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 1.92 2008/11/26 13:10:56 sephe 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 <sys/socketvar2.h>
  82 #include <net/netmsg2.h>
  83
  84 #ifdef SCTP
  85 #include <netinet/sctp_peeloff.h>
  86 #endif /* SCTP */
  87
  88 struct sfbuf_mref {
  89         struct sf_buf   *sf;
  90         int             mref_count;
  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_iresult);
 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, NULL, PRU_PRED);
 291         netmsg_init_abortable(&msg.nm_netmsg, &curthread->td_msgport,
 292                               0,
 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, PCATCH);
 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         fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred, NULL);
 325         tmp = fflag & FASYNC;
 326         fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred, NULL);
 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,
 382                                     &uap->sysmsg_iresult);
 383
 384                 if (error == 0)
 385                         error = copyout(sa, uap->name, sa_len);
 386                 if (error == 0) {
 387                         error = copyout(&sa_len, uap->anamelen,
 388                             sizeof(*uap->anamelen));
 389                 }
 390                 if (sa)
 391                         FREE(sa, M_SONAME);
 392         } else {
 393                 error = kern_accept(uap->s, 0, NULL, 0,
 394                                     &uap->sysmsg_iresult);
 395         }
 396         return (error);
 397 }
 398
 399 /*
 400  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
 401  */
 402 int
 403 sys_extaccept(struct extaccept_args *uap)
 404 {
 405         struct sockaddr *sa = NULL;
 406         int sa_len;
 407         int error;
 408         int fflags = uap->flags & O_FMASK;
 409
 410         if (uap->name) {
 411                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 412                 if (error)
 413                         return (error);
 414
 415                 error = kern_accept(uap->s, fflags, &sa, &sa_len,
 416                                     &uap->sysmsg_iresult);
 417
 418                 if (error == 0)
 419                         error = copyout(sa, uap->name, sa_len);
 420                 if (error == 0) {
 421                         error = copyout(&sa_len, uap->anamelen,
 422                             sizeof(*uap->anamelen));
 423                 }
 424                 if (sa)
 425                         FREE(sa, M_SONAME);
 426         } else {
 427                 error = kern_accept(uap->s, fflags, NULL, 0,
 428                                     &uap->sysmsg_iresult);
 429         }
 430         return (error);
 431 }
 432
 433
 434 /*
 435  * Returns TRUE if predicate satisfied.
 436  */
 437 static boolean_t
 438 soconnected_predicate(struct netmsg *msg0)
 439 {
 440         struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
 441         struct socket *so = msg->nm_so;
 442
 443         /* check predicate */
 444         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 445                 msg->nm_netmsg.nm_lmsg.ms_error = so->so_error;
 446                 return (TRUE);
 447         }
 448
 449         return (FALSE);
 450 }
 451
 452 int
 453 kern_connect(int s, int fflags, struct sockaddr *sa)
 454 {
 455         struct thread *td = curthread;
 456         struct proc *p = td->td_proc;
 457         struct file *fp;
 458         struct socket *so;
 459         int error, interrupted = 0;
 460
 461         error = holdsock(p->p_fd, s, &fp);
 462         if (error)
 463                 return (error);
 464         so = (struct socket *)fp->f_data;
 465
 466         if (fflags & O_FBLOCKING)
 467                 /* fflags &= ~FNONBLOCK; */;
 468         else if (fflags & O_FNONBLOCKING)
 469                 fflags |= FNONBLOCK;
 470         else
 471                 fflags = fp->f_flag;
 472
 473         if (so->so_state & SS_ISCONNECTING) {
 474                 error = EALREADY;
 475                 goto done;
 476         }
 477         error = soconnect(so, sa, td);
 478         if (error)
 479                 goto bad;
 480         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 481                 error = EINPROGRESS;
 482                 goto done;
 483         }
 484         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 485                 struct netmsg_so_notify msg;
 486                 lwkt_port_t port;
 487
 488                 port = so->so_proto->pr_mport(so, sa, NULL, PRU_PRED);
 489                 netmsg_init_abortable(&msg.nm_netmsg,
 490                                       &curthread->td_msgport,
 491                                       0,
 492                                       netmsg_so_notify,
 493                                       netmsg_so_notify_doabort);
 494                 msg.nm_predicate = soconnected_predicate;
 495                 msg.nm_so = so;
 496                 msg.nm_etype = NM_REVENT;
 497                 error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, PCATCH);
 498                 if (error == EINTR || error == ERESTART)
 499                         interrupted = 1;
 500         }
 501         if (error == 0) {
 502                 error = so->so_error;
 503                 so->so_error = 0;
 504         }
 505 bad:
 506         if (!interrupted)
 507                 so->so_state &= ~SS_ISCONNECTING;
 508         if (error == ERESTART)
 509                 error = EINTR;
 510 done:
 511         fdrop(fp);
 512         return (error);
 513 }
 514
 515 /*
 516  * connect_args(int s, caddr_t name, int namelen)
 517  */
 518 int
 519 sys_connect(struct connect_args *uap)
 520 {
 521         struct sockaddr *sa;
 522         int error;
 523
 524         error = getsockaddr(&sa, uap->name, uap->namelen);
 525         if (error)
 526                 return (error);
 527         error = kern_connect(uap->s, 0, sa);
 528         FREE(sa, M_SONAME);
 529
 530         return (error);
 531 }
 532
 533 /*
 534  * connect_args(int s, int fflags, caddr_t name, int namelen)
 535  */
 536 int
 537 sys_extconnect(struct extconnect_args *uap)
 538 {
 539         struct sockaddr *sa;
 540         int error;
 541         int fflags = uap->flags & O_FMASK;
 542
 543         error = getsockaddr(&sa, uap->name, uap->namelen);
 544         if (error)
 545                 return (error);
 546         error = kern_connect(uap->s, fflags, sa);
 547         FREE(sa, M_SONAME);
 548
 549         return (error);
 550 }
 551
 552 int
 553 kern_socketpair(int domain, int type, int protocol, int *sv)
 554 {
 555         struct thread *td = curthread;
 556         struct proc *p = td->td_proc;
 557         struct file *fp1, *fp2;
 558         struct socket *so1, *so2;
 559         int fd1, fd2, error;
 560
 561         KKASSERT(p);
 562         error = socreate(domain, &so1, type, protocol, td);
 563         if (error)
 564                 return (error);
 565         error = socreate(domain, &so2, type, protocol, td);
 566         if (error)
 567                 goto free1;
 568         error = falloc(p, &fp1, &fd1);
 569         if (error)
 570                 goto free2;
 571         sv[0] = fd1;
 572         fp1->f_data = so1;
 573         error = falloc(p, &fp2, &fd2);
 574         if (error)
 575                 goto free3;
 576         fp2->f_data = so2;
 577         sv[1] = fd2;
 578         error = soconnect2(so1, so2);
 579         if (error)
 580                 goto free4;
 581         if (type == SOCK_DGRAM) {
 582                 /*
 583                  * Datagram socket connection is asymmetric.
