Home | History | Annotate | Line # | Download | only in kern
kern_descrip.c revision 1.38
      1 /*	$NetBSD: kern_descrip.c,v 1.38 1996/02/04 02:15:17 christos Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  * (c) UNIX System Laboratories, Inc.
      7  * All or some portions of this file are derived from material licensed
      8  * to the University of California by American Telephone and Telegraph
      9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     10  * the permission of UNIX System Laboratories, Inc.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
     41  */
     42 
     43 #include <sys/param.h>
     44 #include <sys/systm.h>
     45 #include <sys/filedesc.h>
     46 #include <sys/kernel.h>
     47 #include <sys/vnode.h>
     48 #include <sys/proc.h>
     49 #include <sys/file.h>
     50 #include <sys/socket.h>
     51 #include <sys/socketvar.h>
     52 #include <sys/stat.h>
     53 #include <sys/ioctl.h>
     54 #include <sys/fcntl.h>
     55 #include <sys/malloc.h>
     56 #include <sys/syslog.h>
     57 #include <sys/unistd.h>
     58 #include <sys/resourcevar.h>
     59 
     60 #include <sys/mount.h>
     61 #include <sys/syscallargs.h>
     62 
     63 #include <vm/vm.h>
     64 
     65 #include <kern/kern_extern.h>
     66 #include <kern/kern_conf.h>
     67 
     68 /*
     69  * Descriptor management.
     70  */
     71 struct filelist filehead;	/* head of list of open files */
     72 int nfiles;			/* actual number of open files */
     73 
     74 static __inline void fd_used __P((struct filedesc *, int));
     75 static __inline void fd_unused __P((struct filedesc *, int));
     76 int finishdup __P((struct filedesc *, int, int, register_t *));
     77 
     78 static __inline void
     79 fd_used(fdp, fd)
     80 	register struct filedesc *fdp;
     81 	register int fd;
     82 {
     83 
     84 	if (fd > fdp->fd_lastfile)
     85 		fdp->fd_lastfile = fd;
     86 }
     87 
     88 static __inline void
     89 fd_unused(fdp, fd)
     90 	register struct filedesc *fdp;
     91 	register int fd;
     92 {
     93 
     94 	if (fd < fdp->fd_freefile)
     95 		fdp->fd_freefile = fd;
     96 #ifdef DIAGNOSTIC
     97 	if (fd > fdp->fd_lastfile)
     98 		panic("fd_unused: fd_lastfile inconsistent");
     99 #endif
    100 	if (fd == fdp->fd_lastfile) {
    101 		do {
    102 			fd--;
    103 		} while (fd >= 0 && fdp->fd_ofiles[fd] == NULL);
    104 		fdp->fd_lastfile = fd;
    105 	}
    106 }
    107 
    108 /*
    109  * System calls on descriptors.
    110  */
    111 
    112 /*
    113  * Duplicate a file descriptor.
    114  */
    115 /* ARGSUSED */
    116 int
    117 sys_dup(p, v, retval)
    118 	struct proc *p;
    119 	void *v;
    120 	register_t *retval;
    121 {
    122 	struct sys_dup_args /* {
    123 		syscallarg(u_int) fd;
    124 	} */ *uap = v;
    125 	register struct filedesc *fdp = p->p_fd;
    126 	register int old = SCARG(uap, fd);
    127 	int new;
    128 	int error;
    129 
    130 	if ((u_int)old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL)
    131 		return (EBADF);
    132 	if ((error = fdalloc(p, 0, &new)) != 0)
    133 		return (error);
    134 	return (finishdup(fdp, old, new, retval));
    135 }
    136 
    137 /*
    138  * Duplicate a file descriptor to a particular value.
    139  */
    140 /* ARGSUSED */
    141 int
    142 sys_dup2(p, v, retval)
    143 	struct proc *p;
    144 	void *v;
    145 	register_t *retval;
    146 {
    147 	struct sys_dup2_args /* {
    148 		syscallarg(u_int) from;
    149 		syscallarg(u_int) to;
    150 	} */ *uap = v;
    151 	register struct filedesc *fdp = p->p_fd;
    152 	register int old = SCARG(uap, from), new = SCARG(uap, to);
    153 	int i, error;
    154 
    155 	if ((u_int)old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL ||
    156 	    (u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
    157 	    (u_int)new >= maxfiles)
    158 		return (EBADF);
    159 	if (old == new) {
    160 		*retval = new;
    161 		return (0);
    162 	}
    163 	if (new >= fdp->fd_nfiles) {
    164 		if ((error = fdalloc(p, new, &i)) != 0)
    165 			return (error);
    166 		if (new != i)
    167 			panic("dup2: fdalloc");
    168 	} else {
    169 		(void) fdrelease(p, new);
    170 	}
    171 	return (finishdup(fdp, old, new, retval));
    172 }
    173 
    174 /*
    175  * The file control system call.
