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/sysmsg.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 #include <sys/jail.h>
  61 #ifdef KTRACE
  62 #include <sys/ktrace.h>
  63 #endif
  64 #include <vm/vm.h>
  65 #include <vm/vm_object.h>
  66 #include <vm/vm_page.h>
  67 #include <vm/vm_pageout.h>
  68 #include <vm/vm_kern.h>
  69 #include <vm/vm_extern.h>
  70 #include <sys/file2.h>
  71 #include <sys/signalvar.h>
  72 #include <sys/serialize.h>
  73
  74 #include <sys/thread2.h>
  75 #include <sys/msgport2.h>
  76 #include <sys/socketvar2.h>
  77 #include <net/netmsg2.h>
  78 #include <vm/vm_page2.h>
  79
  80 extern int use_soaccept_pred_fast;
  81 extern int use_sendfile_async;
  82 extern int use_soconnect_async;
  83
  84 /*
  85  * System call interface to the socket abstraction.
  86  */
  87
  88 extern  struct fileops socketops;
  89
  90 /*
  91  * socket_args(int domain, int type, int protocol)
  92  */
  93 int
  94 kern_socket(int domain, int type, int protocol, int *res)
  95 {
  96         struct thread *td = curthread;
  97         struct filedesc *fdp = td->td_proc->p_fd;
  98         struct socket *so;
  99         struct file *fp;
 100         int fd, error;
 101         u_int fflags = 0;
 102         int oflags = 0;
 103
 104         KKASSERT(td->td_lwp);
 105
 106         if (type & SOCK_NONBLOCK) {
 107                 type &= ~SOCK_NONBLOCK;
 108                 fflags |= FNONBLOCK;
 109         }
 110         if (type & SOCK_CLOEXEC) {
 111                 type &= ~SOCK_CLOEXEC;
 112                 oflags |= O_CLOEXEC;
 113         }
 114
 115         error = falloc(td->td_lwp, &fp, &fd);
 116         if (error)
 117                 return (error);
 118         error = socreate(domain, &so, type, protocol, td);
 119         if (error) {
 120                 fsetfd(fdp, NULL, fd);
 121         } else {
 122                 fp->f_type = DTYPE_SOCKET;
 123                 fp->f_flag = FREAD | FWRITE | fflags;
 124                 fp->f_ops = &socketops;
 125                 fp->f_data = so;
 126                 if (oflags & O_CLOEXEC)
 127                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 128                 *res = fd;
 129                 fsetfd(fdp, fp, fd);
 130         }
 131         fdrop(fp);
 132         return (error);
 133 }
 134
 135 /*
 136  * MPALMOSTSAFE
 137  */
 138 int
 139 sys_socket(struct sysmsg *sysmsg, const struct socket_args *uap)
 140 {
 141         int error;
 142
 143         error = kern_socket(uap->domain, uap->type, uap->protocol,
 144                             &sysmsg->sysmsg_iresult);
 145
 146         return (error);
 147 }
 148
 149 int
 150 kern_bind(int s, struct sockaddr *sa)
 151 {
 152         struct thread *td = curthread;
 153         struct file *fp;
 154         int error;
 155
 156         error = holdsock(td, s, &fp);
 157         if (error)
 158                 return (error);
 159         error = sobind((struct socket *)fp->f_data, sa, td);
 160         dropfp(td, s, fp);
 161
 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 sysmsg *sysmsg, const 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         if (!prison_remote_ip(curthread, sa)) {
 180                 kfree(sa, M_SONAME);
 181                 return EAFNOSUPPORT;
 182         }
 183         error = kern_bind(uap->s, sa);
 184         kfree(sa, M_SONAME);
 185
 186         return (error);
 187 }
 188
 189 int
 190 kern_listen(int s, int backlog)
 191 {
 192         struct thread *td = curthread;
 193         struct file *fp;
 194         int error;
 195
 196         error = holdsock(td, s, &fp);
 197         if (error)
 198                 return (error);
 199         error = solisten((struct socket *)fp->f_data, backlog, td);
 200         dropfp(td, s, fp);
 201
 202         return (error);
 203 }
 204
 205 /*
 206  * listen_args(int s, int backlog)
 207  *
 208  * MPALMOSTSAFE
 209  */
 210 int
 211 sys_listen(struct sysmsg *sysmsg, const struct listen_args *uap)
 212 {
 213         int error;
 214
 215         error = kern_listen(uap->s, uap->backlog);
 216         return (error);
 217 }
 218
 219 /*
 220  * Returns the accepted socket as well.
 221  *
 222  * NOTE!  The sockets sitting on so_comp/so_incomp might have 0 refs, the
 223  *        pool token is absolutely required to avoid a sofree() race,
 224  *        as well as to avoid tailq handling races.
 225  */
 226 static boolean_t
 227 soaccept_predicate(struct netmsg_so_notify *msg)
 228 {
 229         struct socket *head = msg->base.nm_so;
 230         struct socket *so;
 231
 232         if (head->so_error != 0) {
 233                 msg->base.lmsg.ms_error = head->so_error;
 234                 return (TRUE);
 235         }
 236         lwkt_getpooltoken(head);
 237         if (!TAILQ_EMPTY(&head->so_comp)) {
 238                 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
 239                 so = TAILQ_FIRST(&head->so_comp);
 240                 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
 241                 TAILQ_REMOVE(&head->so_comp, so, so_list);
 242                 head->so_qlen--;
 243                 soclrstate(so, SS_COMP);
 244
 245                 /*
 246                  * Keep a reference before clearing the so_head
 247                  * to avoid racing socket close in netisr.
 248                  */
 249                 soreference(so);
 250                 so->so_head = NULL;
 251
 252                 lwkt_relpooltoken(head);
 253
 254                 msg->base.lmsg.ms_error = 0;
 255                 msg->base.nm_so = so;
 256                 return (TRUE);
 257         }
 258         lwkt_relpooltoken(head);
 259         if (head->so_state & SS_CANTRCVMORE) {
 260                 msg->base.lmsg.ms_error = ECONNABORTED;
 261                 return (TRUE);
 262         }
 263         if (msg->nm_fflags & FNONBLOCK) {
 264                 msg->base.lmsg.ms_error = EWOULDBLOCK;
 265                 return (TRUE);
 266         }
 267
 268         return (FALSE);
 269 }
 270
 271 /*
 272  * The second argument to kern_accept() is a handle to a struct sockaddr.
 273  * This allows kern_accept() to return a pointer to an allocated struct
 274  * sockaddr which must be freed later with FREE().  The caller must
 275  * initialize *name to NULL.