 584                  */
 585                  error = soconnect2(so2, so1);
 586                  if (error)
 587                         goto free4;
 588         }
 589         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 590         fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
 591         fp1->f_ops = fp2->f_ops = &socketops;
 592         fsetfd(p, fp1, fd1);
 593         fsetfd(p, fp2, fd2);
 594         fdrop(fp1);
 595         fdrop(fp2);
 596         return (error);
 597 free4:
 598         fsetfd(p, NULL, fd2);
 599         fdrop(fp2);
 600 free3:
 601         fsetfd(p, NULL, fd1);
 602         fdrop(fp1);
 603 free2:
 604         (void)soclose(so2, 0);
 605 free1:
 606         (void)soclose(so1, 0);
 607         return (error);
 608 }
 609
 610 /*
 611  * socketpair(int domain, int type, int protocol, int *rsv)
 612  */
 613 int
 614 sys_socketpair(struct socketpair_args *uap)
 615 {
 616         int error, sockv[2];
 617
 618         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 619
 620         if (error == 0)
 621                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 622         return (error);
 623 }
 624
 625 int
 626 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 627              struct mbuf *control, int flags, size_t *res)
 628 {
 629         struct thread *td = curthread;
 630         struct lwp *lp = td->td_lwp;
 631         struct proc *p = td->td_proc;
 632         struct file *fp;
 633         size_t len;
 634         int error;
 635         struct socket *so;
 636 #ifdef KTRACE
 637         struct iovec *ktriov = NULL;
 638         struct uio ktruio;
 639 #endif
 640
 641         error = holdsock(p->p_fd, s, &fp);
 642         if (error)
 643                 return (error);
 644 #ifdef KTRACE
 645         if (KTRPOINT(td, KTR_GENIO)) {
 646                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 647
 648                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 649                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 650                 ktruio = *auio;
 651         }
 652 #endif
 653         len = auio->uio_resid;
 654         so = (struct socket *)fp->f_data;
 655         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 656                 if (fp->f_flag & FNONBLOCK)
 657                         flags |= MSG_FNONBLOCKING;
 658         }
 659         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 660         if (error) {
 661                 if (auio->uio_resid != len && (error == ERESTART ||
 662                     error == EINTR || error == EWOULDBLOCK))
 663                         error = 0;
 664                 if (error == EPIPE)
 665                         lwpsignal(p, lp, SIGPIPE);
 666         }
 667 #ifdef KTRACE
 668         if (ktriov != NULL) {
 669                 if (error == 0) {
 670                         ktruio.uio_iov = ktriov;
 671                         ktruio.uio_resid = len - auio->uio_resid;
 672                         ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
 673                 }
 674                 FREE(ktriov, M_TEMP);
 675         }
 676 #endif
 677         if (error == 0)
 678                 *res  = len - auio->uio_resid;
 679         fdrop(fp);
 680         return (error);
 681 }
 682
 683 /*
 684  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
 685  */
 686 int
 687 sys_sendto(struct sendto_args *uap)
 688 {
 689         struct thread *td = curthread;
 690         struct uio auio;
 691         struct iovec aiov;
 692         struct sockaddr *sa = NULL;
 693         int error;
 694
 695         if (uap->to) {
 696                 error = getsockaddr(&sa, uap->to, uap->tolen);
 697                 if (error)
 698                         return (error);
 699         }
 700         aiov.iov_base = uap->buf;
 701         aiov.iov_len = uap->len;
 702         auio.uio_iov = &aiov;
 703         auio.uio_iovcnt = 1;
 704         auio.uio_offset = 0;
 705         auio.uio_resid = uap->len;
 706         auio.uio_segflg = UIO_USERSPACE;
 707         auio.uio_rw = UIO_WRITE;
 708         auio.uio_td = td;
 709
 710         error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
 711                              &uap->sysmsg_szresult);
 712
 713         if (sa)
 714                 FREE(sa, M_SONAME);
 715         return (error);
 716 }
 717
 718 /*
 719  * sendmsg_args(int s, caddr_t msg, int flags)
 720  */
 721 int
 722 sys_sendmsg(struct sendmsg_args *uap)
 723 {
 724         struct thread *td = curthread;
 725         struct msghdr msg;
 726         struct uio auio;
 727         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 728         struct sockaddr *sa = NULL;
 729         struct mbuf *control = NULL;
 730         int error;
 731
 732         error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
 733         if (error)
 734                 return (error);
 735
 736         /*
 737          * Conditionally copyin msg.msg_name.