    176  */
    177 /* ARGSUSED */
    178 int
    179 sys_fcntl(p, v, retval)
    180 	struct proc *p;
    181 	void *v;
    182 	register_t *retval;
    183 {
    184 	register struct sys_fcntl_args /* {
    185 		syscallarg(int) fd;
    186 		syscallarg(int) cmd;
    187 		syscallarg(void *) arg;
    188 	} */ *uap = v;
    189 	int fd = SCARG(uap, fd);
    190 	register struct filedesc *fdp = p->p_fd;
    191 	register struct file *fp;
    192 	struct vnode *vp;
    193 	int i, tmp, error, flg = F_POSIX;
    194 	struct flock fl;
    195 	int newmin;
    196 
    197 	if ((u_int)fd >= fdp->fd_nfiles ||
    198 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    199 		return (EBADF);
    200 	switch (SCARG(uap, cmd)) {
    201 
    202 	case F_DUPFD:
    203 		newmin = (long)SCARG(uap, arg);
    204 		if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
    205 		    (u_int)newmin >= maxfiles)
    206 			return (EINVAL);
    207 		if ((error = fdalloc(p, newmin, &i)) != 0)
    208 			return (error);
    209 		return (finishdup(fdp, fd, i, retval));
    210 
    211 	case F_GETFD:
    212 		*retval = fdp->fd_ofileflags[fd] & UF_EXCLOSE ? 1 : 0;
    213 		return (0);
    214 
    215 	case F_SETFD:
    216 		if ((long)SCARG(uap, arg) & 1)
    217 			fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
    218 		else
    219 			fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
    220 		return (0);
    221 
    222 	case F_GETFL:
    223 		*retval = OFLAGS(fp->f_flag);
    224 		return (0);
    225 
    226 	case F_SETFL:
    227 		fp->f_flag &= ~FCNTLFLAGS;
    228 		fp->f_flag |= FFLAGS((long)SCARG(uap, arg)) & FCNTLFLAGS;
    229 		tmp = fp->f_flag & FNONBLOCK;
    230 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    231 		if (error)
    232 			return (error);
    233 		tmp = fp->f_flag & FASYNC;
    234 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
    235 		if (!error)
    236 			return (0);
    237 		fp->f_flag &= ~FNONBLOCK;
    238 		tmp = 0;
    239 		(void) (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    240 		return (error);
    241 
    242 	case F_GETOWN:
    243 		if (fp->f_type == DTYPE_SOCKET) {
    244 			*retval = ((struct socket *)fp->f_data)->so_pgid;
    245 			return (0);
    246 		}
    247 		error = (*fp->f_ops->fo_ioctl)
    248 			(fp, (int)TIOCGPGRP, (caddr_t)retval, p);
    249 		*retval = -*retval;
    250 		return (error);
    251 
    252 	case F_SETOWN:
    253 		if (fp->f_type == DTYPE_SOCKET) {
    254 			((struct socket *)fp->f_data)->so_pgid =
    255 			    (long)SCARG(uap, arg);
    256 			return (0);
    257 		}
    258 		if ((long)SCARG(uap, arg) <= 0) {
    259 			SCARG(uap, arg) = (void *)(-(long)SCARG(uap, arg));
    260 		} else {
    261 			struct proc *p1 = pfind((long)SCARG(uap, arg));
    262 			if (p1 == 0)
    263 				return (ESRCH);
    264 			SCARG(uap, arg) = (void *)(long)p1->p_pgrp->pg_id;
    265 		}
    266 		return ((*fp->f_ops->fo_ioctl)
    267 			(fp, (int)TIOCSPGRP, (caddr_t)&SCARG(uap, arg), p));
    268 
    269 	case F_SETLKW:
    270 		flg |= F_WAIT;
    271 		/* Fall into F_SETLK */
    272 
    273 	case F_SETLK:
    274 		if (fp->f_type != DTYPE_VNODE)
    275 			return (EBADF);
    276 		vp = (struct vnode *)fp->f_data;
