sys/kern/uipc_syscalls.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * sendfile(2) and related extensions:
   6  * Copyright (c) 1998, David Greenman. All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  *      @(#)uipc_syscalls.c     8.4 (Berkeley) 2/21/94
  33  * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
  34  */
  35
  36 #include "opt_ktrace.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/kernel.h>
  41 #include <sys/sysproto.h>
  42 #include <sys/malloc.h>
  43 #include <sys/filedesc.h>
  44 #include <sys/event.h>
  45 #include <sys/proc.h>
  46 #include <sys/fcntl.h>
  47 #include <sys/file.h>
  48 #include <sys/filio.h>
  49 #include <sys/kern_syscall.h>
  50 #include <sys/mbuf.h>
  51 #include <sys/protosw.h>
  52 #include <sys/sfbuf.h>
  53 #include <sys/socket.h>
  54 #include <sys/socketvar.h>
  55 #include <sys/socketops.h>
  56 #include <sys/uio.h>
  57 #include <sys/vnode.h>
  58 #include <sys/lock.h>
  59 #include <sys/mount.h>
  60 #ifdef KTRACE
  61 #include <sys/ktrace.h>
  62 #endif
  63 #include <vm/vm.h>
  64 #include <vm/vm_object.h>
  65 #include <vm/vm_page.h>
  66 #include <vm/vm_pageout.h>
  67 #include <vm/vm_kern.h>
  68 #include <vm/vm_extern.h>
  69 #include <sys/file2.h>
  70 #include <sys/signalvar.h>
  71 #include <sys/serialize.h>
  72
  73 #include <sys/thread2.h>
  74 #include <sys/msgport2.h>
  75 #include <sys/socketvar2.h>
  76 #include <net/netmsg2.h>
  77 #include <vm/vm_page2.h>
  78
  79 extern int use_soaccept_pred_fast;
  80 extern int use_sendfile_async;
  81 extern int use_soconnect_async;
  82
  83 /*
  84  * System call interface to the socket abstraction.
  85  */
  86
  87 extern  struct fileops socketops;
  88
  89 /*
  90  * socket_args(int domain, int type, int protocol)
  91  */
  92 int
  93 kern_socket(int domain, int type, int protocol, int *res)
  94 {
  95         struct thread *td = curthread;
  96         struct filedesc *fdp = td->td_proc->p_fd;
  97         struct socket *so;
  98         struct file *fp;
  99         int fd, error;
 100         u_int fflags = 0;
 101         int oflags = 0;
 102
 103         KKASSERT(td->td_lwp);
 104
 105         if (type & SOCK_NONBLOCK) {
 106                 type &= ~SOCK_NONBLOCK;
 107                 fflags |= FNONBLOCK;
 108         }
 109         if (type & SOCK_CLOEXEC) {
 110                 type &= ~SOCK_CLOEXEC;
 111                 oflags |= O_CLOEXEC;
 112         }
 113
 114         error = falloc(td->td_lwp, &fp, &fd);
 115         if (error)
 116                 return (error);
 117         error = socreate(domain, &so, type, protocol, td);
 118         if (error) {
 119                 fsetfd(fdp, NULL, fd);
 120         } else {
 121                 fp->f_type = DTYPE_SOCKET;
 122                 fp->f_flag = FREAD | FWRITE | fflags;
 123                 fp->f_ops = &socketops;
 124                 fp->f_data = so;
 125                 if (oflags & O_CLOEXEC)
 126                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 127                 *res = fd;
 128                 fsetfd(fdp, fp, fd);
 129         }
 130         fdrop(fp);
 131         return (error);
 132 }
 133
 134 /*
 135  * MPALMOSTSAFE
 136  */
 137 int
 138 sys_socket(struct socket_args *uap)
 139 {
 140         int error;
 141
 142         error = kern_socket(uap->domain, uap->type, uap->protocol,
 143                             &uap->sysmsg_iresult);
 144
 145         return (error);
 146 }
 147
 148 int
 149 kern_bind(int s, struct sockaddr *sa)
 150 {
 151         struct thread *td = curthread;
 152         struct proc *p = td->td_proc;
 153         struct file *fp;
 154         int error;
 155
 156         KKASSERT(p);
 157         error = holdsock(p->p_fd, s, &fp);
 158         if (error)
 159                 return (error);
 160         error = sobind((struct socket *)fp->f_data, sa, td);
 161         fdrop(fp);
 162         return (error);
 163 }
 164
 165 /*
 166  * bind_args(int s, caddr_t name, int namelen)
 167  *
 168  * MPALMOSTSAFE
 169  */
 170 int
 171 sys_bind(struct bind_args *uap)
 172 {
 173         struct sockaddr *sa;
 174         int error;
 175
 176         error = getsockaddr(&sa, uap->name, uap->namelen);
 177         if (error)
 178                 return (error);
 179         error = kern_bind(uap->s, sa);
 180         kfree(sa, M_SONAME);
 181
 182         return (error);
 183 }
 184
 185 int
 186 kern_listen(int s, int backlog)
 187 {
 188         struct thread *td = curthread;
 189         struct proc *p = td->td_proc;
 190         struct file *fp;
 191         int error;
 192
 193         KKASSERT(p);
 194         error = holdsock(p->p_fd, s, &fp);
 195         if (error)
 196                 return (error);
 197         error = solisten((struct socket *)fp->f_data, backlog, td);
 198         fdrop(fp);
 199         return(error);
 200 }
 201
 202 /*
 203  * listen_args(int s, int backlog)
 204  *
 205  * MPALMOSTSAFE
 206  */
 207 int
 208 sys_listen(struct listen_args *uap)
 209 {
 210         int error;
 211
 212         error = kern_listen(uap->s, uap->backlog);
 213         return (error);
 214 }
 215
 216 /*
 217  * Returns the accepted socket as well.
 218  *
 219  * NOTE!  The sockets sitting on so_comp/so_incomp might have 0 refs, the
 220  *        pool token is absolutely required to avoid a sofree() race,
 221  *        as well as to avoid tailq handling races.
 222  */
 223 static boolean_t
 224 soaccept_predicate(struct netmsg_so_notify *msg)
 225 {
 226         struct socket *head = msg->base.nm_so;
 227         struct socket *so;
 228
 229         if (head->so_error != 0) {
 230                 msg->base.lmsg.ms_error = head->so_error;
 231                 return (TRUE);
 232         }
 233         lwkt_getpooltoken(head);
 234         if (!TAILQ_EMPTY(&head->so_comp)) {
 235                 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
 236                 so = TAILQ_FIRST(&head->so_comp);
 237                 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
 238                 TAILQ_REMOVE(&head->so_comp, so, so_list);
 239                 head->so_qlen--;
 240                 soclrstate(so, SS_COMP);
 241
 242                 /*
 243                  * Keep a reference before clearing the so_head
 244                  * to avoid racing socket close in netisr.
 245                  */
 246                 soreference(so);
 247                 so->so_head = NULL;
 248
 249                 lwkt_relpooltoken(head);
 250
 251                 msg->base.lmsg.ms_error = 0;
 252                 msg->base.nm_so = so;
 253                 return (TRUE);
 254         }
 255         lwkt_relpooltoken(head);
 256         if (head->so_state & SS_CANTRCVMORE) {
 257                 msg->base.lmsg.ms_error = ECONNABORTED;
 258                 return (TRUE);
 259         }
 260         if (msg->nm_fflags & FNONBLOCK) {
 261                 msg->base.lmsg.ms_error = EWOULDBLOCK;
 262                 return (TRUE);
 263         }
 264
 265         return (FALSE);
 266 }
 267
 268 /*
 269  * The second argument to kern_accept() is a handle to a struct sockaddr.
