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 file *fp;
 153         int error;
 154
 155         error = holdsock(td, s, &fp);
 156         if (error)
 157                 return (error);
 158         error = sobind((struct socket *)fp->f_data, sa, td);
 159         dropfp(td, s, fp);
 160
 161         return (error);
 162 }
 163
 164 /*
 165  * bind_args(int s, caddr_t name, int namelen)
 166  *
 167  * MPALMOSTSAFE
 168  */
 169 int
 170 sys_bind(struct bind_args *uap)
 171 {
 172         struct sockaddr *sa;
 173         int error;
 174
 175         error = getsockaddr(&sa, uap->name, uap->namelen);
 176         if (error)
 177                 return (error);
 178         error = kern_bind(uap->s, sa);
 179         kfree(sa, M_SONAME);
 180
 181         return (error);
 182 }
 183
 184 int
 185 kern_listen(int s, int backlog)
 186 {
 187         struct thread *td = curthread;
 188         struct file *fp;
 189         int error;
 190
 191         error = holdsock(td, s, &fp);
 192         if (error)
 193                 return (error);
 194         error = solisten((struct socket *)fp->f_data, backlog, td);
 195         dropfp(td, s, fp);
 196
 197         return (error);
 198 }
 199
 200 /*
 201  * listen_args(int s, int backlog)
 202  *
 203  * MPALMOSTSAFE
 204  */
 205 int
 206 sys_listen(struct listen_args *uap)
 207 {
 208         int error;
 209
 210         error = kern_listen(uap->s, uap->backlog);
 211         return (error);
 212 }
 213
 214 /*
 215  * Returns the accepted socket as well.
 216  *
 217  * NOTE!  The sockets sitting on so_comp/so_incomp might have 0 refs, the
 218  *        pool token is absolutely required to avoid a sofree() race,
 219  *        as well as to avoid tailq handling races.
 220  */
 221 static boolean_t
 222 soaccept_predicate(struct netmsg_so_notify *msg)
 223 {
 224         struct socket *head = msg->base.nm_so;
 225         struct socket *so;
 226
 227         if (head->so_error != 0) {
 228                 msg->base.lmsg.ms_error = head->so_error;
 229                 return (TRUE);
 230         }
 231         lwkt_getpooltoken(head);
 232         if (!TAILQ_EMPTY(&head->so_comp)) {
 233                 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */
 234                 so = TAILQ_FIRST(&head->so_comp);
 235                 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
 236                 TAILQ_REMOVE(&head->so_comp, so, so_list);
 237                 head->so_qlen--;
 238                 soclrstate(so, SS_COMP);
 239
 240                 /*
 241                  * Keep a reference before clearing the so_head
 242                  * to avoid racing socket close in netisr.
 243                  */
 244                 soreference(so);
 245                 so->so_head = NULL;
 246
 247                 lwkt_relpooltoken(head);
 248
 249                 msg->base.lmsg.ms_error = 0;
 250                 msg->base.nm_so = so;
 251                 return (TRUE);
 252         }
 253         lwkt_relpooltoken(head);
 254         if (head->so_state & SS_CANTRCVMORE) {
 255                 msg->base.lmsg.ms_error = ECONNABORTED;
 256                 return (TRUE);
 257         }
 258         if (msg->nm_fflags & FNONBLOCK) {
 259                 msg->base.lmsg.ms_error = EWOULDBLOCK;
 260                 return (TRUE);
 261         }
 262
 263         return (FALSE);
 264 }
 265
 266 /*
 267  * The second argument to kern_accept() is a handle to a struct sockaddr.
 268  * This allows kern_accept() to return a pointer to an allocated struct
 269  * sockaddr which must be freed later with FREE().  The caller must
 270  * initialize *name to NULL.
 271  */
 272 int
 273 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res,
 274     int sockflags)
 275 {
 276         struct thread *td = curthread;
 277         struct filedesc *fdp = td->td_proc->p_fd;
 278         struct file *lfp = NULL;
 279         struct file *nfp = NULL;
 280         struct sockaddr *sa;
 281         struct socket *head, *so;
 282         struct netmsg_so_notify msg;
 283         int fd;
 284         u_int fflag;            /* type must match fp->f_flag */
 285         int error, tmp;
 286
 287         *res = -1;
 288         if (name && namelen && *namelen < 0)
 289                 return (EINVAL);
 290
 291         error = holdsock(td, s, &lfp);
 292         if (error)
 293                 return (error);
 294
 295         error = falloc(td->td_lwp, &nfp, &fd);
 296         if (error) {            /* Probably ran out of file descriptors. */
 297                 fdrop(lfp);
 298                 return (error);
 299         }
 300         head = (struct socket *)lfp->f_data;
 301         if ((head->so_options & SO_ACCEPTCONN) == 0) {
 302                 error = EINVAL;
 303                 goto done;
 304         }
 305
 306         if (fflags & O_FBLOCKING)
 307                 fflags |= lfp->f_flag & ~FNONBLOCK;
 308         else if (fflags & O_FNONBLOCKING)
 309                 fflags |= lfp->f_flag | FNONBLOCK;
 310         else
 311                 fflags = lfp->f_flag;
 312
 313         if (use_soaccept_pred_fast) {
 314                 boolean_t pred;
 315
 316                 /* Initialize necessary parts for soaccept_predicate() */
 317                 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL);
 318                 msg.nm_fflags = fflags;
 319
 320                 lwkt_getpooltoken(head);
 321                 pred = soaccept_predicate(&msg);
 322                 lwkt_relpooltoken(head);
 323
 324                 if (pred) {
 325                         error = msg.base.lmsg.ms_error;
 326                         if (error)
 327                                 goto done;
 328                         else
 329                                 goto accepted;
 330                 }
 331         }
 332
 333         /* optimize for uniprocessor case later XXX JH */
 334         netmsg_init_abortable(&msg.base, head, &curthread->td_msgport,
 335                               0, netmsg_so_notify, netmsg_so_notify_doabort);
 336         msg.nm_predicate = soaccept_predicate;
 337         msg.nm_fflags = fflags;
 338         msg.nm_etype = NM_REVENT;
 339         error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH);
 340         if (error)
 341                 goto done;
 342
 343 accepted:
 344         /*
 345          * At this point we have the connection that's ready to be accepted.
