         int error;
 186
 187         KKASSERT(p);
 188         error = holdsock(p->p_fd, s, &fp);
 189         if (error)
 190                 return (error);
 191         error = solisten((struct socket *)fp->f_data, backlog, td);
 192         fdrop(fp);
 193         return(error);
 194 }
 195
 196 /*
 197  * listen_args(int s, int backlog)
 198  */
 199 int
 200 sys_listen(struct listen_args *uap)
 201 {
 202         int error;
 203
 204         error = kern_listen(uap->s, uap->backlog);
 205         return (error);
 206 }
 207
 208 /*
 209  * Returns the accepted socket as well.
 210  */
 211 static boolean_t
 212 soaccept_predicate(struct netmsg *msg0)
 213 {
 214         struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
 215         struct socket *head = msg->nm_so;
 216
 217         if (head->so_error != 0) {
 218                 msg->nm_netmsg.nm_lmsg.ms_error = head->so_error;
 219                 return (TRUE);
 220         }
 221         if (!TAILQ_EMPTY(&head->so_comp)) {
 222                 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
 223                 msg->nm_so = TAILQ_FIRST(&head->so_comp);
 224                 TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
 225                 head->so_qlen--;
 226
 227                 msg->nm_netmsg.nm_lmsg.ms_error = 0;
 228                 return (TRUE);
 229         }
 230         if (head->so_state & SS_CANTRCVMORE) {
 231                 msg->nm_netmsg.nm_lmsg.ms_error = ECONNABORTED;
 232                 return (TRUE);
 233         }
 234         if (msg->nm_fflags & FNONBLOCK) {
 235                 msg->nm_netmsg.nm_lmsg.ms_error = EWOULDBLOCK;
 236                 return (TRUE);
 237         }
 238
 239         return (FALSE);
 240 }
 241
 242 /*
 243  * The second argument to kern_accept() is a handle to a struct sockaddr.
 244  * This allows kern_accept() to return a pointer to an allocated struct
 245  * sockaddr which must be freed later with FREE().  The caller must
 246  * initialize *name to NULL.
 247  */
 248 int
 249 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
 250 {
 251         struct thread *td = curthread;
 252         struct proc *p = td->td_proc;
 253         struct file *lfp = NULL;
 254         struct file *nfp = NULL;
 255         struct sockaddr *sa;
 256         struct socket *head, *so;
 257         struct netmsg_so_notify msg;
 258         lwkt_port_t port;
 259         int fd;
 260         u_int fflag;            /* type must match fp->f_flag */
 261         int error, tmp;
 262
 263         *res = -1;
 264         if (name && namelen && *namelen < 0)
 265                 return (EINVAL);
 266
 267         error = holdsock(p->p_fd, s, &lfp);
 268         if (error)
 269                 return (error);
 270
 271         error = falloc(p, &nfp, &fd);
 272         if (error) {            /* Probably ran out of file descriptors. */
 273                 fdrop(lfp);
 274                 return (error);
 275         }
 276         head = (struct socket *)lfp->f_data;
 277         if ((head->so_options & SO_ACCEPTCONN) == 0) {
 278                 error = EINVAL;
 279                 goto done;
 280         }
 281
 282         if (fflags & O_FBLOCKING)
 283                 fflags |= lfp->f_flag & ~FNONBLOCK;
 284         else if (fflags & O_FNONBLOCKING)
 285                 fflags |= lfp->f_flag | FNONBLOCK;
 286         else
 287                 fflags = lfp->f_flag;
 288
 289         /* optimize for uniprocessor case later XXX JH */
 290         port = head->so_proto->pr_mport(head, NULL, PRU_PRED);
 291         netmsg_init_abortable(&msg.nm_netmsg, &curthread->td_msgport,
 292                               MSGF_PCATCH,
 293                               netmsg_so_notify,
 294                               netmsg_so_notify_doabort);
 295         msg.nm_predicate = soaccept_predicate;
 296         msg.nm_fflags = fflags;
 297         msg.nm_so = head;
 298         msg.nm_etype = NM_REVENT;
 299         error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg);
 300         if (error)
 301                 goto done;
 302
 303         /*
 304          * At this point we have the connection that's ready to be accepted.
 305          */
 306         so = msg.nm_so;
 307
 308         fflag = lfp->f_flag;
 309
 310         /* connection has been removed from the listen queue */
 311         KNOTE(&head->so_rcv.ssb_sel.si_note, 0);
 312
 313         so->so_state &= ~SS_COMP;
 314         so->so_head = NULL;
 315         if (head->so_sigio != NULL)
 316                 fsetown(fgetown(head->so_sigio), &so->so_sigio);
 317
 318         nfp->f_type = DTYPE_SOCKET;
 319         nfp->f_flag = fflag;
 320         nfp->f_ops = &socketops;
 321         nfp->f_data = so;
 322         /* Sync socket nonblocking/async state with file flags */
 323         tmp = fflag & FNONBLOCK;
 324         (void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred);
 325         tmp = fflag & FASYNC;
 326         (void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred);
 327
 328         sa = NULL;
 329         error = soaccept(so, &sa);
 330
 331         /*
 332          * Set the returned name and namelen as applicable.  Set the returned
 333          * namelen to 0 for older code which might ignore the return value
 334          * from accept.
 335          */
 336         if (error == 0) {
 337                 if (sa && name && namelen) {
 338                         if (*namelen > sa->sa_len)
 339                                 *namelen = sa->sa_len;
 340                         *name = sa;
 341                 } else {
 342                         if (sa)
 343                                 FREE(sa, M_SONAME);
 344                 }
 345         }
 346
 347 done:
 348         /*
 349          * If an error occured clear the reserved descriptor, else associate
 350          * nfp with it.
 351          *
 352          * Note that *res is normally ignored if an error is returned but
 353          * a syscall message will still have access to the result code.