 270  * This allows kern_accept() to return a pointer to an allocated struct
 271  * sockaddr which must be freed later with FREE().  The caller must
 272  * initialize *name to NULL.
 273  */
 274 int
 275 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res,
 276     int sockflags)
 277 {
 278         struct thread *td = curthread;
 279         struct filedesc *fdp = td->td_proc->p_fd;
 280         struct file *lfp = NULL;
 281         struct file *nfp = NULL;
 282         struct sockaddr *sa;
 283         struct socket *head, *so;
 284         struct netmsg_so_notify msg;
 285         int fd;
 286         u_int fflag;            /* type must match fp->f_flag */
 287         int error, tmp;
 288
 289         *res = -1;
 290         if (name && namelen && *namelen < 0)
 291                 return (EINVAL);
 292
 293         error = holdsock(td->td_proc->p_fd, s, &lfp);
 294         if (error)
 295                 return (error);
 296
 297         error = falloc(td->td_lwp, &nfp, &fd);
 298         if (error) {            /* Probably ran out of file descriptors. */
 299                 fdrop(lfp);
 300                 return (error);
 301         }
 302         head = (struct socket *)lfp->f_data;
 303         if ((head->so_options & SO_ACCEPTCONN) == 0) {
 304                 error = EINVAL;
 305                 goto done;
 306         }
 307
 308         if (fflags & O_FBLOCKING)
 309                 fflags |= lfp->f_flag & ~FNONBLOCK;
 310         else if (fflags & O_FNONBLOCKING)
 311                 fflags |= lfp->f_flag | FNONBLOCK;
 312         else
 313                 fflags = lfp->f_flag;
 314
 315         if (use_soaccept_pred_fast) {
 316                 boolean_t pred;
 317
 318                 /* Initialize necessary parts for soaccept_predicate() */
 319                 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
 320                 msg.nm_fflags = fflags;
 321
 322                 lwkt_getpooltoken(head);
 323                 pred = soaccept_predicate(&msg);
 324                 lwkt_relpooltoken(head);
 325
 326                 if (pred) {
 327                         error = msg.base.lmsg.ms_error;
 328                         if (error)
 329                                 goto done;
 330                         else
 331                                 goto accepted;
 332                 }
 333         }
 334
 335         /* optimize for uniprocessor case later XXX JH */
 336         netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
 337                               0, netmsg_so_notify, netmsg_so_notify_doabort);
 338         msg.nm_predicate = soaccept_predicate;
 339         msg.nm_fflags = fflags;
 340         msg.nm_etype = NM_REVENT;
 341         error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
 342         if (error)
 343                 goto done;
 344
 345 accepted:
 346         /*
 347          * At this point we have the connection that's ready to be accepted.
 348          *
 349          * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
 350          *       to eat the ref and turn it into a descriptor.
 351          */
 352         so = msg.base.nm_so;
 353
 354         fflag = lfp->f_flag;
 355
 356         /* connection has been removed from the listen queue */
 357         KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
 358
 359         if (sockflags & SOCK_KERN_NOINHERIT) {
 360                 fflag &= ~(FASYNC | FNONBLOCK);
 361                 if (sockflags & SOCK_NONBLOCK)
 362                         fflag |= FNONBLOCK;
 363         } else {
 364                 if (head->so_sigio != NULL)
 365                         fsetown(fgetown(&head->so_sigio), &so->so_sigio);
 366         }
 367
 368         nfp->f_type = DTYPE_SOCKET;
 369         nfp->f_flag = fflag;
 370         nfp->f_ops = &socketops;
 371         nfp->f_data = so;
 372         /* Sync socket async state with file flags */
 373         tmp = fflag & FASYNC;
 374         fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
 375
 376         sa = NULL;
 377         if (so->so_faddr != NULL) {
 378                 sa = so->so_faddr;
 379                 so->so_faddr = NULL;
 380
 381                 soaccept_generic(so);
 382                 error = 0;
 383         } else {
 384                 error = soaccept(so, &sa);
 385         }
 386
 387         /*
 388          * Set the returned name and namelen as applicable.  Set the returned
 389          * namelen to 0 for older code which might ignore the return value
 390          * from accept.
 391          */
 392         if (error == 0) {
 393                 if (sa && name && namelen) {
 394                         if (*namelen > sa->sa_len)
 395                                 *namelen = sa->sa_len;
 396                         *name = sa;
 397                 } else {
 398                         if (sa)
 399                                 kfree(sa, M_SONAME);
 400                 }
 401         }
 402
 403 done:
 404         /*
 405          * If an error occured clear the reserved descriptor, else associate
 406          * nfp with it.
 407          *
 408          * Note that *res is normally ignored if an error is returned but
 409          * a syscall message will still have access to the result code.
 410          */
 411         if (error) {
 412                 fsetfd(fdp, NULL, fd);
 413         } else {
 414                 if (sockflags & SOCK_CLOEXEC)
 415                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 416                 *res = fd;
 417                 fsetfd(fdp, nfp, fd);
 418         }
 419         fdrop(nfp);
 420         fdrop(lfp);
 421         return (error);
 422 }
 423
 424 /*
 425  * accept(int s, caddr_t name, int *anamelen)
 426  *
 427  * MPALMOSTSAFE
 428  */
 429 int
 430 sys_accept(struct accept_args *uap)
 431 {
 432         struct sockaddr *sa = NULL;
 433         int sa_len;
 434         int error;
 435
 436         if (uap->name) {
 437                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 438                 if (error)
 439                         return (error);
 440
 441                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 442                                     &uap->sysmsg_iresult, 0);
 443
 444                 if (error == 0)
 445                         error = copyout(sa, uap->name, sa_len);
 446                 if (error == 0) {
 447                         error = copyout(&sa_len, uap->anamelen,
 448                             sizeof(*uap->anamelen));
 449                 }
 450                 if (sa)
 451                         kfree(sa, M_SONAME);
 452         } else {
 453                 error = kern_accept(uap->s, 0, NULL, 0,
 454                                     &uap->sysmsg_iresult, 0);
 455         }
 456         return (error);
 457 }
 458
 459 /*
 460  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
 461  *
 462  * MPALMOSTSAFE
 463  */
 464 int
 465 sys_extaccept(struct extaccept_args *uap)
 466 {
 467         struct sockaddr *sa = NULL;
 468         int sa_len;
 469         int error;
 470         int fflags = uap->flags & O_FMASK;
 471
 472         if (uap->name) {
 473                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 474                 if (error)
 475                         return (error);
 476
 477                 error = kern_accept(uap->s, fflags, &sa, &sa_len,
 478                                     &uap->sysmsg_iresult, 0);
 479
 480                 if (error == 0)
 481                         error = copyout(sa, uap->name, sa_len);
 482                 if (error == 0) {
 483                         error = copyout(&sa_len, uap->anamelen,
 484                             sizeof(*uap->anamelen));
 485                 }
 486                 if (sa)
 487                         kfree(sa, M_SONAME);
 488         } else {
 489                 error = kern_accept(uap->s, fflags, NULL, 0,
 490                                     &uap->sysmsg_iresult, 0);
 491         }
 492         return (error);
 493 }
 494
 495 /*
 496  * accept4(int s, caddr_t name, int *anamelen, int flags)
 497  *
 498  * MPALMOSTSAFE
 499  */
 500 int
 501 sys_accept4(struct accept4_args *uap)
 502 {
 503         struct sockaddr *sa = NULL;
 504         int sa_len;
 505         int error;
 506         int sockflags;
 507
 508         if (uap->flags & ~(SOCK_NONBLOCK | SOCK_CLOEXEC))
 509                 return (EINVAL);
 510         sockflags = uap->flags | SOCK_KERN_NOINHERIT;
 511
 512         if (uap->name) {
 513                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 514                 if (error)
 515                         return (error);
 516
 517                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 518                                     &uap->sysmsg_iresult, sockflags);
 519
 520                 if (error == 0)
 521                         error = copyout(sa, uap->name, sa_len);
 522                 if (error == 0) {
 523                         error = copyout(&sa_len, uap->anamelen,
 524                             sizeof(*uap->anamelen));
 525                 }
 526                 if (sa)
 527                         kfree(sa, M_SONAME);
 528         } else {
 529                 error = kern_accept(uap->s, 0, NULL, 0,
 530                                     &uap->sysmsg_iresult, sockflags);
 531         }
 532         return (error);
 533 }
 534
 535 /*
 536  * Returns TRUE if predicate satisfied.
