sys/kern/sys_generic.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  * (c) UNIX System Laboratories, Inc.
   5  * All or some portions of this file are derived from material licensed
   6  * to the University of California by American Telephone and Telegraph
   7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
   8  * the permission of UNIX System Laboratories, Inc.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  * 3. All advertising materials mentioning features or use of this software
  19  *    must display the following acknowledgement:
  20  *      This product includes software developed by the University of
  21  *      California, Berkeley and its contributors.
  22  * 4. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      @(#)sys_generic.c       8.5 (Berkeley) 1/21/94
  39  * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $
  40  * $DragonFly: src/sys/kern/sys_generic.c,v 1.49 2008/05/05 22:09:44 dillon Exp $
  41  */
  42
  43 #include "opt_ktrace.h"
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/sysproto.h>
  48 #include <sys/event.h>
  49 #include <sys/filedesc.h>
  50 #include <sys/filio.h>
  51 #include <sys/fcntl.h>
  52 #include <sys/file.h>
  53 #include <sys/proc.h>
  54 #include <sys/signalvar.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/uio.h>
  57 #include <sys/kernel.h>
  58 #include <sys/kern_syscall.h>
  59 #include <sys/malloc.h>
  60 #include <sys/mapped_ioctl.h>
  61 #include <sys/poll.h>
  62 #include <sys/queue.h>
  63 #include <sys/resourcevar.h>
  64 #include <sys/socketops.h>
  65 #include <sys/sysctl.h>
  66 #include <sys/sysent.h>
  67 #include <sys/buf.h>
  68 #ifdef KTRACE
  69 #include <sys/ktrace.h>
  70 #endif
  71 #include <vm/vm.h>
  72 #include <vm/vm_page.h>
  73
  74 #include <sys/file2.h>
  75 #include <sys/mplock2.h>
  76 #include <sys/spinlock2.h>
  77
  78 #include <machine/limits.h>
  79
  80 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
  81 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer");
  82 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
  83 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
  84
  85 typedef struct kfd_set {
  86         fd_mask fds_bits[2];
  87 } kfd_set;
  88
  89 enum select_copyin_states {
  90     COPYIN_READ, COPYIN_WRITE, COPYIN_EXCEPT, COPYIN_DONE };
  91
  92 struct select_kevent_copyin_args {
  93         kfd_set         *read_set;
  94         kfd_set         *write_set;
  95         kfd_set         *except_set;
  96         int             active_set;     /* One of select_copyin_states */
  97         struct lwp      *lwp;           /* Pointer to our lwp */
  98         int             num_fds;        /* Number of file descriptors (syscall arg) */
  99         int             proc_fds;       /* Processed fd's (wraps) */
 100         int             error;          /* Returned to userland */
 101 };
 102
 103 struct poll_kevent_copyin_args {
 104         struct lwp      *lwp;
 105         struct pollfd   *fds;
 106         int             nfds;
 107         int             pfds;
 108         int             error;
 109 };
 110
 111 static struct lwkt_token mioctl_token = LWKT_TOKEN_INITIALIZER(mioctl_token);
 112
 113 static int      doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
 114                          struct timespec *ts, int *res);
 115 static int      dopoll(int nfds, struct pollfd *fds, struct timespec *ts,
 116                        int *res);
 117 static int      dofileread(int, struct file *, struct uio *, int, size_t *);
 118 static int      dofilewrite(int, struct file *, struct uio *, int, size_t *);
 119
 120 /*
 121  * Read system call.
 122  *
 123  * MPSAFE
 124  */
 125 int
 126 sys_read(struct read_args *uap)
 127 {
 128         struct thread *td = curthread;
 129         struct uio auio;
 130         struct iovec aiov;
 131         int error;
 132
 133         if ((ssize_t)uap->nbyte < 0)
 134                 error = EINVAL;
 135
 136         aiov.iov_base = uap->buf;
 137         aiov.iov_len = uap->nbyte;
 138         auio.uio_iov = &aiov;
 139         auio.uio_iovcnt = 1;
 140         auio.uio_offset = -1;
 141         auio.uio_resid = uap->nbyte;
 142         auio.uio_rw = UIO_READ;
 143         auio.uio_segflg = UIO_USERSPACE;
 144         auio.uio_td = td;
 145
 146         error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
 147         return(error);
 148 }
 149
 150 /*
 151  * Positioned (Pread) read system call
 152  *
 153  * MPSAFE
 154  */
 155 int
 156 sys_extpread(struct extpread_args *uap)
 157 {
 158         struct thread *td = curthread;
 159         struct uio auio;
 160         struct iovec aiov;
 161         int error;
 162         int flags;
 163
 164         if ((ssize_t)uap->nbyte < 0)
 165                 return(EINVAL);
 166
 167         aiov.iov_base = uap->buf;
 168         aiov.iov_len = uap->nbyte;
 169         auio.uio_iov = &aiov;
 170         auio.uio_iovcnt = 1;
 171         auio.uio_offset = uap->offset;
 172         auio.uio_resid = uap->nbyte;
 173         auio.uio_rw = UIO_READ;
 174         auio.uio_segflg = UIO_USERSPACE;
 175         auio.uio_td = td;
 176
 177         flags = uap->flags & O_FMASK;
 178         if (uap->offset != (off_t)-1)
 179                 flags |= O_FOFFSET;
 180
 181         error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
 182         return(error);
 183 }
 184
 185 /*
 186  * Scatter read system call.
 187  *
 188  * MPSAFE
 189  */
 190 int
 191 sys_readv(struct readv_args *uap)
 192 {
 193         struct thread *td = curthread;
 194         struct uio auio;
 195         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 196         int error;
 197
 198         error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
 199                              &auio.uio_resid);
 200         if (error)
 201                 return (error);
 202         auio.uio_iov = iov;
 203         auio.uio_iovcnt = uap->iovcnt;
 204         auio.uio_offset = -1;
 205         auio.uio_rw = UIO_READ;
 206         auio.uio_segflg = UIO_USERSPACE;
 207         auio.uio_td = td;
 208
 209         error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
 210
 211         iovec_free(&iov, aiov);
 212         return (error);
 213 }
 214
 215
 216 /*
 217  * Scatter positioned read system call.