 738          */
 739         if (msg.msg_name) {
 740                 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 741                 if (error)
 742                         return (error);
 743         }
 744
 745         /*
 746          * Populate auio.
 747          */
 748         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 749                              &auio.uio_resid);
 750         if (error)
 751                 goto cleanup2;
 752         auio.uio_iov = iov;
 753         auio.uio_iovcnt = msg.msg_iovlen;
 754         auio.uio_offset = 0;
 755         auio.uio_segflg = UIO_USERSPACE;
 756         auio.uio_rw = UIO_WRITE;
 757         auio.uio_td = td;
 758
 759         /*
 760          * Conditionally copyin msg.msg_control.
 761          */
 762         if (msg.msg_control) {
 763                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 764                     msg.msg_controllen > MLEN) {
 765                         error = EINVAL;
 766                         goto cleanup;
 767                 }
 768                 control = m_get(MB_WAIT, MT_CONTROL);
 769                 if (control == NULL) {
 770                         error = ENOBUFS;
 771                         goto cleanup;
 772                 }
 773                 control->m_len = msg.msg_controllen;
 774                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 775                     msg.msg_controllen);
 776                 if (error) {
 777                         m_free(control);
 778                         goto cleanup;
 779                 }
 780         }
 781
 782         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 783                              &uap->sysmsg_szresult);
 784
 785 cleanup:
 786         iovec_free(&iov, aiov);
 787 cleanup2:
 788         if (sa)
 789                 FREE(sa, M_SONAME);
 790         return (error);
 791 }
 792
 793 /*
 794  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 795  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 796  * Don't forget to FREE() and m_free() these if they are returned.
 797  */
 798 int
 799 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 800              struct mbuf **control, int *flags, size_t *res)
 801 {
 802         struct thread *td = curthread;
 803         struct proc *p = td->td_proc;
 804         struct file *fp;
 805         size_t len;
 806         int error;
 807         int lflags;
 808         struct socket *so;
 809 #ifdef KTRACE
 810         struct iovec *ktriov = NULL;
 811         struct uio ktruio;
 812 #endif
 813
 814         error = holdsock(p->p_fd, s, &fp);
 815         if (error)
 816                 return (error);
 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(td->td_lwp, 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         fdrop(fp);
 859         return (error);
 860 }
 861
 862 /*
 863  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
 864  *                      caddr_t from, int *fromlenaddr)
 865  */
 866 int
 867 sys_recvfrom(struct recvfrom_args *uap)
 868 {
 869         struct thread *td = curthread;
 870         struct uio auio;
 871         struct iovec aiov;
 872         struct sockaddr *sa = NULL;
 873         int error, fromlen;
 874
 875         if (uap->from && uap->fromlenaddr) {
 876                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
 877                 if (error)
 878                         return (error);
 879                 if (fromlen < 0)
 880                         return (EINVAL);
 881         } else {
 882                 fromlen = 0;
 883         }
 884         aiov.iov_base = uap->buf;
 885         aiov.iov_len = uap->len;
 886         auio.uio_iov = &aiov;
 887         auio.uio_iovcnt = 1;
 888         auio.uio_offset = 0;
 889         auio.uio_resid = uap->len;
 890         auio.uio_segflg = UIO_USERSPACE;
 891         auio.uio_rw = UIO_READ;
 892         auio.uio_td = td;
 893
 894         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
 895                              &uap->flags, &uap->sysmsg_szresult);
 896
 897         if (error == 0 && uap->from) {
 898                 /* note: sa may still be NULL */
 899                 if (sa) {
 900                         fromlen = MIN(fromlen, sa->sa_len);
 901                         error = copyout(sa, uap->from, fromlen);
 902                 } else {
 903                         fromlen = 0;
 904                 }
 905                 if (error == 0) {
 906                         error = copyout(&fromlen, uap->fromlenaddr,
 907                                         sizeof(fromlen));
 908                 }
 909         }
 910         if (sa)
 911                 FREE(sa, M_SONAME);
 912
 913         return (error);
 914 }
 915
 916 /*
 917  * recvmsg_args(int s, struct msghdr *msg, int flags)
 918  */
 919 int
 920 sys_recvmsg(struct recvmsg_args *uap)
 921 {
 922         struct thread *td = curthread;
 923         struct msghdr msg;
 924         struct uio auio;
 925         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 926         struct mbuf *m, *control = NULL;
 927         struct sockaddr *sa = NULL;
 928         caddr_t ctlbuf;
 929         socklen_t *ufromlenp, *ucontrollenp;
 930         int error, fromlen, controllen, len, flags, *uflagsp;
 931
 932         /*
 933          * This copyin handles everything except the iovec.
 934          */
 935         error = copyin(uap->msg, &msg, sizeof(msg));
 936         if (error)
 937                 return (error);
 938
 939         if (msg.msg_name && msg.msg_namelen < 0)
 940                 return (EINVAL);
 941         if (msg.msg_control && msg.msg_controllen < 0)
 942                 return (EINVAL);
 943
 944         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 945             msg_namelen));
 946         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 947             msg_controllen));
 948         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
 949             msg_flags));
 950
 951         /*
 952          * Populate auio.