 537  */
 538 static boolean_t
 539 soconnected_predicate(struct netmsg_so_notify *msg)
 540 {
 541         struct socket *so = msg->base.nm_so;
 542
 543         /* check predicate */
 544         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 545                 msg->base.lmsg.ms_error = so->so_error;
 546                 return (TRUE);
 547         }
 548
 549         return (FALSE);
 550 }
 551
 552 int
 553 kern_connect(int s, int fflags, struct sockaddr *sa)
 554 {
 555         struct thread *td = curthread;
 556         struct proc *p = td->td_proc;
 557         struct file *fp;
 558         struct socket *so;
 559         int error, interrupted = 0;
 560
 561         error = holdsock(p->p_fd, s, &fp);
 562         if (error)
 563                 return (error);
 564         so = (struct socket *)fp->f_data;
 565
 566         if (fflags & O_FBLOCKING)
 567                 /* fflags &= ~FNONBLOCK; */;
 568         else if (fflags & O_FNONBLOCKING)
 569                 fflags |= FNONBLOCK;
 570         else
 571                 fflags = fp->f_flag;
 572
 573         if (so->so_state & SS_ISCONNECTING) {
 574                 error = EALREADY;
 575                 goto done;
 576         }
 577         error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE);
 578         if (error)
 579                 goto bad;
 580         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 581                 error = EINPROGRESS;
 582                 goto done;
 583         }
 584         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 585                 struct netmsg_so_notify msg;
 586
 587                 netmsg_init_abortable(&msg.base, so,
 588                                       &curthread->td_msgport,
 589                                       0,
 590                                       netmsg_so_notify,
 591                                       netmsg_so_notify_doabort);
 592                 msg.nm_predicate = soconnected_predicate;
 593                 msg.nm_etype = NM_REVENT;
 594                 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
 595                 if (error == EINTR || error == ERESTART)
 596                         interrupted = 1;
 597         }
 598         if (error == 0) {
 599                 error = so->so_error;
 600                 so->so_error = 0;
 601         }
 602 bad:
 603         if (!interrupted)
 604                 soclrstate(so, SS_ISCONNECTING);
 605         if (error == ERESTART)
 606                 error = EINTR;
 607 done:
 608         fdrop(fp);
 609         return (error);
 610 }
 611
 612 /*
 613  * connect_args(int s, caddr_t name, int namelen)
 614  *
 615  * MPALMOSTSAFE
 616  */
 617 int
 618 sys_connect(struct connect_args *uap)
 619 {
 620         struct sockaddr *sa;
 621         int error;
 622
 623         error = getsockaddr(&sa, uap->name, uap->namelen);
 624         if (error)
 625                 return (error);
 626         error = kern_connect(uap->s, 0, sa);
 627         kfree(sa, M_SONAME);
 628
 629         return (error);
 630 }
 631
 632 /*
 633  * connect_args(int s, int fflags, caddr_t name, int namelen)
 634  *
 635  * MPALMOSTSAFE
 636  */
 637 int
 638 sys_extconnect(struct extconnect_args *uap)
 639 {
 640         struct sockaddr *sa;
 641         int error;
 642         int fflags = uap->flags & O_FMASK;
 643
 644         error = getsockaddr(&sa, uap->name, uap->namelen);
 645         if (error)
 646                 return (error);
 647         error = kern_connect(uap->s, fflags, sa);
 648         kfree(sa, M_SONAME);
 649
 650         return (error);
 651 }
 652
 653 int
 654 kern_socketpair(int domain, int type, int protocol, int *sv)
 655 {
 656         struct thread *td = curthread;
 657         struct filedesc *fdp;
 658         struct file *fp1, *fp2;
 659         struct socket *so1, *so2;
 660         int fd1, fd2, error;
 661         u_int fflags = 0;
 662         int oflags = 0;
 663
 664         if (type & SOCK_NONBLOCK) {
 665                 type &= ~SOCK_NONBLOCK;
 666                 fflags |= FNONBLOCK;
 667         }
 668         if (type & SOCK_CLOEXEC) {
 669                 type &= ~SOCK_CLOEXEC;
 670                 oflags |= O_CLOEXEC;
 671         }
 672
 673         fdp = td->td_proc->p_fd;
 674         error = socreate(domain, &so1, type, protocol, td);
 675         if (error)
 676                 return (error);
 677         error = socreate(domain, &so2, type, protocol, td);
 678         if (error)
 679                 goto free1;
 680         error = falloc(td->td_lwp, &fp1, &fd1);
 681         if (error)
 682                 goto free2;
 683         sv[0] = fd1;
 684         fp1->f_data = so1;
 685         error = falloc(td->td_lwp, &fp2, &fd2);
 686         if (error)
 687                 goto free3;
 688         fp2->f_data = so2;
 689         sv[1] = fd2;
 690         error = soconnect2(so1, so2);
 691         if (error)
 692                 goto free4;
 693         if (type == SOCK_DGRAM) {
 694                 /*
 695                  * Datagram socket connection is asymmetric.
