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                 if (control == NULL) {
 926                         error = ENOBUFS;
 927                         goto cleanup;
 928                 }
 929                 control->m_len = msg.msg_controllen;
 930                 error = copyin(msg.msg_control, mtod(control, caddr_t),
 931                                msg.msg_controllen);
 932                 if (error) {
 933                         m_free(control);
 934                         goto cleanup;
 935                 }
 936         }
 937
 938         error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
 939                              &sysmsg->sysmsg_szresult);
 940
 941 cleanup:
 942         iovec_free(&iov, aiov);
 943 cleanup2:
 944         if (sa)
 945                 kfree(sa, M_SONAME);
 946         return (error);
 947 }
 948
 949 /*
 950  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
 951  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
 952  * Don't forget to FREE() and m_free() these if they are returned.
 953  */
 954 int
 955 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
 956              struct mbuf **control, int *flags, size_t *res)
 957 {
 958         struct thread *td = curthread;
 959         struct file *fp;
 960         size_t len;
 961         int error;
 962         int lflags;
 963         struct socket *so;
 964 #ifdef KTRACE
 965         struct iovec *ktriov = NULL;
 966         struct uio ktruio;
 967 #endif
 968
 969         error = holdsock(td, s, &fp);
 970         if (error)
 971                 return (error);
 972 #ifdef KTRACE
 973         if (KTRPOINT(td, KTR_GENIO)) {
 974                 int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
 975
 976                 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
 977                 bcopy(auio->uio_iov, ktriov, iovlen);
 978                 ktruio = *auio;
 979         }
 980 #endif
 981         len = auio->uio_resid;
 982         so = (struct socket *)fp->f_data;
 983
 984         if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
 985                 if (fp->f_flag & FNONBLOCK) {
 986                         if (flags) {
 987                                 *flags |= MSG_FNONBLOCKING;
 988                         } else {
 989                                 lflags = MSG_FNONBLOCKING;
 990                                 flags = &lflags;
 991                         }
 992                 }
 993         }
 994
 995         error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
 996         if (error) {
 997                 if (auio->uio_resid != len && (error == ERESTART ||
 998                     error == EINTR || error == EWOULDBLOCK))
 999                         error = 0;
1000         }
1001 #ifdef KTRACE
1002         if (ktriov != NULL) {
1003                 if (error == 0) {
1004                         ktruio.uio_iov = ktriov;
1005                         ktruio.uio_resid = len - auio->uio_resid;
1006                         ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
1007                 }
1008                 kfree(ktriov, M_TEMP);
1009         }
1010 #endif
1011         if (error == 0)
1012                 *res = len - auio->uio_resid;
1013         dropfp(td, s, fp);
1014
1015         return (error);
1016 }
1017
1018 /*
1019  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
1020  *                      caddr_t from, int *fromlenaddr)
1021  *
1022  * MPALMOSTSAFE
1023  */
1024 int
1025 sys_recvfrom(struct sysmsg *sysmsg, const struct recvfrom_args *uap)
1026 {
1027         struct thread *td = curthread;
1028         struct uio auio;
1029         struct iovec aiov;
1030         struct sockaddr *sa = NULL;
1031         int error, fromlen;
1032         int flags;
1033
1034         if (uap->from && uap->fromlenaddr) {
1035                 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
1036                 if (error)
1037                         return (error);
1038                 if (fromlen < 0)
1039                         return (EINVAL);
1040         } else {
1041                 fromlen = 0;
1042         }
1043         aiov.iov_base = uap->buf;
1044         aiov.iov_len = uap->len;
1045         auio.uio_iov = &aiov;
1046         auio.uio_iovcnt = 1;
1047         auio.uio_offset = 0;
1048         auio.uio_resid = uap->len;
1049         auio.uio_segflg = UIO_USERSPACE;
1050         auio.uio_rw = UIO_READ;
1051         auio.uio_td = td;
1052         flags = uap->flags;
1053
1054         error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
1055                              &flags, &sysmsg->sysmsg_szresult);
1056
1057         if (error == 0 && uap->from) {
1058                 /* note: sa may still be NULL */
1059                 if (sa) {
1060                         fromlen = MIN(fromlen, sa->sa_len);
1061                         prison_local_ip(curthread, sa);
1062                         error = copyout(sa, uap->from, fromlen);
1063                 } else {
1064                         fromlen = 0;
1065                 }
1066                 if (error == 0) {
1067                         error = copyout(&fromlen, uap->fromlenaddr,
1068                                         sizeof(fromlen));
1069                 }
1070         }
1071         if (sa)
1072                 kfree(sa, M_SONAME);
1073
1074         return (error);
1075 }
1076
1077 /*
1078  * recvmsg_args(int s, struct msghdr *msg, int flags)
1079  *
1080  * MPALMOSTSAFE
1081  */
1082 int
1083 sys_recvmsg(struct sysmsg *sysmsg, const struct recvmsg_args *uap)
1084 {
1085         struct thread *td = curthread;
1086         struct msghdr msg;
1087         struct uio auio;
1088         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1089         struct mbuf *m, *control = NULL;
1090         struct sockaddr *sa = NULL;
1091         caddr_t ctlbuf;
1092         socklen_t *ufromlenp, *ucontrollenp;
1093         int error, fromlen, controllen, len, flags, *uflagsp;
1094
1095         /*
1096          * This copyin handles everything except the iovec.