 218  *
 219  * MPSAFE
 220  */
 221 int
 222 sys_extpreadv(struct extpreadv_args *uap)
 223 {
 224         struct thread *td = curthread;
 225         struct uio auio;
 226         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 227         int error;
 228         int flags;
 229
 230         error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
 231                              &auio.uio_resid);
 232         if (error)
 233                 return (error);
 234         auio.uio_iov = iov;
 235         auio.uio_iovcnt = uap->iovcnt;
 236         auio.uio_offset = uap->offset;
 237         auio.uio_rw = UIO_READ;
 238         auio.uio_segflg = UIO_USERSPACE;
 239         auio.uio_td = td;
 240
 241         flags = uap->flags & O_FMASK;
 242         if (uap->offset != (off_t)-1)
 243                 flags |= O_FOFFSET;
 244
 245         error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
 246
 247         iovec_free(&iov, aiov);
 248         return(error);
 249 }
 250
 251 /*
 252  * MPSAFE
 253  */
 254 int
 255 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
 256 {
 257         struct thread *td = curthread;
 258         struct proc *p = td->td_proc;
 259         struct file *fp;
 260         int error;
 261
 262         KKASSERT(p);
 263
 264         fp = holdfp(p->p_fd, fd, FREAD);
 265         if (fp == NULL)
 266                 return (EBADF);
 267         if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
 268                 error = ESPIPE;
 269         } else {
 270                 error = dofileread(fd, fp, auio, flags, res);
 271         }
 272         fdrop(fp);
 273         return(error);
 274 }
 275
 276 /*
 277  * Common code for readv and preadv that reads data in
 278  * from a file using the passed in uio, offset, and flags.
 279  *
 280  * MPALMOSTSAFE - ktrace needs help
 281  */
 282 static int
 283 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
 284 {
 285         int error;
 286         size_t len;
 287 #ifdef KTRACE
 288         struct thread *td = curthread;
 289         struct iovec *ktriov = NULL;
 290         struct uio ktruio;
 291 #endif
 292
 293 #ifdef KTRACE
 294         /*
 295          * if tracing, save a copy of iovec
 296          */
 297         if (KTRPOINT(td, KTR_GENIO))  {
 298                 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
 299
 300                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 301                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 302                 ktruio = *auio;
 303         }
 304 #endif
 305         len = auio->uio_resid;
 306         error = fo_read(fp, auio, fp->f_cred, flags);
 307         if (error) {
 308                 if (auio->uio_resid != len && (error == ERESTART ||
 309                     error == EINTR || error == EWOULDBLOCK))
 310                         error = 0;
 311         }
 312 #ifdef KTRACE
 313         if (ktriov != NULL) {
 314                 if (error == 0) {
 315                         ktruio.uio_iov = ktriov;
 316                         ktruio.uio_resid = len - auio->uio_resid;
 317                         get_mplock();
 318                         ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
 319                         rel_mplock();
 320                 }
 321                 FREE(ktriov, M_TEMP);
 322         }
 323 #endif
 324         if (error == 0)
 325                 *res = len - auio->uio_resid;
 326
 327         return(error);
 328 }
 329
 330 /*
 331  * Write system call
 332  *
 333  * MPSAFE
 334  */
 335 int
 336 sys_write(struct write_args *uap)
 337 {
 338         struct thread *td = curthread;
 339         struct uio auio;
 340         struct iovec aiov;
 341         int error;
 342
 343         if ((ssize_t)uap->nbyte < 0)
 344                 error = EINVAL;
 345
 346         aiov.iov_base = (void *)(uintptr_t)uap->buf;
 347         aiov.iov_len = uap->nbyte;
 348         auio.uio_iov = &aiov;
 349         auio.uio_iovcnt = 1;
 350         auio.uio_offset = -1;
 351         auio.uio_resid = uap->nbyte;
 352         auio.uio_rw = UIO_WRITE;
 353         auio.uio_segflg = UIO_USERSPACE;
 354         auio.uio_td = td;
 355
 356         error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
 357
 358         return(error);
 359 }
 360
 361 /*
 362  * Pwrite system call
 363  *
 364  * MPSAFE
 365  */
 366 int
 367 sys_extpwrite(struct extpwrite_args *uap)
 368 {
 369         struct thread *td = curthread;
 370         struct uio auio;
 371         struct iovec aiov;
 372         int error;
 373         int flags;
 374
 375         if ((ssize_t)uap->nbyte < 0)
 376                 error = EINVAL;
 377
 378         aiov.iov_base = (void *)(uintptr_t)uap->buf;
 379         aiov.iov_len = uap->nbyte;
 380         auio.uio_iov = &aiov;
 381         auio.uio_iovcnt = 1;
 382         auio.uio_offset = uap->offset;
 383         auio.uio_resid = uap->nbyte;
 384         auio.uio_rw = UIO_WRITE;
 385         auio.uio_segflg = UIO_USERSPACE;
 386         auio.uio_td = td;
 387
 388         flags = uap->flags & O_FMASK;
 389         if (uap->offset != (off_t)-1)
 390                 flags |= O_FOFFSET;
 391         error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
 392         return(error);
 393 }
 394
 395 /*
 396  * MPSAFE
 397  */
 398 int
 399 sys_writev(struct writev_args *uap)
 400 {
 401         struct thread *td = curthread;
 402         struct uio auio;
 403         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 404         int error;
 405
 406         error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
 407                              &auio.uio_resid);
 408         if (error)
 409                 return (error);
 410         auio.uio_iov = iov;
 411         auio.uio_iovcnt = uap->iovcnt;
 412         auio.uio_offset = -1;
 413         auio.uio_rw = UIO_WRITE;
 414         auio.uio_segflg = UIO_USERSPACE;
 415         auio.uio_td = td;
 416
 417         error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
 418
 419         iovec_free(&iov, aiov);
 420         return (error);
 421 }
 422
 423
 424 /*
 425  * Gather positioned write system call
 426  *
 427  * MPSAFE
 428  */
 429 int
 430 sys_extpwritev(struct extpwritev_args *uap)
 431 {
 432         struct thread *td = curthread;
 433         struct uio auio;
 434         struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
 435         int error;
 436         int flags;
 437
 438         error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
 439                              &auio.uio_resid);
 440         if (error)
 441                 return (error);
 442         auio.uio_iov = iov;
 443         auio.uio_iovcnt = uap->iovcnt;
 444         auio.uio_offset = uap->offset;
 445         auio.uio_rw = UIO_WRITE;
 446         auio.uio_segflg = UIO_USERSPACE;
 447         auio.uio_td = td;
 448
 449         flags = uap->flags & O_FMASK;
 450         if (uap->offset != (off_t)-1)
 451                 flags |= O_FOFFSET;
 452
 453         error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
 454
 455         iovec_free(&iov, aiov);
 456         return(error);
 457 }
 458
 459 /*
 460  * MPSAFE
 461  */
 462 int
 463 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
 464 {
 465         struct thread *td = curthread;
 466         struct proc *p = td->td_proc;
 467         struct file *fp;
 468         int error;
 469
 470         KKASSERT(p);
 471
 472         fp = holdfp(p->p_fd, fd, FWRITE);
 473         if (fp == NULL)
 474                 return (EBADF);
 475         else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
 476                 error = ESPIPE;
 477         } else {
 478                 error = dofilewrite(fd, fp, auio, flags, res);
 479         }
 480
 481         fdrop(fp);
 482         return (error);
 483 }
 484
 485 /*
 486  * Common code for writev and pwritev that writes data to
 487  * a file using the passed in uio, offset, and flags.
