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_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, 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         (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, interrupted = 0;
 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 (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, NULL, PRU_PRED);
 485                 netmsg_init_abortable(&msg.nm_netmsg,
 486                                       &curthread->td_msgport,
 487                                       0,
 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, PCATCH);
 494                 if (error == EINTR || error == ERESTART)
 495                         interrupted = 1;
 496         }
 497         if (error == 0) {
 498                 error = so->so_error;
 499                 so->so_error = 0;
 500         }
 501 bad:
 502         if (!interrupted)
 503                 so->so_state &= ~SS_ISCONNECTING;
 504         if (error == ERESTART)
 505                 error = EINTR;
 506 done:
 507         fdrop(fp);
 508         return (error);
 509 }
 510
 511 /*
 512  * connect_args(int s, caddr_t name, int namelen)
 513  */
 514 int
 515 sys_connect(struct connect_args *uap)
 516 {
 517         struct sockaddr *sa;
 518         int error;
 519
 520         error = getsockaddr(&sa, uap->name, uap->namelen);
 521         if (error)
 522                 return (error);
 523         error = kern_connect(uap->s, 0, sa);
 524         FREE(sa, M_SONAME);
 525
 526         return (error);
 527 }
 528
 529 /*
 530  * connect_args(int s, int fflags, caddr_t name, int namelen)
 531  */
 532 int
 533 sys_extconnect(struct extconnect_args *uap)
 534 {
 535         struct sockaddr *sa;
 536         int error;
 537         int fflags = uap->flags & O_FMASK;
 538
 539         error = getsockaddr(&sa, uap->name, uap->namelen);
 540         if (error)
 541                 return (error);
 542         error = kern_connect(uap->s, fflags, sa);
 543         FREE(sa, M_SONAME);
 544
 545         return (error);
 546 }
 547
 548 int
 549 kern_socketpair(int domain, int type, int protocol, int *sv)
 550 {
 551         struct thread *td = curthread;
 552         struct proc *p = td->td_proc;
 553         struct file *fp1, *fp2;
 554         struct socket *so1, *so2;
 555         int fd1, fd2, error;
 556
 557         KKASSERT(p);
 558         error = socreate(domain, &so1, type, protocol, td);
 559         if (error)
 560                 return (error);
 561         error = socreate(domain, &so2, type, protocol, td);
 562         if (error)
 563                 goto free1;
 564         error = falloc(p, &fp1, &fd1);
 565         if (error)
 566                 goto free2;
 567         sv[0] = fd1;
 568         fp1->f_data = so1;
 569         error = falloc(p, &fp2, &fd2);
 570         if (error)
 571                 goto free3;
 572         fp2->f_data = so2;
 573         sv[1] = fd2;
 574         error = soconnect2(so1, so2);
 575         if (error)
 576                 goto free4;
 577         if (type == SOCK_DGRAM) {
 578                 /*
 579                  * Datagram socket connection is asymmetric.
 580                  */
 581                  error = soconnect2(so2, so1);
 582                  if (error)
 583                         goto free4;
 584         }
 585         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 586         fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
 587         fp1->f_ops = fp2->f_ops = &socketops;
 588         fsetfd(p, fp1, fd1);
 589         fsetfd(p, fp2, fd2);
 590         fdrop(fp1);
 591         fdrop(fp2);
 592         return (error);
 593 free4:
 594         fsetfd(p, NULL, fd2);
 595         fdrop(fp2);
 596 free3:
 597         fsetfd(p, NULL, fd1);
 598         fdrop(fp1);
 599 free2:
 600         (void)soclose(so2, 0);
 601 free1:
 602         (void)soclose(so1, 0);
 603         return (error);
 604 }
 605
 606 /*
 607  * socketpair(int domain, int type, int protocol, int *rsv)
 608  */
 609 int
 610 sys_socketpair(struct socketpair_args *uap)
 611 {
 612         int error, sockv[2];
 613
 614         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 615
 616         if (error == 0)
 617                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 618         return (error);
 619 }
 620
 621 int
 622 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 623     struct mbuf *control, int flags, int *res)
 624 {
 625         struct thread *td = curthread;
 626         struct lwp *lp = td->td_lwp;
 627         struct proc *p = td->td_proc;
 628         struct file *fp;
 629         int len, error;
 630         struct socket *so;
 631 #ifdef KTRACE
 632         struct iovec *ktriov = NULL;
 633         struct uio ktruio;
 634 #endif
 635
 636         error = holdsock(p->p_fd, s, &fp);
 637         if (error)
 638                 return (error);
 639         if (auio->uio_resid < 0) {
 640                 error = EINVAL;
 641                 goto done;
 642         }
 643 #ifdef KTRACE
 644         if (KTRPOINT(td, KTR_GENIO)) {
 645                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 646
 647                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 648                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 649                 ktruio = *auio;
 650         }
 651 #endif
 652         len = auio->uio_resid;
 653         so = (struct socket *)fp->f_data;
 654         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 655                 if (fp->f_flag & FNONBLOCK)
 656                         flags |= MSG_FNONBLOCKING;
 657         }
 658         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 659         if (error) {