1097          */
1098         error = copyin(uap->msg, &msg, sizeof(msg));
1099         if (error)
1100                 return (error);
1101
1102         if (msg.msg_name && msg.msg_namelen < 0)
1103                 return (EINVAL);
1104         if (msg.msg_control && msg.msg_controllen < 0)
1105                 return (EINVAL);
1106
1107         ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1108                     msg_namelen));
1109         ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
1110                        msg_controllen));
1111         uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
1112                                                         msg_flags));
1113
1114         /*
1115          * Populate auio.
1116          */
1117         error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
1118                              &auio.uio_resid);
1119         if (error)
1120                 return (error);
1121         auio.uio_iov = iov;
1122         auio.uio_iovcnt = msg.msg_iovlen;
1123         auio.uio_offset = 0;
1124         auio.uio_segflg = UIO_USERSPACE;
1125         auio.uio_rw = UIO_READ;
1126         auio.uio_td = td;
1127
1128         flags = uap->flags;
1129
1130         error = kern_recvmsg(uap->s,
1131                              (msg.msg_name ? &sa : NULL), &auio,
1132                              (msg.msg_control ? &control : NULL), &flags,
1133                              &sysmsg->sysmsg_szresult);
1134
1135         /*
1136          * Conditionally copyout the name and populate the namelen field.
1137          */
1138         if (error == 0 && msg.msg_name) {
1139                 /* note: sa may still be NULL */
1140                 if (sa != NULL) {
1141                         fromlen = MIN(msg.msg_namelen, sa->sa_len);
1142                         prison_local_ip(curthread, sa);
1143                         error = copyout(sa, msg.msg_name, fromlen);
1144                 } else {
1145                         fromlen = 0;
1146                 }
1147                 if (error == 0)
1148                         error = copyout(&fromlen, ufromlenp,
1149                             sizeof(*ufromlenp));
1150         }
1151
1152         /*
1153          * Copyout msg.msg_control and msg.msg_controllen.
1154          */
1155         if (error == 0 && msg.msg_control) {
1156                 len = msg.msg_controllen;
1157                 m = control;
1158                 ctlbuf = (caddr_t)msg.msg_control;
1159
1160                 while(m && len > 0) {
1161                         unsigned int tocopy;
1162
1163                         if (len >= m->m_len) {
1164                                 tocopy = m->m_len;
1165                         } else {
1166                                 msg.msg_flags |= MSG_CTRUNC;
1167                                 tocopy = len;
1168                         }
1169
1170                         error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1171                         if (error)
1172                                 goto cleanup;
1173
1174                         ctlbuf += tocopy;
1175                         len -= tocopy;
1176                         m = m->m_next;
1177                 }
1178                 controllen = ctlbuf - (caddr_t)msg.msg_control;
1179                 error = copyout(&controllen, ucontrollenp,
1180                     sizeof(*ucontrollenp));
1181         }
1182
1183         if (error == 0)
1184                 error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1185
1186 cleanup:
1187         if (sa)
1188                 kfree(sa, M_SONAME);
1189         iovec_free(&iov, aiov);
1190         if (control)
1191                 m_freem(control);
1192         return (error);
1193 }
1194
1195 /*
1196  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1197  * in kernel pointer instead of a userland pointer.  This allows us
1198  * to manipulate socket options in the emulation code.