 953          */
 954         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 955                              &auio.uio_resid);
 956         if (error)
 957                 return (error);
 958         auio.uio_iov = iov;
 959         auio.uio_iovcnt = msg.msg_iovlen;
 960         auio.uio_offset = 0;
 961         auio.uio_segflg = UIO_USERSPACE;
 962         auio.uio_rw = UIO_READ;
 963         auio.uio_td = td;
 964
 965         flags = uap->flags;
 966
 967         error = kern_recvmsg(uap->s,
 968                              (msg.msg_name ? &sa : NULL), &auio,
 969                              (msg.msg_control ? &control : NULL), &flags,
 970                              &uap->sysmsg_szresult);
 971
 972         /*
 973          * Conditionally copyout the name and populate the namelen field.
 974          */
 975         if (error == 0 && msg.msg_name) {
 976                 /* note: sa may still be NULL */
 977                 if (sa != NULL) {
 978                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
 979                         error = copyout(sa, msg.msg_name, fromlen);
 980                 } else {
 981                         fromlen = 0;
 982                 }
 983                 if (error == 0)
 984                         error = copyout(&fromlen, ufromlenp,
 985                             sizeof(*ufromlenp));
 986         }
 987
 988         /*
 989          * Copyout msg.msg_control and msg.msg_controllen.
 990          */
 991         if (error == 0 && msg.msg_control) {
 992                 len = msg.msg_controllen;
 993                 m = control;
 994                 ctlbuf = (caddr_t)msg.msg_control;
 995
 996                 while(m && len > 0) {
 997                         unsigned int tocopy;
 998
 999                         if (len >= m->m_len) {
1000                                 tocopy = m->m_len;
1001                         } else {
1002                                 msg.msg_flags |= MSG_CTRUNC;
1003                                 tocopy = len;
1004                         }
1005
1006                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1007                         if (error)
1008                                 goto cleanup;
1009
1010                         ctlbuf += tocopy;
1011                         len -= tocopy;
1012                         m = m->m_next;
1013                 }
1014                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1015                 error = copyout(&controllen, ucontrollenp,
1016                     sizeof(*ucontrollenp));
1017         }
1018
1019         if (error == 0)
1020                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1021
1022 cleanup:
1023         if (sa)
1024                 FREE(sa, M_SONAME);
1025         iovec_free(&iov, aiov);
1026         if (control)
1027                 m_freem(control);
1028         return (error);
1029 }
1030
1031 /*
1032  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1033  * in kernel pointer instead of a userland pointer.  This allows us
1034  * to manipulate socket options in the emulation code.
1035  */
1036 int
1037 kern_setsockopt(int s, struct sockopt *sopt)
1038 {
1039         struct thread *td = curthread;
1040         struct proc *p = td->td_proc;
1041         struct file *fp;
1042         int error;
1043
1044         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1045                 return (EFAULT);
1046         if (sopt->sopt_valsize < 0)
1047                 return (EINVAL);
1048
1049         error = holdsock(p->p_fd, s, &fp);
1050         if (error)
1051                 return (error);
1052
1053         error = sosetopt((struct socket *)fp->f_data, sopt);
1054         fdrop(fp);
1055         return (error);
1056 }
1057
1058 /*
1059  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1060  */
1061 int
1062 sys_setsockopt(struct setsockopt_args *uap)
1063 {
1064         struct thread *td = curthread;
1065         struct sockopt sopt;
1066         int error;
1067
1068         sopt.sopt_level = uap->level;
1069         sopt.sopt_name = uap->name;
1070         sopt.sopt_valsize = uap->valsize;
1071         sopt.sopt_td = td;
1072         sopt.sopt_val = NULL;
1073
1074         if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1075                 return (EINVAL);
1076         if (uap->val) {
1077                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1078                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1079                 if (error)
1080                         goto out;
1081         }
1082
1083         error = kern_setsockopt(uap->s, &sopt);
1084 out:
1085         if (uap->val)
1086                 kfree(sopt.sopt_val, M_TEMP);
1087         return(error);
1088 }
1089
1090 /*
1091  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1092  * in kernel pointer instead of a userland pointer.  This allows us
1093  * to manipulate socket options in the emulation code.
1094  */
1095 int
1096 kern_getsockopt(int s, struct sockopt *sopt)
1097 {
1098         struct thread *td = curthread;
1099         struct proc *p = td->td_proc;
1100         struct file *fp;
1101         int error;
1102
1103         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1104                 return (EFAULT);
1105         if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE)
1106                 return (EINVAL);
1107
1108         error = holdsock(p->p_fd, s, &fp);
1109         if (error)
1110                 return (error);
1111
1112         error = sogetopt((struct socket *)fp->f_data, sopt);
1113         fdrop(fp);
1114         return (error);
1115 }
1116
1117 /*
1118  * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1119  */
1120 int
1121 sys_getsockopt(struct getsockopt_args *uap)
1122 {
1123         struct thread *td = curthread;
1124         struct  sockopt sopt;
1125         int     error, valsize;
1126
1127         if (uap->val) {
1128                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1129                 if (error)
1130                         return (error);
1131         } else {
1132                 valsize = 0;
1133         }
1134
1135         sopt.sopt_level = uap->level;
1136         sopt.sopt_name = uap->name;
1137         sopt.sopt_valsize = valsize;
1138         sopt.sopt_td = td;
1139         sopt.sopt_val = NULL;
1140
1141         if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1142                 return (EINVAL);
1143         if (uap->val) {
1144                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1145                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1146                 if (error)
1147                         goto out;
1148         }
1149
1150         error = kern_getsockopt(uap->s, &sopt);
1151         if (error)
1152                 goto out;
1153         valsize = sopt.sopt_valsize;
1154         error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1155         if (error)
1156                 goto out;
1157         if (uap->val)
1158                 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1159 out:
1160         if (uap->val)
1161                 kfree(sopt.sopt_val, M_TEMP);
1162         return (error);
1163 }
1164
1165 /*
1166  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1167  * This allows kern_getsockname() to return a pointer to an allocated struct
1168  * sockaddr which must be freed later with FREE().  The caller must
1169  * initialize *name to NULL.