 660                 if (auio->uio_resid != len && (error == ERESTART ||
 661                     error == EINTR || error == EWOULDBLOCK))
 662                         error = 0;
 663                 if (error == EPIPE)
 664                         lwpsignal(p, lp, SIGPIPE);
 665         }
 666 #ifdef KTRACE
 667         if (ktriov != NULL) {
 668                 if (error == 0) {
 669                         ktruio.uio_iov = ktriov;
 670                         ktruio.uio_resid = len - auio->uio_resid;
 671                         ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
 672                 }
 673                 FREE(ktriov, M_TEMP);
 674         }
 675 #endif
 676         if (error == 0)
 677                 *res  = len - auio->uio_resid;
 678 done:
 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_result);
 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_result);
 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, int *res)
 801 {
 802         struct thread *td = curthread;
 803         struct proc *p = td->td_proc;
 804         struct file *fp;
 805         int len, error;
 806         int lflags;
 807         struct socket *so;
 808 #ifdef KTRACE
 809         struct iovec *ktriov = NULL;
 810         struct uio ktruio;
 811 #endif
 812
 813         error = holdsock(p->p_fd, s, &fp);
 814         if (error)
 815                 return (error);
 816         if (auio->uio_resid < 0) {
 817                 error = EINVAL;
 818                 goto done;
 819         }
 820 #ifdef KTRACE
 821         if (KTRPOINT(td, KTR_GENIO)) {
 822                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 823
 824                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 825                 bcopy(auio->uio_iov, ktriov, iovlen);
 826                 ktruio = *auio;
 827         }
 828 #endif
 829         len = auio->uio_resid;
 830         so = (struct socket *)fp->f_data;
 831
 832         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 833                 if (fp->f_flag & FNONBLOCK) {
 834                         if (flags) {
 835                                 *flags |= MSG_FNONBLOCKING;
 836                         } else {
 837                                 lflags = MSG_FNONBLOCKING;
 838                                 flags = &lflags;
 839                         }
 840                 }
 841         }
 842
 843         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 844         if (error) {
 845                 if (auio->uio_resid != len && (error == ERESTART ||
 846                     error == EINTR || error == EWOULDBLOCK))
 847                         error = 0;
 848         }
 849 #ifdef KTRACE
 850         if (ktriov != NULL) {
 851                 if (error == 0) {
 852                         ktruio.uio_iov = ktriov;
 853                         ktruio.uio_resid = len - auio->uio_resid;
 854                         ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
 855                 }
 856                 FREE(ktriov, M_TEMP);
 857         }
 858 #endif
 859         if (error == 0)
 860                 *res = len - auio->uio_resid;
 861 done:
 862         fdrop(fp);
 863         return (error);
 864 }
 865
 866 /*
 867  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
 868  *                      caddr_t from, int *fromlenaddr)
 869  */
 870 int
 871 sys_recvfrom(struct recvfrom_args *uap)
 872 {
 873         struct thread *td = curthread;
 874         struct uio auio;
 875         struct iovec aiov;
 876         struct sockaddr *sa = NULL;
 877         int error, fromlen;
 878
 879         if (uap->from && uap->fromlenaddr) {
 880                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
 881                 if (error)
 882                         return (error);
 883                 if (fromlen < 0)
 884                         return (EINVAL);
 885         } else {
 886                 fromlen = 0;
 887         }
 888         aiov.iov_base = uap->buf;
 889         aiov.iov_len = uap->len;
 890         auio.uio_iov = &aiov;
 891         auio.uio_iovcnt = 1;
 892         auio.uio_offset = 0;
 893         auio.uio_resid = uap->len;
 894         auio.uio_segflg = UIO_USERSPACE;
 895         auio.uio_rw = UIO_READ;
 896         auio.uio_td = td;
 897
 898         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
 899             &uap->flags, &uap->sysmsg_result);
 900
 901         if (error == 0 && uap->from) {
 902                 /* note: sa may still be NULL */
 903                 if (sa) {
 904                         fromlen = MIN(fromlen, sa->sa_len);
 905                         error = copyout(sa, uap->from, fromlen);
 906                 } else {
 907                         fromlen = 0;
 908                 }
 909                 if (error == 0) {
 910                         error = copyout(&fromlen, uap->fromlenaddr,
 911                                         sizeof(fromlen));
 912                 }
 913         }
 914         if (sa)
 915                 FREE(sa, M_SONAME);
 916
 917         return (error);
 918 }
 919
 920 /*
 921  * recvmsg_args(int s, struct msghdr *msg, int flags)
 922  */
 923 int
 924 sys_recvmsg(struct recvmsg_args *uap)
 925 {
 926         struct thread *td = curthread;
 927         struct msghdr msg;
 928         struct uio auio;
 929         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 930         struct mbuf *m, *control = NULL;
 931         struct sockaddr *sa = NULL;
 932         caddr_t ctlbuf;
 933         socklen_t *ufromlenp, *ucontrollenp;
 934         int error, fromlen, controllen, len, flags, *uflagsp;
 935
 936         /*
 937          * This copyin handles everything except the iovec.