 346          *
 347          * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects
 348          *       to eat the ref and turn it into a descriptor.
 349          */
 350         so = msg.base.nm_so;
 351
 352         fflag = lfp->f_flag;
 353
 354         /* connection has been removed from the listen queue */
 355         KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
 356
 357         if (sockflags & SOCK_KERN_NOINHERIT) {
 358                 fflag &= ~(FASYNC | FNONBLOCK);
 359                 if (sockflags & SOCK_NONBLOCK)
 360                         fflag |= FNONBLOCK;
 361         } else {
 362                 if (head->so_sigio != NULL)
 363                         fsetown(fgetown(&head->so_sigio), &so->so_sigio);
 364         }
 365
 366         nfp->f_type = DTYPE_SOCKET;
 367         nfp->f_flag = fflag;
 368         nfp->f_ops = &socketops;
 369         nfp->f_data = so;
 370         /* Sync socket async state with file flags */
 371         tmp = fflag & FASYNC;
 372         fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
 373
 374         sa = NULL;
 375         if (so->so_faddr != NULL) {
 376                 sa = so->so_faddr;
 377                 so->so_faddr = NULL;
 378
 379                 soaccept_generic(so);
 380                 error = 0;
 381         } else {
 382                 error = soaccept(so, &sa);
 383         }
 384
 385         /*
 386          * Set the returned name and namelen as applicable.  Set the returned
 387          * namelen to 0 for older code which might ignore the return value
 388          * from accept.
 389          */
 390         if (error == 0) {
 391                 if (sa && name && namelen) {
 392                         if (*namelen > sa->sa_len)
 393                                 *namelen = sa->sa_len;
 394                         *name = sa;
 395                 } else {
 396                         if (sa)
 397                                 kfree(sa, M_SONAME);
 398                 }
 399         }
 400
 401 done:
 402         /*
 403          * If an error occured clear the reserved descriptor, else associate
 404          * nfp with it.
 405          *
 406          * Note that *res is normally ignored if an error is returned but
 407          * a syscall message will still have access to the result code.
 408          */
 409         if (error) {
 410                 fsetfd(fdp, NULL, fd);
 411         } else {
 412                 if (sockflags & SOCK_CLOEXEC)
 413                         fdp->fd_files[fd].fileflags |= UF_EXCLOSE;
 414                 *res = fd;
 415                 fsetfd(fdp, nfp, fd);
 416         }
 417         fdrop(nfp);
 418         dropfp(td, s, lfp);
 419
 420         return (error);
 421 }
 422
 423 /*
 424  * accept(int s, caddr_t name, int *anamelen)
 425  *
 426  * MPALMOSTSAFE
 427  */
 428 int
 429 sys_accept(struct accept_args *uap)
 430 {
 431         struct sockaddr *sa = NULL;
 432         int sa_len;
 433         int error;
 434
 435         if (uap->name) {
 436                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 437                 if (error)
 438                         return (error);
 439
 440                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 441                                     &uap->sysmsg_iresult, 0);
 442
 443                 if (error == 0)
 444                         error = copyout(sa, uap->name, sa_len);
 445                 if (error == 0) {
 446                         error = copyout(&sa_len, uap->anamelen,
 447                             sizeof(*uap->anamelen));
 448                 }
 449                 if (sa)
 450                         kfree(sa, M_SONAME);
 451         } else {
 452                 error = kern_accept(uap->s, 0, NULL, 0,
 453                                     &uap->sysmsg_iresult, 0);
 454         }
 455         return (error);
 456 }
 457
 458 /*
 459  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
 460  *
 461  * MPALMOSTSAFE
 462  */
 463 int
 464 sys_extaccept(struct extaccept_args *uap)
 465 {
 466         struct sockaddr *sa = NULL;
 467         int sa_len;
 468         int error;
 469         int fflags = uap->flags & O_FMASK;
 470
 471         if (uap->name) {
 472                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 473                 if (error)
 474                         return (error);
 475
 476                 error = kern_accept(uap->s, fflags, &sa, &sa_len,
 477                                     &uap->sysmsg_iresult, 0);
 478
 479                 if (error == 0)
 480                         error = copyout(sa, uap->name, sa_len);
 481                 if (error == 0) {
 482                         error = copyout(&sa_len, uap->anamelen,
 483                             sizeof(*uap->anamelen));
 484                 }
 485                 if (sa)
 486                         kfree(sa, M_SONAME);
 487         } else {
 488                 error = kern_accept(uap->s, fflags, NULL, 0,
 489                                     &uap->sysmsg_iresult, 0);
 490         }
 491         return (error);
 492 }
 493
 494 /*
 495  * accept4(int s, caddr_t name, int *anamelen, int flags)
 496  *
 497  * MPALMOSTSAFE
 498  */
 499 int
 500 sys_accept4(struct accept4_args *uap)
 501 {
 502         struct sockaddr *sa = NULL;
 503         int sa_len;
 504         int error;
 505         int sockflags;
 506
 507         if (uap->flags & ~(SOCK_NONBLOCK | SOCK_CLOEXEC))
 508                 return (EINVAL);
 509         sockflags = uap->flags | SOCK_KERN_NOINHERIT;
 510
 511         if (uap->name) {
 512                 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
 513                 if (error)
 514                         return (error);
 515
 516                 error = kern_accept(uap->s, 0, &sa, &sa_len,
 517                                     &uap->sysmsg_iresult, sockflags);
 518
 519                 if (error == 0)
 520                         error = copyout(sa, uap->name, sa_len);
 521                 if (error == 0) {
 522                         error = copyout(&sa_len, uap->anamelen,
 523                             sizeof(*uap->anamelen));
 524                 }
 525                 if (sa)
 526                         kfree(sa, M_SONAME);
 527         } else {
 528                 error = kern_accept(uap->s, 0, NULL, 0,
 529                                     &uap->sysmsg_iresult, sockflags);
 530         }
 531         return (error);
 532 }
 533
 534 /*
 535  * Returns TRUE if predicate satisfied.
