Home | History | Annotate | Line # | Download | only in kern
sys_generic.c revision 1.44.8.1
      1  1.44.8.1   thorpej /*	$NetBSD: sys_generic.c,v 1.44.8.1 1999/06/21 01:24:04 thorpej Exp $	*/
      2      1.15       cgd 
      3      1.15       cgd /*
      4      1.15       cgd  * Copyright (c) 1982, 1986, 1989, 1993
      5      1.15       cgd  *	The Regents of the University of California.  All rights reserved.
      6      1.15       cgd  * (c) UNIX System Laboratories, Inc.
      7      1.15       cgd  * All or some portions of this file are derived from material licensed
      8      1.15       cgd  * to the University of California by American Telephone and Telegraph
      9      1.15       cgd  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     10      1.15       cgd  * the permission of UNIX System Laboratories, Inc.
     11      1.15       cgd  *
     12      1.15       cgd  * Redistribution and use in source and binary forms, with or without
     13      1.15       cgd  * modification, are permitted provided that the following conditions
     14      1.15       cgd  * are met:
     15      1.15       cgd  * 1. Redistributions of source code must retain the above copyright
     16      1.15       cgd  *    notice, this list of conditions and the following disclaimer.
     17      1.15       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     18      1.15       cgd  *    notice, this list of conditions and the following disclaimer in the
     19      1.15       cgd  *    documentation and/or other materials provided with the distribution.
     20      1.15       cgd  * 3. All advertising materials mentioning features or use of this software
     21      1.15       cgd  *    must display the following acknowledgement:
     22      1.15       cgd  *	This product includes software developed by the University of
     23      1.15       cgd  *	California, Berkeley and its contributors.
     24      1.15       cgd  * 4. Neither the name of the University nor the names of its contributors
     25      1.15       cgd  *    may be used to endorse or promote products derived from this software
     26      1.15       cgd  *    without specific prior written permission.
     27      1.15       cgd  *
     28      1.15       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29      1.15       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30      1.15       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31      1.15       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32      1.15       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33      1.15       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34      1.15       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35      1.15       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36      1.15       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37      1.15       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38      1.15       cgd  * SUCH DAMAGE.
     39      1.15       cgd  *
     40      1.36      fvdl  *	@(#)sys_generic.c	8.9 (Berkeley) 2/14/95
     41      1.15       cgd  */
     42      1.37   thorpej 
     43      1.37   thorpej #include "opt_ktrace.h"
     44      1.15       cgd 
     45      1.15       cgd #include <sys/param.h>
     46      1.15       cgd #include <sys/systm.h>
     47      1.15       cgd #include <sys/filedesc.h>
     48      1.15       cgd #include <sys/ioctl.h>
     49      1.15       cgd #include <sys/file.h>
     50      1.15       cgd #include <sys/proc.h>
     51      1.15       cgd #include <sys/socketvar.h>
     52      1.22  christos #include <sys/signalvar.h>
     53      1.15       cgd #include <sys/uio.h>
     54      1.15       cgd #include <sys/kernel.h>
     55      1.15       cgd #include <sys/stat.h>
     56      1.15       cgd #include <sys/malloc.h>
     57      1.28   mycroft #include <sys/poll.h>
     58      1.15       cgd #ifdef KTRACE
     59      1.15       cgd #include <sys/ktrace.h>
     60      1.15       cgd #endif
     61      1.15       cgd 
     62      1.16       cgd #include <sys/mount.h>
     63      1.16       cgd #include <sys/syscallargs.h>
     64      1.22  christos 
     65      1.25   mycroft int selscan __P((struct proc *, fd_mask *, fd_mask *, int, register_t *));
     66      1.28   mycroft int pollscan __P((struct proc *, struct pollfd *, int, register_t *));
     67      1.22  christos 
     68      1.15       cgd /*
     69      1.15       cgd  * Read system call.
     70      1.15       cgd  */
     71      1.15       cgd /* ARGSUSED */
     72      1.22  christos int
     73      1.21   mycroft sys_read(p, v, retval)
     74      1.15       cgd 	struct proc *p;
     75      1.20   thorpej 	void *v;
     76      1.20   thorpej 	register_t *retval;
     77      1.20   thorpej {
     78      1.21   mycroft 	register struct sys_read_args /* {
     79      1.16       cgd 		syscallarg(int) fd;
     80      1.31       cgd 		syscallarg(void *) buf;
     81      1.31       cgd 		syscallarg(size_t) nbyte;
     82      1.20   thorpej 	} */ *uap = v;
     83      1.34   mycroft 	int fd = SCARG(uap, fd);
     84      1.15       cgd 	register struct file *fp;
     85      1.15       cgd 	register struct filedesc *fdp = p->p_fd;
     86      1.39   thorpej 
     87      1.39   thorpej 	if ((u_int)fd >= fdp->fd_nfiles ||
     88      1.39   thorpej 	    (fp = fdp->fd_ofiles[fd]) == NULL ||
     89  1.44.8.1   thorpej 	    (fp->f_iflags & FIF_WANTCLOSE) != 0 ||
     90      1.39   thorpej 	    (fp->f_flag & FREAD) == 0)
     91      1.39   thorpej 		return (EBADF);
     92      1.39   thorpej 
     93  1.44.8.1   thorpej 	FILE_USE(fp);
     94  1.44.8.1   thorpej 
     95  1.44.8.1   thorpej 	/* dofileread() will unuse the descriptor for us */
     96      1.39   thorpej 	return (dofileread(p, fd, fp, SCARG(uap, buf), SCARG(uap, nbyte),
     97      1.39   thorpej 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
     98      1.39   thorpej }
     99      1.39   thorpej 
    100      1.39   thorpej int
    101      1.39   thorpej dofileread(p, fd, fp, buf, nbyte, offset, flags, retval)
    102      1.39   thorpej 	struct proc *p;
    103      1.39   thorpej 	int fd;
    104      1.39   thorpej 	struct file *fp;
    105      1.39   thorpej 	void *buf;
    106      1.39   thorpej 	size_t nbyte;
    107      1.39   thorpej 	off_t *offset;
    108      1.39   thorpej 	int flags;
    109      1.39   thorpej 	register_t *retval;
    110      1.39   thorpej {
    111      1.15       cgd 	struct uio auio;
    112      1.15       cgd 	struct iovec aiov;
    113      1.15       cgd 	long cnt, error = 0;
    114      1.15       cgd #ifdef KTRACE
    115      1.15       cgd 	struct iovec ktriov;
    116      1.15       cgd #endif
    117      1.15       cgd 
    118      1.39   thorpej 	aiov.iov_base = (caddr_t)buf;
    119      1.39   thorpej 	aiov.iov_len = nbyte;
    120      1.15       cgd 	auio.uio_iov = &aiov;
    121      1.15       cgd 	auio.uio_iovcnt = 1;
    122      1.39   thorpej 	auio.uio_resid = nbyte;
    123      1.15       cgd 	auio.uio_rw = UIO_READ;
    124      1.15       cgd 	auio.uio_segflg = UIO_USERSPACE;
    125      1.15       cgd 	auio.uio_procp = p;
    126      1.40   thorpej 
    127      1.40   thorpej 	/*
    128      1.40   thorpej 	 * Reads return ssize_t because -1 is returned on error.  Therefore
    129      1.40   thorpej 	 * we must restrict the length to SSIZE_MAX to avoid garbage return
    130      1.40   thorpej 	 * values.