 488  *
 489  * MPALMOSTSAFE - ktrace needs help
 490  */
 491 static int
 492 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
 493 {
 494         struct thread *td = curthread;
 495         struct lwp *lp = td->td_lwp;
 496         int error;
 497         size_t len;
 498 #ifdef KTRACE
 499         struct iovec *ktriov = NULL;
 500         struct uio ktruio;
 501 #endif
 502
 503 #ifdef KTRACE
 504         /*
 505          * if tracing, save a copy of iovec and uio
 506          */
 507         if (KTRPOINT(td, KTR_GENIO))  {
 508                 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
 509
 510                 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
 511                 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
 512                 ktruio = *auio;
 513         }
 514 #endif
 515         len = auio->uio_resid;
 516         error = fo_write(fp, auio, fp->f_cred, flags);
 517         if (error) {
 518                 if (auio->uio_resid != len && (error == ERESTART ||
 519                     error == EINTR || error == EWOULDBLOCK))
 520                         error = 0;
 521                 /* Socket layer is responsible for issuing SIGPIPE. */
 522                 if (error == EPIPE) {
 523                         get_mplock();
 524                         lwpsignal(lp->lwp_proc, lp, SIGPIPE);
 525                         rel_mplock();
 526                 }
 527         }
 528 #ifdef KTRACE
 529         if (ktriov != NULL) {
 530                 if (error == 0) {
 531                         ktruio.uio_iov = ktriov;
 532                         ktruio.uio_resid = len - auio->uio_resid;
 533                         get_mplock();
 534                         ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
 535                         rel_mplock();
 536                 }
 537                 FREE(ktriov, M_TEMP);
 538         }
 539 #endif
 540         if (error == 0)
 541                 *res = len - auio->uio_resid;
 542
 543         return(error);
 544 }
 545
 546 /*
 547  * Ioctl system call
 548  *
 549  * MPSAFE
 550  */
 551 int
 552 sys_ioctl(struct ioctl_args *uap)
 553 {
 554         int error;
 555
 556         error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, &uap->sysmsg);
 557         return (error);
 558 }
 559
 560 struct ioctl_map_entry {
 561         const char *subsys;
 562         struct ioctl_map_range *cmd_ranges;
 563         LIST_ENTRY(ioctl_map_entry) entries;
 564 };
 565
 566 /*
 567  * The true heart of all ioctl syscall handlers (native, emulation).
 568  * If map != NULL, it will be searched for a matching entry for com,
 569  * and appropriate conversions/conversion functions will be utilized.
 570  *
 571  * MPSAFE
 572  */
 573 int
 574 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
 575              struct sysmsg *msg)
 576 {
 577         struct thread *td = curthread;
 578         struct proc *p = td->td_proc;
 579         struct ucred *cred;
 580         struct file *fp;
 581         struct ioctl_map_range *iomc = NULL;
 582         int error;
 583         u_int size;
 584         u_long ocom = com;
 585         caddr_t data, memp;
 586         int tmp;
 587 #define STK_PARAMS      128
 588         union {
 589             char stkbuf[STK_PARAMS];
 590             long align;
 591         } ubuf;
 592
 593         KKASSERT(p);
 594         cred = td->td_ucred;
 595
 596         fp = holdfp(p->p_fd, fd, FREAD|FWRITE);
 597         if (fp == NULL)
 598                 return(EBADF);
 599
 600         if (map != NULL) {      /* obey translation map */
 601                 u_long maskcmd;
 602                 struct ioctl_map_entry *e;
 603
 604                 maskcmd = com & map->mask;
 605
 606                 lwkt_gettoken(&mioctl_token);
 607                 LIST_FOREACH(e, &map->mapping, entries) {
 608                         for (iomc = e->cmd_ranges; iomc->start != 0 ||
 609                              iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
 610                              iomc->mapfunc != NULL;
 611                              iomc++) {
 612                                 if (maskcmd >= iomc->start &&
 613                                     maskcmd <= iomc->end)
 614                                         break;
 615                         }
 616
 617                         /* Did we find a match? */
 618                         if (iomc->start != 0 || iomc->maptocmd != 0 ||
 619                             iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
 620                                 break;
 621                 }
 622                 lwkt_reltoken(&mioctl_token);
 623
 624                 if (iomc == NULL ||
 625                     (iomc->start == 0 && iomc->maptocmd == 0
 626                      && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
 627                         kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
 628                                map->sys, fd, maskcmd,
 629                                (int)((maskcmd >> 8) & 0xff),
 630                                (int)(maskcmd & 0xff));
 631                         error = EINVAL;
 632                         goto done;
 633                 }
 634
 635                 /*
 636                  * If it's a non-range one to one mapping, maptocmd should be
 637                  * correct. If it's a ranged one to one mapping, we pass the
 638                  * original value of com, and for a range mapped to a different
 639                  * range, we always need a mapping function to translate the
 640                  * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
 641                  */
 642                 if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
 643                         com = iomc->maptocmd;
 644                 } else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
 645                         if (iomc->mapfunc != NULL)
 646                                 com = iomc->mapfunc(iomc->start, iomc->end,
 647                                                     iomc->start, iomc->end,
 648                                                     com, com);
 649                 } else {
 650                         if (iomc->mapfunc != NULL) {
 651                                 com = iomc->mapfunc(iomc->start, iomc->end,
 652                                                     iomc->maptocmd, iomc->maptoend,
 653                                                     com, ocom);
 654                         } else {
 655                                 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
 656                                        map->sys, fd, maskcmd,
 657                                        (int)((maskcmd >> 8) & 0xff),
 658                                        (int)(maskcmd & 0xff));
 659                                 error = EINVAL;
 660                                 goto done;
 661                         }
 662                 }
 663         }
 664
 665         switch (com) {
 666         case FIONCLEX:
 667                 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
 668                 goto done;
 669         case FIOCLEX:
 670                 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
 671                 goto done;
 672         }
 673
 674         /*
 675          * Interpret high order word to find amount of data to be
 676          * copied to/from the user's address space.