 276  */
 277 int
 278 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res,
 279     int sockflags)
 280 {
 281         struct thread *td = curthread;
 282         struct filedesc *fdp = td->td_proc->p_fd;
 283         struct file *lfp = NULL;
 284         struct file *nfp = NULL;
 285         struct sockaddr *sa;
 286         struct socket *head, *so;
 287         struct netmsg_so_notify msg;
 288         int fd;
 289         u_int fflag;            /* type must match fp->f_flag */
 290         int error, tmp;
 291
 292         *res = -1;
 293         if (name && namelen && *namelen < 0)
 294                 return (EINVAL);
 295
 296         error = holdsock(td, s, &lfp);
 297         if (error)
 298                 return (error);
 299
 300         error = falloc(td->td_lwp, &nfp, &fd);
 301         if (error) {            /* Probably ran out of file descriptors. */
 302                 fdrop(lfp);
 303                 return (error);
 304         }
 305         head = (struct socket *)lfp->f_data;
 306         if ((head->so_options & SO_ACCEPTCONN) == 0) {
 307                 error = EINVAL;
 308                 goto done;
 309         }
 310
 311         if (fflags & O_FBLOCKING)
 312                 fflags |= lfp->f_flag & ~FNONBLOCK;
 313         else if (fflags & O_FNONBLOCKING)
 314                 fflags |= lfp->f_flag | FNONBLOCK;
 315         else
 316                 fflags = lfp->f_flag;
 317
 318         if (use_soaccept_pred_fast) {
 319                 boolean_t pred;
 320
 321                 /* Initialize necessary parts for soaccept_predicate() */
 322                 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
 323                 msg.nm_fflags = fflags;
 324
 325                 lwkt_getpooltoken(head);
 326                 pred = soaccept_predicate(&msg);
 327                 lwkt_relpooltoken(head);
 328
 329                 if (pred) {
 330                         error = msg.base.lmsg.ms_error;
 331                         if (error)
 332                                 goto done;
 333                         else
 334                                 goto accepted;
 335                 }
 336         }
 337
 338         /* optimize for uniprocessor case later XXX JH */
 339         netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
 340                               0, netmsg_so_notify, netmsg_so_notify_doabort);
 341         msg.nm_predicate = soaccept_predicate;
 342         msg.nm_fflags = fflags;
 343         msg.nm_etype = NM_REVENT;
 344         error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
 345         if (error)
 346                 goto done;
 347
 348 accepted:
 349         /*
 350          * At this point we have the connection that's ready to be accepted.
 351          *
 352          * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
 353          *       to eat the ref and turn it into a descriptor.
 354          */
 355         so = msg.base.nm_so;
 356
 357         fflag = lfp->f_flag;
 358
 359         /* connection has been removed from the listen queue */
 360         KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
 361
 362         if (sockflags & SOCK_KERN_NOINHERIT) {
 363                 fflag &= ~(FASYNC | FNONBLOCK);
 364                 if (sockflags & SOCK_NONBLOCK)
 365                         fflag |= FNONBLOCK;
 366         } else {
 367                 if (head->so_sigio != NULL)
 368                         fsetown(fgetown(&head->so_sigio), &so->so_sigio);
 369         }
 370
 371         nfp->f_type = DTYPE_SOCKET;
 372         nfp->f_flag = fflag;
 373         nfp->f_ops = &socketops;
 374         nfp->f_data = so;
 375         /* Sync socket async state with file flags */
 376         tmp = fflag & FASYNC;
 377         fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
 378
 379         sa = NULL;
 380         if (so->so_faddr != NULL) {
 381                 sa = so->so_faddr;
 382                 so->so_faddr = NULL;
 383
 384                 soaccept_generic(so);
 385                 error = 0;
 386         } else {
 387                 error = soaccept(so, &sa);
 388         }
 389
 390         /*
 391          * Set the returned name and namelen as applicable.  Set the returned
 392          * namelen to 0 for older code which might ignore the return value
 393          * from accept.
 394          */
 395         if (error == 0) {
 396                 if (sa && name && namelen) {
 397                         if (*namelen > sa->sa_len)
 398                                 *namelen = sa->sa_len;
 399                         *name = sa;
 400                 } else {
 401                         if (sa)
 402                                 kfree(sa, M_SONAME);
 403                 }
 404         }
 405
 406 done:
 407         /*
 408          * If an error occured clear the reserved descriptor, else associate
 409          * nfp with it.
 410          *
 411          * Note that *res is normally ignored if an error is returned but
 412          * a syscall message will still have access to the result code.
 413          */
 414         if (error) {
 415                 fsetfd(fdp, NULL, fd);
 416         } else {
 417                 if (sockflags & SOCK_CLOEXEC)
 418                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 419                 *res = fd;
 420                 fsetfd(fdp, nfp, fd);
 421         }
 422         fdrop(nfp);
 423         dropfp(td, s, lfp);
 424
 425         return (error);
 426 }
 427
 428 /*
 429  * accept(int s, caddr_t name, int *anamelen)
 430  *
 431  * MPALMOSTSAFE
 432  */
 433 int
 434 sys_accept(struct sysmsg *sysmsg, const struct accept_args *uap)
 435 {
 436         struct sockaddr *sa = NULL;
 437         int sa_len;
 438         int error;
 439
 440         if (uap->name) {
 441                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 442                 if (error)
 443                         return (error);
 444
 445                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 446                                     &sysmsg->sysmsg_iresult, 0);
 447
 448                 if (error == 0) {
 449                         prison_local_ip(curthread, sa);
 450                         error = copyout(sa, uap->name, sa_len);
 451                 }
 452                 if (error == 0) {
 453                         error = copyout(&sa_len, uap->anamelen,
 454                                         sizeof(*uap->anamelen));
 455                 }
 456                 if (sa)
 457                         kfree(sa, M_SONAME);
 458         } else {
 459                 error = kern_accept(uap->s, 0, NULL, 0,
 460                                     &sysmsg->sysmsg_iresult, 0);
 461         }
 462         return (error);
 463 }
 464
 465 /*
 466  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
 467  *
 468  * MPALMOSTSAFE
 469  */
 470 int
 471 sys_extaccept(struct sysmsg *sysmsg, const struct extaccept_args *uap)
 472 {
 473         struct sockaddr *sa = NULL;
 474         int sa_len;
 475         int error;
 476         int fflags = uap->flags & O_FMASK;
 477
 478         if (uap->name) {
 479                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 480                 if (error)
 481                         return (error);
 482
 483                 error = kern_accept(uap->s, fflags, &sa, &sa_len,
 484                                     &sysmsg->sysmsg_iresult, 0);
 485
 486                 if (error == 0) {
 487                         prison_local_ip(curthread, sa);
 488                         error = copyout(sa, uap->name, sa_len);
 489                 }
 490                 if (error == 0) {
 491                         error = copyout(&sa_len, uap->anamelen,
 492                             sizeof(*uap->anamelen));
 493                 }
 494                 if (sa)
 495                         kfree(sa, M_SONAME);
 496         } else {
 497                 error = kern_accept(uap->s, fflags, NULL, 0,
 498                                     &sysmsg->sysmsg_iresult, 0);
 499         }
 500         return (error);
 501 }
 502
 503 /*
 504  * accept4(int s, caddr_t name, int *anamelen, int flags)
 505  *
 506  * MPALMOSTSAFE
 507  */
 508 int
 509 sys_accept4(struct sysmsg *sysmsg, const struct accept4_args *uap)
 510 {
 511         struct sockaddr *sa = NULL;
 512         int sa_len;
 513         int error;
 514         int sockflags;
 515
 516         if (uap->flags & ~(SOCK_NONBLOCK | SOCK_CLOEXEC))
 517                 return (EINVAL);
 518         sockflags = uap->flags | SOCK_KERN_NOINHERIT;
 519
 520         if (uap->name) {
 521                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 522                 if (error)
 523                         return (error);
 524
 525                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 526                                     &sysmsg->sysmsg_iresult, sockflags);
 527
 528                 if (error == 0) {
 529                         prison_local_ip(curthread, sa);
 530                         error = copyout(sa, uap->name, sa_len);
 531                 }
 532                 if (error == 0) {
 533                         error = copyout(&sa_len, uap->anamelen,
 534                                         sizeof(*uap->anamelen));
 535                 }
 536                 if (sa)
 537                         kfree(sa, M_SONAME);
 538         } else {
 539                 error = kern_accept(uap->s, 0, NULL, 0,
 540                                     &sysmsg->sysmsg_iresult, sockflags);
 541         }
 542         return (error);
 543 }
 544
 545 /*
 546  * Returns TRUE if predicate satisfied.