    131      1.40   thorpej 	 */
    132  1.44.8.1   thorpej 	if (auio.uio_resid > SSIZE_MAX) {
    133  1.44.8.1   thorpej 		error = EINVAL;
    134  1.44.8.1   thorpej 		goto out;
    135  1.44.8.1   thorpej 	}
    136      1.40   thorpej 
    137      1.15       cgd #ifdef KTRACE
    138      1.15       cgd 	/*
    139      1.15       cgd 	 * if tracing, save a copy of iovec
    140      1.15       cgd 	 */
    141      1.15       cgd 	if (KTRPOINT(p, KTR_GENIO))
    142      1.15       cgd 		ktriov = aiov;
    143      1.15       cgd #endif
    144      1.38   thorpej 	cnt = auio.uio_resid;
    145      1.39   thorpej 	error = (*fp->f_ops->fo_read)(fp, offset, &auio, fp->f_cred, flags);
    146      1.22  christos 	if (error)
    147      1.15       cgd 		if (auio.uio_resid != cnt && (error == ERESTART ||
    148      1.15       cgd 		    error == EINTR || error == EWOULDBLOCK))
    149      1.15       cgd 			error = 0;
    150      1.15       cgd 	cnt -= auio.uio_resid;
    151      1.15       cgd #ifdef KTRACE
    152      1.15       cgd 	if (KTRPOINT(p, KTR_GENIO) && error == 0)
    153      1.34   mycroft 		ktrgenio(p->p_tracep, fd, UIO_READ, &ktriov, cnt, error);
    154      1.15       cgd #endif
    155      1.15       cgd 	*retval = cnt;
    156  1.44.8.1   thorpej  out:
    157  1.44.8.1   thorpej 	FILE_UNUSE(fp, p);
    158      1.15       cgd 	return (error);
    159      1.15       cgd }
    160      1.15       cgd 
    161      1.15       cgd /*
    162      1.15       cgd  * Scatter read system call.
    163      1.15       cgd  */
    164      1.22  christos int
    165      1.21   mycroft sys_readv(p, v, retval)
    166      1.15       cgd 	struct proc *p;
    167      1.20   thorpej 	void *v;
    168      1.20   thorpej 	register_t *retval;
    169      1.20   thorpej {
    170      1.21   mycroft 	register struct sys_readv_args /* {
    171      1.16       cgd 		syscallarg(int) fd;
    172      1.31       cgd 		syscallarg(const struct iovec *) iovp;
    173      1.34   mycroft 		syscallarg(int) iovcnt;
    174      1.20   thorpej 	} */ *uap = v;
    175      1.34   mycroft 	int fd = SCARG(uap, fd);
    176      1.15       cgd 	register struct file *fp;
    177      1.15       cgd 	register struct filedesc *fdp = p->p_fd;
    178      1.39   thorpej 
    179      1.39   thorpej 	if ((u_int)fd >= fdp->fd_nfiles ||
    180      1.39   thorpej 	    (fp = fdp->fd_ofiles[fd]) == NULL ||
    181  1.44.8.1   thorpej 	    (fp->f_iflags & FIF_WANTCLOSE) != 0 ||
    182      1.39   thorpej 	    (fp->f_flag & FREAD) == 0)
    183      1.39   thorpej 		return (EBADF);
    184      1.39   thorpej 
    185  1.44.8.1   thorpej 	FILE_USE(fp);
    186  1.44.8.1   thorpej 
    187  1.44.8.1   thorpej 	/* dofilereadv() will unuse the descriptor for us */
    188      1.39   thorpej 	return (dofilereadv(p, fd, fp, SCARG(uap, iovp), SCARG(uap, iovcnt),
    189      1.39   thorpej 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
    190      1.39   thorpej }
    191      1.39   thorpej 
    192      1.39   thorpej int
    193      1.39   thorpej dofilereadv(p, fd, fp, iovp, iovcnt, offset, flags, retval)
    194      1.39   thorpej 	struct proc *p;
    195      1.39   thorpej 	int fd;
    196      1.39   thorpej 	struct file *fp;
    197      1.39   thorpej 	const struct iovec *iovp;
    198      1.39   thorpej 	int iovcnt;
    199      1.39   thorpej 	off_t *offset;
    200      1.39   thorpej 	int flags;
    201      1.39   thorpej 	register_t *retval;
    202      1.39   thorpej {
    203      1.15       cgd 	struct uio auio;
    204      1.15       cgd 	register struct iovec *iov;
    205      1.15       cgd 	struct iovec *needfree;
    206      1.15       cgd 	struct iovec aiov[UIO_SMALLIOV];
    207      1.15       cgd 	long i, cnt, error = 0;
    208      1.15       cgd 	u_int iovlen;
    209      1.15       cgd #ifdef KTRACE
    210      1.15       cgd 	struct iovec *ktriov = NULL;
    211      1.15       cgd #endif
    212      1.15       cgd 
    213      1.15       cgd 	/* note: can't use iovlen until iovcnt is validated */
    214      1.42     perry 	iovlen = iovcnt * sizeof(struct iovec);
    215      1.34   mycroft 	if ((u_int)iovcnt > UIO_SMALLIOV) {
    216  1.44.8.1   thorpej 		if ((u_int)iovcnt > IOV_MAX) {
    217  1.44.8.1   thorpej 			error = EINVAL;
    218  1.44.8.1   thorpej 			goto out;
    219  1.44.8.1   thorpej 		}
    220      1.15       cgd 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
    221      1.15       cgd 		needfree = iov;
    222      1.41    kleink 	} else if ((u_int)iovcnt > 0) {
    223      1.15       cgd 		iov = aiov;
    224      1.15       cgd 		needfree = NULL;
    225  1.44.8.1   thorpej 	} else {
    226  1.44.8.1   thorpej 		error = EINVAL;
    227  1.44.8.1   thorpej 		goto out;
    228  1.44.8.1   thorpej 	}
    229      1.41    kleink 
    230      1.15       cgd 	auio.uio_iov = iov;
    231      1.34   mycroft 	auio.uio_iovcnt = iovcnt;
    232      1.15       cgd 	auio.uio_rw = UIO_READ;
    233      1.15       cgd 	auio.uio_segflg = UIO_USERSPACE;
    234      1.15       cgd 	auio.uio_procp = p;
    235      1.39   thorpej 	error = copyin(iovp, iov, iovlen);
    236      1.22  christos 	if (error)
    237      1.15       cgd 		goto done;
    238      1.15       cgd 	auio.uio_resid = 0;
    239      1.34   mycroft 	for (i = 0; i < iovcnt; i++) {
    240      1.15       cgd 		auio.uio_resid += iov->iov_len;
    241      1.40   thorpej 		/*
    242      1.40   thorpej 		 * Reads return ssize_t because -1 is returned on error.
    243      1.40   thorpej 		 * Therefore we must restrict the length to SSIZE_MAX to
    244      1.40   thorpej 		 * avoid garbage return values.