 938          */
 939         error = copyin(uap->msg, &msg, sizeof(msg));
 940         if (error)
 941                 return (error);
 942
 943         if (msg.msg_name && msg.msg_namelen < 0)
 944                 return (EINVAL);
 945         if (msg.msg_control && msg.msg_controllen < 0)
 946                 return (EINVAL);
 947
 948         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 949             msg_namelen));
 950         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
 951             msg_controllen));
 952         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
 953             msg_flags));
 954
 955         /*
 956          * Populate auio.
 957          */
 958         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 959                              &auio.uio_resid);
 960         if (error)
 961                 return (error);
 962         auio.uio_iov = iov;
 963         auio.uio_iovcnt = msg.msg_iovlen;
 964         auio.uio_offset = 0;
 965         auio.uio_segflg = UIO_USERSPACE;
 966         auio.uio_rw = UIO_READ;
 967         auio.uio_td = td;
 968
 969         flags = uap->flags;
 970
 971         error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
 972             msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
 973
 974         /*
 975          * Conditionally copyout the name and populate the namelen field.
 976          */
 977         if (error == 0 && msg.msg_name) {
 978                 /* note: sa may still be NULL */
 979                 if (sa != NULL) {
 980                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
 981                         error = copyout(sa, msg.msg_name, fromlen);
 982                 } else {
 983                         fromlen = 0;
 984                 }
 985                 if (error == 0)
 986                         error = copyout(&fromlen, ufromlenp,
 987                             sizeof(*ufromlenp));
 988         }
 989
 990         /*
 991          * Copyout msg.msg_control and msg.msg_controllen.
 992          */
 993         if (error == 0 && msg.msg_control) {
 994                 len = msg.msg_controllen;
 995                 m = control;
 996                 ctlbuf = (caddr_t)msg.msg_control;
 997
 998                 while(m && len > 0) {
 999                         unsigned int tocopy;
1000
1001                         if (len >= m->m_len) {
1002                                 tocopy = m->m_len;
1003                         } else {
1004                                 msg.msg_flags |= MSG_CTRUNC;
1005                                 tocopy = len;
1006                         }
1007
1008                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1009                         if (error)
1010                                 goto cleanup;
1011
1012                         ctlbuf += tocopy;
1013                         len -= tocopy;
1014                         m = m->m_next;
1015                 }
1016                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1017                 error = copyout(&controllen, ucontrollenp,
1018                     sizeof(*ucontrollenp));
1019         }
1020
1021         if (error == 0)
1022                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1023
1024 cleanup:
1025         if (sa)
1026                 FREE(sa, M_SONAME);
1027         iovec_free(&iov, aiov);
1028         if (control)
1029                 m_freem(control);
1030         return (error);
1031 }
1032
1033 /*
1034  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1035  * in kernel pointer instead of a userland pointer.  This allows us
1036  * to manipulate socket options in the emulation code.
1037  */
1038 int
1039 kern_setsockopt(int s, struct sockopt *sopt)
1040 {
1041         struct thread *td = curthread;
1042         struct proc *p = td->td_proc;
1043         struct file *fp;
1044         int error;
1045
1046         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1047                 return (EFAULT);
1048         if (sopt->sopt_valsize < 0)
1049                 return (EINVAL);
1050
1051         error = holdsock(p->p_fd, s, &fp);
1052         if (error)
1053                 return (error);
1054
1055         error = sosetopt((struct socket *)fp->f_data, sopt);
1056         fdrop(fp);
1057         return (error);
1058 }
1059
1060 /*
1061  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1062  */
1063 int
1064 sys_setsockopt(struct setsockopt_args *uap)
1065 {
1066         struct thread *td = curthread;
1067         struct sockopt sopt;
1068         int error;
1069
1070         sopt.sopt_level = uap->level;
1071         sopt.sopt_name = uap->name;
1072         sopt.sopt_valsize = uap->valsize;
1073         sopt.sopt_td = td;
1074         sopt.sopt_val = NULL;
1075
1076         if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1077                 return (EINVAL);
1078         if (uap->val) {
1079                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1080                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1081                 if (error)
1082                         goto out;
1083         }
1084
1085         error = kern_setsockopt(uap->s, &sopt);
1086 out:
1087         if (uap->val)
1088                 kfree(sopt.sopt_val, M_TEMP);
1089         return(error);
1090 }
1091
1092 /*
1093  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1094  * in kernel pointer instead of a userland pointer.  This allows us
1095  * to manipulate socket options in the emulation code.