 677          */
 678         size = IOCPARM_LEN(com);
 679         if (size > IOCPARM_MAX) {
 680                 error = ENOTTY;
 681                 goto done;
 682         }
 683
 684         if (size > sizeof (ubuf.stkbuf)) {
 685                 memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
 686                 data = memp;
 687         } else {
 688                 memp = NULL;
 689                 data = ubuf.stkbuf;
 690         }
 691         if ((com & IOC_IN) != 0) {
 692                 if (size != 0) {
 693                         error = copyin(uspc_data, data, (size_t)size);
 694                         if (error) {
 695                                 if (memp != NULL)
 696                                         kfree(memp, M_IOCTLOPS);
 697                                 goto done;
 698                         }
 699                 } else {
 700                         *(caddr_t *)data = uspc_data;
 701                 }
 702         } else if ((com & IOC_OUT) != 0 && size) {
 703                 /*
 704                  * Zero the buffer so the user always
 705                  * gets back something deterministic.
 706                  */
 707                 bzero(data, (size_t)size);
 708         } else if ((com & IOC_VOID) != 0) {
 709                 *(caddr_t *)data = uspc_data;
 710         }
 711
 712         switch (com) {
 713         case FIONBIO:
 714                 if ((tmp = *(int *)data))
 715                         atomic_set_int(&fp->f_flag, FNONBLOCK);
 716                 else
 717                         atomic_clear_int(&fp->f_flag, FNONBLOCK);
 718                 error = 0;
 719                 break;
 720
 721         case FIOASYNC:
 722                 if ((tmp = *(int *)data))
 723                         atomic_set_int(&fp->f_flag, FASYNC);
 724                 else
 725                         atomic_clear_int(&fp->f_flag, FASYNC);
 726                 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
 727                 break;
 728
 729         default:
 730                 /*
 731                  *  If there is a override function,
 732                  *  call it instead of directly routing the call
 733                  */
 734                 if (map != NULL && iomc->wrapfunc != NULL)
 735                         error = iomc->wrapfunc(fp, com, ocom, data, cred);
 736                 else
 737                         error = fo_ioctl(fp, com, data, cred, msg);
 738                 /*
 739                  * Copy any data to user, size was
 740                  * already set and checked above.
 741                  */
 742                 if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
 743                         error = copyout(data, uspc_data, (size_t)size);
 744                 break;
 745         }
 746         if (memp != NULL)
 747                 kfree(memp, M_IOCTLOPS);
 748 done:
 749         fdrop(fp);
 750         return(error);
 751 }
 752
 753 /*
 754  * MPSAFE
 755  */
 756 int
 757 mapped_ioctl_register_handler(struct ioctl_map_handler *he)
 758 {
 759         struct ioctl_map_entry *ne;
 760
 761         KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL &&
 762                  he->subsys != NULL && *he->subsys != '\0');
 763
 764         ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP,
 765                      M_WAITOK | M_ZERO);
 766
 767         ne->subsys = he->subsys;
 768         ne->cmd_ranges = he->cmd_ranges;
 769
 770         lwkt_gettoken(&mioctl_token);
 771         LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
 772         lwkt_reltoken(&mioctl_token);
 773
 774         return(0);
 775 }
 776
 777 /*
 778  * MPSAFE
 779  */
 780 int
 781 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
 782 {
 783         struct ioctl_map_entry *ne;
 784         int error = EINVAL;
 785
 786         KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
 787
 788         lwkt_gettoken(&mioctl_token);
 789         LIST_FOREACH(ne, &he->map->mapping, entries) {
 790                 if (ne->cmd_ranges == he->cmd_ranges) {
 791                         LIST_REMOVE(ne, entries);
 792                         kfree(ne, M_IOCTLMAP);
 793                         error = 0;
 794                         break;
 795                 }
 796         }
 797         lwkt_reltoken(&mioctl_token);
 798         return(error);
 799 }
 800
 801 static int      nselcoll;       /* Select collisions since boot */
 802 int     selwait;
 803 SYSCTL_INT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, "");
 804 static int      nseldebug;
 805 SYSCTL_INT(_kern, OID_AUTO, nseldebug, CTLFLAG_RW, &nseldebug, 0, "");
 806
 807 /*
 808  * Select system call.
 809  *
 810  * MPSAFE
 811  */
 812 int
 813 sys_select(struct select_args *uap)
 814 {
 815         struct timeval ktv;
 816         struct timespec *ktsp, kts;
 817         int error;
 818
 819         /*
 820          * Get timeout if any.
 821          */
 822         if (uap->tv != NULL) {
 823                 error = copyin(uap->tv, &ktv, sizeof (ktv));
 824                 if (error)
 825                         return (error);
 826                 TIMEVAL_TO_TIMESPEC(&ktv, &kts);
 827                 ktsp = &kts;
 828         } else {
 829                 ktsp = NULL;
 830         }
 831
 832         /*
 833          * Do real work.
 834          */
 835         error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
 836                          &uap->sysmsg_result);
 837
 838         return (error);
 839 }
 840
 841
 842 /*
 843  * Pselect system call.