 536  */
 537 static boolean_t
 538 soconnected_predicate(struct netmsg_so_notify *msg)
 539 {
 540         struct socket *so = msg->base.nm_so;
 541
 542         /* check predicate */
 543         if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
 544                 msg->base.lmsg.ms_error = so->so_error;
 545                 return (TRUE);
 546         }
 547
 548         return (FALSE);
 549 }
 550
 551 int
 552 kern_connect(int s, int fflags, struct sockaddr *sa)
 553 {
 554         struct thread *td = curthread;
 555         struct file *fp;
 556         struct socket *so;
 557         int error, interrupted = 0;
 558
 559         error = holdsock(td, s, &fp);
 560         if (error)
 561                 return (error);
 562         so = (struct socket *)fp->f_data;
 563
 564         if (fflags & O_FBLOCKING)
 565                 /* fflags &= ~FNONBLOCK; */;
 566         else if (fflags & O_FNONBLOCKING)
 567                 fflags |= FNONBLOCK;
 568         else
 569                 fflags = fp->f_flag;
 570
 571         if (so->so_state & SS_ISCONNECTING) {
 572                 error = EALREADY;
 573                 goto done;
 574         }
 575         error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE);
 576         if (error)
 577                 goto bad;
 578         if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
 579                 error = EINPROGRESS;
 580                 goto done;
 581         }
 582         if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
 583                 struct netmsg_so_notify msg;
 584
 585                 netmsg_init_abortable(&msg.base, so,
 586                                       &curthread->td_msgport,
 587                                       0,
 588                                       netmsg_so_notify,
 589                                       netmsg_so_notify_doabort);
 590                 msg.nm_predicate = soconnected_predicate;
 591                 msg.nm_etype = NM_REVENT;
 592                 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH);
 593                 if (error == EINTR || error == ERESTART)
 594                         interrupted = 1;
 595         }
 596         if (error == 0) {
 597                 error = so->so_error;
 598                 so->so_error = 0;
 599         }
 600 bad:
 601         if (!interrupted)
 602                 soclrstate(so, SS_ISCONNECTING);
 603         if (error == ERESTART)
 604                 error = EINTR;
 605 done:
 606         dropfp(td, s, fp);
 607
 608         return (error);
 609 }
 610
 611 /*
 612  * connect_args(int s, caddr_t name, int namelen)
 613  *
 614  * MPALMOSTSAFE
 615  */
 616 int
 617 sys_connect(struct connect_args *uap)
 618 {
 619         struct sockaddr *sa;
 620         int error;
 621
 622         error = getsockaddr(&sa, uap->name, uap->namelen);
 623         if (error)
 624                 return (error);
 625         error = kern_connect(uap->s, 0, sa);
 626         kfree(sa, M_SONAME);
 627
 628         return (error);
 629 }
 630
 631 /*
 632  * connect_args(int s, int fflags, caddr_t name, int namelen)
 633  *
 634  * MPALMOSTSAFE
 635  */
 636 int
 637 sys_extconnect(struct extconnect_args *uap)
 638 {
 639         struct sockaddr *sa;
 640         int error;
 641         int fflags = uap->flags & O_FMASK;
 642
 643         error = getsockaddr(&sa, uap->name, uap->namelen);
 644         if (error)
 645                 return (error);
 646         error = kern_connect(uap->s, fflags, sa);
 647         kfree(sa, M_SONAME);
 648
 649         return (error);
 650 }
 651
 652 int
 653 kern_socketpair(int domain, int type, int protocol, int *sv)
 654 {
 655         struct thread *td = curthread;
 656         struct filedesc *fdp;
 657         struct file *fp1, *fp2;
 658         struct socket *so1, *so2;
 659         int fd1, fd2, error;
 660         u_int fflags = 0;
 661         int oflags = 0;
 662
 663         if (type & SOCK_NONBLOCK) {
 664                 type &= ~SOCK_NONBLOCK;
 665                 fflags |= FNONBLOCK;
 666         }
 667         if (type & SOCK_CLOEXEC) {
 668                 type &= ~SOCK_CLOEXEC;
 669                 oflags |= O_CLOEXEC;
 670         }
 671
 672         fdp = td->td_proc->p_fd;
 673         error = socreate(domain, &so1, type, protocol, td);
 674         if (error)
 675                 return (error);
 676         error = socreate(domain, &so2, type, protocol, td);
 677         if (error)
 678                 goto free1;
 679         error = falloc(td->td_lwp, &fp1, &fd1);
 680         if (error)
 681                 goto free2;
 682         sv[0] = fd1;
 683         fp1->f_data = so1;
 684         error = falloc(td->td_lwp, &fp2, &fd2);
 685         if (error)
 686                 goto free3;
 687         fp2->f_data = so2;
 688         sv[1] = fd2;
 689         error = soconnect2(so1, so2);
 690         if (error)
 691                 goto free4;
 692         if (type == SOCK_DGRAM) {
 693                 /*
 694                  * Datagram socket connection is asymmetric.