 696                  */
 697                  error = soconnect2(so2, so1);
 698                  if (error)
 699                         goto free4;
 700         }
 701         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 702         fp1->f_flag = fp2->f_flag = FREAD|FWRITE|fflags;
 703         fp1->f_ops = fp2->f_ops = &socketops;
 704         if (oflags & O_CLOEXEC) {
 705                 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE;
 706                 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE;
 707         }
 708         fsetfd(fdp, fp1, fd1);
 709         fsetfd(fdp, fp2, fd2);
 710         fdrop(fp1);
 711         fdrop(fp2);
 712         return (error);
 713 free4:
 714         fsetfd(fdp, NULL, fd2);
 715         fdrop(fp2);
 716 free3:
 717         fsetfd(fdp, NULL, fd1);
 718         fdrop(fp1);
 719 free2:
 720         (void)soclose(so2, 0);
 721 free1:
 722         (void)soclose(so1, 0);
 723         return (error);
 724 }
 725
 726 /*
 727  * socketpair(int domain, int type, int protocol, int *rsv)
 728  */
 729 int
 730 sys_socketpair(struct socketpair_args *uap)
 731 {
 732         int error, sockv[2];
 733
 734         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 735
 736         if (error == 0) {
 737                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 738
 739                 if (error != 0) {
 740                         kern_close(sockv[0]);
 741                         kern_close(sockv[1]);
 742                 }
 743         }
 744
 745         return (error);
 746 }
 747
 748 int
 749 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 750              struct mbuf *control, int flags, size_t *res)
 751 {
 752         struct thread *td = curthread;
 753         struct lwp *lp = td->td_lwp;
 754         struct proc *p = td->td_proc;
 755         struct file *fp;
 756         size_t len;
 757         int error;
 758         struct socket *so;
 759 #ifdef KTRACE
 760         struct iovec *ktriov = NULL;
 761         struct uio ktruio;
 762 #endif
 763
 764         error = holdsock(p->p_fd, s, &fp);
 765         if (error)
 766                 return (error);
 767 #ifdef KTRACE
 768         if (KTRPOINT(td, KTR_GENIO)) {
 769                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 770
 771                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 772                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 773                 ktruio = *auio;
 774         }
 775 #endif
 776         len = auio->uio_resid;
 777         so = (struct socket *)fp->f_data;
 778         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 779                 if (fp->f_flag & FNONBLOCK)
 780                         flags |= MSG_FNONBLOCKING;
 781         }
 782         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 783         if (error) {
 784                 if (auio->uio_resid != len && (error == ERESTART ||
 785                     error == EINTR || error == EWOULDBLOCK))
 786                         error = 0;
 787                 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
 788                     !(so->so_options & SO_NOSIGPIPE))
 789                         lwpsignal(p, lp, SIGPIPE);
 790         }
 791 #ifdef KTRACE
 792         if (ktriov != NULL) {
 793                 if (error == 0) {
 794                         ktruio.uio_iov = ktriov;
 795                         ktruio.uio_resid = len - auio->uio_resid;
 796                         ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
 797                 }
 798                 kfree(ktriov, M_TEMP);
 799         }
 800 #endif
 801         if (error == 0)
 802                 *res  = len - auio->uio_resid;
 803         fdrop(fp);
 804         return (error);
 805 }
 806
 807 /*
 808  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
 809  *
 810  * MPALMOSTSAFE
 811  */
 812 int
 813 sys_sendto(struct sendto_args *uap)
 814 {
 815         struct thread *td = curthread;
 816         struct uio auio;
 817         struct iovec aiov;
 818         struct sockaddr *sa = NULL;
 819         int error;
 820
 821         if (uap->to) {
 822                 error = getsockaddr(&sa, uap->to, uap->tolen);
 823                 if (error)
 824                         return (error);
 825         }
 826         aiov.iov_base = uap->buf;
 827         aiov.iov_len = uap->len;
 828         auio.uio_iov = &aiov;
 829         auio.uio_iovcnt = 1;
 830         auio.uio_offset = 0;
 831         auio.uio_resid = uap->len;
 832         auio.uio_segflg = UIO_USERSPACE;
 833         auio.uio_rw = UIO_WRITE;
 834         auio.uio_td = td;
 835
 836         error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
 837                              &uap->sysmsg_szresult);
 838
 839         if (sa)
 840                 kfree(sa, M_SONAME);
 841         return (error);
 842 }
 843
 844 /*
 845  * sendmsg_args(int s, caddr_t msg, int flags)
 846  *
 847  * MPALMOSTSAFE
 848  */
 849 int
 850 sys_sendmsg(struct sendmsg_args *uap)
 851 {
 852         struct thread *td = curthread;
 853         struct msghdr msg;
 854         struct uio auio;
 855         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 856         struct sockaddr *sa = NULL;
 857         struct mbuf *control = NULL;
 858         int error;
 859
 860         error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
 861         if (error)
 862                 return (error);
 863
 864         /*
 865          * Conditionally copyin msg.msg_name.
 866          */
 867         if (msg.msg_name) {
 868                 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 869                 if (error)
 870                         return (error);
 871         }
 872
 873         /*
 874          * Populate auio.
 875          */
 876         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 877                              &auio.uio_resid);
 878         if (error)
 879                 goto cleanup2;
 880         auio.uio_iov = iov;
 881         auio.uio_iovcnt = msg.msg_iovlen;
 882         auio.uio_offset = 0;
 883         auio.uio_segflg = UIO_USERSPACE;
 884         auio.uio_rw = UIO_WRITE;
 885         auio.uio_td = td;
 886
 887         /*
 888          * Conditionally copyin msg.msg_control.
 889          */
 890         if (msg.msg_control) {
 891                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 892                     msg.msg_controllen > MLEN) {
 893                         error = EINVAL;
 894                         goto cleanup;
 895                 }
 896                 control = m_get(M_WAITOK, MT_CONTROL);
 897                 if (control == NULL) {
 898                         error = ENOBUFS;
 899                         goto cleanup;
 900                 }
 901                 control->m_len = msg.msg_controllen;
 902                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 903                                msg.msg_controllen);
 904                 if (error) {
 905                         m_free(control);
 906                         goto cleanup;
 907                 }
 908         }
 909
 910         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 911                              &uap->sysmsg_szresult);
 912
 913 cleanup:
 914         iovec_free(&iov, aiov);
 915 cleanup2:
 916         if (sa)
 917                 kfree(sa, M_SONAME);
 918         return (error);
 919 }
 920
 921 /*
 922  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 923  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 924  * Don't forget to FREE() and m_free() these if they are returned.