1096  */
1097 int
1098 kern_getsockopt(int s, struct sockopt *sopt)
1099 {
1100         struct thread *td = curthread;
1101         struct proc *p = td->td_proc;
1102         struct file *fp;
1103         int error;
1104
1105         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1106                 return (EFAULT);
1107         if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE)
1108                 return (EINVAL);
1109
1110         error = holdsock(p->p_fd, s, &fp);
1111         if (error)
1112                 return (error);
1113
1114         error = sogetopt((struct socket *)fp->f_data, sopt);
1115         fdrop(fp);
1116         return (error);
1117 }
1118
1119 /*
1120  * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1121  */
1122 int
1123 sys_getsockopt(struct getsockopt_args *uap)
1124 {
1125         struct thread *td = curthread;
1126         struct  sockopt sopt;
1127         int     error, valsize;
1128
1129         if (uap->val) {
1130                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1131                 if (error)
1132                         return (error);
1133         } else {
1134                 valsize = 0;
1135         }
1136
1137         sopt.sopt_level = uap->level;
1138         sopt.sopt_name = uap->name;
1139         sopt.sopt_valsize = valsize;
1140         sopt.sopt_td = td;
1141         sopt.sopt_val = NULL;
1142
1143         if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1144                 return (EINVAL);
1145         if (uap->val) {
1146                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1147                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1148                 if (error)
1149                         goto out;
1150         }
1151
1152         error = kern_getsockopt(uap->s, &sopt);
1153         if (error)
1154                 goto out;
1155         valsize = sopt.sopt_valsize;
1156         error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1157         if (error)
1158                 goto out;
1159         if (uap->val)
1160                 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1161 out:
1162         if (uap->val)
1163                 kfree(sopt.sopt_val, M_TEMP);
1164         return (error);
1165 }
1166
1167 /*
1168  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1169  * This allows kern_getsockname() to return a pointer to an allocated struct
1170  * sockaddr which must be freed later with FREE().  The caller must
1171  * initialize *name to NULL.
1172  */
1173 int
1174 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1175 {
1176         struct thread *td = curthread;
1177         struct proc *p = td->td_proc;
1178         struct file *fp;
1179         struct socket *so;
1180         struct sockaddr *sa = NULL;
1181         int error;
1182
1183         error = holdsock(p->p_fd, s, &fp);
1184         if (error)
1185                 return (error);
1186         if (*namelen < 0) {
1187                 fdrop(fp);
1188                 return (EINVAL);
1189         }
1190         so = (struct socket *)fp->f_data;
1191         error = so_pru_sockaddr(so, &sa);
1192         if (error == 0) {
1193                 if (sa == NULL) {
1194                         *namelen = 0;
1195                 } else {
1196                         *namelen = MIN(*namelen, sa->sa_len);
1197                         *name = sa;
1198                 }
1199         }
1200
1201         fdrop(fp);
1202         return (error);
1203 }
1204
1205 /*
1206  * getsockname_args(int fdes, caddr_t asa, int *alen)
1207  *
1208  * Get socket name.
1209  */
1210 int
1211 sys_getsockname(struct getsockname_args *uap)
1212 {
1213         struct sockaddr *sa = NULL;
1214         int error, sa_len;
1215
1216         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1217         if (error)
1218                 return (error);
1219
1220         error = kern_getsockname(uap->fdes, &sa, &sa_len);
1221
1222         if (error == 0)
1223                 error = copyout(sa, uap->asa, sa_len);
1224         if (error == 0)
1225                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1226         if (sa)
1227                 FREE(sa, M_SONAME);
1228         return (error);
1229 }
1230
1231 /*
1232  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1233  * This allows kern_getpeername() to return a pointer to an allocated struct
1234  * sockaddr which must be freed later with FREE().  The caller must
1235  * initialize *name to NULL.
1236  */
1237 int
1238 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1239 {
1240         struct thread *td = curthread;
1241         struct proc *p = td->td_proc;
1242         struct file *fp;
1243         struct socket *so;
1244         struct sockaddr *sa = NULL;
1245         int error;
1246
1247         error = holdsock(p->p_fd, s, &fp);
1248         if (error)
1249                 return (error);
1250         if (*namelen < 0) {
1251                 fdrop(fp);
1252                 return (EINVAL);
1253         }
1254         so = (struct socket *)fp->f_data;
1255         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1256                 fdrop(fp);
1257                 return (ENOTCONN);
1258         }
1259         error = so_pru_peeraddr(so, &sa);
1260         if (error == 0) {
1261                 if (sa == NULL) {
1262                         *namelen = 0;
1263                 } else {
1264                         *namelen = MIN(*namelen, sa->sa_len);
1265                         *name = sa;
1266                 }
1267         }
1268
1269         fdrop(fp);
1270         return (error);
1271 }
1272
1273 /*
1274  * getpeername_args(int fdes, caddr_t asa, int *alen)
1275  *
1276  * Get name of peer for connected socket.