 354          */
 355         if (error) {
 356                 fsetfd(p, NULL, fd);
 357         } else {
 358                 *res = fd;
 359                 fsetfd(p, nfp, fd);
 360         }
 361         fdrop(nfp);
 362         fdrop(lfp);
 363         return (error);
 364 }
 365
 366 /*
 367  * accept(int s, caddr_t name, int *anamelen)
 368  */
 369 int
 370 sys_accept(struct accept_args *uap)
 371 {
 372         struct sockaddr *sa = NULL;
 373         int sa_len;
 374         int error;
 375
 376         if (uap->name) {
 377                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 378                 if (error)
 379                         return (error);
 380
 381                 error = kern_accept(uap->s, 0, &sa, &sa_len, &uap->sysmsg_result);
 382
 383                 if (error == 0)
 384                         error = copyout(sa, uap->name, sa_len);
 385                 if (error == 0) {
 386                         error = copyout(&sa_len, uap->anamelen,
 387                             sizeof(*uap->anamelen));
 388                 }
 389                 if (sa)
 390                         FREE(sa, M_SONAME);
 391         } else {
 392                 error = kern_accept(uap->s, 0, NULL, 0, &uap->sysmsg_result);
 393         }
 394         return (error);
 395 }
 396
 397 /*
 398  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
 399  */
 400 int
 401 sys_extaccept(struct extaccept_args *uap)
 402 {
 403         struct sockaddr *sa = NULL;
 404         int sa_len;
 405         int error;
 406         int fflags = uap->flags & O_FMASK;
 407
 408         if (uap->name) {
 409                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 410                 if (error)
 411                         return (error);
 412
 413                 error = kern_accept(uap->s, fflags, &sa, &sa_len, &uap->sysmsg_result);
 414
 415                 if (error == 0)
 416                         error = copyout(sa, uap->name, sa_len);
 417                 if (error == 0) {
 418                         error = copyout(&sa_len, uap->anamelen,
 419                             sizeof(*uap->anamelen));
 420                 }
 421                 if (sa)
 422                         FREE(sa, M_SONAME);
 423         } else {
 424                 error = kern_accept(uap->s, fflags, NULL, 0, &uap->sysmsg_result);
 425         }
 426         return (error);
 427 }
 428
 429
 430 /*
 431  * Returns TRUE if predicate satisfied.
 432  */
 433 static boolean_t
 434 soconnected_predicate(struct netmsg *msg0)
 435 {
 436         struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
 437         struct socket *so = msg->nm_so;
 438
 439         /* check predicate */
 440         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 441                 msg->nm_netmsg.nm_lmsg.ms_error = so->so_error;
 442                 return (TRUE);
 443         }
 444
 445         return (FALSE);
 446 }
 447
 448 int
 449 kern_connect(int s, int fflags, struct sockaddr *sa)
 450 {
 451         struct thread *td = curthread;
 452         struct proc *p = td->td_proc;
 453         struct file *fp;
 454         struct socket *so;
 455         int error;
 456
 457         error = holdsock(p->p_fd, s, &fp);
 458         if (error)
 459                 return (error);
 460         so = (struct socket *)fp->f_data;
 461
 462         if (fflags & O_FBLOCKING)
 463                 /* fflags &= ~FNONBLOCK; */;
 464         else if (fflags & O_FNONBLOCKING)
 465                 fflags |= FNONBLOCK;
 466         else
 467                 fflags = fp->f_flag;
 468
 469         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 470                 error = EALREADY;
 471                 goto done;
 472         }
 473         error = soconnect(so, sa, td);
 474         if (error)
 475                 goto bad;
 476         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 477                 error = EINPROGRESS;
 478                 goto done;
 479         }
 480         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 481                 struct netmsg_so_notify msg;
 482                 lwkt_port_t port;
 483
 484                 port = so->so_proto->pr_mport(so, sa, PRU_PRED);
 485                 netmsg_init_abortable(&msg.nm_netmsg,
 486                                       &curthread->td_msgport,
 487                                       MSGF_PCATCH,
 488                                       netmsg_so_notify,
 489                                       netmsg_so_notify_doabort);
 490                 msg.nm_predicate = soconnected_predicate;
 491                 msg.nm_so = so;
 492                 msg.nm_etype = NM_REVENT;
 493                 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg);
 494         }
 495         if (error == 0) {
 496                 error = so->so_error;
 497                 so->so_error = 0;
 498         }
 499 bad:
 500         so->so_state &= ~SS_ISCONNECTING;
 501         if (error == ERESTART)
 502                 error = EINTR;
 503 done:
 504         fdrop(fp);
 505         return (error);
 506 }
 507
 508 /*
 509  * connect_args(int s, caddr_t name, int namelen)
 510  */
 511 int
 512 sys_connect(struct connect_args *uap)
 513 {
 514         struct sockaddr *sa;
 515         int error;
 516
 517         error = getsockaddr(&sa, uap->name, uap->namelen);
 518         if (error)
 519                 return (error);
 520         error = kern_connect(uap->s, 0, sa);
 521         FREE(sa, M_SONAME);
 522
 523         return (error);
 524 }
 525
 526 /*
 527  * connect_args(int s, int fflags, caddr_t name, int namelen)
 528  */
 529 int
 530 sys_extconnect(struct extconnect_args *uap)
 531 {
 532         struct sockaddr *sa;
 533         int error;
 534         int fflags = uap->flags & O_FMASK;
 535
 536         error = getsockaddr(&sa, uap->name, uap->namelen);
 537         if (error)
 538                 return (error);
 539         error = kern_connect(uap->s, fflags, sa);
 540         FREE(sa, M_SONAME);
 541
 542         return (error);
 543 }
 544
 545 int
 546 kern_socketpair(int domain, int type, int protocol, int *sv)
 547 {
 548         struct thread *td = curthread;
 549         struct proc *p = td->td_proc;
 550         struct file *fp1, *fp2;
 551         struct socket *so1, *so2;
 552         int fd1, fd2, error;
 553
 554         KKASSERT(p);
 555         error = socreate(domain, &so1, type, protocol, td);
 556         if (error)
 557                 return (error);
 558         error = socreate(domain, &so2, type, protocol, td);
 559         if (error)
 560                 goto free1;
 561         error = falloc(p, &fp1, &fd1);
 562         if (error)
 563                 goto free2;
 564         sv[0] = fd1;
 565         fp1->f_data = so1;
 566         error = falloc(p, &fp2, &fd2);
 567         if (error)
 568                 goto free3;
 569         fp2->f_data = so2;
 570         sv[1] = fd2;
 571         error = soconnect2(so1, so2);
 572         if (error)
 573                 goto free4;
 574         if (type == SOCK_DGRAM) {
 575                 /*
 576                  * Datagram socket connection is asymmetric.