 925  */
 926 int
 927 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 928              struct mbuf **control, int *flags, size_t *res)
 929 {
 930         struct thread *td = curthread;
 931         struct proc *p = td->td_proc;
 932         struct file *fp;
 933         size_t len;
 934         int error;
 935         int lflags;
 936         struct socket *so;
 937 #ifdef KTRACE
 938         struct iovec *ktriov = NULL;
 939         struct uio ktruio;
 940 #endif
 941
 942         error = holdsock(p->p_fd, s, &fp);
 943         if (error)
 944                 return (error);
 945 #ifdef KTRACE
 946         if (KTRPOINT(td, KTR_GENIO)) {
 947                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 948
 949                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 950                 bcopy(auio->uio_iov, ktriov, iovlen);
 951                 ktruio = *auio;
 952         }
 953 #endif
 954         len = auio->uio_resid;
 955         so = (struct socket *)fp->f_data;
 956
 957         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 958                 if (fp->f_flag & FNONBLOCK) {
 959                         if (flags) {
 960                                 *flags |= MSG_FNONBLOCKING;
 961                         } else {
 962                                 lflags = MSG_FNONBLOCKING;
 963                                 flags = &lflags;
 964                         }
 965                 }
 966         }
 967
 968         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 969         if (error) {
 970                 if (auio->uio_resid != len && (error == ERESTART ||
 971                     error == EINTR || error == EWOULDBLOCK))
 972                         error = 0;
 973         }
 974 #ifdef KTRACE
 975         if (ktriov != NULL) {
 976                 if (error == 0) {
 977                         ktruio.uio_iov = ktriov;
 978                         ktruio.uio_resid = len - auio->uio_resid;
 979                         ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
 980                 }
 981                 kfree(ktriov, M_TEMP);
 982         }
 983 #endif
 984         if (error == 0)
 985                 *res = len - auio->uio_resid;
 986         fdrop(fp);
 987         return (error);
 988 }
 989
 990 /*
 991  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
 992  *                      caddr_t from, int *fromlenaddr)
 993  *
 994  * MPALMOSTSAFE
 995  */
 996 int
 997 sys_recvfrom(struct recvfrom_args *uap)
 998 {
 999         struct thread *td = curthread;
1000         struct uio auio;
1001         struct iovec aiov;
1002         struct sockaddr *sa = NULL;
1003         int error, fromlen;
1004
1005         if (uap->from && uap->fromlenaddr) {
1006                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
1007                 if (error)
1008                         return (error);
1009                 if (fromlen < 0)
1010                         return (EINVAL);
1011         } else {
1012                 fromlen = 0;
1013         }
1014         aiov.iov_base = uap->buf;
1015         aiov.iov_len = uap->len;
1016         auio.uio_iov = &aiov;
1017         auio.uio_iovcnt = 1;
1018         auio.uio_offset = 0;
1019         auio.uio_resid = uap->len;
1020         auio.uio_segflg = UIO_USERSPACE;
1021         auio.uio_rw = UIO_READ;
1022         auio.uio_td = td;
1023
1024         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
1025                              &uap->flags, &uap->sysmsg_szresult);
1026
1027         if (error == 0 && uap->from) {
1028                 /* note: sa may still be NULL */
1029                 if (sa) {
1030                         fromlen = MIN(fromlen, sa->sa_len);
1031                         error = copyout(sa, uap->from, fromlen);
1032                 } else {
1033                         fromlen = 0;
1034                 }
1035                 if (error == 0) {
1036                         error = copyout(&fromlen, uap->fromlenaddr,
1037                                         sizeof(fromlen));
1038                 }
1039         }
1040         if (sa)
1041                 kfree(sa, M_SONAME);
1042
1043         return (error);
1044 }
1045
1046 /*
1047  * recvmsg_args(int s, struct msghdr *msg, int flags)
1048  *
1049  * MPALMOSTSAFE
1050  */
1051 int
1052 sys_recvmsg(struct recvmsg_args *uap)
1053 {
1054         struct thread *td = curthread;
1055         struct msghdr msg;
1056         struct uio auio;
1057         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1058         struct mbuf *m, *control = NULL;
1059         struct sockaddr *sa = NULL;
1060         caddr_t ctlbuf;
1061         socklen_t *ufromlenp, *ucontrollenp;
1062         int error, fromlen, controllen, len, flags, *uflagsp;
1063
1064         /*
1065          * This copyin handles everything except the iovec.
1066          */
1067         error = copyin(uap->msg, &msg, sizeof(msg));
1068         if (error)
1069                 return (error);
1070
1071         if (msg.msg_name && msg.msg_namelen < 0)
1072                 return (EINVAL);
1073         if (msg.msg_control && msg.msg_controllen < 0)
1074                 return (EINVAL);
1075
1076         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1077                     msg_namelen));
1078         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1079                        msg_controllen));
1080         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1081                                                         msg_flags));
1082
1083         /*
1084          * Populate auio.
1085          */
1086         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1087                              &auio.uio_resid);
1088         if (error)
1089                 return (error);
1090         auio.uio_iov = iov;
1091         auio.uio_iovcnt = msg.msg_iovlen;
1092         auio.uio_offset = 0;
1093         auio.uio_segflg = UIO_USERSPACE;
1094         auio.uio_rw = UIO_READ;
1095         auio.uio_td = td;
1096
1097         flags = uap->flags;
1098
1099         error = kern_recvmsg(uap->s,
1100                              (msg.msg_name ? &sa : NULL), &auio,
1101                              (msg.msg_control ? &control : NULL), &flags,
1102                              &uap->sysmsg_szresult);
1103
1104         /*
1105          * Conditionally copyout the name and populate the namelen field.
1106          */
1107         if (error == 0 && msg.msg_name) {
1108                 /* note: sa may still be NULL */
1109                 if (sa != NULL) {
1110                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
1111                         error = copyout(sa, msg.msg_name, fromlen);
1112                 } else {
1113                         fromlen = 0;
1114                 }
1115                 if (error == 0)
1116                         error = copyout(&fromlen, ufromlenp,
1117                             sizeof(*ufromlenp));
1118         }
1119
1120         /*
1121          * Copyout msg.msg_control and msg.msg_controllen.
1122          */
1123         if (error == 0 && msg.msg_control) {
1124                 len = msg.msg_controllen;
1125                 m = control;
1126                 ctlbuf = (caddr_t)msg.msg_control;
1127
1128                 while(m && len > 0) {
1129                         unsigned int tocopy;
1130
1131                         if (len >= m->m_len) {
1132                                 tocopy = m->m_len;
1133                         } else {
1134                                 msg.msg_flags |= MSG_CTRUNC;
1135                                 tocopy = len;
1136                         }
1137
1138                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1139                         if (error)
1140                                 goto cleanup;
1141
1142                         ctlbuf += tocopy;
1143                         len -= tocopy;
1144                         m = m->m_next;
1145                 }
1146                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1147                 error = copyout(&controllen, ucontrollenp,
1148                     sizeof(*ucontrollenp));
1149         }
1150
1151         if (error == 0)
1152                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1153
1154 cleanup:
1155         if (sa)
1156                 kfree(sa, M_SONAME);
1157         iovec_free(&iov, aiov);
1158         if (control)
1159                 m_freem(control);
1160         return (error);
1161 }
1162
1163 /*
1164  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1165  * in kernel pointer instead of a userland pointer.  This allows us
1166  * to manipulate socket options in the emulation code.
1167  */
1168 int
1169 kern_setsockopt(int s, struct sockopt *sopt)
1170 {
1171         struct thread *td = curthread;
1172         struct proc *p = td->td_proc;
1173         struct file *fp;
1174         int error;
1175
1176         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1177                 return (EFAULT);
1178         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1179                 return (EINVAL);
1180         if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1181                 return (EINVAL);
1182
1183         error = holdsock(p->p_fd, s, &fp);
1184         if (error)
1185                 return (error);
1186
1187         error = sosetopt((struct socket *)fp->f_data, sopt);
1188         fdrop(fp);
1189         return (error);
1190 }
1191
1192 /*
1193  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1194  *
1195  * MPALMOSTSAFE
1196  */
1197 int
1198 sys_setsockopt(struct setsockopt_args *uap)
1199 {
1200         struct thread *td = curthread;
1201         struct sockopt sopt;
1202         int error;
1203
1204         sopt.sopt_level = uap->level;
1205         sopt.sopt_name = uap->name;
1206         sopt.sopt_valsize = uap->valsize;
1207         sopt.sopt_td = td;
1208         sopt.sopt_val = NULL;
1209
1210         if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1211                 return (EINVAL);
1212         if (uap->val) {
1213                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1214                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1215                 if (error)
1216                         goto out;
1217         }
1218
1219         error = kern_setsockopt(uap->s, &sopt);
1220 out:
1221         if (uap->val)
1222                 kfree(sopt.sopt_val, M_TEMP);
1223         return(error);
1224 }
1225
1226 /*
1227  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1228  * in kernel pointer instead of a userland pointer.  This allows us
1229  * to manipulate socket options in the emulation code.