 695                  */
 696                  error = soconnect2(so2, so1);
 697                  if (error)
 698                         goto free4;
 699         }
 700         fp1->f_type = fp2->f_type = DTYPE_SOCKET;
 701         fp1->f_flag = fp2->f_flag = FREAD|FWRITE|fflags;
 702         fp1->f_ops = fp2->f_ops = &socketops;
 703         if (oflags & O_CLOEXEC) {
 704                 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE;
 705                 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE;
 706         }
 707         fsetfd(fdp, fp1, fd1);
 708         fsetfd(fdp, fp2, fd2);
 709         fdrop(fp1);
 710         fdrop(fp2);
 711         return (error);
 712 free4:
 713         fsetfd(fdp, NULL, fd2);
 714         fdrop(fp2);
 715 free3:
 716         fsetfd(fdp, NULL, fd1);
 717         fdrop(fp1);
 718 free2:
 719         (void)soclose(so2, 0);
 720 free1:
 721         (void)soclose(so1, 0);
 722         return (error);
 723 }
 724
 725 /*
 726  * socketpair(int domain, int type, int protocol, int *rsv)
 727  */
 728 int
 729 sys_socketpair(struct socketpair_args *uap)
 730 {
 731         int error, sockv[2];
 732
 733         error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
 734
 735         if (error == 0) {
 736                 error = copyout(sockv, uap->rsv, sizeof(sockv));
 737
 738                 if (error != 0) {
 739                         kern_close(sockv[0]);
 740                         kern_close(sockv[1]);
 741                 }
 742         }
 743
 744         return (error);
 745 }
 746
 747 int
 748 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
 749              struct mbuf *control, int flags, size_t *res)
 750 {
 751         struct thread *td = curthread;
 752         struct lwp *lp = td->td_lwp;
 753         struct proc *p = td->td_proc;
 754         struct file *fp;
 755         size_t len;
 756         int error;
 757         struct socket *so;
 758 #ifdef KTRACE
 759         struct iovec *ktriov = NULL;
 760         struct uio ktruio;
 761 #endif
 762
 763         error = holdsock(td, s, &fp);
 764         if (error)
 765                 return (error);
 766 #ifdef KTRACE
 767         if (KTRPOINT(td, KTR_GENIO)) {
 768                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 769
 770                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 771                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 772                 ktruio = *auio;
 773         }
 774 #endif
 775         len = auio->uio_resid;
 776         so = (struct socket *)fp->f_data;
 777         if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 778                 if (fp->f_flag & FNONBLOCK)
 779                         flags |= MSG_FNONBLOCKING;
 780         }
 781         error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
 782         if (error) {
 783                 if (auio->uio_resid != len && (error == ERESTART ||
 784                     error == EINTR || error == EWOULDBLOCK))
 785                         error = 0;
 786                 if (error == EPIPE && !(flags & MSG_NOSIGNAL) &&
 787                     !(so->so_options & SO_NOSIGPIPE))
 788                         lwpsignal(p, lp, SIGPIPE);
 789         }
 790 #ifdef KTRACE
 791         if (ktriov != NULL) {
 792                 if (error == 0) {
 793                         ktruio.uio_iov = ktriov;
 794                         ktruio.uio_resid = len - auio->uio_resid;
 795                         ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
 796                 }
 797                 kfree(ktriov, M_TEMP);
 798         }
 799 #endif
 800         if (error == 0)
 801                 *res  = len - auio->uio_resid;
 802         dropfp(td, s, fp);
 803
 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 file *fp;
 932         size_t len;
 933         int error;
 934         int lflags;
 935         struct socket *so;
 936 #ifdef KTRACE
 937         struct iovec *ktriov = NULL;
 938         struct uio ktruio;
 939 #endif
 940
 941         error = holdsock(td, s, &fp);
 942         if (error)
 943                 return (error);
 944 #ifdef KTRACE
 945         if (KTRPOINT(td, KTR_GENIO)) {
 946                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 947
 948                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 949                 bcopy(auio->uio_iov, ktriov, iovlen);
 950                 ktruio = *auio;
 951         }
 952 #endif
 953         len = auio->uio_resid;
 954         so = (struct socket *)fp->f_data;
 955
 956         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 957                 if (fp->f_flag & FNONBLOCK) {
 958                         if (flags) {
 959                                 *flags |= MSG_FNONBLOCKING;
 960                         } else {
 961                                 lflags = MSG_FNONBLOCKING;
 962                                 flags = &lflags;
 963                         }
 964                 }
 965         }
 966
 967         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 968         if (error) {
 969                 if (auio->uio_resid != len && (error == ERESTART ||
 970                     error == EINTR || error == EWOULDBLOCK))
 971                         error = 0;
 972         }
 973 #ifdef KTRACE
 974         if (ktriov != NULL) {
 975                 if (error == 0) {
 976                         ktruio.uio_iov = ktriov;
 977                         ktruio.uio_resid = len - auio->uio_resid;