 844  *
 845  * MPALMOSTSAFE
 846  */
 847 int
 848 sys_pselect(struct pselect_args *uap)
 849 {
 850         struct thread *td = curthread;
 851         struct lwp *lp = td->td_lwp;
 852         struct timespec *ktsp, kts;
 853         sigset_t sigmask;
 854         int error;
 855
 856         /*
 857          * Get timeout if any.
 858          */
 859         if (uap->ts != NULL) {
 860                 error = copyin(uap->ts, &kts, sizeof (kts));
 861                 if (error)
 862                         return (error);
 863                 ktsp = &kts;
 864         } else {
 865                 ktsp = NULL;
 866         }
 867
 868         /*
 869          * Install temporary signal mask if any provided.
 870          */
 871         if (uap->sigmask != NULL) {
 872                 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
 873                 if (error)
 874                         return (error);
 875                 get_mplock();
 876                 lp->lwp_oldsigmask = lp->lwp_sigmask;
 877                 SIG_CANTMASK(sigmask);
 878                 lp->lwp_sigmask = sigmask;
 879         } else {
 880                 get_mplock();
 881         }
 882
 883         /*
 884          * Do real job.
 885          */
 886         error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
 887                          &uap->sysmsg_result);
 888
 889         if (uap->sigmask != NULL) {
 890                 /* doselect() responsible for turning ERESTART into EINTR */
 891                 KKASSERT(error != ERESTART);
 892                 if (error == EINTR) {
 893                         /*
 894                          * We can't restore the previous signal mask now
 895                          * because it could block the signal that interrupted
 896                          * us.  So make a note to restore it after executing
 897                          * the handler.
 898                          */
 899                         lp->lwp_flag |= LWP_OLDMASK;
 900                 } else {
 901                         /*
 902                          * No handler to run. Restore previous mask immediately.
 903                          */
 904                         lp->lwp_sigmask = lp->lwp_oldsigmask;
 905                 }
 906         }
 907         rel_mplock();
 908
 909         return (error);
 910 }
 911
 912 static int
 913 select_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
 914 {
 915         struct select_kevent_copyin_args *skap = NULL;
 916         struct kevent *kev;
 917         int fd;
 918         kfd_set *fdp = NULL;
 919         short filter = 0;
 920         u_int fflags = 0;
 921
 922         skap = (struct select_kevent_copyin_args *)arg;
 923
 924         if (*events == maxevents)
 925                 return (0);
 926
 927         while (skap->active_set < COPYIN_DONE) {
 928                 switch (skap->active_set) {
 929                 case COPYIN_READ:
 930                         /*
 931                          * Register descriptors for the read filter
 932                          */
 933                         fdp = skap->read_set;
 934                         filter = EVFILT_READ;
 935                         fflags = NOTE_OLDAPI;
 936                         if (fdp)
 937                                 break;
 938                         ++skap->active_set;
 939                         skap->proc_fds = 0;
 940                         /* fall through */
 941                 case COPYIN_WRITE:
 942                         /*
 943                          * Register descriptors for the write filter
 944                          */
 945                         fdp = skap->write_set;
 946                         filter = EVFILT_WRITE;
 947                         fflags = NOTE_OLDAPI;
 948                         if (fdp)
 949                                 break;
 950                         ++skap->active_set;
 951                         skap->proc_fds = 0;
 952                         /* fall through */
 953                 case COPYIN_EXCEPT:
 954                         /*
 955                          * Register descriptors for the exception filter
 956                          */
 957                         fdp = skap->except_set;
 958                         filter = EVFILT_EXCEPT;
 959                         fflags = NOTE_OLDAPI | NOTE_OOB;
 960                         if (fdp)
 961                                 break;
 962                         ++skap->active_set;
 963                         skap->proc_fds = 0;
 964                         /* fall through */
 965                 case COPYIN_DONE:
 966                         /*
 967                          * Nothing left to register
 968                          */
 969                         return(0);
 970                         /* NOT REACHED */
 971                 }
 972
 973                 while (skap->proc_fds < skap->num_fds) {
 974                         fd = skap->proc_fds;
 975                         if (FD_ISSET(fd, fdp)) {
 976                                 kev = &kevp[*events];
 977                                 EV_SET(kev, fd, filter,
 978                                        EV_ADD|EV_ENABLE,
 979                                        fflags, 0,
 980                                        (void *)(uintptr_t)
 981                                         skap->lwp->lwp_kqueue_serial);
 982                                 FD_CLR(fd, fdp);
 983                                 ++*events;
 984
 985                                 if (nseldebug)
 986                                         kprintf("select fd %d filter %d serial %d\n",
 987                                                 fd, filter, skap->lwp->lwp_kqueue_serial);
 988                         }
 989                         ++skap->proc_fds;
 990                         if (*events == maxevents)
 991                                 return (0);
 992                 }
 993                 skap->active_set++;
 994                 skap->proc_fds = 0;
 995         }
 996
 997         return (0);
 998 }
 999
1000 static int
1001 select_copyout(void *arg, struct kevent *kevp, int count, int *res)
1002 {
1003         struct select_kevent_copyin_args *skap;
1004         struct kevent kev;
1005         int i = 0;
1006
1007         skap = (struct select_kevent_copyin_args *)arg;
1008
1009         for (i = 0; i < count; ++i) {
1010                 /*
1011                  * Filter out and delete spurious events
1012                  */
1013                 if ((u_int)(uintptr_t)kevp[i].udata !=
1014                     skap->lwp->lwp_kqueue_serial) {
1015                         kev = kevp[i];
1016                         kev.flags = EV_DISABLE|EV_DELETE;
1017                         kqueue_register(&skap->lwp->lwp_kqueue, &kev);
1018                         if (nseldebug)
1019                                 kprintf("select fd %ju mismatched serial %d\n",
1020                                         (uintmax_t)kevp[i].ident,
1021                                         skap->lwp->lwp_kqueue_serial);
1022                         continue;
1023                 }
1024
1025                 /*
1026                  * Handle errors
1027                  */
1028                 if (kevp[i].flags & EV_ERROR) {
1029                         switch(kevp[i].data) {
1030                         case EBADF:
1031                                 /*
1032                                  * A bad file descriptor is considered a
1033                                  * fatal error for select, bail out.