1230  */
1231 int
1232 kern_getsockopt(int s, struct sockopt *sopt)
1233 {
1234         struct thread *td = curthread;
1235         struct proc *p = td->td_proc;
1236         struct file *fp;
1237         int error;
1238
1239         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1240                 return (EFAULT);
1241         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1242                 return (EINVAL);
1243
1244         error = holdsock(p->p_fd, s, &fp);
1245         if (error)
1246                 return (error);
1247
1248         error = sogetopt((struct socket *)fp->f_data, sopt);
1249         fdrop(fp);
1250         return (error);
1251 }
1252
1253 /*
1254  * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1255  *
1256  * MPALMOSTSAFE
1257  */
1258 int
1259 sys_getsockopt(struct getsockopt_args *uap)
1260 {
1261         struct thread *td = curthread;
1262         struct sockopt sopt;
1263         int error, valsize, valszmax, mflag = 0;
1264
1265         if (uap->val) {
1266                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1267                 if (error)
1268                         return (error);
1269         } else {
1270                 valsize = 0;
1271         }
1272
1273         sopt.sopt_level = uap->level;
1274         sopt.sopt_name = uap->name;
1275         sopt.sopt_valsize = valsize;
1276         sopt.sopt_td = td;
1277         sopt.sopt_val = NULL;
1278
1279         if (td->td_proc->p_ucred->cr_uid == 0) {
1280                 valszmax = SOMAXOPT_SIZE0;
1281                 mflag = M_NULLOK;
1282         } else {
1283                 valszmax = SOMAXOPT_SIZE;
1284         }
1285         if (sopt.sopt_valsize > valszmax) /* unsigned */
1286                 return (EINVAL);
1287         if (uap->val) {
1288                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP,
1289                     M_WAITOK | mflag);
1290                 if (sopt.sopt_val == NULL)
1291                         return (ENOBUFS);
1292                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1293                 if (error)
1294                         goto out;
1295         }
1296
1297         error = kern_getsockopt(uap->s, &sopt);
1298         if (error)
1299                 goto out;
1300         valsize = sopt.sopt_valsize;
1301         error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1302         if (error)
1303                 goto out;
1304         if (uap->val)
1305                 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1306 out:
1307         if (uap->val)
1308                 kfree(sopt.sopt_val, M_TEMP);
1309         return (error);
1310 }
1311
1312 /*
1313  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1314  * This allows kern_getsockname() to return a pointer to an allocated struct
1315  * sockaddr which must be freed later with FREE().  The caller must
1316  * initialize *name to NULL.
1317  */
1318 int
1319 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1320 {
1321         struct thread *td = curthread;
1322         struct proc *p = td->td_proc;
1323         struct file *fp;
1324         struct socket *so;
1325         struct sockaddr *sa = NULL;
1326         int error;
1327
1328         error = holdsock(p->p_fd, s, &fp);
1329         if (error)
1330                 return (error);
1331         if (*namelen < 0) {
1332                 fdrop(fp);
1333                 return (EINVAL);
1334         }
1335         so = (struct socket *)fp->f_data;
1336         error = so_pru_sockaddr(so, &sa);
1337         if (error == 0) {
1338                 if (sa == NULL) {
1339                         *namelen = 0;
1340                 } else {
1341                         *namelen = MIN(*namelen, sa->sa_len);
1342                         *name = sa;
1343                 }
1344         }
1345
1346         fdrop(fp);
1347         return (error);
1348 }
1349
1350 /*
1351  * getsockname_args(int fdes, caddr_t asa, int *alen)
1352  *
1353  * Get socket name.
1354  *
1355  * MPALMOSTSAFE
1356  */
1357 int
1358 sys_getsockname(struct getsockname_args *uap)
1359 {
1360         struct sockaddr *sa = NULL;
1361         int error, sa_len;
1362
1363         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1364         if (error)
1365                 return (error);
1366
1367         error = kern_getsockname(uap->fdes, &sa, &sa_len);
1368
1369         if (error == 0)
1370                 error = copyout(sa, uap->asa, sa_len);
1371         if (error == 0)
1372                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1373         if (sa)
1374                 kfree(sa, M_SONAME);
1375         return (error);
1376 }
1377
1378 /*
1379  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1380  * This allows kern_getpeername() to return a pointer to an allocated struct
1381  * sockaddr which must be freed later with FREE().  The caller must
1382  * initialize *name to NULL.
1383  */
1384 int
1385 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1386 {
1387         struct thread *td = curthread;
1388         struct proc *p = td->td_proc;
1389         struct file *fp;
1390         struct socket *so;
1391         struct sockaddr *sa = NULL;
1392         int error;
1393
1394         error = holdsock(p->p_fd, s, &fp);
1395         if (error)
1396                 return (error);
1397         if (*namelen < 0) {
1398                 fdrop(fp);
1399                 return (EINVAL);
1400         }
1401         so = (struct socket *)fp->f_data;
1402         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1403                 fdrop(fp);
1404                 return (ENOTCONN);
1405         }
1406         error = so_pru_peeraddr(so, &sa);
1407         if (error == 0) {
1408                 if (sa == NULL) {
1409                         *namelen = 0;
1410                 } else {
1411                         *namelen = MIN(*namelen, sa->sa_len);
1412                         *name = sa;
1413                 }
1414         }
1415
1416         fdrop(fp);
1417         return (error);
1418 }
1419
1420 /*
1421  * getpeername_args(int fdes, caddr_t asa, int *alen)
1422  *
1423  * Get name of peer for connected socket.
1424  *
1425  * MPALMOSTSAFE
1426  */
1427 int
1428 sys_getpeername(struct getpeername_args *uap)
1429 {
1430         struct sockaddr *sa = NULL;
1431         int error, sa_len;
1432
1433         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1434         if (error)
1435                 return (error);
1436
1437         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1438
1439         if (error == 0)
1440                 error = copyout(sa, uap->asa, sa_len);
1441         if (error == 0)
1442                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1443         if (sa)
1444                 kfree(sa, M_SONAME);
1445         return (error);
1446 }
1447
1448 int
1449 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1450 {
1451         struct sockaddr *sa;
1452         int error;
1453
1454         *namp = NULL;
1455         if (len > SOCK_MAXADDRLEN)
1456                 return ENAMETOOLONG;
1457         if (len < offsetof(struct sockaddr, sa_data[0]))
1458                 return EDOM;
1459         sa = kmalloc(len, M_SONAME, M_WAITOK);
1460         error = copyin(uaddr, sa, len);
1461         if (error) {
1462                 kfree(sa, M_SONAME);
1463         } else {
1464 #if BYTE_ORDER != BIG_ENDIAN
1465                 /*
1466                  * The bind(), connect(), and sendto() syscalls were not
1467                  * versioned for COMPAT_43.  Thus, this check must stay.
1468                  */
1469                 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1470                         sa->sa_family = sa->sa_len;
1471 #endif
1472                 sa->sa_len = len;
1473                 *namp = sa;
1474         }
1475         return error;
1476 }
1477
1478 /*
1479  * Detach a mapped page and release resources back to the system.
1480  * We must release our wiring and if the object is ripped out
1481  * from under the vm_page we become responsible for freeing the
1482  * page.