    245      1.40   thorpej 		 */
    246      1.40   thorpej 		if (iov->iov_len > SSIZE_MAX || auio.uio_resid > SSIZE_MAX) {
    247      1.15       cgd 			error = EINVAL;
    248      1.15       cgd 			goto done;
    249      1.15       cgd 		}
    250      1.15       cgd 		iov++;
    251      1.15       cgd 	}
    252      1.15       cgd #ifdef KTRACE
    253      1.15       cgd 	/*
    254      1.15       cgd 	 * if tracing, save a copy of iovec
    255      1.15       cgd 	 */
    256      1.15       cgd 	if (KTRPOINT(p, KTR_GENIO))  {
    257      1.15       cgd 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
    258      1.44     perry 		memcpy((caddr_t)ktriov, (caddr_t)auio.uio_iov, iovlen);
    259      1.15       cgd 	}
    260      1.15       cgd #endif
    261      1.15       cgd 	cnt = auio.uio_resid;
    262      1.39   thorpej 	error = (*fp->f_ops->fo_read)(fp, offset, &auio, fp->f_cred, flags);
    263      1.22  christos 	if (error)
    264      1.15       cgd 		if (auio.uio_resid != cnt && (error == ERESTART ||
    265      1.15       cgd 		    error == EINTR || error == EWOULDBLOCK))
    266      1.15       cgd 			error = 0;
    267      1.15       cgd 	cnt -= auio.uio_resid;
    268      1.15       cgd #ifdef KTRACE
    269      1.34   mycroft 	if (KTRPOINT(p, KTR_GENIO))
    270      1.34   mycroft 		if (error == 0) {
    271      1.34   mycroft 			ktrgenio(p->p_tracep, fd, UIO_READ, ktriov, cnt,
    272      1.34   mycroft 			    error);
    273      1.15       cgd 		FREE(ktriov, M_TEMP);
    274      1.15       cgd 	}
    275      1.15       cgd #endif
    276      1.15       cgd 	*retval = cnt;
    277  1.44.8.1   thorpej  done:
    278      1.15       cgd 	if (needfree)
    279      1.15       cgd 		FREE(needfree, M_IOV);
    280  1.44.8.1   thorpej  out:
    281  1.44.8.1   thorpej 	FILE_UNUSE(fp, p);
    282      1.15       cgd 	return (error);
    283      1.15       cgd }
    284      1.15       cgd 
    285      1.15       cgd /*
    286      1.15       cgd  * Write system call
    287      1.15       cgd  */
    288      1.22  christos int
    289      1.21   mycroft sys_write(p, v, retval)
    290      1.15       cgd 	struct proc *p;
    291      1.20   thorpej 	void *v;
    292      1.20   thorpej 	register_t *retval;
    293      1.20   thorpej {
    294      1.21   mycroft 	register struct sys_write_args /* {
    295      1.16       cgd 		syscallarg(int) fd;
    296      1.31       cgd 		syscallarg(const void *) buf;
    297      1.31       cgd 		syscallarg(size_t) nbyte;
    298      1.20   thorpej 	} */ *uap = v;
    299      1.34   mycroft 	int fd = SCARG(uap, fd);
    300      1.15       cgd 	register struct file *fp;
    301      1.15       cgd 	register struct filedesc *fdp = p->p_fd;
    302      1.39   thorpej 
    303      1.39   thorpej 	if ((u_int)fd >= fdp->fd_nfiles ||
    304      1.39   thorpej 	    (fp = fdp->fd_ofiles[fd]) == NULL ||
    305  1.44.8.1   thorpej 	    (fp->f_iflags & FIF_WANTCLOSE) != 0 ||
    306      1.39   thorpej 	    (fp->f_flag & FWRITE) == 0)
    307      1.39   thorpej 		return (EBADF);
    308      1.39   thorpej 
    309  1.44.8.1   thorpej 	FILE_USE(fp);
    310  1.44.8.1   thorpej 
    311  1.44.8.1   thorpej 	/* dofilewrite() will unuse the descriptor for us */
    312      1.39   thorpej 	return (dofilewrite(p, fd, fp, SCARG(uap, buf), SCARG(uap, nbyte),
    313      1.39   thorpej 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
    314      1.39   thorpej }
    315      1.39   thorpej 
    316      1.39   thorpej int
    317      1.39   thorpej dofilewrite(p, fd, fp, buf, nbyte, offset, flags, retval)
    318      1.39   thorpej 	struct proc *p;
    319      1.39   thorpej 	int fd;
    320      1.39   thorpej 	struct file *fp;
    321      1.39   thorpej 	const void *buf;
    322      1.39   thorpej 	size_t nbyte;
    323      1.39   thorpej 	off_t *offset;
    324      1.39   thorpej 	int flags;
    325      1.39   thorpej 	register_t *retval;
    326      1.39   thorpej {
    327      1.15       cgd 	struct uio auio;
    328      1.15       cgd 	struct iovec aiov;
    329      1.15       cgd 	long cnt, error = 0;
    330      1.15       cgd #ifdef KTRACE
    331      1.15       cgd 	struct iovec ktriov;
    332      1.15       cgd #endif
    333      1.15       cgd 
    334      1.39   thorpej 	aiov.iov_base = (caddr_t)buf;		/* XXX kills const */
    335      1.39   thorpej 	aiov.iov_len = nbyte;
    336      1.15       cgd 	auio.uio_iov = &aiov;
    337      1.15       cgd 	auio.uio_iovcnt = 1;
    338      1.39   thorpej 	auio.uio_resid = nbyte;
    339      1.15       cgd 	auio.uio_rw = UIO_WRITE;
    340      1.15       cgd 	auio.uio_segflg = UIO_USERSPACE;
    341      1.15       cgd 	auio.uio_procp = p;
    342      1.40   thorpej 
    343      1.40   thorpej 	/*
    344      1.40   thorpej 	 * Writes return ssize_t because -1 is returned on error.  Therefore
    345      1.40   thorpej 	 * we must restrict the length to SSIZE_MAX to avoid garbage return
    346      1.40   thorpej 	 * values.