1199  */
1200 int
1201 kern_setsockopt(int s, struct sockopt *sopt)
1202 {
1203         struct thread *td = curthread;
1204         struct file *fp;
1205         int error;
1206
1207         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1208                 return (EFAULT);
1209         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1210                 return (EINVAL);
1211         if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1212                 return (EINVAL);
1213
1214         error = holdsock(td, s, &fp);
1215         if (error)
1216                 return (error);
1217
1218         error = sosetopt((struct socket *)fp->f_data, sopt);
1219         dropfp(td, s, fp);
1220
1221         return (error);
1222 }
1223
1224 /*
1225  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1226  *
1227  * MPALMOSTSAFE
1228  */
1229 int
1230 sys_setsockopt(struct sysmsg *sysmsg, const struct setsockopt_args *uap)
1231 {
1232         struct thread *td = curthread;
1233         struct sockopt sopt;
1234         int error;
1235
1236         sopt.sopt_level = uap->level;
1237         sopt.sopt_name = uap->name;
1238         sopt.sopt_valsize = uap->valsize;
1239         sopt.sopt_td = td;
1240         sopt.sopt_val = NULL;
1241
1242         if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */
1243                 return (EINVAL);
1244         if (uap->val) {
1245                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1246                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1247                 if (error)
1248                         goto out;
1249         }
1250
1251         error = kern_setsockopt(uap->s, &sopt);
1252 out:
1253         if (uap->val)
1254                 kfree(sopt.sopt_val, M_TEMP);
1255         return(error);
1256 }
1257
1258 /*
1259  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1260  * in kernel pointer instead of a userland pointer.  This allows us
1261  * to manipulate socket options in the emulation code.
1262  */
1263 int
1264 kern_getsockopt(int s, struct sockopt *sopt)
1265 {
1266         struct thread *td = curthread;
1267         struct file *fp;
1268         int error;
1269
1270         if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1271                 return (EFAULT);
1272         if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1273                 return (EINVAL);
1274
1275         error = holdsock(td, s, &fp);
1276         if (error)
1277                 return (error);
1278
1279         error = sogetopt((struct socket *)fp->f_data, sopt);
1280         dropfp(td, s, fp);
1281
1282         return (error);
1283 }
1284
1285 /*
1286  * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1287  *
1288  * MPALMOSTSAFE
1289  */
1290 int
1291 sys_getsockopt(struct sysmsg *sysmsg, const struct getsockopt_args *uap)
1292 {
1293         struct thread *td = curthread;
1294         struct sockopt sopt;
1295         int error, valsize, valszmax, mflag = 0;
1296
1297         if (uap->val) {
1298                 error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1299                 if (error)
1300                         return (error);
1301         } else {
1302                 valsize = 0;
1303         }
1304
1305         sopt.sopt_level = uap->level;
1306         sopt.sopt_name = uap->name;
1307         sopt.sopt_valsize = valsize;
1308         sopt.sopt_td = td;
1309         sopt.sopt_val = NULL;
1310
1311         if (td->td_proc->p_ucred->cr_uid == 0) {
1312                 valszmax = SOMAXOPT_SIZE0;
1313                 mflag = M_NULLOK;
1314         } else {
1315                 valszmax = SOMAXOPT_SIZE;
1316         }
1317         if (sopt.sopt_valsize > valszmax) /* unsigned */
1318                 return (EINVAL);
1319         if (uap->val) {
1320                 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP,
1321                     M_WAITOK | mflag);
1322                 if (sopt.sopt_val == NULL)
1323                         return (ENOBUFS);
1324                 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1325                 if (error)
1326                         goto out;
1327         }
1328
1329         error = kern_getsockopt(uap->s, &sopt);
1330         if (error)
1331                 goto out;
1332         valsize = sopt.sopt_valsize;
1333         error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1334         if (error)
1335                 goto out;
1336         if (uap->val)
1337                 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1338 out:
1339         if (uap->val)
1340                 kfree(sopt.sopt_val, M_TEMP);
1341         return (error);
1342 }
1343
1344 /*
1345  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1346  * This allows kern_getsockname() to return a pointer to an allocated struct
1347  * sockaddr which must be freed later with FREE().  The caller must
1348  * initialize *name to NULL.