1170  */
1171 int
1172 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1173 {
1174         struct thread *td = curthread;
1175         struct proc *p = td->td_proc;
1176         struct file *fp;
1177         struct socket *so;
1178         struct sockaddr *sa = NULL;
1179         int error;
1180
1181         error = holdsock(p->p_fd, s, &fp);
1182         if (error)
1183                 return (error);
1184         if (*namelen < 0) {
1185                 fdrop(fp);
1186                 return (EINVAL);
1187         }
1188         so = (struct socket *)fp->f_data;
1189         error = so_pru_sockaddr(so, &sa);
1190         if (error == 0) {
1191                 if (sa == NULL) {
1192                         *namelen = 0;
1193                 } else {
1194                         *namelen = MIN(*namelen, sa->sa_len);
1195                         *name = sa;
1196                 }
1197         }
1198
1199         fdrop(fp);
1200         return (error);
1201 }
1202
1203 /*
1204  * getsockname_args(int fdes, caddr_t asa, int *alen)
1205  *
1206  * Get socket name.
1207  */
1208 int
1209 sys_getsockname(struct getsockname_args *uap)
1210 {
1211         struct sockaddr *sa = NULL;
1212         int error, sa_len;
1213
1214         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1215         if (error)
1216                 return (error);
1217
1218         error = kern_getsockname(uap->fdes, &sa, &sa_len);
1219
1220         if (error == 0)
1221                 error = copyout(sa, uap->asa, sa_len);
1222         if (error == 0)
1223                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1224         if (sa)
1225                 FREE(sa, M_SONAME);
1226         return (error);
1227 }
1228
1229 /*
1230  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1231  * This allows kern_getpeername() to return a pointer to an allocated struct
1232  * sockaddr which must be freed later with FREE().  The caller must
1233  * initialize *name to NULL.
1234  */
1235 int
1236 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1237 {
1238         struct thread *td = curthread;
1239         struct proc *p = td->td_proc;
1240         struct file *fp;
1241         struct socket *so;
1242         struct sockaddr *sa = NULL;
1243         int error;
1244
1245         error = holdsock(p->p_fd, s, &fp);
1246         if (error)
1247                 return (error);
1248         if (*namelen < 0) {
1249                 fdrop(fp);
1250                 return (EINVAL);
1251         }
1252         so = (struct socket *)fp->f_data;
1253         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1254                 fdrop(fp);
1255                 return (ENOTCONN);
1256         }
1257         error = so_pru_peeraddr(so, &sa);
1258         if (error == 0) {
1259                 if (sa == NULL) {
1260                         *namelen = 0;
1261                 } else {
1262                         *namelen = MIN(*namelen, sa->sa_len);
1263                         *name = sa;
1264                 }
1265         }
1266
1267         fdrop(fp);
1268         return (error);
1269 }
1270
1271 /*
1272  * getpeername_args(int fdes, caddr_t asa, int *alen)
1273  *
1274  * Get name of peer for connected socket.
1275  */
1276 int
1277 sys_getpeername(struct getpeername_args *uap)
1278 {
1279         struct sockaddr *sa = NULL;
1280         int error, sa_len;
1281
1282         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1283         if (error)
1284                 return (error);
1285
1286         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1287
1288         if (error == 0)
1289                 error = copyout(sa, uap->asa, sa_len);
1290         if (error == 0)
1291                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1292         if (sa)
1293                 FREE(sa, M_SONAME);
1294         return (error);
1295 }
1296
1297 int
1298 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1299 {
1300         struct sockaddr *sa;
1301         int error;
1302
1303         *namp = NULL;
1304         if (len > SOCK_MAXADDRLEN)
1305                 return ENAMETOOLONG;
1306         if (len < offsetof(struct sockaddr, sa_data[0]))
1307                 return EDOM;
1308         MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1309         error = copyin(uaddr, sa, len);
1310         if (error) {
1311                 FREE(sa, M_SONAME);
1312         } else {
1313 #if BYTE_ORDER != BIG_ENDIAN
1314                 /*
1315                  * The bind(), connect(), and sendto() syscalls were not
1316                  * versioned for COMPAT_43.  Thus, this check must stay.
1317                  */
1318                 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1319                         sa->sa_family = sa->sa_len;
1320 #endif
1321                 sa->sa_len = len;
1322                 *namp = sa;
1323         }
1324         return error;
1325 }
1326
1327 /*
1328  * Detach a mapped page and release resources back to the system.
1329  * We must release our wiring and if the object is ripped out
1330  * from under the vm_page we become responsible for freeing the
1331  * page.  These routines must be MPSAFE.
1332  *
1333  * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1334  *
1335  * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1336  */
1337 static void
1338 sf_buf_mref(void *arg)
1339 {
1340         struct sfbuf_mref *sfm = arg;
1341
1342         /*
1343          * We must already hold a ref so there is no race to 0, just
1344          * atomically increment the count.