1277  */
1278 int
1279 sys_getpeername(struct getpeername_args *uap)
1280 {
1281         struct sockaddr *sa = NULL;
1282         int error, sa_len;
1283
1284         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1285         if (error)
1286                 return (error);
1287
1288         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1289
1290         if (error == 0)
1291                 error = copyout(sa, uap->asa, sa_len);
1292         if (error == 0)
1293                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1294         if (sa)
1295                 FREE(sa, M_SONAME);
1296         return (error);
1297 }
1298
1299 int
1300 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1301 {
1302         struct sockaddr *sa;
1303         int error;
1304
1305         *namp = NULL;
1306         if (len > SOCK_MAXADDRLEN)
1307                 return ENAMETOOLONG;
1308         if (len < offsetof(struct sockaddr, sa_data[0]))
1309                 return EDOM;
1310         MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1311         error = copyin(uaddr, sa, len);
1312         if (error) {
1313                 FREE(sa, M_SONAME);
1314         } else {
1315 #if BYTE_ORDER != BIG_ENDIAN
1316                 /*
1317                  * The bind(), connect(), and sendto() syscalls were not
1318                  * versioned for COMPAT_43.  Thus, this check must stay.
1319                  */
1320                 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1321                         sa->sa_family = sa->sa_len;
1322 #endif
1323                 sa->sa_len = len;
1324                 *namp = sa;
1325         }
1326         return error;
1327 }
1328
1329 /*
1330  * Detach a mapped page and release resources back to the system.
1331  * We must release our wiring and if the object is ripped out
1332  * from under the vm_page we become responsible for freeing the
1333  * page.  These routines must be MPSAFE.
1334  *
1335  * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1336  *
1337  * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1338  */
1339 static void
1340 sf_buf_mref(void *arg)
1341 {
1342         struct sfbuf_mref *sfm = arg;
1343
1344         /*
1345          * We must already hold a ref so there is no race to 0, just
1346          * atomically increment the count.
1347          */
1348         atomic_add_int(&sfm->mref_count, 1);
1349 }
1350
1351 static void
1352 sf_buf_mfree(void *arg)
1353 {
1354         struct sfbuf_mref *sfm = arg;
1355         vm_page_t m;
1356
1357         KKASSERT(sfm->mref_count > 0);
1358         if (atomic_fetchadd_int(&sfm->mref_count, -1) == 1) {
1359                 /*
1360                  * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1361                  */
1362                 get_mplock();
1363                 crit_enter();
1364                 m = sf_buf_page(sfm->sf);
1365                 sf_buf_free(sfm->sf);
1366                 vm_page_unwire(m, 0);
1367                 if (m->wire_count == 0 && m->object == NULL)
1368                         vm_page_try_to_free(m);
1369                 crit_exit();
1370                 rel_mplock();
1371                 kfree(sfm, M_SENDFILE);
1372         }
1373 }
1374
1375 /*
1376  * sendfile(2).
1377  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1378  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1379  *
1380  * Send a file specified by 'fd' and starting at 'offset' to a socket
1381  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1382  * nbytes == 0. Optionally add a header and/or trailer to the socket
1383  * output. If specified, write the total number of bytes sent into *sbytes.
1384  *
1385  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1386  * the headers to count against the remaining bytes to be sent from
1387  * the file descriptor.  We may wish to implement a compatibility syscall
1388  * in the future.
1389  */
1390 int
1391 sys_sendfile(struct sendfile_args *uap)
1392 {
1393         struct thread *td = curthread;
1394         struct proc *p = td->td_proc;
1395         struct file *fp;
1396         struct vnode *vp = NULL;
1397         struct sf_hdtr hdtr;
1398         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1399         struct uio auio;
1400         struct mbuf *mheader = NULL;
1401         off_t hdtr_size = 0, sbytes;
1402         int error, hbytes = 0, tbytes;
1403
1404         KKASSERT(p);
1405
1406         /*
1407          * Do argument checking. Must be a regular file in, stream
1408          * type and connected socket out, positive offset.
1409          */
1410         fp = holdfp(p->p_fd, uap->fd, FREAD);
1411         if (fp == NULL) {
1412                 return (EBADF);
1413         }
1414         if (fp->f_type != DTYPE_VNODE) {
1415                 fdrop(fp);
1416                 return (EINVAL);
1417         }
1418         vp = (struct vnode *)fp->f_data;
1419         vref(vp);
1420         fdrop(fp);
1421
1422         /*
1423          * If specified, get the pointer to the sf_hdtr struct for
1424          * any headers/trailers.