1349  */
1350 int
1351 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1352 {
1353         struct thread *td = curthread;
1354         struct file *fp;
1355         struct socket *so;
1356         struct sockaddr *sa = NULL;
1357         int error;
1358
1359         error = holdsock(td, s, &fp);
1360         if (error)
1361                 return (error);
1362         if (*namelen < 0) {
1363                 fdrop(fp);
1364                 return (EINVAL);
1365         }
1366         so = (struct socket *)fp->f_data;
1367         error = so_pru_sockaddr(so, &sa);
1368         if (error == 0) {
1369                 if (sa == NULL) {
1370                         *namelen = 0;
1371                 } else {
1372                         *namelen = MIN(*namelen, sa->sa_len);
1373                         *name = sa;
1374                 }
1375         }
1376         dropfp(td, s, fp);
1377
1378         return (error);
1379 }
1380
1381 /*
1382  * getsockname_args(int fdes, caddr_t asa, int *alen)
1383  *
1384  * Get socket name.
1385  *
1386  * MPALMOSTSAFE
1387  */
1388 int
1389 sys_getsockname(struct sysmsg *sysmsg, const struct getsockname_args *uap)
1390 {
1391         struct sockaddr *sa = NULL;
1392         struct sockaddr satmp;
1393         int error, sa_len_in, sa_len_out;
1394
1395         error = copyin(uap->alen, &sa_len_in, sizeof(sa_len_in));
1396         if (error)
1397                 return (error);
1398
1399         sa_len_out = sa_len_in;
1400         error = kern_getsockname(uap->fdes, &sa, &sa_len_out);
1401
1402         if (error == 0) {
1403                 if (sa) {
1404                         prison_local_ip(curthread, sa);
1405                         error = copyout(sa, uap->asa, sa_len_out);
1406                 } else {
1407                         /*
1408                          * unnamed uipc sockets don't bother storing
1409                          * sockaddr, simulate an AF_LOCAL sockaddr.
1410                          */
1411                         sa_len_out = sizeof(satmp);
1412                         if (sa_len_out > sa_len_in)
1413                                 sa_len_out = sa_len_in;
1414                         if (sa_len_out < 0)
1415                                 sa_len_out = 0;
1416                         bzero(&satmp, sizeof(satmp));
1417                         satmp.sa_len = sa_len_out;
1418                         satmp.sa_family = AF_LOCAL;
1419                         error = copyout(&satmp, uap->asa, sa_len_out);
1420                 }
1421         }
1422         if (error == 0 && sa_len_out != sa_len_in)
1423                 error = copyout(&sa_len_out, uap->alen, sizeof(*uap->alen));
1424         if (sa)
1425                 kfree(sa, M_SONAME);
1426         return (error);
1427 }
1428
1429 /*
1430  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1431  * This allows kern_getpeername() to return a pointer to an allocated struct
1432  * sockaddr which must be freed later with FREE().  The caller must
1433  * initialize *name to NULL.
1434  */
1435 int
1436 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1437 {
1438         struct thread *td = curthread;
1439         struct file *fp;
1440         struct socket *so;
1441         struct sockaddr *sa = NULL;
1442         int error;
1443
1444         error = holdsock(td, s, &fp);
1445         if (error)
1446                 return (error);
1447         if (*namelen < 0) {
1448                 fdrop(fp);
1449                 return (EINVAL);
1450         }
1451         so = (struct socket *)fp->f_data;
1452         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1453                 fdrop(fp);
1454                 return (ENOTCONN);
1455         }
1456         error = so_pru_peeraddr(so, &sa);
1457         if (error == 0) {
1458                 if (sa == NULL) {
1459                         *namelen = 0;
1460                 } else {
1461                         *namelen = MIN(*namelen, sa->sa_len);
1462                         *name = sa;
1463                 }
1464         }
1465         dropfp(td, s, fp);
1466
1467         return (error);
1468 }
1469
1470 /*
1471  * getpeername_args(int fdes, caddr_t asa, int *alen)
1472  *
1473  * Get name of peer for connected socket.