    347      1.40   thorpej 	 */
    348  1.44.8.1   thorpej 	if (auio.uio_resid > SSIZE_MAX) {
    349  1.44.8.1   thorpej 		error = EINVAL;
    350  1.44.8.1   thorpej 		goto out;
    351  1.44.8.1   thorpej 	}
    352      1.40   thorpej 
    353      1.15       cgd #ifdef KTRACE
    354      1.15       cgd 	/*
    355      1.15       cgd 	 * if tracing, save a copy of iovec
    356      1.15       cgd 	 */
    357      1.15       cgd 	if (KTRPOINT(p, KTR_GENIO))
    358      1.15       cgd 		ktriov = aiov;
    359      1.15       cgd #endif
    360      1.38   thorpej 	cnt = auio.uio_resid;
    361      1.39   thorpej 	error = (*fp->f_ops->fo_write)(fp, offset, &auio, fp->f_cred, flags);
    362      1.22  christos 	if (error) {
    363      1.15       cgd 		if (auio.uio_resid != cnt && (error == ERESTART ||
    364      1.15       cgd 		    error == EINTR || error == EWOULDBLOCK))
    365      1.15       cgd 			error = 0;
    366      1.15       cgd 		if (error == EPIPE)
    367      1.15       cgd 			psignal(p, SIGPIPE);
    368      1.15       cgd 	}
    369      1.15       cgd 	cnt -= auio.uio_resid;
    370      1.15       cgd #ifdef KTRACE
    371      1.15       cgd 	if (KTRPOINT(p, KTR_GENIO) && error == 0)
    372      1.34   mycroft 		ktrgenio(p->p_tracep, fd, UIO_WRITE, &ktriov, cnt, error);
    373      1.15       cgd #endif
    374      1.15       cgd 	*retval = cnt;
    375  1.44.8.1   thorpej  out:
    376  1.44.8.1   thorpej 	FILE_UNUSE(fp, p);
    377      1.15       cgd 	return (error);
    378      1.15       cgd }
    379      1.15       cgd 
    380      1.15       cgd /*
    381      1.15       cgd  * Gather write system call
    382      1.15       cgd  */
    383      1.22  christos int
    384      1.21   mycroft sys_writev(p, v, retval)
    385      1.15       cgd 	struct proc *p;
    386      1.20   thorpej 	void *v;
    387      1.20   thorpej 	register_t *retval;
    388      1.20   thorpej {
    389      1.21   mycroft 	register struct sys_writev_args /* {
    390      1.16       cgd 		syscallarg(int) fd;
    391      1.31       cgd 		syscallarg(const struct iovec *) iovp;
    392      1.39   thorpej 		syscallarg(int) iovcnt;
    393      1.20   thorpej 	} */ *uap = v;
    394      1.34   mycroft 	int fd = SCARG(uap, fd);
    395      1.15       cgd 	register struct file *fp;
    396      1.15       cgd 	register struct filedesc *fdp = p->p_fd;
    397      1.39   thorpej 
    398      1.39   thorpej 	if ((u_int)fd >= fdp->fd_nfiles ||
    399      1.39   thorpej 	    (fp = fdp->fd_ofiles[fd]) == NULL ||
    400  1.44.8.1   thorpej 	    (fp->f_iflags & FIF_WANTCLOSE) != 0 ||
    401      1.39   thorpej 	    (fp->f_flag & FWRITE) == 0)
    402      1.39   thorpej 		return (EBADF);
    403      1.39   thorpej 
    404  1.44.8.1   thorpej 	FILE_USE(fp);
    405  1.44.8.1   thorpej 
    406  1.44.8.1   thorpej 	/* dofilewritev() will unuse the descriptor for us */
    407      1.39   thorpej 	return (dofilewritev(p, fd, fp, SCARG(uap, iovp), SCARG(uap, iovcnt),
    408      1.39   thorpej 	    &fp->f_offset, FOF_UPDATE_OFFSET, retval));
    409      1.39   thorpej }
    410      1.39   thorpej 
    411      1.39   thorpej int
    412      1.39   thorpej dofilewritev(p, fd, fp, iovp, iovcnt, offset, flags, retval)
    413      1.39   thorpej 	struct proc *p;
    414      1.39   thorpej 	int fd;
    415      1.39   thorpej 	struct file *fp;
    416      1.39   thorpej 	const struct iovec *iovp;
    417      1.39   thorpej 	int iovcnt;
    418      1.39   thorpej 	off_t *offset;
    419      1.39   thorpej 	int flags;
    420      1.39   thorpej 	register_t *retval;
    421      1.39   thorpej {
    422      1.15       cgd 	struct uio auio;
    423      1.15       cgd 	register struct iovec *iov;
    424      1.15       cgd 	struct iovec *needfree;
    425      1.15       cgd 	struct iovec aiov[UIO_SMALLIOV];
    426      1.15       cgd 	long i, cnt, error = 0;
    427      1.15       cgd 	u_int iovlen;
    428      1.15       cgd #ifdef KTRACE
    429      1.15       cgd 	struct iovec *ktriov = NULL;
    430      1.15       cgd #endif
    431      1.15       cgd 
    432      1.15       cgd 	/* note: can't use iovlen until iovcnt is validated */
    433      1.42     perry 	iovlen = iovcnt * sizeof(struct iovec);
    434      1.34   mycroft 	if ((u_int)iovcnt > UIO_SMALLIOV) {
    435      1.43    kleink 		if ((u_int)iovcnt > IOV_MAX)
    436      1.15       cgd 			return (EINVAL);
    437      1.15       cgd 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
    438      1.15       cgd 		needfree = iov;
    439      1.41    kleink 	} else if ((u_int)iovcnt > 0) {
    440      1.15       cgd 		iov = aiov;
    441      1.15       cgd 		needfree = NULL;
    442  1.44.8.1   thorpej 	} else {
    443  1.44.8.1   thorpej 		error = EINVAL;
    444  1.44.8.1   thorpej 		goto out;
    445  1.44.8.1   thorpej 	}
    446      1.41    kleink 
    447      1.15       cgd 	auio.uio_iov = iov;
    448      1.34   mycroft 	auio.uio_iovcnt = iovcnt;
    449      1.15       cgd 	auio.uio_rw = UIO_WRITE;
    450      1.15       cgd 	auio.uio_segflg = UIO_USERSPACE;
    451      1.15       cgd 	auio.uio_procp = p;
    452      1.39   thorpej 	error = copyin(iovp, iov, iovlen);
    453      1.22  christos 	if (error)
    454      1.15       cgd 		goto done;
    455      1.15       cgd 	auio.uio_resid = 0;
    456      1.34   mycroft 	for (i = 0; i < iovcnt; i++) {
    457      1.15       cgd 		auio.uio_resid += iov->iov_len;
    458      1.40   thorpej 		/*
    459      1.40   thorpej 		 * Writes return ssize_t because -1 is returned on error.
    460      1.40   thorpej 		 * Therefore we must restrict the length to SSIZE_MAX to
    461      1.40   thorpej 		 * avoid garbage return values.