 547  */
 548 static boolean_t
 549 soconnected_predicate(struct netmsg_so_notify *msg)
 550 {
 551         struct socket *so = msg->base.nm_so;
 552
 553         /* check predicate */
 554         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 555                 msg->base.lmsg.ms_error = so->so_error;
 556                 return (TRUE);
 557         }
 558
 559         return (FALSE);
 560 }
 561
 562 int
 563 kern_connect(int s, int fflags, struct sockaddr *sa)
 564 {
 565         struct thread *td = curthread;
 566         struct file *fp;
 567         struct socket *so;
 568         int error, interrupted = 0;
 569
 570         error = holdsock(td, s, &fp);
 571         if (error)
 572                 return (error);
 573         so = (struct socket *)fp->f_data;
 574
 575         if (fflags & O_FBLOCKING)
 576                 /* fflags &= ~FNONBLOCK; */;
 577         else if (fflags & O_FNONBLOCKING)
 578                 fflags |= FNONBLOCK;
 579         else
 580                 fflags = fp->f_flag;
 581
 582         if (so->so_state & SS_ISCONNECTING) {
 583                 error = EALREADY;
 584                 goto done;
 585         }
 586         error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE);
 587         if (error)
 588                 goto bad;
 589         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 590                 error = EINPROGRESS;
 591                 goto done;
 592         }
 593         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 594                 struct netmsg_so_notify msg;
 595
 596                 netmsg_init_abortable(&msg.base, so,
 597                                       &curthread->td_msgport,
 598                                       0,
 599                                       netmsg_so_notify,
 600                                       netmsg_so_notify_doabort);
 601                 msg.nm_predicate = soconnected_predicate;
 602                 msg.nm_etype = NM_REVENT;
 603                 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
 604                 if (error == EINTR || error == ERESTART)
 605                         interrupted = 1;
 606         }
 607         if (error == 0) {
 608                 error = so->so_error;
 609                 so->so_error = 0;
 610         }
 611 bad:
 612         if (!interrupted)
 613                 soclrstate(so, SS_ISCONNECTING);
 614         if (error == ERESTART)
 615                 error = EINTR;
 616 done:
 617         dropfp(td, s, fp);
 618
 619         return (error);
 620 }
 621
 622 /*
 623  * connect_args(int s, caddr_t name, int namelen)
 624  *
 625  * MPALMOSTSAFE
 626  */
 627 int
 628 sys_connect(struct sysmsg *sysmsg, const struct connect_args *uap)
 629 {
 630         struct sockaddr *sa;
 631         int error;
 632
 633         error = getsockaddr(&sa, uap->name, uap->namelen);
 634         if (error)
 635                 return (error);
 636         if (!prison_remote_ip(curthread, sa)) {
 637                 kfree(sa, M_SONAME);
 638                 return EAFNOSUPPORT;
 639         }
 640         error = kern_connect(uap->s, 0, sa);
 641         kfree(sa, M_SONAME);
 642
 643         return (error);
 644 }
 645
 646 /*
 647  * connect_args(int s, int fflags, caddr_t name, int namelen)
 648  *
 649  * MPALMOSTSAFE
 650  */
 651 int
 652 sys_extconnect(struct sysmsg *sysmsg, const struct extconnect_args *uap)
 653 {
 654         struct sockaddr *sa;
 655         int error;
 656         int fflags = uap->flags & O_FMASK;
 657
 658         error = getsockaddr(&sa, uap->name, uap->namelen);
 659         if (error)
 660                 return (error);
 661         if (!prison_remote_ip(curthread, sa)) {
 662                 kfree(sa, M_SONAME);
 663                 return EAFNOSUPPORT;
 664         }
 665         error = kern_connect(uap->s, fflags, sa);
 666         kfree(sa, M_SONAME);
 667
 668         return (error);
 669 }
 670
 671 int
 672 kern_socketpair(int domain, int type, int protocol, int *sv)
 673 {
 674         struct thread *td = curthread;
 675         struct filedesc *fdp;
 676         struct file *fp1, *fp2;
 677         struct socket *so1, *so2;
 678         struct ucred *cred = curthread->td_ucred;
 679         int fd1, fd2, error;
 680         u_int fflags = 0;
 681         int oflags = 0;
 682
 683         if (type & SOCK_NONBLOCK) {
 684                 type &= ~SOCK_NONBLOCK;
 685                 fflags |= FNONBLOCK;
 686         }
 687         if (type & SOCK_CLOEXEC) {
 688                 type &= ~SOCK_CLOEXEC;
 689                 oflags |= O_CLOEXEC;
 690         }
 691
 692         fdp = td->td_proc->p_fd;
 693         error = socreate(domain, &so1, type, protocol, td);
 694         if (error)
 695                 return (error);
 696         error = socreate(domain, &so2, type, protocol, td);
 697         if (error)
 698                 goto free1;
 699         error = falloc(td->td_lwp, &fp1, &fd1);
 700         if (error)
 701                 goto free2;
 702         sv[0] = fd1;
 703         fp1->f_data = so1;
 704         error = falloc(td->td_lwp, &fp2, &fd2);
 705         if (error)
 706                 goto free3;
 707         fp2->f_data = so2;
 708         sv[1] = fd2;
 709         error = soconnect2(so1, so2, cred);
 710         if (error)
 711                 goto free4;
 712         if (type == SOCK_DGRAM) {
 713                 /*
 714                  * Datagram socket connection is asymmetric.