    277 		/* Copy in the lock structure */
    278 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    279 		    sizeof (fl));
    280 		if (error)
    281 			return (error);
    282 		if (fl.l_whence == SEEK_CUR)
    283 			fl.l_start += fp->f_offset;
    284 		switch (fl.l_type) {
    285 
    286 		case F_RDLCK:
    287 			if ((fp->f_flag & FREAD) == 0)
    288 				return (EBADF);
    289 			p->p_flag |= P_ADVLOCK;
    290 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    291 
    292 		case F_WRLCK:
    293 			if ((fp->f_flag & FWRITE) == 0)
    294 				return (EBADF);
    295 			p->p_flag |= P_ADVLOCK;
    296 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    297 
    298 		case F_UNLCK:
    299 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &fl,
    300 				F_POSIX));
    301 
    302 		default:
    303 			return (EINVAL);
    304 		}
    305 
    306 	case F_GETLK:
    307 		if (fp->f_type != DTYPE_VNODE)
    308 			return (EBADF);
    309 		vp = (struct vnode *)fp->f_data;
    310 		/* Copy in the lock structure */
    311 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    312 		    sizeof (fl));
    313 		if (error)
    314 			return (error);
    315 		if (fl.l_whence == SEEK_CUR)
    316 			fl.l_start += fp->f_offset;
    317 		error = VOP_ADVLOCK(vp, (caddr_t)p, F_GETLK, &fl, F_POSIX);
    318 		if (error)
    319 			return (error);
    320 		return (copyout((caddr_t)&fl, (caddr_t)SCARG(uap, arg),
    321 		    sizeof (fl)));
    322 
    323 	default:
    324 		return (EINVAL);
    325 	}
    326 	/* NOTREACHED */
    327 }
    328 
    329 /*
    330  * Common code for dup, dup2, and fcntl(F_DUPFD).
    331  */
    332 int
    333 finishdup(fdp, old, new, retval)
    334 	register struct filedesc *fdp;
    335 	register int old, new;
    336 	register_t *retval;
    337 {
    338 	register struct file *fp;
    339 
    340 	fp = fdp->fd_ofiles[old];
    341 	fdp->fd_ofiles[new] = fp;
    342 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
    343 	fp->f_count++;
    344 	fd_used(fdp, new);
    345 	*retval = new;
    346 	return (0);
    347 }
    348 
    349 int
    350 fdrelease(p, fd)
    351 	struct proc *p;
    352 	int fd;
    353 {
    354 	register struct filedesc *fdp = p->p_fd;
    355 	register struct file **fpp, *fp;
    356 	register char *pf;
    357 
    358 	fpp = &fdp->fd_ofiles[fd];
    359 	fp = *fpp;
    360 	if (fp == NULL)
    361 		return (EBADF);
    362 	pf = &fdp->fd_ofileflags[fd];
    363 	if (*pf & UF_MAPPED)
    364 		(void) munmapfd(p, fd);
    365 	*fpp = NULL;
    366 	*pf = 0;
    367 	fd_unused(fdp, fd);
    368 	return (closef(fp, p));
    369 }
    370 
    371 /*
    372  * Close a file descriptor.
    373  */
    374 /* ARGSUSED */
    375 int
    376 sys_close(p, v, retval)
    377 	struct proc *p;
    378 	void *v;
    379 	register_t *retval;
    380 {
    381 	struct sys_close_args /* {
    382 		syscallarg(int) fd;
    383 	} */ *uap = v;
    384 	int fd = SCARG(uap, fd);
    385 	register struct filedesc *fdp = p->p_fd;
    386 
    387 	if ((u_int)fd >= fdp->fd_nfiles)
    388 		return (EBADF);
    389 	return (fdrelease(p, fd));
    390 }
    391 
    392 /*
    393  * Return status information about a file descriptor.