 978                         ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
 979                 }
 980                 kfree(ktriov, M_TEMP);
 981         }
 982 #endif
 983         if (error == 0)
 984                 *res = len - auio->uio_resid;
 985         dropfp(td, s, fp);
 986
 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 file *fp;
1173         int error;
1174
1175         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1176                 return (EFAULT);
1177         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1178                 return (EINVAL);
1179         if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1180                 return (EINVAL);
1181
1182         error = holdsock(td, s, &fp);
1183         if (error)
1184                 return (error);
1185
1186         error = sosetopt((struct socket *)fp->f_data, sopt);
1187         dropfp(td, s, fp);
1188
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 file *fp;
1236         int error;
1237
1238         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1239                 return (EFAULT);
1240         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1241                 return (EINVAL);
1242
1243         error = holdsock(td, s, &fp);
1244         if (error)
1245                 return (error);
1246
1247         error = sogetopt((struct socket *)fp->f_data, sopt);
1248         dropfp(td, s, fp);
1249
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 file *fp;
1323         struct socket *so;
1324         struct sockaddr *sa = NULL;
1325         int error;
1326
1327         error = holdsock(td, s, &fp);
1328         if (error)
1329                 return (error);
1330         if (*namelen < 0) {
1331                 fdrop(fp);
1332                 return (EINVAL);
1333         }
1334         so = (struct socket *)fp->f_data;
1335         error = so_pru_sockaddr(so, &sa);
1336         if (error == 0) {
1337                 if (sa == NULL) {
1338                         *namelen = 0;
1339                 } else {
1340                         *namelen = MIN(*namelen, sa->sa_len);
1341                         *name = sa;
1342                 }
1343         }
1344         dropfp(td, s, fp);
1345
1346         return (error);
1347 }
1348
1349 /*
1350  * getsockname_args(int fdes, caddr_t asa, int *alen)
1351  *
1352  * Get socket name.
1353  *
1354  * MPALMOSTSAFE
1355  */
1356 int
1357 sys_getsockname(struct getsockname_args *uap)
1358 {
1359         struct sockaddr *sa = NULL;
1360         struct sockaddr satmp;
1361         int error, sa_len_in, sa_len_out;
1362
1363         error = copyin(uap->alen, &sa_len_in, sizeof(sa_len_in));
1364         if (error)
1365                 return (error);
1366
1367         sa_len_out = sa_len_in;
1368         error = kern_getsockname(uap->fdes, &sa, &sa_len_out);
1369
1370         if (error == 0) {
1371                 if (sa) {
1372                         error = copyout(sa, uap->asa, sa_len_out);
1373                 } else {
1374                         /*
1375                          * unnamed uipc sockets don't bother storing
1376                          * sockaddr, simulate an AF_LOCAL sockaddr.
1377                          */
1378                         sa_len_out = sizeof(satmp);
1379                         if (sa_len_out > sa_len_in)
1380                                 sa_len_out = sa_len_in;
1381                         if (sa_len_out < 0)
1382                                 sa_len_out = 0;
1383                         bzero(&satmp, sizeof(satmp));
1384                         satmp.sa_len = sa_len_out;
1385                         satmp.sa_family = AF_LOCAL;
1386                         error = copyout(&satmp, uap->asa, sa_len_out);
1387                 }
1388         }
1389         if (error == 0 && sa_len_out != sa_len_in)
1390                 error = copyout(&sa_len_out, uap->alen, sizeof(*uap->alen));
1391         if (sa)
1392                 kfree(sa, M_SONAME);
1393         return (error);
1394 }
1395
1396 /*
1397  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1398  * This allows kern_getpeername() to return a pointer to an allocated struct
1399  * sockaddr which must be freed later with FREE().  The caller must
1400  * initialize *name to NULL.
1401  */
1402 int
1403 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1404 {
1405         struct thread *td = curthread;
1406         struct file *fp;
1407         struct socket *so;
1408         struct sockaddr *sa = NULL;
1409         int error;
1410
1411         error = holdsock(td, s, &fp);
1412         if (error)
1413                 return (error);
1414         if (*namelen < 0) {
1415                 fdrop(fp);
1416                 return (EINVAL);
1417         }
1418         so = (struct socket *)fp->f_data;
1419         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1420                 fdrop(fp);
1421                 return (ENOTCONN);
1422         }
1423         error = so_pru_peeraddr(so, &sa);
1424         if (error == 0) {
1425                 if (sa == NULL) {
1426                         *namelen = 0;
1427                 } else {
1428                         *namelen = MIN(*namelen, sa->sa_len);
1429                         *name = sa;
1430                 }
1431         }
1432         dropfp(td, s, fp);
1433
1434         return (error);
1435 }
1436
1437 /*
1438  * getpeername_args(int fdes, caddr_t asa, int *alen)
1439  *
1440  * Get name of peer for connected socket.