1345          */
1346         atomic_add_int(&sfm->mref_count, 1);
1347 }
1348
1349 static void
1350 sf_buf_mfree(void *arg)
1351 {
1352         struct sfbuf_mref *sfm = arg;
1353         vm_page_t m;
1354
1355         KKASSERT(sfm->mref_count > 0);
1356         if (atomic_fetchadd_int(&sfm->mref_count, -1) == 1) {
1357                 /*
1358                  * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1359                  */
1360                 get_mplock();
1361                 crit_enter();
1362                 m = sf_buf_page(sfm->sf);
1363                 sf_buf_free(sfm->sf);
1364                 vm_page_unwire(m, 0);
1365                 if (m->wire_count == 0 && m->object == NULL)
1366                         vm_page_try_to_free(m);
1367                 crit_exit();
1368                 rel_mplock();
1369                 kfree(sfm, M_SENDFILE);
1370         }
1371 }
1372
1373 /*
1374  * sendfile(2).
1375  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1376  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1377  *
1378  * Send a file specified by 'fd' and starting at 'offset' to a socket
1379  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1380  * nbytes == 0. Optionally add a header and/or trailer to the socket
1381  * output. If specified, write the total number of bytes sent into *sbytes.
1382  *
1383  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1384  * the headers to count against the remaining bytes to be sent from
1385  * the file descriptor.  We may wish to implement a compatibility syscall
1386  * in the future.
1387  */
1388 int
1389 sys_sendfile(struct sendfile_args *uap)
1390 {
1391         struct thread *td = curthread;
1392         struct proc *p = td->td_proc;
1393         struct file *fp;
1394         struct vnode *vp = NULL;
1395         struct sf_hdtr hdtr;
1396         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1397         struct uio auio;
1398         struct mbuf *mheader = NULL;
1399         size_t hbytes = 0;
1400         size_t tbytes;
1401         off_t hdtr_size = 0;
1402         off_t sbytes;
1403         int error;
1404
1405         KKASSERT(p);
1406
1407         /*
1408          * Do argument checking. Must be a regular file in, stream
1409          * type and connected socket out, positive offset.
1410          */
1411         fp = holdfp(p->p_fd, uap->fd, FREAD);
1412         if (fp == NULL) {
1413                 return (EBADF);
1414         }
1415         if (fp->f_type != DTYPE_VNODE) {
1416                 fdrop(fp);
1417                 return (EINVAL);
1418         }
1419         vp = (struct vnode *)fp->f_data;
1420         vref(vp);
1421         fdrop(fp);
1422
1423         /*
1424          * If specified, get the pointer to the sf_hdtr struct for
1425          * any headers/trailers.
1426          */
1427         if (uap->hdtr) {
1428                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1429                 if (error)
1430                         goto done;
1431                 /*
1432                  * Send any headers.
1433                  */
1434                 if (hdtr.headers) {
1435                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1436                                              hdtr.hdr_cnt, &hbytes);
1437                         if (error)
1438                                 goto done;
1439                         auio.uio_iov = iov;
1440                         auio.uio_iovcnt = hdtr.hdr_cnt;
1441                         auio.uio_offset = 0;
1442                         auio.uio_segflg = UIO_USERSPACE;
1443                         auio.uio_rw = UIO_WRITE;
1444                         auio.uio_td = td;
1445                         auio.uio_resid = hbytes;
1446
1447                         mheader = m_uiomove(&auio);
1448
1449                         iovec_free(&iov, aiov);
1450                         if (mheader == NULL)
1451                                 goto done;
1452                 }
1453         }
1454
1455         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1456             &sbytes, uap->flags);
1457         if (error)
1458                 goto done;
1459
1460         /*
1461          * Send trailers. Wimp out and use writev(2).
1462          */
1463         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1464                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1465                                      hdtr.trl_cnt, &auio.uio_resid);
1466                 if (error)
1467                         goto done;
1468                 auio.uio_iov = iov;
1469                 auio.uio_iovcnt = hdtr.trl_cnt;
1470                 auio.uio_offset = 0;
1471                 auio.uio_segflg = UIO_USERSPACE;
1472                 auio.uio_rw = UIO_WRITE;
1473                 auio.uio_td = td;
1474
1475                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1476
1477                 iovec_free(&iov, aiov);
1478                 if (error)
1479                         goto done;
1480                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1481         }
1482
1483 done:
1484         if (uap->sbytes != NULL) {
1485                 sbytes += hdtr_size;
1486                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1487         }
1488         if (vp)
1489                 vrele(vp);
1490         return (error);
1491 }
1492
1493 int
1494 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1495     struct mbuf *mheader, off_t *sbytes, int flags)
1496 {
1497         struct thread *td = curthread;
1498         struct proc *p = td->td_proc;
1499         struct vm_object *obj;
1500         struct socket *so;
1501         struct file *fp;
1502         struct mbuf *m;
1503         struct sf_buf *sf;
1504         struct sfbuf_mref *sfm;
1505         struct vm_page *pg;
1506         off_t off, xfsize;
1507         off_t hbytes = 0;
1508         int error = 0;
1509
1510         if (vp->v_type != VREG) {
1511                 error = EINVAL;
1512                 goto done0;
1513         }
1514         if ((obj = vp->v_object) == NULL) {
1515                 error = EINVAL;
1516                 goto done0;
1517         }
1518         error = holdsock(p->p_fd, sfd, &fp);
1519         if (error)
1520                 goto done0;
1521         so = (struct socket *)fp->f_data;
1522         if (so->so_type != SOCK_STREAM) {
1523                 error = EINVAL;
1524                 goto done;
1525         }
1526         if ((so->so_state & SS_ISCONNECTED) == 0) {
1527                 error = ENOTCONN;
1528                 goto done;
1529         }
1530         if (offset < 0) {
1531                 error = EINVAL;
1532                 goto done;
1533         }
1534
1535         *sbytes = 0;
1536         /*
1537          * Protect against multiple writers to the socket.