1483  *
1484  * MPSAFE
1485  */
1486 static void
1487 sf_buf_mfree(void *arg)
1488 {
1489         struct sf_buf *sf = arg;
1490         vm_page_t m;
1491
1492         m = sf_buf_page(sf);
1493         if (sf_buf_free(sf)) {
1494                 /* sf invalid now */
1495                 vm_page_sbusy_drop(m);
1496 #if 0
1497                 if (m->object == NULL &&
1498                     m->wire_count == 0 &&
1499                     (m->flags & PG_NEED_COMMIT) == 0) {
1500                         vm_page_free(m);
1501                 } else {
1502                         vm_page_wakeup(m);
1503                 }
1504 #endif
1505         }
1506 }
1507
1508 /*
1509  * sendfile(2).
1510  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1511  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1512  *
1513  * Send a file specified by 'fd' and starting at 'offset' to a socket
1514  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1515  * nbytes == 0. Optionally add a header and/or trailer to the socket
1516  * output. If specified, write the total number of bytes sent into *sbytes.
1517  *
1518  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1519  * the headers to count against the remaining bytes to be sent from
1520  * the file descriptor.  We may wish to implement a compatibility syscall
1521  * in the future.
1522  *
1523  * MPALMOSTSAFE
1524  */
1525 int
1526 sys_sendfile(struct sendfile_args *uap)
1527 {
1528         struct thread *td = curthread;
1529         struct proc *p = td->td_proc;
1530         struct file *fp;
1531         struct vnode *vp = NULL;
1532         struct sf_hdtr hdtr;
1533         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1534         struct uio auio;
1535         struct mbuf *mheader = NULL;
1536         size_t hbytes = 0;
1537         size_t tbytes;
1538         off_t hdtr_size = 0;
1539         off_t sbytes;
1540         int error;
1541
1542         KKASSERT(p);
1543
1544         /*
1545          * Do argument checking. Must be a regular file in, stream
1546          * type and connected socket out, positive offset.
1547          */
1548         fp = holdfp(p->p_fd, uap->fd, FREAD);
1549         if (fp == NULL) {
1550                 return (EBADF);
1551         }
1552         if (fp->f_type != DTYPE_VNODE) {
1553                 fdrop(fp);
1554                 return (EINVAL);
1555         }
1556         vp = (struct vnode *)fp->f_data;
1557         vref(vp);
1558         fdrop(fp);
1559
1560         /*
1561          * If specified, get the pointer to the sf_hdtr struct for
1562          * any headers/trailers.
1563          */
1564         if (uap->hdtr) {
1565                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1566                 if (error)
1567                         goto done;
1568                 /*
1569                  * Send any headers.
1570                  */
1571                 if (hdtr.headers) {
1572                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1573                                              hdtr.hdr_cnt, &hbytes);
1574                         if (error)
1575                                 goto done;
1576                         auio.uio_iov = iov;
1577                         auio.uio_iovcnt = hdtr.hdr_cnt;
1578                         auio.uio_offset = 0;
1579                         auio.uio_segflg = UIO_USERSPACE;
1580                         auio.uio_rw = UIO_WRITE;
1581                         auio.uio_td = td;
1582                         auio.uio_resid = hbytes;
1583
1584                         mheader = m_uiomove(&auio);
1585
1586                         iovec_free(&iov, aiov);
1587                         if (mheader == NULL)
1588                                 goto done;
1589                 }
1590         }
1591
1592         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1593                               &sbytes, uap->flags);
1594         if (error)
1595                 goto done;
1596
1597         /*
1598          * Send trailers. Wimp out and use writev(2).
1599          */
1600         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1601                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1602                                      hdtr.trl_cnt, &auio.uio_resid);
1603                 if (error)
1604                         goto done;
1605                 auio.uio_iov = iov;
1606                 auio.uio_iovcnt = hdtr.trl_cnt;
1607                 auio.uio_offset = 0;
1608                 auio.uio_segflg = UIO_USERSPACE;
1609                 auio.uio_rw = UIO_WRITE;
1610                 auio.uio_td = td;
1611
1612                 tbytes = 0;     /* avoid gcc warnings */
1613                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1614
1615                 iovec_free(&iov, aiov);
1616                 if (error)
1617                         goto done;
1618                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1619         }
1620
1621 done:
1622         if (vp)
1623                 vrele(vp);
1624         if (uap->sbytes != NULL) {
1625                 sbytes += hdtr_size;
1626                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1627         }
1628         return (error);
1629 }
1630
1631 int
1632 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1633               struct mbuf *mheader, off_t *sbytes, int flags)
1634 {
1635         struct thread *td = curthread;
1636         struct proc *p = td->td_proc;
1637         struct vm_object *obj;
1638         struct socket *so;
1639         struct file *fp;
1640         struct mbuf *m, *mp;
1641         struct sf_buf *sf;
1642         struct vm_page *pg;
1643         off_t off, xfsize, xbytes;
1644         off_t hbytes = 0;
1645         int error = 0;
1646
1647         if (vp->v_type != VREG) {
1648                 error = EINVAL;
1649                 goto done0;
1650         }
1651         if ((obj = vp->v_object) == NULL) {
1652                 error = EINVAL;
1653                 goto done0;
1654         }
1655         error = holdsock(p->p_fd, sfd, &fp);
1656         if (error)
1657                 goto done0;
1658         so = (struct socket *)fp->f_data;
1659         if (so->so_type != SOCK_STREAM) {
1660                 error = EINVAL;
1661                 goto done1;
1662         }
1663         if ((so->so_state & SS_ISCONNECTED) == 0) {
1664                 error = ENOTCONN;
1665                 goto done1;
1666         }
1667         if (offset < 0) {
1668                 error = EINVAL;
1669                 goto done1;
1670         }
1671
1672         /*
1673          * preallocation is required for asynchronous passing of mbufs,
1674          * otherwise we can wind up building up an infinite number of
1675          * mbufs during the asynchronous latency.
1676          */
1677         if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) {
1678                 error = EINVAL;
1679                 goto done1;
1680         }
1681
1682         *sbytes = 0;
1683         xbytes = 0;
1684
1685         /*
1686          * Protect against multiple writers to the socket.
1687          * We need at least a shared lock on the VM object
1688          */
1689         ssb_lock(&so->so_snd, M_WAITOK);
1690         vm_object_hold_shared(obj);
1691
1692         /*
1693          * Loop through the pages in the file, starting with the requested
1694          * offset. Get a file page (do I/O if necessary), map the file page
1695          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1696          * it on the socket.
1697          */
1698         for (off = offset; ;
1699              off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) {
1700                 vm_pindex_t pindex;
1701                 vm_offset_t pgoff;
1702                 long space;
1703                 int loops;
1704
1705                 pindex = OFF_TO_IDX(off);
1706                 loops = 0;
1707
1708 retry_lookup:
1709                 /*
1710                  * Calculate the amount to transfer. Not to exceed a page,
1711                  * the EOF, or the passed in nbytes.
1712                  */
1713                 xfsize = vp->v_filesize - off;
1714                 if (xfsize > PAGE_SIZE)
1715                         xfsize = PAGE_SIZE;
1716                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1717                 if (PAGE_SIZE - pgoff < xfsize)
1718                         xfsize = PAGE_SIZE - pgoff;
1719                 if (nbytes && xfsize > (nbytes - xbytes))
1720                         xfsize = nbytes - xbytes;
1721                 if (xfsize <= 0)
1722                         break;
1723                 /*
1724                  * Optimize the non-blocking case by looking at the socket space
1725                  * before going to the extra work of constituting the sf_buf.