 577                  */
 578                  error = soconnect2(so2, so1);
 579                  if (error)
 580                         goto free4;
 581         }
 582         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 583         fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
 584         fp1->f_ops = fp2->f_ops = &socketops;
 585         fsetfd(p, fp1, fd1);
 586         fsetfd(p, fp2, fd2);
 587         fdrop(fp1);
 588         fdrop(fp2);
 589         return (error);
 590 free4:
 591         fsetfd(p, NULL, fd2);
 592         fdrop(fp2);
 593 free3:
 594         fsetfd(p, NULL, fd1);
 595         fdrop(fp1);
 596 free2:
 597         (void)soclose(so2, 0);
 598 free1:
 599         (void)soclose(so1, 0);
 600         return (error);
 601 }
 602
 603 /*
 604  * socketpair(int domain, int type, int protocol, int *rsv)
 605  */
 606 int
 607 sys_socketpair(struct socketpair_args *uap)
 608 {
 609         int error, sockv[2];
 610
 611         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 612
 613         if (error == 0)
 614                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 615         return (error);
 616 }
 617
 618 int
 619 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 620     struct mbuf *control, int flags, int *res)
 621 {
 622         struct thread *td = curthread;
 623         struct lwp *lp = td->td_lwp;
 624         struct proc *p = td->td_proc;
 625         struct file *fp;
 626         int len, error;
 627         struct socket *so;
 628 #ifdef KTRACE
 629         struct iovec *ktriov = NULL;
 630         struct uio ktruio;
 631 #endif
 632
 633         error = holdsock(p->p_fd, s, &fp);
 634         if (error)
 635                 return (error);
 636         if (auio->uio_resid < 0) {
 637                 error = EINVAL;
 638                 goto done;
 639         }
 640 #ifdef KTRACE
 641         if (KTRPOINT(td, KTR_GENIO)) {
 642                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 643
 644                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 645                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 646                 ktruio = *auio;
 647         }
 648 #endif
 649         len = auio->uio_resid;
 650         so = (struct socket *)fp->f_data;
 651         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 652                 if (fp->f_flag & FNONBLOCK)
 653                         flags |= MSG_FNONBLOCKING;
 654         }
 655         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 656         if (error) {
 657                 if (auio->uio_resid != len && (error == ERESTART ||
 658                     error == EINTR || error == EWOULDBLOCK))
 659                         error = 0;
 660                 if (error == EPIPE)
 661                         lwpsignal(p, lp, SIGPIPE);
 662         }
 663 #ifdef KTRACE
 664         if (ktriov != NULL) {
 665                 if (error == 0) {
 666                         ktruio.uio_iov = ktriov;
 667                         ktruio.uio_resid = len - auio->uio_resid;
 668                         ktrgenio(p, s, UIO_WRITE, &ktruio, error);
 669                 }
 670                 FREE(ktriov, M_TEMP);
 671         }
 672 #endif
 673         if (error == 0)
 674                 *res  = len - auio->uio_resid;
 675 done:
 676         fdrop(fp);
 677         return (error);
 678 }
 679
 680 /*
 681  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
 682  */
 683 int
 684 sys_sendto(struct sendto_args *uap)
 685 {
 686         struct thread *td = curthread;
 687         struct uio auio;
 688         struct iovec aiov;
 689         struct sockaddr *sa = NULL;
 690         int error;
 691
 692         if (uap->to) {
 693                 error = getsockaddr(&sa, uap->to, uap->tolen);
 694                 if (error)
 695                         return (error);
 696         }
 697         aiov.iov_base = uap->buf;
 698         aiov.iov_len = uap->len;
 699         auio.uio_iov = &aiov;
 700         auio.uio_iovcnt = 1;
 701         auio.uio_offset = 0;
 702         auio.uio_resid = uap->len;
 703         auio.uio_segflg = UIO_USERSPACE;
 704         auio.uio_rw = UIO_WRITE;
 705         auio.uio_td = td;
 706
 707         error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
 708             &uap->sysmsg_result);
 709
 710         if (sa)
 711                 FREE(sa, M_SONAME);
 712         return (error);
 713 }
 714
 715 /*
 716  * sendmsg_args(int s, caddr_t msg, int flags)
 717  */
 718 int
 719 sys_sendmsg(struct sendmsg_args *uap)
 720 {
 721         struct thread *td = curthread;
 722         struct msghdr msg;
 723         struct uio auio;
 724         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 725         struct sockaddr *sa = NULL;
 726         struct mbuf *control = NULL;
 727         int error;
 728
 729         error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
 730         if (error)
 731                 return (error);
 732
 733         /*
 734          * Conditionally copyin msg.msg_name.
 735          */
 736         if (msg.msg_name) {
 737                 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 738                 if (error)
 739                         return (error);
 740         }
 741
 742         /*
 743          * Populate auio.
 744          */
 745         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 746                              &auio.uio_resid);
 747         if (error)
 748                 goto cleanup2;
 749         auio.uio_iov = iov;
 750         auio.uio_iovcnt = msg.msg_iovlen;
 751         auio.uio_offset = 0;
 752         auio.uio_segflg = UIO_USERSPACE;
 753         auio.uio_rw = UIO_WRITE;
 754         auio.uio_td = td;
 755
 756         /*
 757          * Conditionally copyin msg.msg_control.
 758          */
 759         if (msg.msg_control) {
 760                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 761                     msg.msg_controllen > MLEN) {
 762                         error = EINVAL;
 763                         goto cleanup;
 764                 }
 765                 control = m_get(MB_WAIT, MT_CONTROL);
 766                 if (control == NULL) {
 767                         error = ENOBUFS;
 768                         goto cleanup;
 769                 }
 770                 control->m_len = msg.msg_controllen;
 771                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 772                     msg.msg_controllen);
 773                 if (error) {
 774                         m_free(control);
 775                         goto cleanup;
 776                 }
 777         }
 778
 779         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 780             &uap->sysmsg_result);
 781
 782 cleanup:
 783         iovec_free(&iov, aiov);
 784 cleanup2:
 785         if (sa)
 786                 FREE(sa, M_SONAME);
 787         return (error);
 788 }
 789
 790 /*
 791  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 792  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 793  * Don't forget to FREE() and m_free() these if they are returned.