1441  *
1442  * MPALMOSTSAFE
1443  */
1444 int
1445 sys_getpeername(struct getpeername_args *uap)
1446 {
1447         struct sockaddr *sa = NULL;
1448         int error, sa_len;
1449
1450         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1451         if (error)
1452                 return (error);
1453
1454         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1455
1456         if (error == 0)
1457                 error = copyout(sa, uap->asa, sa_len);
1458         if (error == 0)
1459                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1460         if (sa)
1461                 kfree(sa, M_SONAME);
1462         return (error);
1463 }
1464
1465 int
1466 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1467 {
1468         struct sockaddr *sa;
1469         int error;
1470
1471         *namp = NULL;
1472         if (len > SOCK_MAXADDRLEN)
1473                 return ENAMETOOLONG;
1474         if (len < offsetof(struct sockaddr, sa_data[0]))
1475                 return EDOM;
1476         sa = kmalloc(len, M_SONAME, M_WAITOK);
1477         error = copyin(uaddr, sa, len);
1478         if (error) {
1479                 kfree(sa, M_SONAME);
1480         } else {
1481 #if BYTE_ORDER != BIG_ENDIAN
1482                 /*
1483                  * The bind(), connect(), and sendto() syscalls were not
1484                  * versioned for COMPAT_43.  Thus, this check must stay.
1485                  */
1486                 if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1487                         sa->sa_family = sa->sa_len;
1488 #endif
1489                 sa->sa_len = len;
1490                 *namp = sa;
1491         }
1492         return error;
1493 }
1494
1495 /*
1496  * Detach a mapped page and release resources back to the system.
1497  * We must release our wiring and if the object is ripped out
1498  * from under the vm_page we become responsible for freeing the
1499  * page.
1500  *
1501  * MPSAFE
1502  */
1503 static void
1504 sf_buf_mfree(void *arg)
1505 {
1506         struct sf_buf *sf = arg;
1507         vm_page_t m;
1508
1509         m = sf_buf_page(sf);
1510         if (sf_buf_free(sf)) {
1511                 /* sf invalid now */
1512                 vm_page_sbusy_drop(m);
1513 #if 0
1514                 if (m->object == NULL &&
1515                     m->wire_count == 0 &&
1516                     (m->flags & PG_NEED_COMMIT) == 0) {
1517                         vm_page_free(m);
1518                 } else {
1519                         vm_page_wakeup(m);
1520                 }
1521 #endif
1522         }
1523 }
1524
1525 /*
1526  * sendfile(2).
1527  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1528  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1529  *
1530  * Send a file specified by 'fd' and starting at 'offset' to a socket
1531  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1532  * nbytes == 0. Optionally add a header and/or trailer to the socket
1533  * output. If specified, write the total number of bytes sent into *sbytes.
1534  *
1535  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1536  * the headers to count against the remaining bytes to be sent from
1537  * the file descriptor.  We may wish to implement a compatibility syscall
1538  * in the future.
1539  *
1540  * MPALMOSTSAFE
1541  */
1542 int
1543 sys_sendfile(struct sendfile_args *uap)
1544 {
1545         struct thread *td = curthread;
1546         struct file *fp;
1547         struct vnode *vp = NULL;
1548         struct sf_hdtr hdtr;
1549         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1550         struct uio auio;
1551         struct mbuf *mheader = NULL;
1552         size_t hbytes = 0;
1553         size_t tbytes;
1554         off_t hdtr_size = 0;
1555         off_t sbytes;
1556         int error;
1557
1558         /*
1559          * Do argument checking. Must be a regular file in, stream
1560          * type and connected socket out, positive offset.
1561          */
1562         fp = holdfp(td, uap->fd, FREAD);
1563         if (fp == NULL) {
1564                 return (EBADF);
1565         }
1566         if (fp->f_type != DTYPE_VNODE) {
1567                 fdrop(fp);
1568                 return (EINVAL);
1569         }
1570         vp = (struct vnode *)fp->f_data;
1571         vref(vp);
1572         dropfp(td, uap->fd, fp);
1573
1574         /*
1575          * If specified, get the pointer to the sf_hdtr struct for
1576          * any headers/trailers.
1577          */
1578         if (uap->hdtr) {
1579                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1580                 if (error)
1581                         goto done;
1582                 /*
1583                  * Send any headers.
1584                  */
1585                 if (hdtr.headers) {
1586                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1587                                              hdtr.hdr_cnt, &hbytes);
1588                         if (error)
1589                                 goto done;
1590                         auio.uio_iov = iov;
1591                         auio.uio_iovcnt = hdtr.hdr_cnt;
1592                         auio.uio_offset = 0;
1593                         auio.uio_segflg = UIO_USERSPACE;
1594                         auio.uio_rw = UIO_WRITE;
1595                         auio.uio_td = td;
1596                         auio.uio_resid = hbytes;
1597
1598                         mheader = m_uiomove(&auio);
1599
1600                         iovec_free(&iov, aiov);
1601                         if (mheader == NULL)
1602                                 goto done;
1603                 }
1604         }
1605
1606         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1607                               &sbytes, uap->flags);
1608         if (error)
1609                 goto done;
1610
1611         /*
1612          * Send trailers. Wimp out and use writev(2).