1425          */
1426         if (uap->hdtr) {
1427                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1428                 if (error)
1429                         goto done;
1430                 /*
1431                  * Send any headers.
1432                  */
1433                 if (hdtr.headers) {
1434                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1435                                              hdtr.hdr_cnt, &hbytes);
1436                         if (error)
1437                                 goto done;
1438                         auio.uio_iov = iov;
1439                         auio.uio_iovcnt = hdtr.hdr_cnt;
1440                         auio.uio_offset = 0;
1441                         auio.uio_segflg = UIO_USERSPACE;
1442                         auio.uio_rw = UIO_WRITE;
1443                         auio.uio_td = td;
1444                         auio.uio_resid = hbytes;
1445
1446                         mheader = m_uiomove(&auio);
1447
1448                         iovec_free(&iov, aiov);
1449                         if (mheader == NULL)
1450                                 goto done;
1451                 }
1452         }
1453
1454         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1455             &sbytes, uap->flags);
1456         if (error)
1457                 goto done;
1458
1459         /*
1460          * Send trailers. Wimp out and use writev(2).
1461          */
1462         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1463                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1464                                      hdtr.trl_cnt, &auio.uio_resid);
1465                 if (error)
1466                         goto done;
1467                 auio.uio_iov = iov;
1468                 auio.uio_iovcnt = hdtr.trl_cnt;
1469                 auio.uio_offset = 0;
1470                 auio.uio_segflg = UIO_USERSPACE;
1471                 auio.uio_rw = UIO_WRITE;
1472                 auio.uio_td = td;
1473
1474                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1475
1476                 iovec_free(&iov, aiov);
1477                 if (error)
1478                         goto done;
1479                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1480         }
1481
1482 done:
1483         if (uap->sbytes != NULL) {
1484                 sbytes += hdtr_size;
1485                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1486         }
1487         if (vp)
1488                 vrele(vp);
1489         return (error);
1490 }
1491
1492 int
1493 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1494     struct mbuf *mheader, off_t *sbytes, int flags)
1495 {
1496         struct thread *td = curthread;
1497         struct proc *p = td->td_proc;
1498         struct vm_object *obj;
1499         struct socket *so;
1500         struct file *fp;
1501         struct mbuf *m;
1502         struct sf_buf *sf;
1503         struct sfbuf_mref *sfm;
1504         struct vm_page *pg;
1505         off_t off, xfsize;
1506         off_t hbytes = 0;
1507         int error = 0;
1508
1509         if (vp->v_type != VREG) {
1510                 error = EINVAL;
1511                 goto done0;
1512         }
1513         if ((obj = vp->v_object) == NULL) {
1514                 error = EINVAL;
1515                 goto done0;
1516         }
1517         error = holdsock(p->p_fd, sfd, &fp);
1518         if (error)
1519                 goto done0;
1520         so = (struct socket *)fp->f_data;
1521         if (so->so_type != SOCK_STREAM) {
1522                 error = EINVAL;
1523                 goto done;
1524         }
1525         if ((so->so_state & SS_ISCONNECTED) == 0) {
1526                 error = ENOTCONN;
1527                 goto done;
1528         }
1529         if (offset < 0) {
1530                 error = EINVAL;
1531                 goto done;
1532         }
1533
1534         *sbytes = 0;
1535         /*
1536          * Protect against multiple writers to the socket.
1537          */
1538         ssb_lock(&so->so_snd, M_WAITOK);
1539
1540         /*
1541          * Loop through the pages in the file, starting with the requested
1542          * offset. Get a file page (do I/O if necessary), map the file page
1543          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1544          * it on the socket.
1545          */
1546         for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1547                 vm_pindex_t pindex;
1548                 vm_offset_t pgoff;
1549
1550                 pindex = OFF_TO_IDX(off);
1551 retry_lookup:
1552                 /*
1553                  * Calculate the amount to transfer. Not to exceed a page,
1554                  * the EOF, or the passed in nbytes.
1555                  */
1556                 xfsize = vp->v_filesize - off;
1557                 if (xfsize > PAGE_SIZE)
1558                         xfsize = PAGE_SIZE;
1559                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1560                 if (PAGE_SIZE - pgoff < xfsize)
1561                         xfsize = PAGE_SIZE - pgoff;
1562                 if (nbytes && xfsize > (nbytes - *sbytes))
1563                         xfsize = nbytes - *sbytes;
1564                 if (xfsize <= 0)
1565                         break;
1566                 /*
1567                  * Optimize the non-blocking case by looking at the socket space
1568                  * before going to the extra work of constituting the sf_buf.
1569                  */
1570                 if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1571                         if (so->so_state & SS_CANTSENDMORE)
1572                                 error = EPIPE;
1573                         else
1574                                 error = EAGAIN;
1575                         ssb_unlock(&so->so_snd);
1576                         goto done;
1577                 }
1578                 /*
1579                  * Attempt to look up the page.