 794  */
 795 int
 796 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 797     struct mbuf **control, int *flags, int *res)
 798 {
 799         struct thread *td = curthread;
 800         struct proc *p = td->td_proc;
 801         struct file *fp;
 802         int len, error;
 803         int lflags;
 804         struct socket *so;
 805 #ifdef KTRACE
 806         struct iovec *ktriov = NULL;
 807         struct uio ktruio;
 808 #endif
 809
 810         error = holdsock(p->p_fd, s, &fp);
 811         if (error)
 812                 return (error);
 813         if (auio->uio_resid < 0) {
 814                 error = EINVAL;
 815                 goto done;
 816         }
 817 #ifdef KTRACE
 818         if (KTRPOINT(td, KTR_GENIO)) {
 819                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 820
 821                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 822                 bcopy(auio->uio_iov, ktriov, iovlen);
 823                 ktruio = *auio;
 824         }
 825 #endif
 826         len = auio->uio_resid;
 827         so = (struct socket *)fp->f_data;
 828
 829         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 830                 if (fp->f_flag & FNONBLOCK) {
 831                         if (flags) {
 832                                 *flags |= MSG_FNONBLOCKING;
 833                         } else {
 834                                 lflags = MSG_FNONBLOCKING;
 835                                 flags = &lflags;
 836                         }
 837                 }
 838         }
 839
 840         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 841         if (error) {
 842                 if (auio->uio_resid != len && (error == ERESTART ||
 843                     error == EINTR || error == EWOULDBLOCK))
 844                         error = 0;
 845         }
 846 #ifdef KTRACE
 847         if (ktriov != NULL) {
 848                 if (error == 0) {
 849                         ktruio.uio_iov = ktriov;
 850                         ktruio.uio_resid = len - auio->uio_resid;
 851                         ktrgenio(p, s, UIO_READ, &ktruio, error);
 852                 }
 853                 FREE(ktriov, M_TEMP);
 854         }
 855 #endif
 856         if (error == 0)
 857                 *res = len - auio->uio_resid;
 858 done:
 859         fdrop(fp);
 860         return (error);
 861 }
 862
 863 /*
 864  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
 865  *                      caddr_t from, int *fromlenaddr)
 866  */
 867 int
 868 sys_recvfrom(struct recvfrom_args *uap)
 869 {
 870         struct thread *td = curthread;
 871         struct uio auio;
 872         struct iovec aiov;
 873         struct sockaddr *sa = NULL;
 874         int error, fromlen;
 875
 876         if (uap->from && uap->fromlenaddr) {
 877                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
 878                 if (error)
 879                         return (error);
 880                 if (fromlen < 0)
 881                         return (EINVAL);
 882         } else {
 883                 fromlen = 0;
 884         }
 885         aiov.iov_base = uap->buf;
 886         aiov.iov_len = uap->len;
 887         auio.uio_iov = &aiov;
 888         auio.uio_iovcnt = 1;
 889         auio.uio_offset = 0;
 890         auio.uio_resid = uap->len;
 891         auio.uio_segflg = UIO_USERSPACE;
 892         auio.uio_rw = UIO_READ;
 893         auio.uio_td = td;
 894
 895         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
 896             &uap->flags, &uap->sysmsg_result);
 897
 898         if (error == 0 && uap->from) {
 899                 /* note: sa may still be NULL */
 900                 if (sa) {
 901                         fromlen = MIN(fromlen, sa->sa_len);
 902                         error = copyout(sa, uap->from, fromlen);
 903                 } else {
 904                         fromlen = 0;
 905                 }
 906                 if (error == 0) {
 907                         error = copyout(&fromlen, uap->fromlenaddr,
 908                                         sizeof(fromlen));
 909                 }
 910         }
 911         if (sa)
 912                 FREE(sa, M_SONAME);
 913
 914         return (error);
 915 }
 916
 917 /*
 918  * recvmsg_args(int s, struct msghdr *msg, int flags)
 919  */
 920 int
 921 sys_recvmsg(struct recvmsg_args *uap)
 922 {
 923         struct thread *td = curthread;
 924         struct msghdr msg;
 925         struct uio auio;
 926         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 927         struct mbuf *m, *control = NULL;
 928         struct sockaddr *sa = NULL;
 929         caddr_t ctlbuf;
 930         socklen_t *ufromlenp, *ucontrollenp;
 931         int error, fromlen, controllen, len, flags, *uflagsp;
 932
 933         /*
 934          * This copyin handles everything except the iovec.
 935          */
 936         error = copyin(uap->msg, &msg, sizeof(msg));
 937         if (error)
 938                 return (error);
 939
 940         if (msg.msg_name && msg.msg_namelen < 0)
 941                 return (EINVAL);
 942         if (msg.msg_control && msg.msg_controllen < 0)
 943                 return (EINVAL);
 944
 945         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 946             msg_namelen));
 947         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 948             msg_controllen));
 949         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
 950             msg_flags));
 951
 952         /*
 953          * Populate auio.
 954          */
 955         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 956                              &auio.uio_resid);
 957         if (error)
 958                 return (error);
 959         auio.uio_iov = iov;
 960         auio.uio_iovcnt = msg.msg_iovlen;
 961         auio.uio_offset = 0;
 962         auio.uio_segflg = UIO_USERSPACE;
 963         auio.uio_rw = UIO_READ;
 964         auio.uio_td = td;
 965
 966         flags = uap->flags;
 967
 968         error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
 969             msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
 970
 971         /*
 972          * Conditionally copyout the name and populate the namelen field.
 973          */
 974         if (error == 0 && msg.msg_name) {
 975                 /* note: sa may still be NULL */
 976                 if (sa != NULL) {
 977                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
 978                         error = copyout(sa, msg.msg_name, fromlen);
 979                 } else
 980                         fromlen = 0;
 981                 if (error == 0)
 982                         error = copyout(&fromlen, ufromlenp,
 983                             sizeof(*ufromlenp));
 984         }
 985
 986         /*
 987          * Copyout msg.msg_control and msg.msg_controllen.
 988          */
 989         if (error == 0 && msg.msg_control) {
 990                 len = msg.msg_controllen;
 991                 m = control;
 992                 ctlbuf = (caddr_t)msg.msg_control;
 993
 994                 while(m && len > 0) {
 995                         unsigned int tocopy;
 996
 997                         if (len >= m->m_len) {
 998                                 tocopy = m->m_len;
 999                         } else {
1000                                 msg.msg_flags |= MSG_CTRUNC;
1001                                 tocopy = len;
1002                         }
1003
1004                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1005                         if (error)
1006                                 goto cleanup;
1007
1008                         ctlbuf += tocopy;
1009                         len -= tocopy;
1010                         m = m->m_next;
1011                 }
1012                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1013                 error = copyout(&controllen, ucontrollenp,
1014                     sizeof(*ucontrollenp));
1015         }
1016
1017         if (error == 0)
1018                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1019
1020 cleanup:
1021         if (sa)
1022                 FREE(sa, M_SONAME);
1023         iovec_free(&iov, aiov);
1024         if (control)
1025                 m_freem(control);
1026         return (error);
1027 }
1028
1029 /*
1030  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1031  * in kernel pointer instead of a userland pointer.  This allows us
1032  * to manipulate socket options in the emulation code.