1613          */
1614         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1615                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1616                                      hdtr.trl_cnt, &auio.uio_resid);
1617                 if (error)
1618                         goto done;
1619                 auio.uio_iov = iov;
1620                 auio.uio_iovcnt = hdtr.trl_cnt;
1621                 auio.uio_offset = 0;
1622                 auio.uio_segflg = UIO_USERSPACE;
1623                 auio.uio_rw = UIO_WRITE;
1624                 auio.uio_td = td;
1625
1626                 tbytes = 0;     /* avoid gcc warnings */
1627                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1628
1629                 iovec_free(&iov, aiov);
1630                 if (error)
1631                         goto done;
1632                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1633         }
1634
1635 done:
1636         if (vp)
1637                 vrele(vp);
1638         if (uap->sbytes != NULL) {
1639                 sbytes += hdtr_size;
1640                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1641         }
1642         return (error);
1643 }
1644
1645 int
1646 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1647               struct mbuf *mheader, off_t *sbytes, int flags)
1648 {
1649         struct thread *td = curthread;
1650         struct vm_object *obj;
1651         struct socket *so;
1652         struct file *fp;
1653         struct mbuf *m, *mp;
1654         struct sf_buf *sf;
1655         struct vm_page *pg;
1656         off_t off, xfsize, xbytes;
1657         off_t hbytes = 0;
1658         int error = 0;
1659
1660         if (vp->v_type != VREG) {
1661                 error = EINVAL;
1662                 goto done0;
1663         }
1664         if ((obj = vp->v_object) == NULL) {
1665                 error = EINVAL;
1666                 goto done0;
1667         }
1668         error = holdsock(td, sfd, &fp);
1669         if (error)
1670                 goto done0;
1671         so = (struct socket *)fp->f_data;
1672         if (so->so_type != SOCK_STREAM) {
1673                 error = EINVAL;
1674                 goto done1;
1675         }
1676         if ((so->so_state & SS_ISCONNECTED) == 0) {
1677                 error = ENOTCONN;
1678                 goto done1;
1679         }
1680         if (offset < 0) {
1681                 error = EINVAL;
1682                 goto done1;
1683         }
1684
1685         /*
1686          * preallocation is required for asynchronous passing of mbufs,
1687          * otherwise we can wind up building up an infinite number of
1688          * mbufs during the asynchronous latency.
1689          */
1690         if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) {
1691                 error = EINVAL;
1692                 goto done1;
1693         }
1694
1695         *sbytes = 0;
1696         xbytes = 0;
1697
1698         /*
1699          * Protect against multiple writers to the socket.
1700          * We need at least a shared lock on the VM object
1701          */
1702         ssb_lock(&so->so_snd, M_WAITOK);
1703         vm_object_hold_shared(obj);
1704
1705         /*
1706          * Loop through the pages in the file, starting with the requested
1707          * offset. Get a file page (do I/O if necessary), map the file page
1708          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1709          * it on the socket.
1710          */
1711         for (off = offset; ;
1712              off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) {
1713                 vm_pindex_t pindex;
1714                 vm_offset_t pgoff;
1715                 long space;
1716                 int loops;
1717
1718                 pindex = OFF_TO_IDX(off);
1719                 loops = 0;
1720
1721 retry_lookup:
1722                 /*
1723                  * Calculate the amount to transfer. Not to exceed a page,
1724                  * the EOF, or the passed in nbytes.
1725                  */
1726                 xfsize = vp->v_filesize - off;
1727                 if (xfsize > PAGE_SIZE)
1728                         xfsize = PAGE_SIZE;
1729                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1730                 if (PAGE_SIZE - pgoff < xfsize)
1731                         xfsize = PAGE_SIZE - pgoff;
1732                 if (nbytes && xfsize > (nbytes - xbytes))
1733                         xfsize = nbytes - xbytes;
1734                 if (xfsize <= 0)
1735                         break;
1736                 /*
1737                  * Optimize the non-blocking case by looking at the socket space
1738                  * before going to the extra work of constituting the sf_buf.
1739                  */
1740                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1741                         space = ssb_space_prealloc(&so->so_snd);
1742                 else
1743                         space = ssb_space(&so->so_snd);
1744
1745                 if ((fp->f_flag & FNONBLOCK) && space <= 0) {
1746                         if (so->so_state & SS_CANTSENDMORE)
1747                                 error = EPIPE;
1748                         else
1749                                 error = EAGAIN;
1750                         goto done;
1751                 }
1752
1753                 /*
1754                  * Attempt to look up the page.
1755                  *
1756                  * Try to find the data using a shared vm_object token and
1757                  * vm_page_lookup_sbusy_try() first.
1758                  *
1759                  * If data is missing, use a UIO_NOCOPY VOP_READ to load
1760                  * the missing data and loop back up.  We avoid all sorts
1761                  * of problems by not trying to hold onto the page during
1762                  * the I/O.
1763                  *
1764                  * NOTE: The soft-busy will temporary block filesystem
1765                  *       truncation operations when a file is removed
1766                  *       while the sendfile is running.