1034                                  */
1035                                 skap->error = EBADF;
1036                                 *res = 0;
1037                                 return (1);
1038                                 break;
1039                         default:
1040                                 /*
1041                                  * Select silently swallows any unknown errors
1042                                  * for descriptors in the read or write sets.
1043                                  *
1044                                  * ALWAYS filter out EOPNOTSUPP errors from
1045                                  * filters (at least until all filters support
1046                                  * EVFILT_EXCEPT)
1047                                  */
1048                                 if (kevp[i].filter != EVFILT_READ &&
1049                                     kevp[i].filter != EVFILT_WRITE &&
1050                                     kevp[i].data != EOPNOTSUPP) {
1051                                         skap->error = kevp[i].data;
1052                                         *res = 0;
1053                                         return (1);
1054                                 }
1055                                 break;
1056                         }
1057                         if (nseldebug)
1058                                 kprintf("select fd %ju filter %d error %jd\n",
1059                                         (uintmax_t)kevp[i].ident,
1060                                         kevp[i].filter,
1061                                         (intmax_t)kevp[i].data);
1062                         continue;
1063                 }
1064
1065                 switch (kevp[i].filter) {
1066                 case EVFILT_READ:
1067                         FD_SET(kevp[i].ident, skap->read_set);
1068                         break;
1069                 case EVFILT_WRITE:
1070                         FD_SET(kevp[i].ident, skap->write_set);
1071                         break;
1072                 case EVFILT_EXCEPT:
1073                         FD_SET(kevp[i].ident, skap->except_set);
1074                         break;
1075                 }
1076
1077                 ++*res;
1078         }
1079
1080         return (0);
1081 }
1082
1083 /*
1084  * Copy select bits in from userland.  Allocate kernel memory if the
1085  * set is large.
1086  */
1087 static int
1088 getbits(int bytes, fd_set *in_set, kfd_set **out_set, kfd_set *tmp_set)
1089 {
1090         int error;
1091
1092         if (in_set) {
1093                 if (bytes < sizeof(*tmp_set))
1094                         *out_set = tmp_set;
1095                 else
1096                         *out_set = kmalloc(bytes, M_SELECT, M_WAITOK);
1097                 error = copyin(in_set, *out_set, bytes);
1098         } else {
1099                 *out_set = NULL;
1100                 error = 0;
1101         }
1102         return (error);
1103 }
1104
1105 /*
1106  * Copy returned select bits back out to userland.
1107  */
1108 static int
1109 putbits(int bytes, kfd_set *in_set, fd_set *out_set)
1110 {
1111         int error;
1112
1113         if (in_set) {
1114                 error = copyout(in_set, out_set, bytes);
1115         } else {
1116                 error = 0;
1117         }
1118         return (error);
1119 }
1120
1121 /*
1122  * Common code for sys_select() and sys_pselect().
1123  *
1124  * in, out and ex are userland pointers.  ts must point to validated
1125  * kernel-side timeout value or NULL for infinite timeout.  res must
1126  * point to syscall return value.
1127  */
1128 static int
1129 doselect(int nd, fd_set *read, fd_set *write, fd_set *except,
1130          struct timespec *ts, int *res)
1131 {
1132         struct proc *p = curproc;
1133         struct select_kevent_copyin_args *kap, ka;
1134         int bytes, error;
1135         kfd_set read_tmp;
1136         kfd_set write_tmp;
1137         kfd_set except_tmp;
1138
1139         *res = 0;
1140         if (nd < 0)
1141                 return (EINVAL);
1142         if (nd > p->p_fd->fd_nfiles)            /* limit kmalloc */
1143                 nd = p->p_fd->fd_nfiles;
1144
1145         kap = &ka;
1146         kap->lwp = curthread->td_lwp;
1147         kap->num_fds = nd;
1148         kap->proc_fds = 0;
1149         kap->error = 0;
1150         kap->active_set = COPYIN_READ;
1151
1152         /*
1153          * Calculate bytes based on the number of __fd_mask[] array entries
1154          * multiplied by the size of __fd_mask.
1155          */
1156         bytes = howmany(nd, __NFDBITS) * sizeof(__fd_mask);
1157
1158         /* kap->read_set = NULL; not needed */
1159         kap->write_set = NULL;
1160         kap->except_set = NULL;
1161
1162         error = getbits(bytes, read, &kap->read_set, &read_tmp);
1163         if (error == 0)
1164                 error = getbits(bytes, write, &kap->write_set, &write_tmp);
1165         if (error == 0)
1166                 error = getbits(bytes, except, &kap->except_set, &except_tmp);
1167         if (error)
1168                 goto done;
1169
1170         /*
1171          * NOTE: Make sure the max events passed to kern_kevent() is
1172          *       effectively unlimited.  (nd * 3) accomplishes this.
1173          *
1174          *       (*res) continues to increment as returned events are
1175          *       loaded in.
1176          */
1177         error = kern_kevent(&kap->lwp->lwp_kqueue, 0x7FFFFFFF, res, kap,
1178                             select_copyin, select_copyout, ts);
1179         if (error == 0)
1180                 error = putbits(bytes, kap->read_set, read);
1181         if (error == 0)
1182                 error = putbits(bytes, kap->write_set, write);
1183         if (error == 0)
1184                 error = putbits(bytes, kap->except_set, except);
1185
1186         /*
1187          * An error from an individual event that should be passed
1188          * back to userland (EBADF)
1189          */
1190         if (kap->error)
1191                 error = kap->error;
1192
1193         /*
1194          * Clean up.
1195          */
1196 done:
1197         if (kap->read_set && kap->read_set != &read_tmp)
1198                 kfree(kap->read_set, M_SELECT);
1199         if (kap->write_set && kap->write_set != &write_tmp)
1200                 kfree(kap->write_set, M_SELECT);
1201         if (kap->except_set && kap->except_set != &except_tmp)
1202                 kfree(kap->except_set, M_SELECT);
1203
1204         kap->lwp->lwp_kqueue_serial += kap->num_fds;
1205
1206         return (error);
1207 }
1208
1209 /*
1210  * Poll system call.