    462      1.40   thorpej 		 */
    463      1.40   thorpej 		if (iov->iov_len > SSIZE_MAX || auio.uio_resid > SSIZE_MAX) {
    464      1.15       cgd 			error = EINVAL;
    465      1.15       cgd 			goto done;
    466      1.15       cgd 		}
    467      1.15       cgd 		iov++;
    468      1.15       cgd 	}
    469      1.15       cgd #ifdef KTRACE
    470      1.15       cgd 	/*
    471      1.15       cgd 	 * if tracing, save a copy of iovec
    472      1.15       cgd 	 */
    473      1.15       cgd 	if (KTRPOINT(p, KTR_GENIO))  {
    474      1.15       cgd 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
    475      1.44     perry 		memcpy((caddr_t)ktriov, (caddr_t)auio.uio_iov, iovlen);
    476      1.15       cgd 	}
    477      1.15       cgd #endif
    478      1.15       cgd 	cnt = auio.uio_resid;
    479      1.39   thorpej 	error = (*fp->f_ops->fo_write)(fp, offset, &auio, fp->f_cred, flags);
    480      1.22  christos 	if (error) {
    481      1.15       cgd 		if (auio.uio_resid != cnt && (error == ERESTART ||
    482      1.15       cgd 		    error == EINTR || error == EWOULDBLOCK))
    483      1.15       cgd 			error = 0;
    484      1.15       cgd 		if (error == EPIPE)
    485      1.15       cgd 			psignal(p, SIGPIPE);
    486      1.15       cgd 	}
    487      1.15       cgd 	cnt -= auio.uio_resid;
    488      1.15       cgd #ifdef KTRACE
    489      1.34   mycroft 	if (KTRPOINT(p, KTR_GENIO))
    490      1.34   mycroft 		if (error == 0) {
    491      1.34   mycroft 			ktrgenio(p->p_tracep, fd, UIO_WRITE, ktriov, cnt,
    492      1.34   mycroft 			    error);
    493      1.15       cgd 		FREE(ktriov, M_TEMP);
    494      1.15       cgd 	}
    495      1.15       cgd #endif
    496      1.15       cgd 	*retval = cnt;
    497  1.44.8.1   thorpej  done:
    498      1.15       cgd 	if (needfree)
    499      1.15       cgd 		FREE(needfree, M_IOV);
    500  1.44.8.1   thorpej  out:
    501  1.44.8.1   thorpej 	FILE_UNUSE(fp, p);
    502      1.15       cgd 	return (error);
    503      1.15       cgd }
    504      1.15       cgd 
    505      1.15       cgd /*
    506      1.15       cgd  * Ioctl system call
    507      1.15       cgd  */
    508      1.15       cgd /* ARGSUSED */
    509      1.22  christos int
    510      1.21   mycroft sys_ioctl(p, v, retval)
    511      1.15       cgd 	struct proc *p;
    512      1.20   thorpej 	void *v;
    513      1.20   thorpej 	register_t *retval;
    514      1.20   thorpej {
    515      1.21   mycroft 	register struct sys_ioctl_args /* {
    516      1.16       cgd 		syscallarg(int) fd;
    517      1.17       cgd 		syscallarg(u_long) com;
    518      1.16       cgd 		syscallarg(caddr_t) data;
    519      1.20   thorpej 	} */ *uap = v;
    520      1.15       cgd 	register struct file *fp;
    521      1.15       cgd 	register struct filedesc *fdp;
    522      1.17       cgd 	register u_long com;
    523  1.44.8.1   thorpej 	register int error = 0;
    524      1.15       cgd 	register u_int size;
    525      1.15       cgd 	caddr_t data, memp;
    526      1.15       cgd 	int tmp;
    527      1.15       cgd #define STK_PARAMS	128
    528      1.15       cgd 	char stkbuf[STK_PARAMS];
    529      1.15       cgd 
    530      1.15       cgd 	fdp = p->p_fd;
    531      1.16       cgd 	if ((u_int)SCARG(uap, fd) >= fdp->fd_nfiles ||
    532  1.44.8.1   thorpej 	    (fp = fdp->fd_ofiles[SCARG(uap, fd)]) == NULL ||
    533  1.44.8.1   thorpej 	    (fp->f_iflags & FIF_WANTCLOSE) != 0)
    534      1.15       cgd 		return (EBADF);
    535      1.15       cgd 
    536  1.44.8.1   thorpej 	FILE_USE(fp);
    537  1.44.8.1   thorpej 
    538  1.44.8.1   thorpej 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
    539  1.44.8.1   thorpej 		error = EBADF;
    540  1.44.8.1   thorpej 		goto out;
    541  1.44.8.1   thorpej 	}
    542      1.15       cgd 
    543      1.16       cgd 	switch (com = SCARG(uap, com)) {
    544      1.15       cgd 	case FIONCLEX:
    545      1.16       cgd 		fdp->fd_ofileflags[SCARG(uap, fd)] &= ~UF_EXCLOSE;
    546  1.44.8.1   thorpej 		goto out;
    547  1.44.8.1   thorpej 
    548      1.15       cgd 	case FIOCLEX:
    549      1.16       cgd 		fdp->fd_ofileflags[SCARG(uap, fd)] |= UF_EXCLOSE;
    550  1.44.8.1   thorpej 		goto out;
    551      1.15       cgd 	}
    552      1.15       cgd 
    553      1.15       cgd 	/*
    554      1.15       cgd 	 * Interpret high order word to find amount of data to be
    555      1.15       cgd 	 * copied to/from the user's address space.
    556      1.15       cgd 	 */
    557      1.15       cgd 	size = IOCPARM_LEN(com);
    558  1.44.8.1   thorpej 	if (size > IOCPARM_MAX) {
    559  1.44.8.1   thorpej 		error = ENOTTY;
    560  1.44.8.1   thorpej 		goto out;
    561  1.44.8.1   thorpej 	}
    562      1.15       cgd 	memp = NULL;
    563      1.42     perry 	if (size > sizeof(stkbuf)) {
    564      1.15       cgd 		memp = (caddr_t)malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
    565      1.15       cgd 		data = memp;
    566      1.15       cgd 	} else
    567      1.15       cgd 		data = stkbuf;
    568      1.15       cgd 	if (com&IOC_IN) {
    569      1.15       cgd 		if (size) {
    570      1.31       cgd 			error = copyin(SCARG(uap, data), data, size);
    571      1.15       cgd 			if (error) {
    572      1.15       cgd 				if (memp)
    573      1.15       cgd 					free(memp, M_IOCTLOPS);
    574  1.44.8.1   thorpej 				goto out;
    575      1.15       cgd 			}
    576      1.15       cgd 		} else
    577      1.16       cgd 			*(caddr_t *)data = SCARG(uap, data);
    578      1.15       cgd 	} else if ((com&IOC_OUT) && size)
    579      1.15       cgd 		/*
    580      1.15       cgd 		 * Zero the buffer so the user always
    581      1.15       cgd 		 * gets back something deterministic.
    582      1.15       cgd 		 */
    583      1.44     perry 		memset(data, 0, size);
    584      1.15       cgd 	else if (com&IOC_VOID)
    585      1.16       cgd 		*(caddr_t *)data = SCARG(uap, data);
    586      1.15       cgd 
    587      1.15       cgd 	switch (com) {
    588      1.15       cgd 
    589      1.15       cgd 	case FIONBIO:
    590      1.22  christos 		if ((tmp = *(int *)data) != 0)
    591      1.15       cgd 			fp->f_flag |= FNONBLOCK;
    592      1.15       cgd 		else
    593      1.15       cgd 			fp->f_flag &= ~FNONBLOCK;
    594      1.15       cgd 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    595      1.15       cgd 		break;
    596      1.15       cgd 
    597      1.15       cgd 	case FIOASYNC:
    598      1.22  christos 		if ((tmp = *(int *)data) != 0)
    599      1.15       cgd 			fp->f_flag |= FASYNC;
    600      1.15       cgd 		else
    601      1.15       cgd 			fp->f_flag &= ~FASYNC;
    602      1.15       cgd 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
    603      1.15       cgd 		break;
    604      1.15       cgd 
    605      1.15       cgd 	case FIOSETOWN:
    606      1.15       cgd 		tmp = *(int *)data;
    607      1.15       cgd 		if (fp->f_type == DTYPE_SOCKET) {
    608      1.15       cgd 			((struct socket *)fp->f_data)->so_pgid = tmp;
    609      1.15       cgd 			error = 0;
    610      1.15       cgd 			break;
    611      1.15       cgd 		}
    612      1.15       cgd 		if (tmp <= 0) {
    613      1.15       cgd 			tmp = -tmp;
    614      1.15       cgd 		} else {
    615      1.15       cgd 			struct proc *p1 = pfind(tmp);
    616      1.15       cgd 			if (p1 == 0) {
    617      1.15       cgd 				error = ESRCH;
    618      1.15       cgd 				break;
    619      1.15       cgd 			}
    620      1.15       cgd 			tmp = p1->p_pgrp->pg_id;
    621      1.15       cgd 		}
    622      1.15       cgd 		error = (*fp->f_ops->fo_ioctl)
    623      1.24       cgd 			(fp, TIOCSPGRP, (caddr_t)&tmp, p);
    624      1.15       cgd 		break;
    625      1.15       cgd 
    626      1.15       cgd 	case FIOGETOWN:
    627      1.15       cgd 		if (fp->f_type == DTYPE_SOCKET) {
    628      1.15       cgd 			error = 0;
    629      1.15       cgd 			*(int *)data = ((struct socket *)fp->f_data)->so_pgid;
    630      1.15       cgd 			break;
    631      1.15       cgd 		}
    632      1.17       cgd 		error = (*fp->f_ops->fo_ioctl)(fp, TIOCGPGRP, data, p);
    633      1.15       cgd 		*(int *)data = -*(int *)data;
    634      1.15       cgd 		break;
    635      1.15       cgd 
    636      1.15       cgd 	default:
    637      1.15       cgd 		error = (*fp->f_ops->fo_ioctl)(fp, com, data, p);
    638      1.15       cgd 		/*
    639      1.15       cgd 		 * Copy any data to user, size was
    640      1.15       cgd 		 * already set and checked above.