1033  */
1034 int
1035 kern_setsockopt(int s, struct sockopt *sopt)
1036 {
1037         struct thread *td = curthread;
1038         struct proc *p = td->td_proc;
1039         struct file *fp;
1040         int error;
1041
1042         if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1043                 return (EFAULT);
1044         if (sopt->sopt_valsize < 0)
1045                 return (EINVAL);
1046
1047         error = holdsock(p->p_fd, s, &fp);
1048         if (error)
1049                 return (error);
1050
1051         error = sosetopt((struct socket *)fp->f_data, sopt);
1052         fdrop(fp);
1053         return (error);
1054 }
1055
1056 /*
1057  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1058  */
1059 int
1060 sys_setsockopt(struct setsockopt_args *uap)
1061 {
1062         struct thread *td = curthread;
1063         struct sockopt sopt;
1064         int error;
1065
1066         sopt.sopt_level = uap->level;
1067         sopt.sopt_name = uap->name;
1068         sopt.sopt_val = uap->val;
1069         sopt.sopt_valsize = uap->valsize;
1070         sopt.sopt_td = td;
1071
1072         error = kern_setsockopt(uap->s, &sopt);
1073         return(error);
1074 }
1075
1076 /*
1077  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1078  * in kernel pointer instead of a userland pointer.  This allows us
1079  * to manipulate socket options in the emulation code.
1080  */
1081 int
1082 kern_getsockopt(int s, struct sockopt *sopt)
1083 {
1084         struct thread *td = curthread;
1085         struct proc *p = td->td_proc;
1086         struct file *fp;
1087         int error;
1088
1089         if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1090                 return (EFAULT);
1091         if (sopt->sopt_valsize < 0)
1092                 return (EINVAL);
1093
1094         error = holdsock(p->p_fd, s, &fp);
1095         if (error)
1096                 return (error);
1097
1098         error = sogetopt((struct socket *)fp->f_data, sopt);
1099         fdrop(fp);
1100         return (error);
1101 }
1102
1103 /*
1104  * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1105  */
1106 int
1107 sys_getsockopt(struct getsockopt_args *uap)
1108 {
1109         struct thread *td = curthread;
1110         struct  sockopt sopt;
1111         int     error, valsize;
1112
1113         if (uap->val) {
1114                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1115                 if (error)
1116                         return (error);
1117                 if (valsize < 0)
1118                         return (EINVAL);
1119         } else {
1120                 valsize = 0;
1121         }
1122
1123         sopt.sopt_level = uap->level;
1124         sopt.sopt_name = uap->name;
1125         sopt.sopt_val = uap->val;
1126         sopt.sopt_valsize = valsize;
1127         sopt.sopt_td = td;
1128
1129         error = kern_getsockopt(uap->s, &sopt);
1130         if (error == 0) {
1131                 valsize = sopt.sopt_valsize;
1132                 error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1133         }
1134         return (error);
1135 }
1136
1137 /*
1138  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1139  * This allows kern_getsockname() to return a pointer to an allocated struct
1140  * sockaddr which must be freed later with FREE().  The caller must
1141  * initialize *name to NULL.
1142  */
1143 int
1144 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1145 {
1146         struct thread *td = curthread;
1147         struct proc *p = td->td_proc;
1148         struct file *fp;
1149         struct socket *so;
1150         struct sockaddr *sa = NULL;
1151         int error;
1152
1153         error = holdsock(p->p_fd, s, &fp);
1154         if (error)
1155                 return (error);
1156         if (*namelen < 0) {
1157                 fdrop(fp);
1158                 return (EINVAL);
1159         }
1160         so = (struct socket *)fp->f_data;
1161         error = so_pru_sockaddr(so, &sa);
1162         if (error == 0) {
1163                 if (sa == 0) {
1164                         *namelen = 0;
1165                 } else {
1166                         *namelen = MIN(*namelen, sa->sa_len);
1167                         *name = sa;
1168                 }
1169         }
1170
1171         fdrop(fp);
1172         return (error);
1173 }
1174
1175 /*
1176  * getsockname_args(int fdes, caddr_t asa, int *alen)
1177  *
1178  * Get socket name.
1179  */
1180 int
1181 sys_getsockname(struct getsockname_args *uap)
1182 {
1183         struct sockaddr *sa = NULL;
1184         int error, sa_len;
1185
1186         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1187         if (error)
1188                 return (error);
1189
1190         error = kern_getsockname(uap->fdes, &sa, &sa_len);
1191
1192         if (error == 0)
1193                 error = copyout(sa, uap->asa, sa_len);
1194         if (error == 0)
1195                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1196         if (sa)
1197                 FREE(sa, M_SONAME);
1198         return (error);
1199 }
1200
1201 /*
1202  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1203  * This allows kern_getpeername() to return a pointer to an allocated struct
1204  * sockaddr which must be freed later with FREE().  The caller must
1205  * initialize *name to NULL.
1206  */
1207 int
1208 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1209 {
1210         struct thread *td = curthread;
1211         struct proc *p = td->td_proc;
1212         struct file *fp;
1213         struct socket *so;
1214         struct sockaddr *sa = NULL;
1215         int error;
1216
1217         error = holdsock(p->p_fd, s, &fp);
1218         if (error)
1219                 return (error);
1220         if (*namelen < 0) {
1221                 fdrop(fp);
1222                 return (EINVAL);
1223         }
1224         so = (struct socket *)fp->f_data;
1225         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1226                 fdrop(fp);
1227                 return (ENOTCONN);
1228         }
1229         error = so_pru_peeraddr(so, &sa);
1230         if (error == 0) {
1231                 if (sa == 0) {
1232                         *namelen = 0;
1233                 } else {
1234                         *namelen = MIN(*namelen, sa->sa_len);
1235                         *name = sa;
1236                 }
1237         }
1238
1239         fdrop(fp);
1240         return (error);
1241 }
1242
1243 /*
1244  * getpeername_args(int fdes, caddr_t asa, int *alen)
1245  *
1246  * Get name of peer for connected socket.