 715                  */
 716                  error = soconnect2(so2, so1, cred);
 717                  if (error)
 718                         goto free4;
 719         }
 720         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 721         fp1->f_flag = fp2->f_flag = FREAD|FWRITE|fflags;
 722         fp1->f_ops = fp2->f_ops = &socketops;
 723         if (oflags & O_CLOEXEC) {
 724                 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE;
 725                 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE;
 726         }
 727         fsetfd(fdp, fp1, fd1);
 728         fsetfd(fdp, fp2, fd2);
 729         fdrop(fp1);
 730         fdrop(fp2);
 731         return (error);
 732 free4:
 733         fsetfd(fdp, NULL, fd2);
 734         fdrop(fp2);
 735 free3:
 736         fsetfd(fdp, NULL, fd1);
 737         fdrop(fp1);
 738 free2:
 739         (void)soclose(so2, 0);
 740 free1:
 741         (void)soclose(so1, 0);
 742         return (error);
 743 }
 744
 745 /*
 746  * socketpair(int domain, int type, int protocol, int *rsv)
 747  */
 748 int
 749 sys_socketpair(struct sysmsg *sysmsg, const struct socketpair_args *uap)
 750 {
 751         int error, sockv[2];
 752
 753         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 754
 755         if (error == 0) {
 756                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 757
 758                 if (error != 0) {
 759                         kern_close(sockv[0]);
 760                         kern_close(sockv[1]);
 761                 }
 762         }
 763
 764         return (error);
 765 }
 766
 767 int
 768 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 769              struct mbuf *control, int flags, size_t *res)
 770 {
 771         struct thread *td = curthread;
 772         struct lwp *lp = td->td_lwp;
 773         struct proc *p = td->td_proc;
 774         struct file *fp;
 775         size_t len;
 776         int error;
 777         struct socket *so;
 778 #ifdef KTRACE
 779         struct iovec *ktriov = NULL;
 780         struct uio ktruio;
 781 #endif
 782
 783         error = holdsock(td, s, &fp);
 784         if (error)
 785                 return (error);
 786 #ifdef KTRACE
 787         if (KTRPOINT(td, KTR_GENIO)) {
 788                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 789
 790                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 791                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 792                 ktruio = *auio;
 793         }
 794 #endif
 795         len = auio->uio_resid;
 796         so = (struct socket *)fp->f_data;
 797         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 798                 if (fp->f_flag & FNONBLOCK)
 799                         flags |= MSG_FNONBLOCKING;
 800         }
 801         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 802         if (error) {
 803                 if (auio->uio_resid != len && (error == ERESTART ||
 804                     error == EINTR || error == EWOULDBLOCK))
 805                         error = 0;
 806                 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
 807                     !(so->so_options & SO_NOSIGPIPE))
 808                         lwpsignal(p, lp, SIGPIPE);
 809         }
 810 #ifdef KTRACE
 811         if (ktriov != NULL) {
 812                 if (error == 0) {
 813                         ktruio.uio_iov = ktriov;
 814                         ktruio.uio_resid = len - auio->uio_resid;
 815                         ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
 816                 }
 817                 kfree(ktriov, M_TEMP);
 818         }
 819 #endif
 820         if (error == 0)
 821                 *res  = len - auio->uio_resid;
 822         dropfp(td, s, fp);
 823
 824         return (error);
 825 }
 826
 827 /*
 828  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
 829  *
 830  * MPALMOSTSAFE
 831  */
 832 int
 833 sys_sendto(struct sysmsg *sysmsg, const struct sendto_args *uap)
 834 {
 835         struct thread *td = curthread;
 836         struct uio auio;
 837         struct iovec aiov;
 838         struct sockaddr *sa = NULL;
 839         int error;
 840
 841         if (uap->to) {
 842                 error = getsockaddr(&sa, uap->to, uap->tolen);
 843                 if (error)
 844                         return (error);
 845                 if (!prison_remote_ip(curthread, sa)) {
 846                         kfree(sa, M_SONAME);
 847                         return EAFNOSUPPORT;
 848                 }
 849         }
 850         aiov.iov_base = uap->buf;
 851         aiov.iov_len = uap->len;
 852         auio.uio_iov = &aiov;
 853         auio.uio_iovcnt = 1;
 854         auio.uio_offset = 0;
 855         auio.uio_resid = uap->len;
 856         auio.uio_segflg = UIO_USERSPACE;
 857         auio.uio_rw = UIO_WRITE;
 858         auio.uio_td = td;
 859
 860         error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
 861                              &sysmsg->sysmsg_szresult);
 862
 863         if (sa)
 864                 kfree(sa, M_SONAME);
 865         return (error);
 866 }
 867
 868 /*
 869  * sendmsg_args(int s, caddr_t msg, int flags)
 870  *
 871  * MPALMOSTSAFE
 872  */
 873 int
 874 sys_sendmsg(struct sysmsg *sysmsg, const struct sendmsg_args *uap)
 875 {
 876         struct thread *td = curthread;
 877         struct msghdr msg;
 878         struct uio auio;
 879         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 880         struct sockaddr *sa = NULL;
 881         struct mbuf *control = NULL;
 882         int error;
 883
 884         error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
 885         if (error)
 886                 return (error);
 887
 888         /*
 889          * Conditionally copyin msg.msg_name.
 890          */
 891         if (msg.msg_name) {
 892                 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
 893                 if (error)
 894                         return (error);
 895                 if (!prison_remote_ip(curthread, sa)) {
 896                         kfree(sa, M_SONAME);
 897                         return EAFNOSUPPORT;
 898                 }
 899         }
 900
 901         /*
 902          * Populate auio.
 903          */
 904         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
 905                              &auio.uio_resid);
 906         if (error)
 907                 goto cleanup2;
 908         auio.uio_iov = iov;
 909         auio.uio_iovcnt = msg.msg_iovlen;
 910         auio.uio_offset = 0;
 911         auio.uio_segflg = UIO_USERSPACE;
 912         auio.uio_rw = UIO_WRITE;
 913         auio.uio_td = td;
 914
 915         /*
 916          * Conditionally copyin msg.msg_control.
 917          */
 918         if (msg.msg_control) {
 919                 if (msg.msg_controllen < sizeof(struct cmsghdr) ||
 920                     msg.msg_controllen > MLEN) {
 921                         error = EINVAL;
 922                         goto cleanup;
 923                 }
 924                 control = m_get(M_WAITOK, MT_CONTROL);
 925                 control->m_len = msg.msg_controllen;
 926                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 927                                msg.msg_controllen);
 928                 if (error) {
 929                         m_free(control);
 930                         goto cleanup;
 931                 }
 932         }
 933
 934         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 935                              &sysmsg->sysmsg_szresult);
 936
 937 cleanup:
 938         iovec_free(&iov, aiov);
 939 cleanup2:
 940         if (sa)
 941                 kfree(sa, M_SONAME);
 942         return (error);
 943 }
 944
 945 /*
 946  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 947  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 948  * Don't forget to FREE() and m_free() these if they are returned.