1767                  */
1768                 pg = vm_page_lookup_sbusy_try(obj, pindex, pgoff, xfsize);
1769                 if (pg == NULL) {
1770                         struct uio auio;
1771                         struct iovec aiov;
1772                         int bsize;
1773
1774                         if (++loops > 100000) {
1775                                 kprintf("sendfile: VOP operation failed "
1776                                         "to retain page\n");
1777                                 error = EIO;
1778                                 goto done;
1779                         }
1780
1781                         vm_object_drop(obj);
1782                         bsize = vp->v_mount->mnt_stat.f_iosize;
1783                         auio.uio_iov = &aiov;
1784                         auio.uio_iovcnt = 1;
1785                         aiov.iov_base = 0;
1786                         aiov.iov_len = MAXBSIZE;
1787                         auio.uio_resid = MAXBSIZE;
1788                         auio.uio_offset = trunc_page(off);
1789                         auio.uio_segflg = UIO_NOCOPY;
1790                         auio.uio_rw = UIO_READ;
1791                         auio.uio_td = td;
1792
1793                         vn_lock(vp, LK_SHARED | LK_RETRY);
1794                         error = VOP_READ(vp, &auio,
1795                                          IO_VMIO | ((MAXBSIZE / bsize) << 16),
1796                                          td->td_ucred);
1797                         vn_unlock(vp);
1798                         vm_object_hold_shared(obj);
1799
1800                         if (error)
1801                                 goto done;
1802                         goto retry_lookup;
1803                 }
1804
1805                 /*
1806                  * Get a sendfile buf. We usually wait as long as necessary,
1807                  * but this wait can be interrupted.
1808                  */
1809                 if ((sf = sf_buf_alloc(pg)) == NULL) {
1810                         vm_page_sbusy_drop(pg);
1811                         /* vm_page_try_to_free(pg); */
1812                         error = EINTR;
1813                         goto done;
1814                 }
1815
1816                 /*
1817                  * Get an mbuf header and set it up as having external storage.
1818                  */
1819                 MGETHDR(m, M_WAITOK, MT_DATA);
1820                 if (m == NULL) {
1821                         error = ENOBUFS;
1822                         vm_page_sbusy_drop(pg);
1823                         /* vm_page_try_to_free(pg); */
1824                         sf_buf_free(sf);
1825                         goto done;
1826                 }
1827
1828                 m->m_ext.ext_free = sf_buf_mfree;
1829                 m->m_ext.ext_ref = sf_buf_ref;
1830                 m->m_ext.ext_arg = sf;
1831                 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1832                 m->m_ext.ext_size = PAGE_SIZE;
1833                 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1834                 m->m_flags |= M_EXT;
1835                 m->m_pkthdr.len = m->m_len = xfsize;
1836                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1837
1838                 if (mheader != NULL) {
1839                         hbytes = mheader->m_pkthdr.len;
1840                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1841                         m_cat(mheader, m);
1842                         m = mheader;
1843                         mheader = NULL;
1844                 } else {
1845                         hbytes = 0;
1846                 }
1847
1848                 /*
1849                  * Add the buffer to the socket buffer chain.
1850                  */
1851                 crit_enter();
1852 retry_space:
1853                 /*
1854                  * Make sure that the socket is still able to take more data.
1855                  * CANTSENDMORE being true usually means that the connection
1856                  * was closed. so_error is true when an error was sensed after
1857                  * a previous send.
1858                  * The state is checked after the page mapping and buffer
1859                  * allocation above since those operations may block and make
1860                  * any socket checks stale. From this point forward, nothing
1861                  * blocks before the pru_send (or more accurately, any blocking
1862                  * results in a loop back to here to re-check).
1863                  */
1864                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1865                         if (so->so_state & SS_CANTSENDMORE) {
1866                                 error = EPIPE;
1867                         } else {
1868                                 error = so->so_error;
1869                                 so->so_error = 0;
1870                         }
1871                         m_freem(m);
1872                         crit_exit();
1873                         goto done;
1874                 }
1875                 /*
1876                  * Wait for socket space to become available. We do this just
1877                  * after checking the connection state above in order to avoid
1878                  * a race condition with ssb_wait().
1879                  */
1880                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1881                         space = ssb_space_prealloc(&so->so_snd);
1882                 else
1883                         space = ssb_space(&so->so_snd);
1884
1885                 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1886                         if (fp->f_flag & FNONBLOCK) {
1887                                 m_freem(m);
1888                                 crit_exit();
1889                                 error = EAGAIN;
1890                                 goto done;
1891                         }
1892                         error = ssb_wait(&so->so_snd);
1893                         /*
1894                          * An error from ssb_wait usually indicates that we've
1895                          * been interrupted by a signal. If we've sent anything
1896                          * then return bytes sent, otherwise return the error.
1897                          */
1898                         if (error) {
1899                                 m_freem(m);
1900                                 crit_exit();
1901                                 goto done;
1902                         }
1903                         goto retry_space;
1904                 }
1905
1906                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1907                         for (mp = m; mp != NULL; mp = mp->m_next)
1908                                 ssb_preallocstream(&so->so_snd, mp);
1909                 }
1910                 if (use_sendfile_async)
1911                         error = so_pru_senda(so, 0, m, NULL, NULL, td);
1912                 else
1913                         error = so_pru_send(so, 0, m, NULL, NULL, td);
1914
1915                 crit_exit();
1916                 if (error)
1917                         goto done;
1918         }
1919         if (mheader != NULL) {
1920                 *sbytes += mheader->m_pkthdr.len;
1921
1922                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1923                         for (mp = mheader; mp != NULL; mp = mp->m_next)
1924                                 ssb_preallocstream(&so->so_snd, mp);
1925                 }
1926                 if (use_sendfile_async)
1927                         error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1928                 else
1929                         error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1930
1931                 mheader = NULL;
1932         }
1933 done:
1934         vm_object_drop(obj);
1935         ssb_unlock(&so->so_snd);
1936 done1:
1937         dropfp(td, sfd, fp);
1938 done0:
1939         if (mheader != NULL)
1940                 m_freem(mheader);
1941         return (error);
1942 }