    641      1.15       cgd 		 */
    642      1.15       cgd 		if (error == 0 && (com&IOC_OUT) && size)
    643      1.31       cgd 			error = copyout(data, SCARG(uap, data), size);
    644      1.15       cgd 		break;
    645      1.15       cgd 	}
    646      1.15       cgd 	if (memp)
    647      1.15       cgd 		free(memp, M_IOCTLOPS);
    648  1.44.8.1   thorpej  out:
    649  1.44.8.1   thorpej 	FILE_UNUSE(fp, p);
    650      1.15       cgd 	return (error);
    651      1.15       cgd }
    652      1.15       cgd 
    653      1.15       cgd int	selwait, nselcoll;
    654      1.15       cgd 
    655      1.15       cgd /*
    656      1.15       cgd  * Select system call.
    657      1.15       cgd  */
    658      1.22  christos int
    659      1.21   mycroft sys_select(p, v, retval)
    660      1.15       cgd 	register struct proc *p;
    661      1.20   thorpej 	void *v;
    662      1.20   thorpej 	register_t *retval;
    663      1.20   thorpej {
    664      1.21   mycroft 	register struct sys_select_args /* {
    665      1.35   thorpej 		syscallarg(int) nd;
    666      1.16       cgd 		syscallarg(fd_set *) in;
    667      1.16       cgd 		syscallarg(fd_set *) ou;
    668      1.16       cgd 		syscallarg(fd_set *) ex;
    669      1.16       cgd 		syscallarg(struct timeval *) tv;
    670      1.20   thorpej 	} */ *uap = v;
    671      1.25   mycroft 	caddr_t bits;
    672      1.26       cgd 	char smallbits[howmany(FD_SETSIZE, NFDBITS) * sizeof(fd_mask) * 6];
    673      1.15       cgd 	struct timeval atv;
    674      1.15       cgd 	int s, ncoll, error = 0, timo;
    675      1.27   mycroft 	size_t ni;
    676      1.15       cgd 
    677      1.35   thorpej 	if (SCARG(uap, nd) < 0)
    678      1.35   thorpej 		return (EINVAL);
    679      1.16       cgd 	if (SCARG(uap, nd) > p->p_fd->fd_nfiles) {
    680      1.16       cgd 		/* forgiving; slightly wrong */
    681      1.16       cgd 		SCARG(uap, nd) = p->p_fd->fd_nfiles;
    682      1.16       cgd 	}
    683      1.16       cgd 	ni = howmany(SCARG(uap, nd), NFDBITS) * sizeof(fd_mask);
    684      1.27   mycroft 	if (ni * 6 > sizeof(smallbits))
    685      1.25   mycroft 		bits = malloc(ni * 6, M_TEMP, M_WAITOK);
    686      1.25   mycroft 	else
    687      1.26       cgd 		bits = smallbits;
    688      1.15       cgd 
    689      1.15       cgd #define	getbits(name, x) \
    690      1.25   mycroft 	if (SCARG(uap, name)) { \
    691      1.31       cgd 		error = copyin(SCARG(uap, name), bits + ni * x, ni); \
    692      1.25   mycroft 		if (error) \
    693      1.25   mycroft 			goto done; \
    694      1.25   mycroft 	} else \
    695      1.44     perry 		memset(bits + ni * x, 0, ni);
    696      1.15       cgd 	getbits(in, 0);
    697      1.15       cgd 	getbits(ou, 1);
    698      1.15       cgd 	getbits(ex, 2);
    699      1.15       cgd #undef	getbits
    700      1.15       cgd 
    701      1.16       cgd 	if (SCARG(uap, tv)) {
    702      1.31       cgd 		error = copyin(SCARG(uap, tv), (caddr_t)&atv,
    703      1.42     perry 			sizeof(atv));
    704      1.15       cgd 		if (error)
    705      1.15       cgd 			goto done;
    706      1.15       cgd 		if (itimerfix(&atv)) {
    707      1.15       cgd 			error = EINVAL;
    708      1.15       cgd 			goto done;
    709      1.15       cgd 		}
    710      1.15       cgd 		s = splclock();
    711      1.19   mycroft 		timeradd(&atv, &time, &atv);
    712      1.15       cgd 		timo = hzto(&atv);
    713      1.15       cgd 		/*
    714      1.15       cgd 		 * Avoid inadvertently sleeping forever.