 949  */
 950 int
 951 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 952              struct mbuf **control, int *flags, size_t *res)
 953 {
 954         struct thread *td = curthread;
 955         struct file *fp;
 956         size_t len;
 957         int error;
 958         int lflags;
 959         struct socket *so;
 960 #ifdef KTRACE
 961         struct iovec *ktriov = NULL;
 962         struct uio ktruio;
 963 #endif
 964
 965         error = holdsock(td, s, &fp);
 966         if (error)
 967                 return (error);
 968 #ifdef KTRACE
 969         if (KTRPOINT(td, KTR_GENIO)) {
 970                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 971
 972                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 973                 bcopy(auio->uio_iov, ktriov, iovlen);
 974                 ktruio = *auio;
 975         }
 976 #endif
 977         len = auio->uio_resid;
 978         so = (struct socket *)fp->f_data;
 979
 980         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 981                 if (fp->f_flag & FNONBLOCK) {
 982                         if (flags) {
 983                                 *flags |= MSG_FNONBLOCKING;
 984                         } else {
 985                                 lflags = MSG_FNONBLOCKING;
 986                                 flags = &lflags;
 987                         }
 988                 }
 989         }
 990
 991         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 992         if (error) {
 993                 if (auio->uio_resid != len && (error == ERESTART ||
 994                     error == EINTR || error == EWOULDBLOCK))
 995                         error = 0;
 996         }
 997 #ifdef KTRACE
 998         if (ktriov != NULL) {
 999                 if (error == 0) {
1000                         ktruio.uio_iov = ktriov;
1001                         ktruio.uio_resid = len - auio->uio_resid;
1002                         ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
1003                 }
1004                 kfree(ktriov, M_TEMP);
1005         }
1006 #endif
1007         if (error == 0)
1008                 *res = len - auio->uio_resid;
1009         dropfp(td, s, fp);
1010
1011         return (error);
1012 }
1013
1014 /*
1015  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
1016  *                      caddr_t from, int *fromlenaddr)
1017  *
1018  * MPALMOSTSAFE
1019  */
1020 int
1021 sys_recvfrom(struct sysmsg *sysmsg, const struct recvfrom_args *uap)
1022 {
1023         struct thread *td = curthread;
1024         struct uio auio;
1025         struct iovec aiov;
1026         struct sockaddr *sa = NULL;
1027         int error, fromlen;
1028         int flags;
1029
1030         if (uap->from && uap->fromlenaddr) {
1031                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
1032                 if (error)
1033                         return (error);
1034                 if (fromlen < 0)
1035                         return (EINVAL);
1036         } else {
1037                 fromlen = 0;
1038         }
1039         aiov.iov_base = uap->buf;
1040         aiov.iov_len = uap->len;
1041         auio.uio_iov = &aiov;
1042         auio.uio_iovcnt = 1;
1043         auio.uio_offset = 0;
1044         auio.uio_resid = uap->len;
1045         auio.uio_segflg = UIO_USERSPACE;
1046         auio.uio_rw = UIO_READ;
1047         auio.uio_td = td;
1048         flags = uap->flags;
1049
1050         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
1051                              &flags, &sysmsg->sysmsg_szresult);
1052
1053         if (error == 0 && uap->from) {
1054                 /* note: sa may still be NULL */
1055                 if (sa) {
1056                         fromlen = MIN(fromlen, sa->sa_len);
1057                         prison_local_ip(curthread, sa);
1058                         error = copyout(sa, uap->from, fromlen);
1059                 } else {
1060                         fromlen = 0;
1061                 }
1062                 if (error == 0) {
1063                         error = copyout(&fromlen, uap->fromlenaddr,
1064                                         sizeof(fromlen));
1065                 }
1066         }
1067         if (sa)
1068                 kfree(sa, M_SONAME);
1069
1070         return (error);
1071 }
1072
1073 /*
1074  * recvmsg_args(int s, struct msghdr *msg, int flags)
1075  *
1076  * MPALMOSTSAFE
1077  */
1078 int
1079 sys_recvmsg(struct sysmsg *sysmsg, const struct recvmsg_args *uap)
1080 {
1081         struct thread *td = curthread;
1082         struct msghdr msg;
1083         struct uio auio;
1084         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1085         struct mbuf *m, *control = NULL;
1086         struct sockaddr *sa = NULL;
1087         caddr_t ctlbuf;
1088         socklen_t *ufromlenp, *ucontrollenp;
1089         int error, fromlen, controllen, len, flags, *uflagsp;
1090
1091         /*
1092          * This copyin handles everything except the iovec.
1093          */
1094         error = copyin(uap->msg, &msg, sizeof(msg));
1095         if (error)
1096                 return (error);
1097
1098         if (msg.msg_name && msg.msg_namelen < 0)
1099                 return (EINVAL);
1100         if (msg.msg_control && msg.msg_controllen < 0)
1101                 return (EINVAL);
1102
1103         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1104                     msg_namelen));
1105         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1106                        msg_controllen));
1107         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1108                                                         msg_flags));
1109
1110         /*
1111          * Populate auio.
1112          */
1113         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1114                              &auio.uio_resid);
1115         if (error)
1116                 return (error);
1117         auio.uio_iov = iov;
1118         auio.uio_iovcnt = msg.msg_iovlen;
1119         auio.uio_offset = 0;
1120         auio.uio_segflg = UIO_USERSPACE;
1121         auio.uio_rw = UIO_READ;
1122         auio.uio_td = td;
1123
1124         flags = uap->flags;
1125
1126         error = kern_recvmsg(uap->s,
1127                              (msg.msg_name ? &sa : NULL), &auio,
1128                              (msg.msg_control ? &control : NULL), &flags,
1129                              &sysmsg->sysmsg_szresult);
1130
1131         /*
1132          * Conditionally copyout the name and populate the namelen field.
1133          */
1134         if (error == 0 && msg.msg_name) {
1135                 /* note: sa may still be NULL */
1136                 if (sa != NULL) {
1137                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
1138                         prison_local_ip(curthread, sa);
1139                         error = copyout(sa, msg.msg_name, fromlen);
1140                 } else {
1141                         fromlen = 0;
1142                 }
1143                 if (error == 0)
1144                         error = copyout(&fromlen, ufromlenp,
1145                             sizeof(*ufromlenp));
1146         }
1147
1148         /*
1149          * Copyout msg.msg_control and msg.msg_controllen.
1150          */
1151         if (error == 0 && msg.msg_control) {
1152                 len = msg.msg_controllen;
1153                 m = control;
1154                 ctlbuf = (caddr_t)msg.msg_control;
1155
1156                 while(m && len > 0) {
1157                         unsigned int tocopy;
1158
1159                         if (len >= m->m_len) {
1160                                 tocopy = m->m_len;
1161                         } else {
1162                                 msg.msg_flags |= MSG_CTRUNC;
1163                                 tocopy = len;
1164                         }
1165
1166                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1167                         if (error)
1168                                 goto cleanup;
1169
1170                         ctlbuf += tocopy;
1171                         len -= tocopy;
1172                         m = m->m_next;
1173                 }
1174                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1175                 error = copyout(&controllen, ucontrollenp,
1176                     sizeof(*ucontrollenp));
1177         }
1178
1179         if (error == 0)
1180                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1181
1182 cleanup:
1183         if (sa)
1184                 kfree(sa, M_SONAME);
1185         iovec_free(&iov, aiov);
1186         if (control)
1187                 m_freem(control);
1188         return (error);
1189 }
1190
1191 /*
1192  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1193  * in kernel pointer instead of a userland pointer.  This allows us
1194  * to manipulate socket options in the emulation code.
1195  */
1196 int
1197 kern_setsockopt(int s, struct sockopt *sopt)
1198 {
1199         struct thread *td = curthread;
1200         struct file *fp;
1201         int error;
1202
1203         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1204                 return (EFAULT);
1205         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1206                 return (EINVAL);
1207         if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1208                 return (EINVAL);
1209
1210         error = holdsock(td, s, &fp);
1211         if (error)
1212                 return (error);
1213
1214         error = sosetopt((struct socket *)fp->f_data, sopt);
1215         dropfp(td, s, fp);
1216
1217         return (error);
1218 }
1219
1220 /*
1221  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1222  *
1223  * MPALMOSTSAFE
1224  */
1225 int
1226 sys_setsockopt(struct sysmsg *sysmsg, const struct setsockopt_args *uap)
1227 {
1228         struct thread *td = curthread;
1229         struct sockopt sopt;
1230         int error;
1231
1232         sopt.sopt_level = uap->level;
1233         sopt.sopt_name = uap->name;
1234         sopt.sopt_valsize = uap->valsize;
1235         sopt.sopt_td = td;
1236         sopt.sopt_val = NULL;
1237
1238         if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1239                 return (EINVAL);
1240         if (uap->val) {
1241                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1242                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1243                 if (error)
1244                         goto out;
1245         }
1246
1247         error = kern_setsockopt(uap->s, &sopt);
1248 out:
1249         if (uap->val)
1250                 kfree(sopt.sopt_val, M_TEMP);
1251         return(error);
1252 }
1253
1254 /*
1255  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1256  * in kernel pointer instead of a userland pointer.  This allows us
1257  * to manipulate socket options in the emulation code.