1580                  *
1581                  *      Allocate if not found, wait and loop if busy, then
1582                  *      wire the page.  critical section protection is
1583                  *      required to maintain the object association (an
1584                  *      interrupt can free the page) through to the
1585                  *      vm_page_wire() call.
1586                  */
1587                 crit_enter();
1588                 pg = vm_page_lookup(obj, pindex);
1589                 if (pg == NULL) {
1590                         pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1591                         if (pg == NULL) {
1592                                 vm_wait(0);
1593                                 crit_exit();
1594                                 goto retry_lookup;
1595                         }
1596                         vm_page_wakeup(pg);
1597                 } else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1598                         crit_exit();
1599                         goto retry_lookup;
1600                 }
1601                 vm_page_wire(pg);
1602                 crit_exit();
1603
1604                 /*
1605                  * If page is not valid for what we need, initiate I/O
1606                  */
1607
1608                 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1609                         struct uio auio;
1610                         struct iovec aiov;
1611                         int bsize;
1612
1613                         /*
1614                          * Ensure that our page is still around when the I/O
1615                          * completes.
1616                          */
1617                         vm_page_io_start(pg);
1618
1619                         /*
1620                          * Get the page from backing store.
1621                          */
1622                         bsize = vp->v_mount->mnt_stat.f_iosize;
1623                         auio.uio_iov = &aiov;
1624                         auio.uio_iovcnt = 1;
1625                         aiov.iov_base = 0;
1626                         aiov.iov_len = MAXBSIZE;
1627                         auio.uio_resid = MAXBSIZE;
1628                         auio.uio_offset = trunc_page(off);
1629                         auio.uio_segflg = UIO_NOCOPY;
1630                         auio.uio_rw = UIO_READ;
1631                         auio.uio_td = td;
1632                         vn_lock(vp, LK_SHARED | LK_RETRY);
1633                         error = VOP_READ(vp, &auio,
1634                                     IO_VMIO | ((MAXBSIZE / bsize) << 16),
1635                                     p->p_ucred);
1636                         vn_unlock(vp);
1637                         vm_page_flag_clear(pg, PG_ZERO);
1638                         vm_page_io_finish(pg);
1639                         if (error) {
1640                                 crit_enter();
1641                                 vm_page_unwire(pg, 0);
1642                                 vm_page_try_to_free(pg);
1643                                 crit_exit();
1644                                 ssb_unlock(&so->so_snd);
1645                                 goto done;
1646                         }
1647                 }
1648
1649
1650                 /*
1651                  * Get a sendfile buf. We usually wait as long as necessary,
1652                  * but this wait can be interrupted.
1653                  */
1654                 if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1655                         crit_enter();
1656                         vm_page_unwire(pg, 0);
1657                         vm_page_try_to_free(pg);
1658                         crit_exit();
1659                         ssb_unlock(&so->so_snd);
1660                         error = EINTR;
1661                         goto done;
1662                 }
1663
1664                 /*
1665                  * Get an mbuf header and set it up as having external storage.
1666                  */
1667                 MGETHDR(m, MB_WAIT, MT_DATA);
1668                 if (m == NULL) {
1669                         error = ENOBUFS;
1670                         sf_buf_free(sf);
1671                         ssb_unlock(&so->so_snd);
1672                         goto done;
1673                 }
1674
1675                 /*
1676                  * sfm is a temporary hack, use a per-cpu cache for this.
1677                  */
1678                 sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1679                 sfm->sf = sf;
1680                 sfm->mref_count = 1;
1681
1682                 m->m_ext.ext_free = sf_buf_mfree;
1683                 m->m_ext.ext_ref = sf_buf_mref;
1684                 m->m_ext.ext_arg = sfm;
1685                 m->m_ext.ext_buf = (void *)sf->kva;
1686                 m->m_ext.ext_size = PAGE_SIZE;
1687                 m->m_data = (char *) sf->kva + pgoff;
1688                 m->m_flags |= M_EXT;
1689                 m->m_pkthdr.len = m->m_len = xfsize;
1690                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1691
1692                 if (mheader != NULL) {
1693                         hbytes = mheader->m_pkthdr.len;
1694                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1695                         m_cat(mheader, m);
1696                         m = mheader;
1697                         mheader = NULL;
1698                 } else
1699                         hbytes = 0;
1700
1701                 /*
1702                  * Add the buffer to the socket buffer chain.
1703                  */
1704                 crit_enter();
1705 retry_space:
1706                 /*
1707                  * Make sure that the socket is still able to take more data.
1708                  * CANTSENDMORE being true usually means that the connection
1709                  * was closed. so_error is true when an error was sensed after
1710                  * a previous send.
1711                  * The state is checked after the page mapping and buffer
1712                  * allocation above since those operations may block and make
1713                  * any socket checks stale. From this point forward, nothing
1714                  * blocks before the pru_send (or more accurately, any blocking
1715                  * results in a loop back to here to re-check).