1726                  */
1727                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1728                         space = ssb_space_prealloc(&so->so_snd);
1729                 else
1730                         space = ssb_space(&so->so_snd);
1731
1732                 if ((fp->f_flag & FNONBLOCK) && space <= 0) {
1733                         if (so->so_state & SS_CANTSENDMORE)
1734                                 error = EPIPE;
1735                         else
1736                                 error = EAGAIN;
1737                         goto done;
1738                 }
1739
1740                 /*
1741                  * Attempt to look up the page.
1742                  *
1743                  * Try to find the data using a shared vm_object token and
1744                  * vm_page_lookup_sbusy_try() first.
1745                  *
1746                  * If data is missing, use a UIO_NOCOPY VOP_READ to load
1747                  * the missing data and loop back up.  We avoid all sorts
1748                  * of problems by not trying to hold onto the page during
1749                  * the I/O.
1750                  *
1751                  * NOTE: The soft-busy will temporary block filesystem
1752                  *       truncation operations when a file is removed
1753                  *       while the sendfile is running.
1754                  */
1755                 pg = vm_page_lookup_sbusy_try(obj, pindex, pgoff, xfsize);
1756                 if (pg == NULL) {
1757                         struct uio auio;
1758                         struct iovec aiov;
1759                         int bsize;
1760
1761                         if (++loops > 100000) {
1762                                 kprintf("sendfile: VOP operation failed "
1763                                         "to retain page\n");
1764                                 error = EIO;
1765                                 goto done;
1766                         }
1767
1768                         vm_object_drop(obj);
1769                         bsize = vp->v_mount->mnt_stat.f_iosize;
1770                         auio.uio_iov = &aiov;
1771                         auio.uio_iovcnt = 1;
1772                         aiov.iov_base = 0;
1773                         aiov.iov_len = MAXBSIZE;
1774                         auio.uio_resid = MAXBSIZE;
1775                         auio.uio_offset = trunc_page(off);
1776                         auio.uio_segflg = UIO_NOCOPY;
1777                         auio.uio_rw = UIO_READ;
1778                         auio.uio_td = td;
1779
1780                         vn_lock(vp, LK_SHARED | LK_RETRY);
1781                         error = VOP_READ(vp, &auio,
1782                                          IO_VMIO | ((MAXBSIZE / bsize) << 16),
1783                                          td->td_ucred);
1784                         vn_unlock(vp);
1785                         vm_object_hold_shared(obj);
1786
1787                         if (error)
1788                                 goto done;
1789                         goto retry_lookup;
1790                 }
1791
1792                 /*
1793                  * Get a sendfile buf. We usually wait as long as necessary,
1794                  * but this wait can be interrupted.
1795                  */
1796                 if ((sf = sf_buf_alloc(pg)) == NULL) {
1797                         vm_page_sbusy_drop(pg);
1798                         /* vm_page_try_to_free(pg); */
1799                         error = EINTR;
1800                         goto done;
1801                 }
1802
1803                 /*
1804                  * Get an mbuf header and set it up as having external storage.
1805                  */
1806                 MGETHDR(m, M_WAITOK, MT_DATA);
1807                 if (m == NULL) {
1808                         error = ENOBUFS;
1809                         vm_page_sbusy_drop(pg);
1810                         /* vm_page_try_to_free(pg); */
1811                         sf_buf_free(sf);
1812                         goto done;
1813                 }
1814
1815                 m->m_ext.ext_free = sf_buf_mfree;
1816                 m->m_ext.ext_ref = sf_buf_ref;
1817                 m->m_ext.ext_arg = sf;
1818                 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1819                 m->m_ext.ext_size = PAGE_SIZE;
1820                 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1821                 m->m_flags |= M_EXT;
1822                 m->m_pkthdr.len = m->m_len = xfsize;
1823                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1824
1825                 if (mheader != NULL) {
1826                         hbytes = mheader->m_pkthdr.len;
1827                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1828                         m_cat(mheader, m);
1829                         m = mheader;
1830                         mheader = NULL;
1831                 } else {
1832                         hbytes = 0;
1833                 }
1834
1835                 /*
1836                  * Add the buffer to the socket buffer chain.
1837                  */
1838                 crit_enter();
1839 retry_space:
1840                 /*
1841                  * Make sure that the socket is still able to take more data.
1842                  * CANTSENDMORE being true usually means that the connection
1843                  * was closed. so_error is true when an error was sensed after
1844                  * a previous send.
1845                  * The state is checked after the page mapping and buffer
1846                  * allocation above since those operations may block and make
1847                  * any socket checks stale. From this point forward, nothing
1848                  * blocks before the pru_send (or more accurately, any blocking
1849                  * results in a loop back to here to re-check).
1850                  */
1851                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1852                         if (so->so_state & SS_CANTSENDMORE) {
1853                                 error = EPIPE;
1854                         } else {
1855                                 error = so->so_error;
1856                                 so->so_error = 0;
1857                         }
1858                         m_freem(m);
1859                         crit_exit();
1860                         goto done;
1861                 }
1862                 /*
1863                  * Wait for socket space to become available. We do this just
1864                  * after checking the connection state above in order to avoid
1865                  * a race condition with ssb_wait().
1866                  */
1867                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1868                         space = ssb_space_prealloc(&so->so_snd);
1869                 else
1870                         space = ssb_space(&so->so_snd);
1871
1872                 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1873                         if (fp->f_flag & FNONBLOCK) {
1874                                 m_freem(m);
1875                                 crit_exit();
1876                                 error = EAGAIN;
1877                                 goto done;
1878                         }
1879                         error = ssb_wait(&so->so_snd);
1880                         /*
1881                          * An error from ssb_wait usually indicates that we've
1882                          * been interrupted by a signal. If we've sent anything
1883                          * then return bytes sent, otherwise return the error.
1884                          */
1885                         if (error) {
1886                                 m_freem(m);
1887                                 crit_exit();
1888                                 goto done;
1889                         }
1890                         goto retry_space;
1891                 }
1892
1893                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1894                         for (mp = m; mp != NULL; mp = mp->m_next)
1895                                 ssb_preallocstream(&so->so_snd, mp);
1896                 }
1897                 if (use_sendfile_async)
1898                         error = so_pru_senda(so, 0, m, NULL, NULL, td);
1899                 else
1900                         error = so_pru_send(so, 0, m, NULL, NULL, td);
1901
1902                 crit_exit();
1903                 if (error)
1904                         goto done;
1905         }
1906         if (mheader != NULL) {
1907                 *sbytes += mheader->m_pkthdr.len;
1908
1909                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1910                         for (mp = mheader; mp != NULL; mp = mp->m_next)
1911                                 ssb_preallocstream(&so->so_snd, mp);
1912                 }
1913                 if (use_sendfile_async)
1914                         error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1915                 else
1916                         error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1917
1918                 mheader = NULL;
1919         }
1920 done:
1921         vm_object_drop(obj);
1922         ssb_unlock(&so->so_snd);
1923 done1:
1924         fdrop(fp);
1925 done0:
1926         if (mheader != NULL)
1927                 m_freem(mheader);
1928         return (error);
1929 }