1258  */
1259 int
1260 kern_getsockopt(int s, struct sockopt *sopt)
1261 {
1262         struct thread *td = curthread;
1263         struct file *fp;
1264         int error;
1265
1266         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1267                 return (EFAULT);
1268         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1269                 return (EINVAL);
1270
1271         error = holdsock(td, s, &fp);
1272         if (error)
1273                 return (error);
1274
1275         error = sogetopt((struct socket *)fp->f_data, sopt);
1276         dropfp(td, s, fp);
1277
1278         return (error);
1279 }
1280
1281 /*
1282  * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1283  *
1284  * MPALMOSTSAFE
1285  */
1286 int
1287 sys_getsockopt(struct sysmsg *sysmsg, const struct getsockopt_args *uap)
1288 {
1289         struct thread *td = curthread;
1290         struct sockopt sopt;
1291         int error, valsize, valszmax, mflag = 0;
1292
1293         if (uap->val) {
1294                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1295                 if (error)
1296                         return (error);
1297         } else {
1298                 valsize = 0;
1299         }
1300
1301         sopt.sopt_level = uap->level;
1302         sopt.sopt_name = uap->name;
1303         sopt.sopt_valsize = valsize;
1304         sopt.sopt_td = td;
1305         sopt.sopt_val = NULL;
1306
1307         if (td->td_proc->p_ucred->cr_uid == 0) {
1308                 valszmax = SOMAXOPT_SIZE0;
1309                 mflag = M_NULLOK;
1310         } else {
1311                 valszmax = SOMAXOPT_SIZE;
1312         }
1313         if (sopt.sopt_valsize > valszmax) /* unsigned */
1314                 return (EINVAL);
1315         if (uap->val) {
1316                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP,
1317                     M_WAITOK | mflag);
1318                 if (sopt.sopt_val == NULL)
1319                         return (ENOBUFS);
1320                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1321                 if (error)
1322                         goto out;
1323         }
1324
1325         error = kern_getsockopt(uap->s, &sopt);
1326         if (error)
1327                 goto out;
1328         valsize = sopt.sopt_valsize;
1329         error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1330         if (error)
1331                 goto out;
1332         if (uap->val)
1333                 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1334 out:
1335         if (uap->val)
1336                 kfree(sopt.sopt_val, M_TEMP);
1337         return (error);
1338 }
1339
1340 /*
1341  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1342  * This allows kern_getsockname() to return a pointer to an allocated struct
1343  * sockaddr which must be freed later with FREE().  The caller must
1344  * initialize *name to NULL.
1345  */
1346 int
1347 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1348 {
1349         struct thread *td = curthread;
1350         struct file *fp;
1351         struct socket *so;
1352         struct sockaddr *sa = NULL;
1353         int error;
1354
1355         error = holdsock(td, s, &fp);
1356         if (error)
1357                 return (error);
1358         if (*namelen < 0) {
1359                 fdrop(fp);
1360                 return (EINVAL);
1361         }
1362         so = (struct socket *)fp->f_data;
1363         error = so_pru_sockaddr(so, &sa);
1364         if (error == 0) {
1365                 if (sa == NULL) {
1366                         *namelen = 0;
1367                 } else {
1368                         *namelen = MIN(*namelen, sa->sa_len);
1369                         *name = sa;
1370                 }
1371         }
1372         dropfp(td, s, fp);
1373
1374         return (error);
1375 }
1376
1377 /*
1378  * getsockname_args(int fdes, caddr_t asa, int *alen)
1379  *
1380  * Get socket name.
1381  *
1382  * MPALMOSTSAFE
1383  */
1384 int
1385 sys_getsockname(struct sysmsg *sysmsg, const struct getsockname_args *uap)
1386 {
1387         struct sockaddr *sa = NULL;
1388         struct sockaddr satmp;
1389         int error, sa_len_in, sa_len_out;
1390
1391         error = copyin(uap->alen, &sa_len_in, sizeof(sa_len_in));
1392         if (error)
1393                 return (error);
1394
1395         sa_len_out = sa_len_in;
1396         error = kern_getsockname(uap->fdes, &sa, &sa_len_out);
1397
1398         if (error == 0) {
1399                 if (sa) {
1400                         prison_local_ip(curthread, sa);
1401                         error = copyout(sa, uap->asa, sa_len_out);
1402                 } else {
1403                         /*
1404                          * unnamed uipc sockets don't bother storing
1405                          * sockaddr, simulate an AF_LOCAL sockaddr.
1406                          */
1407                         sa_len_out = sizeof(satmp);
1408                         if (sa_len_out > sa_len_in)
1409                                 sa_len_out = sa_len_in;
1410                         if (sa_len_out < 0)
1411                                 sa_len_out = 0;
1412                         bzero(&satmp, sizeof(satmp));
1413                         satmp.sa_len = sa_len_out;
1414                         satmp.sa_family = AF_LOCAL;
1415                         error = copyout(&satmp, uap->asa, sa_len_out);
1416                 }
1417         }
1418         if (error == 0 && sa_len_out != sa_len_in)
1419                 error = copyout(&sa_len_out, uap->alen, sizeof(*uap->alen));
1420         if (sa)
1421                 kfree(sa, M_SONAME);
1422         return (error);
1423 }
1424
1425 /*
1426  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1427  * This allows kern_getpeername() to return a pointer to an allocated struct
1428  * sockaddr which must be freed later with FREE().  The caller must
1429  * initialize *name to NULL.
1430  */
1431 int
1432 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1433 {
1434         struct thread *td = curthread;
1435         struct file *fp;
1436         struct socket *so;
1437         struct sockaddr *sa = NULL;
1438         int error;
1439
1440         error = holdsock(td, s, &fp);
1441         if (error)
1442                 return (error);
1443         if (*namelen < 0) {
1444                 fdrop(fp);
1445                 return (EINVAL);
1446         }
1447         so = (struct socket *)fp->f_data;
1448         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1449                 fdrop(fp);
1450                 return (ENOTCONN);
1451         }
1452         error = so_pru_peeraddr(so, &sa);
1453         if (error == 0) {
1454                 if (sa == NULL) {
1455                         *namelen = 0;
1456                 } else {
1457                         *namelen = MIN(*namelen, sa->sa_len);
1458                         *name = sa;
1459                 }
1460         }
1461         dropfp(td, s, fp);
1462
1463         return (error);
1464 }
1465
1466 /*
1467  * getpeername_args(int fdes, caddr_t asa, int *alen)
1468  *
1469  * Get name of peer for connected socket.
1470  *
1471  * MPALMOSTSAFE
1472  */
1473 int
1474 sys_getpeername(struct sysmsg *sysmsg, const struct getpeername_args *uap)
1475 {
1476         struct sockaddr *sa = NULL;
1477         int error, sa_len;
1478
1479         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1480         if (error)
1481                 return (error);
1482
1483         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1484
1485         if (error == 0) {
1486                 prison_local_ip(curthread, sa);
1487                 error = copyout(sa, uap->asa, sa_len);
1488         }
1489         if (error == 0)
1490                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1491         if (sa)
1492                 kfree(sa, M_SONAME);
1493         return (error);
1494 }
1495
1496 int
1497 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1498 {
1499         struct sockaddr *sa;
1500         int error;
1501
1502         *namp = NULL;
1503         if (len > SOCK_MAXADDRLEN)
1504                 return ENAMETOOLONG;
1505         if (len < offsetof(struct sockaddr, sa_data[0]))
1506                 return EDOM;
1507         sa = kmalloc(len, M_SONAME, M_WAITOK);
1508         error = copyin(uaddr, sa, len);
1509         if (error) {
1510                 kfree(sa, M_SONAME);
1511         } else {
1512                 sa->sa_len = len;
1513                 *namp = sa;
1514         }
1515         return error;
1516 }
1517
1518 /*
1519  * Detach a mapped page and release resources back to the system.