1474  *
1475  * MPALMOSTSAFE
1476  */
1477 int
1478 sys_getpeername(struct sysmsg *sysmsg, const struct getpeername_args *uap)
1479 {
1480         struct sockaddr *sa = NULL;
1481         int error, sa_len;
1482
1483         error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1484         if (error)
1485                 return (error);
1486
1487         error = kern_getpeername(uap->fdes, &sa, &sa_len);
1488
1489         if (error == 0) {
1490                 prison_local_ip(curthread, sa);
1491                 error = copyout(sa, uap->asa, sa_len);
1492         }
1493         if (error == 0)
1494                 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1495         if (sa)
1496                 kfree(sa, M_SONAME);
1497         return (error);
1498 }
1499
1500 int
1501 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1502 {
1503         struct sockaddr *sa;
1504         int error;
1505
1506         *namp = NULL;
1507         if (len > SOCK_MAXADDRLEN)
1508                 return ENAMETOOLONG;
1509         if (len < offsetof(struct sockaddr, sa_data[0]))
1510                 return EDOM;
1511         sa = kmalloc(len, M_SONAME, M_WAITOK);
1512         error = copyin(uaddr, sa, len);
1513         if (error) {
1514                 kfree(sa, M_SONAME);
1515         } else {
1516                 sa->sa_len = len;
1517                 *namp = sa;
1518         }
1519         return error;
1520 }
1521
1522 /*
1523  * Detach a mapped page and release resources back to the system.
1524  * We must release our wiring and if the object is ripped out
1525  * from under the vm_page we become responsible for freeing the
1526  * page.
1527  *
1528  * MPSAFE
1529  */
1530 static void
1531 sf_buf_mfree(void *arg)
1532 {
1533         struct sf_buf *sf = arg;
1534         vm_page_t m;
1535
1536         m = sf_buf_page(sf);
1537         if (sf_buf_free(sf)) {
1538                 /* sf invalid now */
1539                 vm_page_sbusy_drop(m);
1540 #if 0
1541                 if (m->object == NULL &&
1542                     m->wire_count == 0 &&
1543                     (m->flags & PG_NEED_COMMIT) == 0) {
1544                         vm_page_free(m);
1545                 } else {
1546                         vm_page_wakeup(m);
1547                 }
1548 #endif
1549         }
1550 }
1551
1552 /*
1553  * sendfile(2).
1554  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1555  *       struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1556  *
1557  * Send a file specified by 'fd' and starting at 'offset' to a socket
1558  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1559  * nbytes == 0. Optionally add a header and/or trailer to the socket
1560  * output. If specified, write the total number of bytes sent into *sbytes.
1561  *
1562  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1563  * the headers to count against the remaining bytes to be sent from
1564  * the file descriptor.  We may wish to implement a compatibility syscall
1565  * in the future.
1566  *
1567  * MPALMOSTSAFE
1568  */
1569 int
1570 sys_sendfile(struct sysmsg *sysmsg, const struct sendfile_args *uap)
1571 {
1572         struct thread *td = curthread;
1573         struct file *fp;
1574         struct vnode *vp = NULL;
1575         struct sf_hdtr hdtr;
1576         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1577         struct uio auio;
1578         struct mbuf *mheader = NULL;
1579         size_t hbytes = 0;
1580         size_t tbytes;
1581         off_t hdtr_size = 0;
1582         off_t sbytes;
1583         int error;
1584
1585         /*
1586          * Do argument checking. Must be a regular file in, stream
1587          * type and connected socket out, positive offset.
1588          */
1589         fp = holdfp(td, uap->fd, FREAD);
1590         if (fp == NULL) {
1591                 return (EBADF);
1592         }
1593         if (fp->f_type != DTYPE_VNODE) {
1594                 fdrop(fp);
1595                 return (EINVAL);
1596         }
1597         vp = (struct vnode *)fp->f_data;
1598         vref(vp);
1599         dropfp(td, uap->fd, fp);
1600
1601         /*
1602          * If specified, get the pointer to the sf_hdtr struct for
1603          * any headers/trailers.
1604          */
1605         if (uap->hdtr) {
1606                 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1607                 if (error)
1608                         goto done;
1609                 /*
1610                  * Send any headers.