1211  *
1212  * MPSAFE
1213  */
1214 int
1215 sys_poll(struct poll_args *uap)
1216 {
1217         struct timespec ts, *tsp;
1218         int error;
1219
1220         if (uap->timeout != INFTIM) {
1221                 ts.tv_sec = uap->timeout / 1000;
1222                 ts.tv_nsec = (uap->timeout % 1000) * 1000 * 1000;
1223                 tsp = &ts;
1224         } else {
1225                 tsp = NULL;
1226         }
1227
1228         error = dopoll(uap->nfds, uap->fds, tsp, &uap->sysmsg_result);
1229
1230         return (error);
1231 }
1232
1233 static int
1234 poll_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1235 {
1236         struct poll_kevent_copyin_args *pkap;
1237         struct pollfd *pfd;
1238         struct kevent *kev;
1239         int kev_count;
1240
1241         pkap = (struct poll_kevent_copyin_args *)arg;
1242
1243         while (pkap->pfds < pkap->nfds) {
1244                 pfd = &pkap->fds[pkap->pfds];
1245
1246                 /* Clear return events */
1247                 pfd->revents = 0;
1248
1249                 /* Do not check if fd is equal to -1 */
1250                 if (pfd->fd == -1) {
1251                         ++pkap->pfds;
1252                         continue;
1253                 }
1254
1255                 kev_count = 0;
1256                 if (pfd->events & (POLLIN | POLLRDNORM))
1257                         kev_count++;
1258                 if (pfd->events & (POLLOUT | POLLWRNORM))
1259                         kev_count++;
1260                 if (pfd->events & (POLLPRI | POLLRDBAND))
1261                         kev_count++;
1262
1263                 if (*events + kev_count > maxevents)
1264                         return (0);
1265
1266                 /*
1267                  * NOTE: A combined serial number and poll array index is
1268                  * stored in kev->udata.
1269                  */
1270                 kev = &kevp[*events];
1271                 if (pfd->events & (POLLIN | POLLRDNORM)) {
1272                         EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1273                                NOTE_OLDAPI, 0, (void *)(uintptr_t)
1274                                 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1275                 }
1276                 if (pfd->events & (POLLOUT | POLLWRNORM)) {
1277                         EV_SET(kev++, pfd->fd, EVFILT_WRITE, EV_ADD|EV_ENABLE,
1278                                NOTE_OLDAPI, 0, (void *)(uintptr_t)
1279                                 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1280                 }
1281                 if (pfd->events & (POLLPRI | POLLRDBAND)) {
1282                         EV_SET(kev++, pfd->fd, EVFILT_EXCEPT, EV_ADD|EV_ENABLE,
1283                                NOTE_OLDAPI | NOTE_OOB, 0,
1284                                (void *)(uintptr_t)
1285                                 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1286                 }
1287
1288                 if (nseldebug) {
1289                         kprintf("poll index %d/%d fd %d events %08x serial %d\n",
1290                                 pkap->pfds, pkap->nfds-1, pfd->fd, pfd->events,
1291                                 pkap->lwp->lwp_kqueue_serial);
1292                 }
1293
1294                 ++pkap->pfds;
1295                 (*events) += kev_count;
1296         }
1297
1298         return (0);
1299 }
1300
1301 static int
1302 poll_copyout(void *arg, struct kevent *kevp, int count, int *res)
1303 {
1304         struct poll_kevent_copyin_args *pkap;
1305         struct pollfd *pfd;
1306         struct kevent kev;
1307         int count_res;
1308         int i;
1309         u_int pi;
1310
1311         pkap = (struct poll_kevent_copyin_args *)arg;
1312
1313         for (i = 0; i < count; ++i) {
1314                 /*
1315                  * Extract the poll array index and delete spurious events.
1316                  * We can easily tell if the serial number is incorrect
1317                  * by checking whether the extracted index is out of range.
1318                  */
1319                 pi = (u_int)(uintptr_t)kevp[i].udata -
1320                      (u_int)pkap->lwp->lwp_kqueue_serial;
1321
1322                 if (pi >= pkap->nfds) {
1323                         kev = kevp[i];
1324                         kev.flags = EV_DISABLE|EV_DELETE;
1325                         kqueue_register(&pkap->lwp->lwp_kqueue, &kev);
1326                         if (nseldebug)
1327                                 kprintf("poll index %d out of range against serial %d\n",
1328                                         pi, pkap->lwp->lwp_kqueue_serial);
1329                         continue;
1330                 }
1331                 pfd = &pkap->fds[pi];
1332                 if (kevp[i].ident == pfd->fd) {
1333                         /*
1334                          * A single descriptor may generate an error against
1335                          * more than one filter, make sure to set the
1336                          * appropriate flags but do not increment (*res)
1337                          * more than once.
1338                          */
1339                         count_res = (pfd->revents == 0);
1340                         if (kevp[i].flags & EV_ERROR) {
1341                                 switch(kevp[i].data) {
1342                                 case EBADF:
1343                                         /* Bad file descriptor */
1344                                         if (count_res)
1345                                                 ++*res;
1346                                         pfd->revents |= POLLNVAL;
1347                                         break;
1348                                 default:
1349                                         /*
1350                                          * Poll silently swallows any unknown
1351                                          * errors except in the case of POLLPRI
1352                                          * (OOB/urgent data).
1353                                          *
1354                                          * ALWAYS filter out EOPNOTSUPP errors
1355                                          * from filters, common applications
1356                                          * set POLLPRI|POLLRDBAND and most
1357                                          * filters do not support EVFILT_EXCEPT.
1358                                          */
1359                                         if (kevp[i].filter != EVFILT_READ &&
1360                                             kevp[i].filter != EVFILT_WRITE &&
1361                                             kevp[i].data != EOPNOTSUPP) {
1362                                                 if (count_res == 0)
1363                                                         ++*res;
1364                                                 pfd->revents |= POLLERR;
1365                                         }
1366                                         break;
1367                                 }
1368                                 if (nseldebug) {
1369                                         kprintf("poll index %d fd %d "
1370                                                 "filter %d error %jd\n",
1371                                                 pi, pfd->fd,
1372                                                 kevp[i].filter,
1373                                                 (intmax_t)kevp[i].data);
1374                                 }
1375                                 continue;
1376                         }
1377
1378                         switch (kevp[i].filter) {
1379                         case EVFILT_READ:
1380 #if 0
1381                                 /*
1382                                  * EOF on the read side can indicate a
1383                                  * half-closed situation and not necessarily
1384                                  * a disconnect, so depend on the user
1385                                  * issuing a read() and getting 0 bytes back.