1520  * We must release our wiring and if the object is ripped out
1521  * from under the vm_page we become responsible for freeing the
1522  * page.
1523  *
1524  * MPSAFE
1525  */
1526 static void
1527 sf_buf_mfree(void *arg)
1528 {
1529         struct sf_buf *sf = arg;
1530         vm_page_t m;
1531
1532         m = sf_buf_page(sf);
1533         if (sf_buf_free(sf)) {
1534                 /* sf invalid now */
1535                 vm_page_sbusy_drop(m);
1536 #if 0
1537                 if (m->object == NULL &&
1538                     m->wire_count == 0 &&
1539                     (m->flags & PG_NEED_COMMIT) == 0) {
1540                         vm_page_free(m);
1541                 } else {
1542                         vm_page_wakeup(m);
1543                 }
1544 #endif
1545         }
1546 }
1547
1548 /*
1549  * sendfile(2).
1550  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1551  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1552  *
1553  * Send a file specified by 'fd' and starting at 'offset' to a socket
1554  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1555  * nbytes == 0. Optionally add a header and/or trailer to the socket
1556  * output. If specified, write the total number of bytes sent into *sbytes.
1557  *
1558  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1559  * the headers to count against the remaining bytes to be sent from
1560  * the file descriptor.  We may wish to implement a compatibility syscall
1561  * in the future.
1562  *
1563  * MPALMOSTSAFE
1564  */
1565 int
1566 sys_sendfile(struct sysmsg *sysmsg, const struct sendfile_args *uap)
1567 {
1568         struct thread *td = curthread;
1569         struct file *fp;
1570         struct vnode *vp = NULL;
1571         struct sf_hdtr hdtr;
1572         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1573         struct uio auio;
1574         struct mbuf *mheader = NULL;
1575         size_t hbytes = 0;
1576         size_t tbytes;
1577         off_t hdtr_size = 0;
1578         off_t sbytes;
1579         int error;
1580
1581         /*
1582          * Do argument checking. Must be a regular file in, stream
1583          * type and connected socket out, positive offset.
1584          */
1585         fp = holdfp(td, uap->fd, FREAD);
1586         if (fp == NULL) {
1587                 return (EBADF);
1588         }
1589         if (fp->f_type != DTYPE_VNODE) {
1590                 fdrop(fp);
1591                 return (EINVAL);
1592         }
1593         vp = (struct vnode *)fp->f_data;
1594         vref(vp);
1595         dropfp(td, uap->fd, fp);
1596
1597         /*
1598          * If specified, get the pointer to the sf_hdtr struct for
1599          * any headers/trailers.
1600          */
1601         if (uap->hdtr) {
1602                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1603                 if (error)
1604                         goto done;
1605                 /*
1606                  * Send any headers.
1607                  */
1608                 if (hdtr.headers) {
1609                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1610                                              hdtr.hdr_cnt, &hbytes);
1611                         if (error)
1612                                 goto done;
1613                         auio.uio_iov = iov;
1614                         auio.uio_iovcnt = hdtr.hdr_cnt;
1615                         auio.uio_offset = 0;
1616                         auio.uio_segflg = UIO_USERSPACE;
1617                         auio.uio_rw = UIO_WRITE;
1618                         auio.uio_td = td;
1619                         auio.uio_resid = hbytes;
1620
1621                         mheader = m_uiomove(&auio);
1622
1623                         iovec_free(&iov, aiov);
1624                         if (mheader == NULL)
1625                                 goto done;
1626                 }
1627         }
1628
1629         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1630                               &sbytes, uap->flags);
1631         if (error)
1632                 goto done;
1633
1634         /*
1635          * Send trailers. Wimp out and use writev(2).
1636          */
1637         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1638                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1639                                      hdtr.trl_cnt, &auio.uio_resid);
1640                 if (error)
1641                         goto done;
1642                 auio.uio_iov = iov;
1643                 auio.uio_iovcnt = hdtr.trl_cnt;
1644                 auio.uio_offset = 0;
1645                 auio.uio_segflg = UIO_USERSPACE;
1646                 auio.uio_rw = UIO_WRITE;
1647                 auio.uio_td = td;
1648
1649                 tbytes = 0;     /* avoid gcc warnings */
1650                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1651
1652                 iovec_free(&iov, aiov);
1653                 if (error)
1654                         goto done;
1655                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1656         }
1657
1658 done:
1659         if (vp)
1660                 vrele(vp);
1661         if (uap->sbytes != NULL) {
1662                 sbytes += hdtr_size;
1663                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1664         }
1665         return (error);
1666 }
1667
1668 int
1669 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1670               struct mbuf *mheader, off_t *sbytes, int flags)
1671 {
1672         struct thread *td = curthread;
1673         struct vm_object *obj;
1674         struct socket *so;
1675         struct file *fp;
1676         struct mbuf *m, *mp;
1677         struct sf_buf *sf;
1678         struct vm_page *pg;
1679         off_t off, xfsize, xbytes;
1680         off_t hbytes = 0;
1681         int error = 0;
1682
1683         if (vp->v_type != VREG) {
1684                 error = EINVAL;
1685                 goto done0;
1686         }
1687         if ((obj = vp->v_object) == NULL) {
1688                 error = EINVAL;
1689                 goto done0;
1690         }
1691         error = holdsock(td, sfd, &fp);
1692         if (error)
1693                 goto done0;
1694         so = (struct socket *)fp->f_data;
1695         if (so->so_type != SOCK_STREAM) {
1696                 error = EINVAL;
1697                 goto done1;
1698         }
1699         if ((so->so_state & SS_ISCONNECTED) == 0) {
1700                 error = ENOTCONN;
1701                 goto done1;
1702         }
1703         if (offset < 0) {
1704                 error = EINVAL;
1705                 goto done1;
1706         }
1707
1708         /*
1709          * preallocation is required for asynchronous passing of mbufs,
1710          * otherwise we can wind up building up an infinite number of
1711          * mbufs during the asynchronous latency.
1712          */
1713         if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) {
1714                 error = EINVAL;
1715                 goto done1;
1716         }
1717
1718         *sbytes = 0;
1719         xbytes = 0;
1720
1721         /*
1722          * Protect against multiple writers to the socket.
1723          * We need at least a shared lock on the VM object
1724          */
1725         ssb_lock(&so->so_snd, M_WAITOK);
1726         vm_object_hold_shared(obj);
1727
1728         /*
1729          * Loop through the pages in the file, starting with the requested
1730          * offset. Get a file page (do I/O if necessary), map the file page
1731          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1732          * it on the socket.
1733          */
1734         for (off = offset; ;
1735              off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) {
1736                 vm_pindex_t pindex;
1737                 vm_offset_t pgoff;
1738                 long space;
1739                 int loops;
1740
1741                 pindex = OFF_TO_IDX(off);
1742                 loops = 0;
1743
1744 retry_lookup:
1745                 /*
1746                  * Calculate the amount to transfer. Not to exceed a page,
1747                  * the EOF, or the passed in nbytes.