1611                  */
1612                 if (hdtr.headers) {
1613                         error = iovec_copyin(hdtr.headers, &iov, aiov,
1614                                              hdtr.hdr_cnt, &hbytes);
1615                         if (error)
1616                                 goto done;
1617                         auio.uio_iov = iov;
1618                         auio.uio_iovcnt = hdtr.hdr_cnt;
1619                         auio.uio_offset = 0;
1620                         auio.uio_segflg = UIO_USERSPACE;
1621                         auio.uio_rw = UIO_WRITE;
1622                         auio.uio_td = td;
1623                         auio.uio_resid = hbytes;
1624
1625                         mheader = m_uiomove(&auio);
1626
1627                         iovec_free(&iov, aiov);
1628                         if (mheader == NULL)
1629                                 goto done;
1630                 }
1631         }
1632
1633         error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1634                               &sbytes, uap->flags);
1635         if (error)
1636                 goto done;
1637
1638         /*
1639          * Send trailers. Wimp out and use writev(2).
1640          */
1641         if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1642                 error = iovec_copyin(hdtr.trailers, &iov, aiov,
1643                                      hdtr.trl_cnt, &auio.uio_resid);
1644                 if (error)
1645                         goto done;
1646                 auio.uio_iov = iov;
1647                 auio.uio_iovcnt = hdtr.trl_cnt;
1648                 auio.uio_offset = 0;
1649                 auio.uio_segflg = UIO_USERSPACE;
1650                 auio.uio_rw = UIO_WRITE;
1651                 auio.uio_td = td;
1652
1653                 tbytes = 0;     /* avoid gcc warnings */
1654                 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1655
1656                 iovec_free(&iov, aiov);
1657                 if (error)
1658                         goto done;
1659                 hdtr_size += tbytes;    /* trailer bytes successfully sent */
1660         }
1661
1662 done:
1663         if (vp)
1664                 vrele(vp);
1665         if (uap->sbytes != NULL) {
1666                 sbytes += hdtr_size;
1667                 copyout(&sbytes, uap->sbytes, sizeof(off_t));
1668         }
1669         return (error);
1670 }
1671
1672 int
1673 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1674               struct mbuf *mheader, off_t *sbytes, int flags)
1675 {
1676         struct thread *td = curthread;
1677         struct vm_object *obj;
1678         struct socket *so;
1679         struct file *fp;
1680         struct mbuf *m, *mp;
1681         struct sf_buf *sf;
1682         struct vm_page *pg;
1683         off_t off, xfsize, xbytes;
1684         off_t hbytes = 0;
1685         int error = 0;
1686
1687         if (vp->v_type != VREG) {
1688                 error = EINVAL;
1689                 goto done0;
1690         }
1691         if ((obj = vp->v_object) == NULL) {
1692                 error = EINVAL;
1693                 goto done0;
1694         }
1695         error = holdsock(td, sfd, &fp);
1696         if (error)
1697                 goto done0;
1698         so = (struct socket *)fp->f_data;
1699         if (so->so_type != SOCK_STREAM) {
1700                 error = EINVAL;
1701                 goto done1;
1702         }
1703         if ((so->so_state & SS_ISCONNECTED) == 0) {
1704                 error = ENOTCONN;
1705                 goto done1;
1706         }
1707         if (offset < 0) {
1708                 error = EINVAL;
1709                 goto done1;
1710         }
1711
1712         /*
1713          * preallocation is required for asynchronous passing of mbufs,
1714          * otherwise we can wind up building up an infinite number of
1715          * mbufs during the asynchronous latency.
1716          */
1717         if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) {
1718                 error = EINVAL;
1719                 goto done1;
1720         }
1721
1722         *sbytes = 0;
1723         xbytes = 0;
1724
1725         /*
1726          * Protect against multiple writers to the socket.
1727          * We need at least a shared lock on the VM object
1728          */
1729         ssb_lock(&so->so_snd, M_WAITOK);
1730         vm_object_hold_shared(obj);
1731
1732         /*
1733          * Loop through the pages in the file, starting with the requested
1734          * offset. Get a file page (do I/O if necessary), map the file page
1735          * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1736          * it on the socket.