1386                                  */
1387                                 if (kevp[i].flags & EV_EOF)
1388                                         pfd->revents |= POLLHUP;
1389 #endif
1390                                 if (pfd->events & POLLIN)
1391                                         pfd->revents |= POLLIN;
1392                                 if (pfd->events & POLLRDNORM)
1393                                         pfd->revents |= POLLRDNORM;
1394                                 break;
1395                         case EVFILT_WRITE:
1396                                 /*
1397                                  * As per the OpenGroup POLLHUP is mutually
1398                                  * exclusive with the writability flags.  I
1399                                  * consider this a bit broken but...
1400                                  *
1401                                  * In this case a disconnect is implied even
1402                                  * for a half-closed (write side) situation.
1403                                  */
1404                                 if (kevp[i].flags & EV_EOF) {
1405                                         pfd->revents |= POLLHUP;
1406                                 } else {
1407                                         if (pfd->events & POLLOUT)
1408                                                 pfd->revents |= POLLOUT;
1409                                         if (pfd->events & POLLWRNORM)
1410                                                 pfd->revents |= POLLWRNORM;
1411                                 }
1412                                 break;
1413                         case EVFILT_EXCEPT:
1414                                 /*
1415                                  * EV_EOF should never be tagged for this
1416                                  * filter.
1417                                  */
1418                                 if (pfd->events & POLLPRI)
1419                                         pfd->revents |= POLLPRI;
1420                                 if (pfd->events & POLLRDBAND)
1421                                         pfd->revents |= POLLRDBAND;
1422                                 break;
1423                         }
1424
1425                         if (nseldebug) {
1426                                 kprintf("poll index %d/%d fd %d revents %08x\n",
1427                                         pi, pkap->nfds, pfd->fd, pfd->revents);
1428                         }
1429
1430                         if (count_res && pfd->revents)
1431                                 ++*res;
1432                 } else {
1433                         if (nseldebug) {
1434                                 kprintf("poll index %d mismatch %ju/%d\n",
1435                                         pi, (uintmax_t)kevp[i].ident, pfd->fd);
1436                         }
1437                 }
1438         }
1439
1440         return (0);
1441 }
1442
1443 static int
1444 dopoll(int nfds, struct pollfd *fds, struct timespec *ts, int *res)
1445 {
1446         struct poll_kevent_copyin_args ka;
1447         struct pollfd sfds[64];
1448         int bytes;
1449         int error;
1450
1451         *res = 0;
1452         if (nfds < 0)
1453                 return (EINVAL);
1454
1455         /*
1456          * This is a bit arbitrary but we need to limit internal kmallocs.
1457          */
1458         if (nfds > maxfilesperproc * 2)
1459                 nfds = maxfilesperproc * 2;
1460         bytes = sizeof(struct pollfd) * nfds;
1461
1462         ka.lwp = curthread->td_lwp;
1463         ka.nfds = nfds;
1464         ka.pfds = 0;
1465         ka.error = 0;
1466
1467         if (ka.nfds < 64)
1468                 ka.fds = sfds;
1469         else
1470                 ka.fds = kmalloc(bytes, M_SELECT, M_WAITOK);
1471
1472         error = copyin(fds, ka.fds, bytes);
1473         if (error == 0)
1474                 error = kern_kevent(&ka.lwp->lwp_kqueue, 0x7FFFFFFF, res, &ka,
1475                                     poll_copyin, poll_copyout, ts);
1476
1477         if (error == 0)
1478                 error = copyout(ka.fds, fds, bytes);
1479
1480         if (ka.fds != sfds)
1481                 kfree(ka.fds, M_SELECT);
1482
1483         ka.lwp->lwp_kqueue_serial += nfds;
1484
1485         return (error);
1486 }
1487
1488 static int
1489 socket_wait_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1490 {
1491         return (0);
1492 }
1493
1494 static int
1495 socket_wait_copyout(void *arg, struct kevent *kevp, int count, int *res)
1496 {
1497         ++*res;
1498         return (0);
1499 }
1500
1501 extern  struct fileops socketops;
1502
1503 /*
1504  * NOTE: Callers of socket_wait() must already have a reference on the
1505  *       socket.
1506  */
1507 int
1508 socket_wait(struct socket *so, struct timespec *ts, int *res)
1509 {
1510         struct thread *td = curthread;
1511         struct file *fp;
1512         struct kqueue kq;
1513         struct kevent kev;
1514         int error, fd;
1515
1516         if ((error = falloc(td->td_lwp, &fp, &fd)) != 0)
1517                 return (error);
1518
1519         fp->f_type = DTYPE_SOCKET;
1520         fp->f_flag = FREAD | FWRITE;
1521         fp->f_ops = &socketops;
1522         fp->f_data = so;
1523         fsetfd(td->td_lwp->lwp_proc->p_fd, fp, fd);
1524
1525         kqueue_init(&kq, td->td_lwp->lwp_proc->p_fd);
1526         EV_SET(&kev, fd, EVFILT_READ, EV_ADD|EV_ENABLE, 0, 0, NULL);
1527         if ((error = kqueue_register(&kq, &kev)) != 0) {
1528                 fdrop(fp);
1529                 return (error);
1530         }
1531
1532         error = kern_kevent(&kq, 1, res, NULL, socket_wait_copyin,
1533                             socket_wait_copyout, ts);
1534
1535         EV_SET(&kev, fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
1536         kqueue_register(&kq, &kev);
1537         fp->f_ops = &badfileops;
1538         fdrop(fp);
1539
1540         return (error);
1541 }
1542
1543 /*
1544  * OpenBSD poll system call.
1545  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1546  *
1547  * MPSAFE
1548  */
1549 int
1550 sys_openbsd_poll(struct openbsd_poll_args *uap)
1551 {
1552         return (sys_poll((struct poll_args *)uap));
1553 }
1554
1555 /*ARGSUSED*/
1556 int
1557 seltrue(cdev_t dev, int events)
1558 {
1559         return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1560 }