    715      1.15       cgd 		 */
    716      1.15       cgd 		if (timo == 0)
    717      1.15       cgd 			timo = 1;
    718      1.15       cgd 		splx(s);
    719      1.15       cgd 	} else
    720      1.15       cgd 		timo = 0;
    721      1.15       cgd retry:
    722      1.15       cgd 	ncoll = nselcoll;
    723      1.15       cgd 	p->p_flag |= P_SELECT;
    724      1.25   mycroft 	error = selscan(p, (fd_mask *)(bits + ni * 0),
    725      1.25   mycroft 			   (fd_mask *)(bits + ni * 3), SCARG(uap, nd), retval);
    726      1.15       cgd 	if (error || *retval)
    727      1.15       cgd 		goto done;
    728      1.15       cgd 	s = splhigh();
    729      1.27   mycroft 	if (timo && timercmp(&time, &atv, >=)) {
    730      1.15       cgd 		splx(s);
    731      1.15       cgd 		goto done;
    732      1.15       cgd 	}
    733      1.15       cgd 	if ((p->p_flag & P_SELECT) == 0 || nselcoll != ncoll) {
    734      1.15       cgd 		splx(s);
    735      1.15       cgd 		goto retry;
    736      1.15       cgd 	}
    737      1.15       cgd 	p->p_flag &= ~P_SELECT;
    738      1.15       cgd 	error = tsleep((caddr_t)&selwait, PSOCK | PCATCH, "select", timo);
    739      1.15       cgd 	splx(s);
    740      1.15       cgd 	if (error == 0)
    741      1.15       cgd 		goto retry;
    742      1.15       cgd done:
    743      1.15       cgd 	p->p_flag &= ~P_SELECT;
    744      1.15       cgd 	/* select is not restarted after signals... */
    745      1.15       cgd 	if (error == ERESTART)
    746      1.15       cgd 		error = EINTR;
    747      1.15       cgd 	if (error == EWOULDBLOCK)
    748      1.15       cgd 		error = 0;
    749      1.27   mycroft 	if (error == 0) {
    750      1.15       cgd #define	putbits(name, x) \
    751      1.27   mycroft 		if (SCARG(uap, name)) { \
    752      1.31       cgd 			error = copyout(bits + ni * x, SCARG(uap, name), ni); \
    753      1.27   mycroft 			if (error) \
    754      1.27   mycroft 				goto out; \
    755      1.27   mycroft 		}
    756      1.25   mycroft 		putbits(in, 3);
    757      1.25   mycroft 		putbits(ou, 4);
    758      1.25   mycroft 		putbits(ex, 5);
    759      1.15       cgd #undef putbits
    760      1.15       cgd 	}
    761      1.27   mycroft out:
    762      1.27   mycroft 	if (ni * 6 > sizeof(smallbits))
    763      1.25   mycroft 		free(bits, M_TEMP);
    764      1.15       cgd 	return (error);
    765      1.15       cgd }
    766      1.15       cgd 
    767      1.22  christos int
    768      1.25   mycroft selscan(p, ibitp, obitp, nfd, retval)
    769      1.15       cgd 	struct proc *p;
    770      1.25   mycroft 	fd_mask *ibitp, *obitp;
    771      1.16       cgd 	int nfd;
    772      1.16       cgd 	register_t *retval;
    773      1.15       cgd {
    774      1.15       cgd 	register struct filedesc *fdp = p->p_fd;
    775      1.15       cgd 	register int msk, i, j, fd;
    776      1.25   mycroft 	register fd_mask ibits, obits;
    777      1.15       cgd 	struct file *fp;
    778      1.15       cgd 	int n = 0;
    779      1.28   mycroft 	static int flag[3] = { POLLRDNORM | POLLHUP | POLLERR,
    780      1.28   mycroft 			       POLLWRNORM | POLLHUP | POLLERR,
    781      1.28   mycroft 			       POLLRDBAND };
    782      1.15       cgd 
    783      1.15       cgd 	for (msk = 0; msk < 3; msk++) {
    784      1.15       cgd 		for (i = 0; i < nfd; i += NFDBITS) {
    785      1.25   mycroft 			ibits = *ibitp++;
    786      1.25   mycroft 			obits = 0;
    787      1.25   mycroft 			while ((j = ffs(ibits)) && (fd = i + --j) < nfd) {
    788      1.25   mycroft 				ibits &= ~(1 << j);
    789      1.15       cgd 				fp = fdp->fd_ofiles[fd];
    790  1.44.8.1   thorpej 				if (fp == NULL ||
    791  1.44.8.1   thorpej 				    (fp->f_iflags & FIF_WANTCLOSE) != 0)
    792      1.15       cgd 					return (EBADF);
    793  1.44.8.1   thorpej 				FILE_USE(fp);
    794      1.28   mycroft 				if ((*fp->f_ops->fo_poll)(fp, flag[msk], p)) {
    795      1.25   mycroft 					obits |= (1 << j);
    796      1.15       cgd 					n++;
    797      1.15       cgd 				}
    798  1.44.8.1   thorpej 				FILE_UNUSE(fp, p);
    799      1.15       cgd 			}
    800      1.25   mycroft 			*obitp++ = obits;
    801      1.15       cgd 		}
    802      1.15       cgd 	}
    803      1.15       cgd 	*retval = n;
    804      1.15       cgd 	return (0);
    805      1.15       cgd }
    806      1.15       cgd 
    807      1.28   mycroft /*
    808      1.28   mycroft  * Poll system call.
    809      1.28   mycroft  */
    810      1.28   mycroft int
    811      1.28   mycroft sys_poll(p, v, retval)
    812      1.28   mycroft 	register struct proc *p;
    813      1.28   mycroft 	void *v;
    814      1.28   mycroft 	register_t *retval;
    815      1.28   mycroft {
    816      1.28   mycroft 	register struct sys_poll_args /* {
    817      1.28   mycroft 		syscallarg(struct pollfd *) fds;
    818      1.29   mycroft 		syscallarg(u_int) nfds;
    819      1.28   mycroft 		syscallarg(int) timeout;
    820      1.28   mycroft 	} */ *uap = v;
    821      1.28   mycroft 	caddr_t bits;
    822      1.28   mycroft 	char smallbits[32 * sizeof(struct pollfd)];
    823      1.28   mycroft 	struct timeval atv;
    824      1.28   mycroft 	int s, ncoll, error = 0, timo;
    825      1.28   mycroft 	size_t ni;
    826      1.28   mycroft 
    827      1.28   mycroft 	if (SCARG(uap, nfds) > p->p_fd->fd_nfiles) {
    828      1.28   mycroft 		/* forgiving; slightly wrong */
    829      1.28   mycroft 		SCARG(uap, nfds) = p->p_fd->fd_nfiles;
    830      1.28   mycroft 	}
    831      1.28   mycroft 	ni = SCARG(uap, nfds) * sizeof(struct pollfd);
    832      1.28   mycroft 	if (ni > sizeof(smallbits))
    833      1.28   mycroft 		bits = malloc(ni, M_TEMP, M_WAITOK);
    834      1.28   mycroft 	else
    835      1.28   mycroft 		bits = smallbits;
    836      1.28   mycroft 
    837      1.31       cgd 	error = copyin(SCARG(uap, fds), bits, ni);
    838      1.28   mycroft 	if (error)
    839      1.28   mycroft 		goto done;
    840      1.28   mycroft 
    841      1.30   mycroft 	if (SCARG(uap, timeout) != INFTIM) {
    842      1.28   mycroft 		atv.tv_sec = SCARG(uap, timeout) / 1000;
    843      1.28   mycroft 		atv.tv_usec = (SCARG(uap, timeout) % 1000) * 1000;
    844      1.28   mycroft 		if (itimerfix(&atv)) {
    845      1.28   mycroft 			error = EINVAL;
    846      1.28   mycroft 			goto done;
    847      1.28   mycroft 		}
    848      1.28   mycroft 		s = splclock();
    849      1.28   mycroft 		timeradd(&atv, &time, &atv);
    850      1.28   mycroft 		timo = hzto(&atv);
    851      1.28   mycroft 		/*
    852      1.28   mycroft 		 * Avoid inadvertently sleeping forever.