1737          */
1738         for (off = offset; ;
1739              off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) {
1740                 vm_pindex_t pindex;
1741                 vm_offset_t pgoff;
1742                 long space;
1743                 int loops;
1744
1745                 pindex = OFF_TO_IDX(off);
1746                 loops = 0;
1747
1748 retry_lookup:
1749                 /*
1750                  * Calculate the amount to transfer. Not to exceed a page,
1751                  * the EOF, or the passed in nbytes.
1752                  */
1753                 xfsize = vp->v_filesize - off;
1754                 if (xfsize > PAGE_SIZE)
1755                         xfsize = PAGE_SIZE;
1756                 pgoff = (vm_offset_t)(off & PAGE_MASK);
1757                 if (PAGE_SIZE - pgoff < xfsize)
1758                         xfsize = PAGE_SIZE - pgoff;
1759                 if (nbytes && xfsize > (nbytes - xbytes))
1760                         xfsize = nbytes - xbytes;
1761                 if (xfsize <= 0)
1762                         break;
1763                 /*
1764                  * Optimize the non-blocking case by looking at the socket space
1765                  * before going to the extra work of constituting the sf_buf.
1766                  */
1767                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1768                         space = ssb_space_prealloc(&so->so_snd);
1769                 else
1770                         space = ssb_space(&so->so_snd);
1771
1772                 if ((fp->f_flag & FNONBLOCK) && space <= 0) {
1773                         if (so->so_state & SS_CANTSENDMORE)
1774                                 error = EPIPE;
1775                         else
1776                                 error = EAGAIN;
1777                         goto done;
1778                 }
1779
1780                 /*
1781                  * Attempt to look up the page.
1782                  *
1783                  * Try to find the data using a shared vm_object token and
1784                  * vm_page_lookup_sbusy_try() first.
1785                  *
1786                  * If data is missing, use a UIO_NOCOPY VOP_READ to load
1787                  * the missing data and loop back up.  We avoid all sorts
1788                  * of problems by not trying to hold onto the page during
1789                  * the I/O.
1790                  *
1791                  * NOTE: The soft-busy will temporary block filesystem
1792                  *       truncation operations when a file is removed
1793                  *       while the sendfile is running.
1794                  */
1795                 pg = vm_page_lookup_sbusy_try(obj, pindex, pgoff, xfsize);
1796                 if (pg == NULL) {
1797                         struct uio auio;
1798                         struct iovec aiov;
1799                         int bsize;
1800
1801                         if (++loops > 100000) {
1802                                 kprintf("sendfile: VOP operation failed "
1803                                         "to retain page\n");
1804                                 error = EIO;
1805                                 goto done;
1806                         }
1807
1808                         vm_object_drop(obj);
1809                         bsize = vp->v_mount->mnt_stat.f_iosize;
1810                         auio.uio_iov = &aiov;
1811                         auio.uio_iovcnt = 1;
1812                         aiov.iov_base = 0;
1813                         aiov.iov_len = MAXBSIZE;
1814                         auio.uio_resid = MAXBSIZE;
1815                         auio.uio_offset = trunc_page(off);
1816                         auio.uio_segflg = UIO_NOCOPY;
1817                         auio.uio_rw = UIO_READ;
1818                         auio.uio_td = td;
1819
1820                         vn_lock(vp, LK_SHARED | LK_RETRY);
1821                         error = VOP_READ_FP(vp, &auio,
1822                                          IO_VMIO | ((MAXBSIZE / bsize) << 16),
1823                                          td->td_ucred, fp);
1824                         vn_unlock(vp);
1825                         vm_object_hold_shared(obj);
1826
1827                         if (error)
1828                                 goto done;
1829                         goto retry_lookup;
1830                 }
1831
1832                 /*
1833                  * Get a sendfile buf. We usually wait as long as necessary,
1834                  * but this wait can be interrupted.
1835                  */
1836                 if ((sf = sf_buf_alloc(pg)) == NULL) {
1837                         vm_page_sbusy_drop(pg);
1838                         /* vm_page_try_to_free(pg); */
1839                         error = EINTR;
1840                         goto done;
1841                 }
1842
1843                 /*
1844                  * Get an mbuf header and set it up as having external storage.