1247  */
1248 int
1249 sys_getpeername(struct getpeername_args *uap)
1250 {
1251         struct sockaddr *sa = NULL;
1252         int error, sa_len;
1253
1254         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1255         if (error)
1256                 return (error);
1257
1258         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1259
1260         if (error == 0)
1261                 error = copyout(sa, uap->asa, sa_len);
1262         if (error == 0)
1263                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1264         if (sa)
1265                 FREE(sa, M_SONAME);
1266         return (error);
1267 }
1268
1269 int
1270 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1271 {
1272         struct sockaddr *sa;
1273         int error;
1274
1275         *namp = NULL;
1276         if (len > SOCK_MAXADDRLEN)
1277                 return ENAMETOOLONG;
1278         if (len < offsetof(struct sockaddr, sa_data[0]))
1279                 return EDOM;
1280         MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1281         error = copyin(uaddr, sa, len);
1282         if (error) {
1283                 FREE(sa, M_SONAME);
1284         } else {
1285 #if BYTE_ORDER != BIG_ENDIAN
1286                 /*
1287                  * The bind(), connect(), and sendto() syscalls were not
1288                  * versioned for COMPAT_43.  Thus, this check must stay.
1289                  */
1290                 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1291                         sa->sa_family = sa->sa_len;
1292 #endif
1293                 sa->sa_len = len;
1294                 *namp = sa;
1295         }
1296         return error;
1297 }
1298
1299 /*
1300  * Detach a mapped page and release resources back to the system.
1301  * We must release our wiring and if the object is ripped out
1302  * from under the vm_page we become responsible for freeing the
1303  * page.  These routines must be MPSAFE.
1304  *
1305  * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1306  *
1307  * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1308  */
1309 static void
1310 sf_buf_mref(void *arg)
1311 {
1312         struct sfbuf_mref *sfm = arg;
1313
1314         /*
1315          * We must already hold a ref so there is no race to 0, just
1316          * atomically increment the count.
1317          */
1318         atomic_add_int(&sfm->mref_count, 1);
1319 }
1320
1321 static void
1322 sf_buf_mfree(void *arg)
1323 {
1324         struct sfbuf_mref *sfm = arg;
1325         vm_page_t m;
1326
1327         KKASSERT(sfm->mref_count > 0);
1328         if (sfm->mref_count == 1) {
1329                 /*
1330                  * We are the only holder so no further locking is required,
1331                  * the sfbuf can simply be freed.
1332                  */
1333                 sfm->mref_count = 0;
1334                 goto freeit;
1335         } else {
1336                 /*
1337                  * There may be other holders, we must obtain the serializer
1338                  * to protect against a sf_buf_mfree() race to 0.  An atomic
1339                  * operation is still required for races against
1340                  * sf_buf_mref().
1341                  *
1342                  * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1343                  */
1344                 lwkt_serialize_enter(&sfm->serializer);
1345                 atomic_subtract_int(&sfm->mref_count, 1);
1346                 if (sfm->mref_count == 0) {
1347                         lwkt_serialize_exit(&sfm->serializer);
1348 freeit:
1349                         get_mplock();
1350                         crit_enter();
1351                         m = sf_buf_page(sfm->sf);
1352                         sf_buf_free(sfm->sf);
1353                         vm_page_unwire(m, 0);
1354                         if (m->wire_count == 0 && m->object == NULL)
1355                                 vm_page_try_to_free(m);
1356                         crit_exit();
1357                         rel_mplock();
1358                         kfree(sfm, M_SENDFILE);
1359                 } else {
1360                         lwkt_serialize_exit(&sfm->serializer);
1361                 }
1362         }
1363 }
1364
1365 /*
1366  * sendfile(2).
1367  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1368  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1369  *
1370  * Send a file specified by 'fd' and starting at 'offset' to a socket
1371  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1372  * nbytes == 0. Optionally add a header and/or trailer to the socket
1373  * output. If specified, write the total number of bytes sent into *sbytes.
1374  *
1375  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1376  * the headers to count against the remaining bytes to be sent from
1377  * the file descriptor.  We may wish to implement a compatibility syscall
1378  * in the future.
1379  */
1380 int
1381 sys_sendfile(struct sendfile_args *uap)
1382 {
1383         struct thread *td = curthread;
1384         struct proc *p = td->td_proc;
1385         struct file *fp;
1386         struct vnode *vp = NULL;
1387         struct sf_hdtr hdtr;
1388         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1389         struct uio auio;
1390         struct mbuf *mheader = NULL;
1391         off_t hdtr_size = 0, sbytes;
1392         int error, hbytes = 0, tbytes;
1393
1394         KKASSERT(p);
1395
1396         /*
1397          * Do argument checking. Must be a regular file in, stream
1398          * type and connected socket out, positive offset.
1399          */
1400         fp = holdfp(p->p_fd, uap->fd, FREAD);
1401         if (fp == NULL) {
1402                 return (EBADF);
1403         }
1404         if (fp->f_type != DTYPE_VNODE) {
1405                 fdrop(fp);
1406                 return (EINVAL);
1407         }
1408         vp = (struct vnode *)fp->f_data;
1409         vref(vp);
1410         fdrop(fp);
1411
1412         /*
1413          * If specified, get the pointer to the sf_hdtr struct for
1414          * any headers/trailers.
1415          */
1416         if (uap->hdtr) {
1417                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1418                 if (error)
1419                         goto done;
1420                 /*
1421                  * Send any headers.
1422                  */
1423                 if (hdtr.headers) {
1424                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1425                                              hdtr.hdr_cnt, &hbytes);
1426                         if (error)
1427                                 goto done;
1428                         auio.uio_iov = iov;
1429                         auio.uio_iovcnt = hdtr.hdr_cnt;
1430                         auio.uio_offset = 0;
1431                         auio.uio_segflg = UIO_USERSPACE;
1432                         auio.uio_rw = UIO_WRITE;
1433                         auio.uio_td = td;
1434                         auio.uio_resid = hbytes;
1435
1436                         mheader = m_uiomove(&auio);
1437
1438                         iovec_free(&iov, aiov);
1439                         if (mheader == NULL)
1440                                 goto done;
1441                 }
1442         }
1443
1444         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1445             &sbytes, uap->flags);
1446         if (error)
1447                 goto done;
1448
1449         /*
1450          * Send trailers. Wimp out and use writev(2).