    853      1.28   mycroft 		 */
    854      1.28   mycroft 		if (timo == 0)
    855      1.28   mycroft 			timo = 1;
    856      1.28   mycroft 		splx(s);
    857      1.28   mycroft 	} else
    858      1.28   mycroft 		timo = 0;
    859      1.28   mycroft retry:
    860      1.28   mycroft 	ncoll = nselcoll;
    861      1.28   mycroft 	p->p_flag |= P_SELECT;
    862      1.28   mycroft 	error = pollscan(p, (struct pollfd *)bits, SCARG(uap, nfds), retval);
    863      1.28   mycroft 	if (error || *retval)
    864      1.28   mycroft 		goto done;
    865      1.28   mycroft 	s = splhigh();
    866      1.28   mycroft 	if (timo && timercmp(&time, &atv, >=)) {
    867      1.28   mycroft 		splx(s);
    868      1.28   mycroft 		goto done;
    869      1.28   mycroft 	}
    870      1.28   mycroft 	if ((p->p_flag & P_SELECT) == 0 || nselcoll != ncoll) {
    871      1.28   mycroft 		splx(s);
    872      1.28   mycroft 		goto retry;
    873      1.28   mycroft 	}
    874      1.28   mycroft 	p->p_flag &= ~P_SELECT;
    875      1.28   mycroft 	error = tsleep((caddr_t)&selwait, PSOCK | PCATCH, "select", timo);
    876      1.28   mycroft 	splx(s);
    877      1.28   mycroft 	if (error == 0)
    878      1.28   mycroft 		goto retry;
    879      1.28   mycroft done:
    880      1.28   mycroft 	p->p_flag &= ~P_SELECT;
    881      1.28   mycroft 	/* poll is not restarted after signals... */
    882      1.28   mycroft 	if (error == ERESTART)
    883      1.28   mycroft 		error = EINTR;
    884      1.28   mycroft 	if (error == EWOULDBLOCK)
    885      1.28   mycroft 		error = 0;
    886      1.28   mycroft 	if (error == 0) {
    887      1.31       cgd 		error = copyout(bits, SCARG(uap, fds), ni);
    888      1.28   mycroft 		if (error)
    889      1.28   mycroft 			goto out;
    890      1.28   mycroft 	}
    891      1.28   mycroft out:
    892      1.28   mycroft 	if (ni > sizeof(smallbits))
    893      1.28   mycroft 		free(bits, M_TEMP);
    894      1.28   mycroft 	return (error);
    895      1.28   mycroft }
    896      1.28   mycroft 
    897      1.28   mycroft int
    898      1.28   mycroft pollscan(p, fds, nfd, retval)
    899      1.28   mycroft 	struct proc *p;
    900      1.28   mycroft 	struct pollfd *fds;
    901      1.28   mycroft 	int nfd;
    902      1.28   mycroft 	register_t *retval;
    903      1.28   mycroft {
    904      1.28   mycroft 	register struct filedesc *fdp = p->p_fd;
    905      1.28   mycroft 	int i;
    906      1.28   mycroft 	struct file *fp;
    907      1.28   mycroft 	int n = 0;
    908      1.28   mycroft 
    909      1.28   mycroft 	for (i = 0; i < nfd; i++, fds++) {
    910      1.33       mrg 		if ((u_int)fds->fd >= fdp->fd_nfiles) {
    911      1.28   mycroft 			fds->revents = POLLNVAL;
    912      1.28   mycroft 			n++;
    913      1.28   mycroft 		} else {
    914      1.32       mrg 			fp = fdp->fd_ofiles[fds->fd];
    915  1.44.8.1   thorpej 			if (fp == NULL ||
    916  1.44.8.1   thorpej 			    (fp->f_iflags & FIF_WANTCLOSE) != 0) {
    917      1.32       mrg 				fds->revents = POLLNVAL;
    918      1.28   mycroft 				n++;
    919      1.32       mrg 			} else {
    920  1.44.8.1   thorpej 				FILE_USE(fp);
    921      1.32       mrg 				fds->revents = (*fp->f_ops->fo_poll)(fp,
    922      1.32       mrg 				    fds->events | POLLERR | POLLHUP, p);
    923      1.32       mrg 				if (fds->revents != 0)
    924      1.32       mrg 					n++;
    925  1.44.8.1   thorpej 				FILE_UNUSE(fp, p);
    926      1.32       mrg 			}
    927      1.28   mycroft 		}
    928      1.28   mycroft 	}
    929      1.28   mycroft 	*retval = n;
    930      1.28   mycroft 	return (0);
    931      1.28   mycroft }
    932      1.28   mycroft 
    933      1.15       cgd /*ARGSUSED*/
    934      1.22  christos int
    935      1.28   mycroft seltrue(dev, events, p)
    936      1.15       cgd 	dev_t dev;
    937      1.28   mycroft 	int events;
    938      1.15       cgd 	struct proc *p;
    939      1.15       cgd {
    940      1.15       cgd 
    941      1.28   mycroft 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
    942      1.15       cgd }
    943      1.15       cgd 
    944      1.15       cgd /*
    945      1.15       cgd  * Record a select request.
    946      1.15       cgd  */
    947      1.15       cgd void
    948      1.15       cgd selrecord(selector, sip)
    949      1.15       cgd 	struct proc *selector;
    950      1.15       cgd 	struct selinfo *sip;
    951      1.15       cgd {
    952      1.15       cgd 	struct proc *p;
    953      1.15       cgd 	pid_t mypid;
    954      1.15       cgd 
    955      1.15       cgd 	mypid = selector->p_pid;
    956      1.15       cgd 	if (sip->si_pid == mypid)
    957      1.15       cgd 		return;
    958      1.15       cgd 	if (sip->si_pid && (p = pfind(sip->si_pid)) &&
    959      1.15       cgd 	    p->p_wchan == (caddr_t)&selwait)
    960      1.15       cgd 		sip->si_flags |= SI_COLL;
    961      1.15       cgd 	else
    962      1.15       cgd 		sip->si_pid = mypid;
    963      1.15       cgd }
    964      1.15       cgd 
    965      1.15       cgd /*
    966      1.15       cgd  * Do a wakeup when a selectable event occurs.
    967      1.15       cgd  */
    968      1.15       cgd void
    969      1.15       cgd selwakeup(sip)
    970      1.15       cgd 	register struct selinfo *sip;
    971      1.15       cgd {
    972      1.15       cgd 	register struct proc *p;
    973      1.15       cgd 	int s;
    974      1.15       cgd 
    975      1.15       cgd 	if (sip->si_pid == 0)
    976      1.15       cgd 		return;
    977      1.15       cgd 	if (sip->si_flags & SI_COLL) {
    978      1.15       cgd 		nselcoll++;
    979      1.15       cgd 		sip->si_flags &= ~SI_COLL;
    980      1.15       cgd 		wakeup((caddr_t)&selwait);
    981      1.15       cgd 	}
    982      1.15       cgd 	p = pfind(sip->si_pid);
    983      1.15       cgd 	sip->si_pid = 0;
    984      1.15       cgd 	if (p != NULL) {
    985      1.15       cgd 		s = splhigh();
    986      1.15       cgd 		if (p->p_wchan == (caddr_t)&selwait) {
    987      1.15       cgd 			if (p->p_stat == SSLEEP)
    988      1.15       cgd 				setrunnable(p);
    989      1.15       cgd 			else
    990      1.15       cgd 				unsleep(p);
    991      1.15       cgd 		} else if (p->p_flag & P_SELECT)
    992      1.15       cgd 			p->p_flag &= ~P_SELECT;
    993      1.15       cgd 		splx(s);
    994      1.15       cgd 	}
    995      1.15       cgd }
    996