1845                  */
1846                 MGETHDR(m, M_WAITOK, MT_DATA);
1847                 if (m == NULL) {
1848                         error = ENOBUFS;
1849                         vm_page_sbusy_drop(pg);
1850                         /* vm_page_try_to_free(pg); */
1851                         sf_buf_free(sf);
1852                         goto done;
1853                 }
1854
1855                 m->m_ext.ext_free = sf_buf_mfree;
1856                 m->m_ext.ext_ref = sf_buf_ref;
1857                 m->m_ext.ext_arg = sf;
1858                 m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1859                 m->m_ext.ext_size = PAGE_SIZE;
1860                 m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1861                 m->m_flags |= M_EXT;
1862                 m->m_pkthdr.len = m->m_len = xfsize;
1863                 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1864
1865                 if (mheader != NULL) {
1866                         hbytes = mheader->m_pkthdr.len;
1867                         mheader->m_pkthdr.len += m->m_pkthdr.len;
1868                         m_cat(mheader, m);
1869                         m = mheader;
1870                         mheader = NULL;
1871                 } else {
1872                         hbytes = 0;
1873                 }
1874
1875                 /*
1876                  * Add the buffer to the socket buffer chain.
1877                  */
1878                 crit_enter();
1879 retry_space:
1880                 /*
1881                  * Make sure that the socket is still able to take more data.
1882                  * CANTSENDMORE being true usually means that the connection
1883                  * was closed. so_error is true when an error was sensed after
1884                  * a previous send.
1885                  * The state is checked after the page mapping and buffer
1886                  * allocation above since those operations may block and make
1887                  * any socket checks stale. From this point forward, nothing
1888                  * blocks before the pru_send (or more accurately, any blocking
1889                  * results in a loop back to here to re-check).
1890                  */
1891                 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1892                         if (so->so_state & SS_CANTSENDMORE) {
1893                                 error = EPIPE;
1894                         } else {
1895                                 error = so->so_error;
1896                                 so->so_error = 0;
1897                         }
1898                         m_freem(m);
1899                         crit_exit();
1900                         goto done;
1901                 }
1902                 /*
1903                  * Wait for socket space to become available. We do this just
1904                  * after checking the connection state above in order to avoid
1905                  * a race condition with ssb_wait().
1906                  */
1907                 if (so->so_snd.ssb_flags & SSB_PREALLOC)
1908                         space = ssb_space_prealloc(&so->so_snd);
1909                 else
1910                         space = ssb_space(&so->so_snd);
1911
1912                 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) {
1913                         if (fp->f_flag & FNONBLOCK) {
1914                                 m_freem(m);
1915                                 crit_exit();
1916                                 error = EAGAIN;
1917                                 goto done;
1918                         }
1919                         error = ssb_wait(&so->so_snd);
1920                         /*
1921                          * An error from ssb_wait usually indicates that we've
1922                          * been interrupted by a signal. If we've sent anything
1923                          * then return bytes sent, otherwise return the error.
1924                          */
1925                         if (error) {
1926                                 m_freem(m);
1927                                 crit_exit();
1928                                 goto done;
1929                         }
1930                         goto retry_space;
1931                 }
1932
1933                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1934                         for (mp = m; mp != NULL; mp = mp->m_next)
1935                                 ssb_preallocstream(&so->so_snd, mp);
1936                 }
1937                 if (use_sendfile_async)
1938                         error = so_pru_senda(so, 0, m, NULL, NULL, td);
1939                 else
1940                         error = so_pru_send(so, 0, m, NULL, NULL, td);
1941
1942                 crit_exit();
1943                 if (error)
1944                         goto done;
1945         }
1946         if (mheader != NULL) {
1947                 *sbytes += mheader->m_pkthdr.len;
1948
1949                 if (so->so_snd.ssb_flags & SSB_PREALLOC) {
1950                         for (mp = mheader; mp != NULL; mp = mp->m_next)
1951                                 ssb_preallocstream(&so->so_snd, mp);
1952                 }
1953                 if (use_sendfile_async)
1954                         error = so_pru_senda(so, 0, mheader, NULL, NULL, td);
1955                 else
1956                         error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1957
1958                 mheader = NULL;
1959         }
1960 done:
1961         vm_object_drop(obj);
1962         ssb_unlock(&so->so_snd);
1963 done1:
1964         dropfp(td, sfd, fp);
1965 done0:
1966         if (mheader != NULL)
1967                 m_freem(mheader);
1968         return (error);
1969 }