1451          */
1452         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1453                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1454                                      hdtr.trl_cnt, &auio.uio_resid);
1455                 if (error)
1456                         goto done;
1457                 auio.uio_iov = iov;
1458                 auio.uio_iovcnt = hdtr.trl_cnt;
1459                 auio.uio_offset = 0;
1460                 auio.uio_segflg = UIO_USERSPACE;
1461                 auio.uio_rw = UIO_WRITE;
1462                 auio.uio_td = td;
1463
1464                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1465
1466                 iovec_free(&iov, aiov);
1467                 if (error)
1468                         goto done;
1469                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1470         }
1471
1472 done:
1473         if (uap->sbytes != NULL) {
1474                 sbytes += hdtr_size;
1475                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1476         }
1477         if (vp)
1478                 vrele(vp);
1479         return (error);
1480 }
1481
1482 int
1483 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1484     struct mbuf *mheader, off_t *sbytes, int flags)
1485 {
1486         struct thread *td = curthread;
1487         struct proc *p = td->td_proc;
1488         struct vm_object *obj;
1489         struct socket *so;
1490         struct file *fp;
1491         struct mbuf *m;
1492         struct sf_buf *sf;
1493         struct sfbuf_mref *sfm;
1494         struct vm_page *pg;
1495         off_t off, xfsize;
1496         off_t hbytes = 0;
1497         int error = 0;
1498
1499         if (vp->v_type != VREG) {
1500                 error = EINVAL;
1501                 goto done0;
1502         }
1503         if ((obj = vp->v_object) == NULL) {
1504                 error = EINVAL;
1505                 goto done0;
1506         }
1507         error = holdsock(p->p_fd, sfd, &fp);
1508         if (error)
1509                 goto done0;
1510         so = (struct socket *)fp->f_data;
1511         if (so->so_type != SOCK_STREAM) {
1512                 error = EINVAL;
1513                 goto done;
1514         }
1515         if ((so->so_state & SS_ISCONNECTED) == 0) {
1516                 error = ENOTCONN;
1517                 goto done;
1518         }
1519         if (offset < 0) {
1520                 error = EINVAL;
1521                 goto done;
1522         }
1523
1524         *sbytes = 0;
1525         /*
1526          * Protect against multiple writers to the socket.
1527          */
1528         ssb_lock(&so->so_snd, M_WAITOK);
1529
1530         /*
1531          * Loop through the pages in the file, starting with the requested
1532          * offset. Get a file page (do I/O if necessary), map the file page
1533          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1534          * it on the socket.
1535          */
1536         for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1537                 vm_pindex_t pindex;
1538                 vm_offset_t pgoff;
1539
1540                 pindex = OFF_TO_IDX(off);
1541 retry_lookup:
1542                 /*
1543                  * Calculate the amount to transfer. Not to exceed a page,
1544                  * the EOF, or the passed in nbytes.
1545                  */
1546                 xfsize = vp->v_filesize - off;
1547                 if (xfsize > PAGE_SIZE)
1548                         xfsize = PAGE_SIZE;
1549                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1550                 if (PAGE_SIZE - pgoff < xfsize)
1551                         xfsize = PAGE_SIZE - pgoff;
1552                 if (nbytes && xfsize > (nbytes - *sbytes))
1553                         xfsize = nbytes - *sbytes;
1554                 if (xfsize <= 0)
1555                         break;
1556                 /*
1557                  * Optimize the non-blocking case by looking at the socket space
1558                  * before going to the extra work of constituting the sf_buf.
1559                  */
1560                 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1561                         if (so->so_state & SS_CANTSENDMORE)
1562                                 error = EPIPE;
1563                         else
1564                                 error = EAGAIN;
1565                         ssb_unlock(&so->so_snd);
1566                         goto done;
1567                 }
1568                 /*
1569                  * Attempt to look up the page.
1570                  *
1571                  *      Allocate if not found, wait and loop if busy, then
1572                  *      wire the page.  critical section protection is
1573                  *      required to maintain the object association (an
1574                  *      interrupt can free the page) through to the
1575                  *      vm_page_wire() call.
1576                  */
1577                 crit_enter();
1578                 pg = vm_page_lookup(obj, pindex);
1579                 if (pg == NULL) {
1580                         pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1581                         if (pg == NULL) {
1582                                 vm_wait();
1583                                 crit_exit();
1584                                 goto retry_lookup;
1585                         }
1586                         vm_page_wakeup(pg);
1587                 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1588                         crit_exit();
1589                         goto retry_lookup;
1590                 }
1591                 vm_page_wire(pg);
1592                 crit_exit();
1593
1594                 /*
1595                  * If page is not valid for what we need, initiate I/O
1596                  */
1597
1598                 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1599                         struct uio auio;
1600                         struct iovec aiov;
1601                         int bsize;
1602
1603                         /*
1604                          * Ensure that our page is still around when the I/O
1605                          * completes.
1606                          */
1607                         vm_page_io_start(pg);
1608
1609                         /*
1610                          * Get the page from backing store.
1611                          */
1612                         bsize = vp->v_mount->mnt_stat.f_iosize;
1613                         auio.uio_iov = &aiov;
1614                         auio.uio_iovcnt = 1;
1615                         aiov.iov_base = 0;
1616                         aiov.iov_len = MAXBSIZE;
1617                         auio.uio_resid = MAXBSIZE;
1618                         auio.uio_offset = trunc_page(off);
1619                         auio.uio_segflg = UIO_NOCOPY;
1620                         auio.uio_rw = UIO_READ;
1621                         auio.uio_td = td;
1622                         vn_lock(vp, LK_SHARED | LK_RETRY);
1623                         error = VOP_READ(vp, &auio,
1624                                     IO_VMIO | ((MAXBSIZE / bsize) << 16),
1625                                     p->p_ucred);
1626                         vn_unlock(vp);
1627                         vm_page_flag_clear(pg, PG_ZERO);
1628                         vm_page_io_finish(pg);
1629                         if (error) {
1630                                 crit_enter();
1631                                 vm_page_unwire(pg, 0);
1632                                 vm_page_try_to_free(pg);
1633                                 crit_exit();
1634                                 ssb_unlock(&so->so_snd);
1635                                 goto done;
1636                         }
1637                 }
1638
1639
1640                 /*
1641                  * Get a sendfile buf. We usually wait as long as necessary,
1642                  * but this wait can be interrupted.
1643                  */
1644                 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1645                         crit_enter();
1646                         vm_page_unwire(pg, 0);
1647                         vm_page_try_to_free(pg);
1648                         crit_exit();
1649                         ssb_unlock(&so->so_snd);
1650                         error = EINTR;
1651                         goto done;
1652                 }
1653
1654                 /*
1655                  * Get an mbuf header and set it up as having external storage.