1538          */
1539         ssb_lock(&so->so_snd, M_WAITOK);
1540
1541         /*
1542          * Loop through the pages in the file, starting with the requested
1543          * offset. Get a file page (do I/O if necessary), map the file page
1544          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1545          * it on the socket.
1546          */
1547         for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1548                 vm_pindex_t pindex;
1549                 vm_offset_t pgoff;
1550
1551                 pindex = OFF_TO_IDX(off);
1552 retry_lookup:
1553                 /*
1554                  * Calculate the amount to transfer. Not to exceed a page,
1555                  * the EOF, or the passed in nbytes.
1556                  */
1557                 xfsize = vp->v_filesize - off;
1558                 if (xfsize > PAGE_SIZE)
1559                         xfsize = PAGE_SIZE;
1560                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1561                 if (PAGE_SIZE - pgoff < xfsize)
1562                         xfsize = PAGE_SIZE - pgoff;
1563                 if (nbytes && xfsize > (nbytes - *sbytes))
1564                         xfsize = nbytes - *sbytes;
1565                 if (xfsize <= 0)
1566                         break;
1567                 /*
1568                  * Optimize the non-blocking case by looking at the socket space
1569                  * before going to the extra work of constituting the sf_buf.
1570                  */
1571                 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1572                         if (so->so_state & SS_CANTSENDMORE)
1573                                 error = EPIPE;
1574                         else
1575                                 error = EAGAIN;
1576                         ssb_unlock(&so->so_snd);
1577                         goto done;
1578                 }
1579                 /*
1580                  * Attempt to look up the page.
1581                  *
1582                  *      Allocate if not found, wait and loop if busy, then
1583                  *      wire the page.  critical section protection is
1584                  *      required to maintain the object association (an
1585                  *      interrupt can free the page) through to the
1586                  *      vm_page_wire() call.
1587                  */
1588                 crit_enter();
1589                 pg = vm_page_lookup(obj, pindex);
1590                 if (pg == NULL) {
1591                         pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1592                         if (pg == NULL) {
1593                                 vm_wait(0);
1594                                 crit_exit();
1595                                 goto retry_lookup;
1596                         }
1597                         vm_page_wakeup(pg);
1598                 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1599                         crit_exit();
1600                         goto retry_lookup;
1601                 }
1602                 vm_page_wire(pg);
1603                 crit_exit();
1604
1605                 /*
1606                  * If page is not valid for what we need, initiate I/O
1607                  */
1608
1609                 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1610                         struct uio auio;
1611                         struct iovec aiov;
1612                         int bsize;
1613
1614                         /*
1615                          * Ensure that our page is still around when the I/O
1616                          * completes.
1617                          */
1618                         vm_page_io_start(pg);
1619
1620                         /*
1621                          * Get the page from backing store.
1622                          */
1623                         bsize = vp->v_mount->mnt_stat.f_iosize;
1624                         auio.uio_iov = &aiov;
1625                         auio.uio_iovcnt = 1;
1626                         aiov.iov_base = 0;
1627                         aiov.iov_len = MAXBSIZE;
1628                         auio.uio_resid = MAXBSIZE;
1629                         auio.uio_offset = trunc_page(off);
1630                         auio.uio_segflg = UIO_NOCOPY;
1631                         auio.uio_rw = UIO_READ;
1632                         auio.uio_td = td;
1633                         vn_lock(vp, LK_SHARED | LK_RETRY);
1634                         error = VOP_READ(vp, &auio,
1635                                     IO_VMIO | ((MAXBSIZE / bsize) << 16),
1636                                     p->p_ucred);
1637                         vn_unlock(vp);
1638                         vm_page_flag_clear(pg, PG_ZERO);
1639                         vm_page_io_finish(pg);
1640                         if (error) {
1641                                 crit_enter();
1642                                 vm_page_unwire(pg, 0);
1643                                 vm_page_try_to_free(pg);
1644                                 crit_exit();
1645                                 ssb_unlock(&so->so_snd);
1646                                 goto done;
1647                         }
1648                 }
1649
1650
1651                 /*
1652                  * Get a sendfile buf. We usually wait as long as necessary,
1653                  * but this wait can be interrupted.
1654                  */
1655                 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1656                         crit_enter();
1657                         vm_page_unwire(pg, 0);
1658                         vm_page_try_to_free(pg);
1659                         crit_exit();
1660                         ssb_unlock(&so->so_snd);
1661                         error = EINTR;
1662                         goto done;
1663                 }
1664
1665                 /*
1666                  * Get an mbuf header and set it up as having external storage.
1667                  */
1668                 MGETHDR(m, MB_WAIT, MT_DATA);
1669                 if (m == NULL) {
1670                         error = ENOBUFS;
1671                         sf_buf_free(sf);
1672                         ssb_unlock(&so->so_snd);
1673                         goto done;
1674                 }
1675
1676                 /*
1677                  * sfm is a temporary hack, use a per-cpu cache for this.
1678                  */
1679                 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1680                 sfm->sf = sf;
1681                 sfm->mref_count = 1;
1682
1683                 m->m_ext.ext_free = sf_buf_mfree;
1684                 m->m_ext.ext_ref = sf_buf_mref;
1685                 m->m_ext.ext_arg = sfm;
1686                 m->m_ext.ext_buf = (void *)sf->kva;
1687                 m->m_ext.ext_size = PAGE_SIZE;
1688                 m->m_data = (char *) sf->kva + pgoff;
1689                 m->m_flags |= M_EXT;
1690                 m->m_pkthdr.len = m->m_len = xfsize;
1691                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1692
1693                 if (mheader != NULL) {
1694                         hbytes = mheader->m_pkthdr.len;
1695                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1696                         m_cat(mheader, m);
1697                         m = mheader;
1698                         mheader = NULL;
1699                 } else
1700                         hbytes = 0;
1701
1702                 /*
1703                  * Add the buffer to the socket buffer chain.