1748                  */
1749                 xfsize = vp->v_filesize - off;
1750                 if (xfsize > PAGE_SIZE)
1751                         xfsize = PAGE_SIZE;
1752                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1753                 if (PAGE_SIZE - pgoff < xfsize)
1754                         xfsize = PAGE_SIZE - pgoff;
1755                 if (nbytes && xfsize > (nbytes - xbytes))
1756                         xfsize = nbytes - xbytes;
1757                 if (xfsize <= 0)
1758                         break;
1759                 /*
1760                  * Optimize the non-blocking case by looking at the socket space
1761                  * before going to the extra work of constituting the sf_buf.
1762                  */
1763                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1764                         space = ssb_space_prealloc(&so->so_snd);
1765                 else
1766                         space = ssb_space(&so->so_snd);
1767
1768                 if ((fp->f_flag & FNONBLOCK) && space <= 0) {
1769                         if (so->so_state & SS_CANTSENDMORE)
1770                                 error = EPIPE;
1771                         else
1772                                 error = EAGAIN;
1773                         goto done;
1774                 }
1775
1776                 /*
1777                  * Attempt to look up the page.
1778                  *
1779                  * Try to find the data using a shared vm_object token and
1780                  * vm_page_lookup_sbusy_try() first.
1781                  *
1782                  * If data is missing, use a UIO_NOCOPY VOP_READ to load
1783                  * the missing data and loop back up.  We avoid all sorts
1784                  * of problems by not trying to hold onto the page during
1785                  * the I/O.
1786                  *
1787                  * NOTE: The soft-busy will temporary block filesystem
1788                  *       truncation operations when a file is removed
1789                  *       while the sendfile is running.
1790                  */
1791                 pg = vm_page_lookup_sbusy_try(obj, pindex, pgoff, xfsize);
1792                 if (pg == NULL) {
1793                         struct uio auio;
1794                         struct iovec aiov;
1795                         int bsize;
1796
1797                         if (++loops > 100000) {
1798                                 kprintf("sendfile: VOP operation failed "
1799                                         "to retain page\n");
1800                                 error = EIO;
1801                                 goto done;
1802                         }
1803
1804                         vm_object_drop(obj);
1805                         bsize = vp->v_mount->mnt_stat.f_iosize;
1806                         auio.uio_iov = &aiov;
1807                         auio.uio_iovcnt = 1;
1808                         aiov.iov_base = 0;
1809                         aiov.iov_len = MAXBSIZE;
1810                         auio.uio_resid = MAXBSIZE;
1811                         auio.uio_offset = trunc_page(off);
1812                         auio.uio_segflg = UIO_NOCOPY;
1813                         auio.uio_rw = UIO_READ;
1814                         auio.uio_td = td;
1815
1816                         vn_lock(vp, LK_SHARED | LK_RETRY);
1817                         error = VOP_READ_FP(vp, &auio,
1818                                          IO_VMIO | ((MAXBSIZE / bsize) << 16),
1819                                          td->td_ucred, fp);
1820                         vn_unlock(vp);
1821                         vm_object_hold_shared(obj);
1822
1823                         if (error)
1824                                 goto done;
1825                         goto retry_lookup;
1826                 }
1827
1828                 /*
1829                  * Get a sendfile buf. We usually wait as long as necessary,
1830                  * but this wait can be interrupted.
1831                  */
1832                 if ((sf = sf_buf_alloc(pg)) == NULL) {
1833                         vm_page_sbusy_drop(pg);
1834                         /* vm_page_try_to_free(pg); */
1835                         error = EINTR;
1836                         goto done;
1837                 }
1838
1839                 /*
1840                  * Get an mbuf header and set it up as having external storage.
1841                  */
1842                 MGETHDR(m, M_WAITOK, MT_DATA);
1843                 m->m_ext.ext_free = sf_buf_mfree;
1844                 m->m_ext.ext_ref = sf_buf_ref;
1845                 m->m_ext.ext_arg = sf;
1846                 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1847                 m->m_ext.ext_size = PAGE_SIZE;
1848                 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1849                 m->m_flags |= M_EXT;
1850                 m->m_pkthdr.len = m->m_len = xfsize;
1851                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1852
1853                 if (mheader != NULL) {
1854                         hbytes = mheader->m_pkthdr.len;
1855                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1856                         m_cat(mheader, m);
1857                         m = mheader;
1858                         mheader = NULL;
1859                 } else {
1860                         hbytes = 0;
1861                 }
1862
1863                 /*
1864                  * Add the buffer to the socket buffer chain.
1865                  */
1866                 crit_enter();
1867 retry_space:
1868                 /*
1869                  * Make sure that the socket is still able to take more data.
1870                  * CANTSENDMORE being true usually means that the connection
1871                  * was closed. so_error is true when an error was sensed after
1872                  * a previous send.
1873                  * The state is checked after the page mapping and buffer
1874                  * allocation above since those operations may block and make
1875                  * any socket checks stale. From this point forward, nothing
1876                  * blocks before the pru_send (or more accurately, any blocking
1877                  * results in a loop back to here to re-check).
1878                  */
1879                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1880                         if (so->so_state & SS_CANTSENDMORE) {
1881                                 error = EPIPE;
1882                         } else {
1883                                 error = so->so_error;
1884                                 so->so_error = 0;
1885                         }
1886                         m_freem(m);
1887                         crit_exit();
1888                         goto done;
1889                 }
1890                 /*
1891                  * Wait for socket space to become available. We do this just
1892                  * after checking the connection state above in order to avoid
1893                  * a race condition with ssb_wait().
1894                  */
1895                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1896                         space = ssb_space_prealloc(&so->so_snd);
1897                 else
1898                         space = ssb_space(&so->so_snd);
1899
1900                 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1901                         if (fp->f_flag & FNONBLOCK) {
1902                                 m_freem(m);
1903                                 crit_exit();
1904                                 error = EAGAIN;
1905                                 goto done;
1906                         }
1907                         error = ssb_wait(&so->so_snd);
1908                         /*
1909                          * An error from ssb_wait usually indicates that we've
1910                          * been interrupted by a signal. If we've sent anything
1911                          * then return bytes sent, otherwise return the error.
1912                          */
1913                         if (error) {
1914                                 m_freem(m);
1915                                 crit_exit();
1916                                 goto done;
1917                         }
1918                         goto retry_space;
1919                 }
1920
1921                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1922                         for (mp = m; mp != NULL; mp = mp->m_next)
1923                                 ssb_preallocstream(&so->so_snd, mp);
1924                 }
1925                 if (use_sendfile_async)
1926                         error = so_pru_senda(so, 0, m, NULL, NULL, td);
1927                 else
1928                         error = so_pru_send(so, 0, m, NULL, NULL, td);
1929
1930                 crit_exit();
1931                 if (error)
1932                         goto done;
1933         }
1934         if (mheader != NULL) {
1935                 *sbytes += mheader->m_pkthdr.len;
1936
1937                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1938                         for (mp = mheader; mp != NULL; mp = mp->m_next)
1939                                 ssb_preallocstream(&so->so_snd, mp);
1940                 }
1941                 if (use_sendfile_async)
1942                         error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1943                 else
1944                         error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1945
1946                 mheader = NULL;
1947         }
1948 done:
1949         vm_object_drop(obj);
1950         ssb_unlock(&so->so_snd);
1951 done1:
1952         dropfp(td, sfd, fp);
1953 done0:
1954         if (mheader != NULL)
1955                 m_freem(mheader);
1956         return (error);
1957 }