1656                  */
1657                 MGETHDR(m, MB_WAIT, MT_DATA);
1658                 if (m == NULL) {
1659                         error = ENOBUFS;
1660                         sf_buf_free(sf);
1661                         ssb_unlock(&so->so_snd);
1662                         goto done;
1663                 }
1664
1665                 /*
1666                  * sfm is a temporary hack, use a per-cpu cache for this.
1667                  */
1668                 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1669                 sfm->sf = sf;
1670                 sfm->mref_count = 1;
1671                 lwkt_serialize_init(&sfm->serializer);
1672
1673                 m->m_ext.ext_free = sf_buf_mfree;
1674                 m->m_ext.ext_ref = sf_buf_mref;
1675                 m->m_ext.ext_arg = sfm;
1676                 m->m_ext.ext_buf = (void *)sf->kva;
1677                 m->m_ext.ext_size = PAGE_SIZE;
1678                 m->m_data = (char *) sf->kva + pgoff;
1679                 m->m_flags |= M_EXT;
1680                 m->m_pkthdr.len = m->m_len = xfsize;
1681                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1682
1683                 if (mheader != NULL) {
1684                         hbytes = mheader->m_pkthdr.len;
1685                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1686                         m_cat(mheader, m);
1687                         m = mheader;
1688                         mheader = NULL;
1689                 } else
1690                         hbytes = 0;
1691
1692                 /*
1693                  * Add the buffer to the socket buffer chain.
1694                  */
1695                 crit_enter();
1696 retry_space:
1697                 /*
1698                  * Make sure that the socket is still able to take more data.
1699                  * CANTSENDMORE being true usually means that the connection
1700                  * was closed. so_error is true when an error was sensed after
1701                  * a previous send.
1702                  * The state is checked after the page mapping and buffer
1703                  * allocation above since those operations may block and make
1704                  * any socket checks stale. From this point forward, nothing
1705                  * blocks before the pru_send (or more accurately, any blocking
1706                  * results in a loop back to here to re-check).
1707                  */
1708                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1709                         if (so->so_state & SS_CANTSENDMORE) {
1710                                 error = EPIPE;
1711                         } else {
1712                                 error = so->so_error;
1713                                 so->so_error = 0;
1714                         }
1715                         m_freem(m);
1716                         ssb_unlock(&so->so_snd);
1717                         crit_exit();
1718                         goto done;
1719                 }
1720                 /*
1721                  * Wait for socket space to become available. We do this just
1722                  * after checking the connection state above in order to avoid
1723                  * a race condition with ssb_wait().
1724                  */
1725                 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1726                         if (fp->f_flag & FNONBLOCK) {
1727                                 m_freem(m);
1728                                 ssb_unlock(&so->so_snd);
1729                                 crit_exit();
1730                                 error = EAGAIN;
1731                                 goto done;
1732                         }
1733                         error = ssb_wait(&so->so_snd);
1734                         /*
1735                          * An error from ssb_wait usually indicates that we've
1736                          * been interrupted by a signal. If we've sent anything
1737                          * then return bytes sent, otherwise return the error.
1738                          */
1739                         if (error) {
1740                                 m_freem(m);
1741                                 ssb_unlock(&so->so_snd);
1742                                 crit_exit();
1743                                 goto done;
1744                         }
1745                         goto retry_space;
1746                 }
1747                 error = so_pru_send(so, 0, m, NULL, NULL, td);
1748                 crit_exit();
1749                 if (error) {
1750                         ssb_unlock(&so->so_snd);
1751                         goto done;
1752                 }
1753         }
1754         if (mheader != NULL) {
1755                 *sbytes += mheader->m_pkthdr.len;
1756                 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1757                 mheader = NULL;
1758         }
1759         ssb_unlock(&so->so_snd);
1760
1761 done:
1762         fdrop(fp);
1763 done0:
1764         if (mheader != NULL)
1765                 m_freem(mheader);
1766         return (error);
1767 }
1768
1769 int
1770 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1771 {
1772 #ifdef SCTP
1773         struct thread *td = curthread;
1774         struct proc *p = td->td_proc;
1775         struct file *lfp = NULL;
1776         struct file *nfp = NULL;
1777         int error;
1778         struct socket *head, *so;
1779         caddr_t assoc_id;
1780         int fd;
1781         short fflag;            /* type must match fp->f_flag */
1782
1783         assoc_id = uap->name;
1784         error = holdsock(p->p_fd, uap->sd, &lfp);
1785         if (error) {
1786                 return (error);
1787         }
1788         crit_enter();
1789         head = (struct socket *)lfp->f_data;
1790         error = sctp_can_peel_off(head, assoc_id);
1791         if (error) {
1792                 crit_exit();
1793                 goto done;
1794         }
1795         /*
1796          * At this point we know we do have a assoc to pull
1797          * we proceed to get the fd setup. This may block
1798          * but that is ok.
1799          */
1800
1801         fflag = lfp->f_flag;
1802         error = falloc(p, &nfp, &fd);
1803         if (error) {
1804                 /*
1805                  * Probably ran out of file descriptors. Put the
1806                  * unaccepted connection back onto the queue and
1807                  * do another wakeup so some other process might
1808                  * have a chance at it.
1809                  */
1810                 crit_exit();
1811                 goto done;
1812         }
1813         uap->sysmsg_result = fd;
1814
1815         so = sctp_get_peeloff(head, assoc_id, &error);
1816         if (so == NULL) {
1817                 /*
1818                  * Either someone else peeled it off OR
1819                  * we can't get a socket.
1820                  */
1821                 goto noconnection;
1822         }
1823         so->so_state &= ~SS_COMP;
1824         so->so_state &= ~SS_NOFDREF;
1825         so->so_head = NULL;
1826         if (head->so_sigio != NULL)
1827                 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1828
1829         nfp->f_type = DTYPE_SOCKET;
1830         nfp->f_flag = fflag;
1831         nfp->f_ops = &socketops;
1832         nfp->f_data = so;
1833
1834 noconnection:
1835         /*
1836          * Assign the file pointer to the reserved descriptor, or clear
1837          * the reserved descriptor if an error occured.
1838          */
1839         if (error)
1840                 fsetfd(p, NULL, fd);
1841         else
1842                 fsetfd(p, nfp, fd);
1843         crit_exit();
1844         /*
1845          * Release explicitly held references before returning.
1846          */
1847 done:
1848         if (nfp != NULL)
1849                 fdrop(nfp);
1850         fdrop(lfp);
1851         return (error);
1852 #else /* SCTP */
1853         return(EOPNOTSUPP);
1854 #endif /* SCTP */
1855 }