1704                  */
1705                 crit_enter();
1706 retry_space:
1707                 /*
1708                  * Make sure that the socket is still able to take more data.
1709                  * CANTSENDMORE being true usually means that the connection
1710                  * was closed. so_error is true when an error was sensed after
1711                  * a previous send.
1712                  * The state is checked after the page mapping and buffer
1713                  * allocation above since those operations may block and make
1714                  * any socket checks stale. From this point forward, nothing
1715                  * blocks before the pru_send (or more accurately, any blocking
1716                  * results in a loop back to here to re-check).
1717                  */
1718                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1719                         if (so->so_state & SS_CANTSENDMORE) {
1720                                 error = EPIPE;
1721                         } else {
1722                                 error = so->so_error;
1723                                 so->so_error = 0;
1724                         }
1725                         m_freem(m);
1726                         ssb_unlock(&so->so_snd);
1727                         crit_exit();
1728                         goto done;
1729                 }
1730                 /*
1731                  * Wait for socket space to become available. We do this just
1732                  * after checking the connection state above in order to avoid
1733                  * a race condition with ssb_wait().
1734                  */
1735                 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1736                         if (fp->f_flag & FNONBLOCK) {
1737                                 m_freem(m);
1738                                 ssb_unlock(&so->so_snd);
1739                                 crit_exit();
1740                                 error = EAGAIN;
1741                                 goto done;
1742                         }
1743                         error = ssb_wait(&so->so_snd);
1744                         /*
1745                          * An error from ssb_wait usually indicates that we've
1746                          * been interrupted by a signal. If we've sent anything
1747                          * then return bytes sent, otherwise return the error.
1748                          */
1749                         if (error) {
1750                                 m_freem(m);
1751                                 ssb_unlock(&so->so_snd);
1752                                 crit_exit();
1753                                 goto done;
1754                         }
1755                         goto retry_space;
1756                 }
1757                 error = so_pru_send(so, 0, m, NULL, NULL, td);
1758                 crit_exit();
1759                 if (error) {
1760                         ssb_unlock(&so->so_snd);
1761                         goto done;
1762                 }
1763         }
1764         if (mheader != NULL) {
1765                 *sbytes += mheader->m_pkthdr.len;
1766                 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1767                 mheader = NULL;
1768         }
1769         ssb_unlock(&so->so_snd);
1770
1771 done:
1772         fdrop(fp);
1773 done0:
1774         if (mheader != NULL)
1775                 m_freem(mheader);
1776         return (error);
1777 }
1778
1779 int
1780 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1781 {
1782 #ifdef SCTP
1783         struct thread *td = curthread;
1784         struct proc *p = td->td_proc;
1785         struct file *lfp = NULL;
1786         struct file *nfp = NULL;
1787         int error;
1788         struct socket *head, *so;
1789         caddr_t assoc_id;
1790         int fd;
1791         short fflag;            /* type must match fp->f_flag */
1792
1793         assoc_id = uap->name;
1794         error = holdsock(p->p_fd, uap->sd, &lfp);
1795         if (error) {
1796                 return (error);
1797         }
1798         crit_enter();
1799         head = (struct socket *)lfp->f_data;
1800         error = sctp_can_peel_off(head, assoc_id);
1801         if (error) {
1802                 crit_exit();
1803                 goto done;
1804         }
1805         /*
1806          * At this point we know we do have a assoc to pull
1807          * we proceed to get the fd setup. This may block
1808          * but that is ok.
1809          */
1810
1811         fflag = lfp->f_flag;
1812         error = falloc(p, &nfp, &fd);
1813         if (error) {
1814                 /*
1815                  * Probably ran out of file descriptors. Put the
1816                  * unaccepted connection back onto the queue and
1817                  * do another wakeup so some other process might
1818                  * have a chance at it.
1819                  */
1820                 crit_exit();
1821                 goto done;
1822         }
1823         uap->sysmsg_iresult = fd;
1824
1825         so = sctp_get_peeloff(head, assoc_id, &error);
1826         if (so == NULL) {
1827                 /*
1828                  * Either someone else peeled it off OR
1829                  * we can't get a socket.
1830                  */
1831                 goto noconnection;
1832         }
1833         so->so_state &= ~SS_COMP;
1834         so->so_state &= ~SS_NOFDREF;
1835         so->so_head = NULL;
1836         if (head->so_sigio != NULL)
1837                 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1838
1839         nfp->f_type = DTYPE_SOCKET;
1840         nfp->f_flag = fflag;
1841         nfp->f_ops = &socketops;
1842         nfp->f_data = so;
1843
1844 noconnection:
1845         /*
1846          * Assign the file pointer to the reserved descriptor, or clear
1847          * the reserved descriptor if an error occured.
1848          */
1849         if (error)
1850                 fsetfd(p, NULL, fd);
1851         else
1852                 fsetfd(p, nfp, fd);
1853         crit_exit();
1854         /*
1855          * Release explicitly held references before returning.
1856          */
1857 done:
1858         if (nfp != NULL)
1859                 fdrop(nfp);
1860         fdrop(lfp);
1861         return (error);
1862 #else /* SCTP */
1863         return(EOPNOTSUPP);
1864 #endif /* SCTP */
1865 }