1716                  */
1717                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1718                         if (so->so_state & SS_CANTSENDMORE) {
1719                                 error = EPIPE;
1720                         } else {
1721                                 error = so->so_error;
1722                                 so->so_error = 0;
1723                         }
1724                         m_freem(m);
1725                         ssb_unlock(&so->so_snd);
1726                         crit_exit();
1727                         goto done;
1728                 }
1729                 /*
1730                  * Wait for socket space to become available. We do this just
1731                  * after checking the connection state above in order to avoid
1732                  * a race condition with ssb_wait().
1733                  */
1734                 if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1735                         if (fp->f_flag & FNONBLOCK) {
1736                                 m_freem(m);
1737                                 ssb_unlock(&so->so_snd);
1738                                 crit_exit();
1739                                 error = EAGAIN;
1740                                 goto done;
1741                         }
1742                         error = ssb_wait(&so->so_snd);
1743                         /*
1744                          * An error from ssb_wait usually indicates that we've
1745                          * been interrupted by a signal. If we've sent anything
1746                          * then return bytes sent, otherwise return the error.
1747                          */
1748                         if (error) {
1749                                 m_freem(m);
1750                                 ssb_unlock(&so->so_snd);
1751                                 crit_exit();
1752                                 goto done;
1753                         }
1754                         goto retry_space;
1755                 }
1756                 error = so_pru_send(so, 0, m, NULL, NULL, td);
1757                 crit_exit();
1758                 if (error) {
1759                         ssb_unlock(&so->so_snd);
1760                         goto done;
1761                 }
1762         }
1763         if (mheader != NULL) {
1764                 *sbytes += mheader->m_pkthdr.len;
1765                 error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1766                 mheader = NULL;
1767         }
1768         ssb_unlock(&so->so_snd);
1769
1770 done:
1771         fdrop(fp);
1772 done0:
1773         if (mheader != NULL)
1774                 m_freem(mheader);
1775         return (error);
1776 }
1777
1778 int
1779 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1780 {
1781 #ifdef SCTP
1782         struct thread *td = curthread;
1783         struct proc *p = td->td_proc;
1784         struct file *lfp = NULL;
1785         struct file *nfp = NULL;
1786         int error;
1787         struct socket *head, *so;
1788         caddr_t assoc_id;
1789         int fd;
1790         short fflag;            /* type must match fp->f_flag */
1791
1792         assoc_id = uap->name;
1793         error = holdsock(p->p_fd, uap->sd, &lfp);
1794         if (error) {
1795                 return (error);
1796         }
1797         crit_enter();
1798         head = (struct socket *)lfp->f_data;
1799         error = sctp_can_peel_off(head, assoc_id);
1800         if (error) {
1801                 crit_exit();
1802                 goto done;
1803         }
1804         /*
1805          * At this point we know we do have a assoc to pull
1806          * we proceed to get the fd setup. This may block
1807          * but that is ok.
1808          */
1809
1810         fflag = lfp->f_flag;
1811         error = falloc(p, &nfp, &fd);
1812         if (error) {
1813                 /*
1814                  * Probably ran out of file descriptors. Put the
1815                  * unaccepted connection back onto the queue and
1816                  * do another wakeup so some other process might
1817                  * have a chance at it.
1818                  */
1819                 crit_exit();
1820                 goto done;
1821         }
1822         uap->sysmsg_result = fd;
1823
1824         so = sctp_get_peeloff(head, assoc_id, &error);
1825         if (so == NULL) {
1826                 /*
1827                  * Either someone else peeled it off OR
1828                  * we can't get a socket.
1829                  */
1830                 goto noconnection;
1831         }
1832         so->so_state &= ~SS_COMP;
1833         so->so_state &= ~SS_NOFDREF;
1834         so->so_head = NULL;
1835         if (head->so_sigio != NULL)
1836                 fsetown(fgetown(head->so_sigio), &so->so_sigio);
1837
1838         nfp->f_type = DTYPE_SOCKET;
1839         nfp->f_flag = fflag;
1840         nfp->f_ops = &socketops;
1841         nfp->f_data = so;
1842
1843 noconnection:
1844         /*
1845          * Assign the file pointer to the reserved descriptor, or clear
1846          * the reserved descriptor if an error occured.
1847          */
1848         if (error)
1849                 fsetfd(p, NULL, fd);
1850         else
1851                 fsetfd(p, nfp, fd);
1852         crit_exit();
1853         /*
1854          * Release explicitly held references before returning.
1855          */
1856 done:
1857         if (nfp != NULL)
1858                 fdrop(nfp);
1859         fdrop(lfp);
1860         return (error);
1861 #else /* SCTP */
1862         return(EOPNOTSUPP);
1863 #endif /* SCTP */
1864 }