    394  */
    395 /* ARGSUSED */
    396 int
    397 sys_fstat(p, v, retval)
    398 	struct proc *p;
    399 	void *v;
    400 	register_t *retval;
    401 {
    402 	register struct sys_fstat_args /* {
    403 		syscallarg(int) fd;
    404 		syscallarg(struct stat *) sb;
    405 	} */ *uap = v;
    406 	int fd = SCARG(uap, fd);
    407 	register struct filedesc *fdp = p->p_fd;
    408 	register struct file *fp;
    409 	struct stat ub;
    410 	int error;
    411 
    412 	if ((u_int)fd >= fdp->fd_nfiles ||
    413 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    414 		return (EBADF);
    415 	switch (fp->f_type) {
    416 
    417 	case DTYPE_VNODE:
    418 		error = vn_stat((struct vnode *)fp->f_data, &ub, p);
    419 		break;
    420 
    421 	case DTYPE_SOCKET:
    422 		error = soo_stat((struct socket *)fp->f_data, &ub);
    423 		break;
    424 
    425 	default:
    426 		panic("fstat");
    427 		/*NOTREACHED*/
    428 	}
    429 	if (error == 0)
    430 		error = copyout((caddr_t)&ub, (caddr_t)SCARG(uap, sb),
    431 		    sizeof (ub));
    432 	return (error);
    433 }
    434 
    435 /*
    436  * Return pathconf information about a file descriptor.
    437  */
    438 /* ARGSUSED */
    439 int
    440 sys_fpathconf(p, v, retval)
    441 	struct proc *p;
    442 	void *v;
    443 	register_t *retval;
    444 {
    445 	register struct sys_fpathconf_args /* {
    446 		syscallarg(int) fd;
    447 		syscallarg(int) name;
    448 	} */ *uap = v;
    449 	int fd = SCARG(uap, fd);
    450 	struct filedesc *fdp = p->p_fd;
    451 	struct file *fp;
    452 	struct vnode *vp;
    453 
    454 	if ((u_int)fd >= fdp->fd_nfiles ||
    455 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    456 		return (EBADF);
    457 	switch (fp->f_type) {
    458 
    459 	case DTYPE_SOCKET:
    460 		if (SCARG(uap, name) != _PC_PIPE_BUF)
    461 			return (EINVAL);
    462 		*retval = PIPE_BUF;
    463 		return (0);
    464 
    465 	case DTYPE_VNODE:
    466 		vp = (struct vnode *)fp->f_data;
    467 		return (VOP_PATHCONF(vp, SCARG(uap, name), retval));
    468 
    469 	default:
    470 		panic("fpathconf");
    471 	}
    472 	/*NOTREACHED*/
    473 }
    474 
    475 /*
    476  * Allocate a file descriptor for the process.
    477  */
    478 int fdexpand;
    479 
    480 int
    481 fdalloc(p, want, result)
    482 	struct proc *p;
    483 	int want;
    484 	int *result;
    485 {
    486 	register struct filedesc *fdp = p->p_fd;
    487 	register int i;
    488 	int lim, last, nfiles;
    489 	struct file **newofile;
    490 	char *newofileflags;
    491 
    492 	/*
    493 	 * Search for a free descriptor starting at the higher
    494 	 * of want or fd_freefile.  If that fails, consider
    495 	 * expanding the ofile array.
    496 	 */
    497 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
    498 	for (;;) {
    499 		last = min(fdp->fd_nfiles, lim);
    500 		if ((i = want) < fdp->fd_freefile)
    501 			i = fdp->fd_freefile;
    502 		for (; i < last; i++) {
    503 			if (fdp->fd_ofiles[i] == NULL) {
    504 				fd_used(fdp, i);
    505 				if (want <= fdp->fd_freefile)
    506 					fdp->fd_freefile = i;
    507 				*result = i;
    508 				return (0);
    509 			}
    510 		}
    511 
    512 		/*
    513 		 * No space in current array.  Expand?
    514 		 */
    515 		if (fdp->fd_nfiles >= lim)
    516 			return (EMFILE);
    517 		if (fdp->fd_nfiles < NDEXTENT)
    518 			nfiles = NDEXTENT;
    519 		else
    520 			nfiles = 2 * fdp->fd_nfiles;
    521 		MALLOC(newofile, struct file **, nfiles * OFILESIZE,
    522 		    M_FILEDESC, M_WAITOK);
    523 		newofileflags = (char *) &newofile[nfiles];
    524 		/*
    525 		 * Copy the existing ofile and ofileflags arrays
    526 		 * and zero the new portion of each array.
    527 		 */
    528 		bcopy(fdp->fd_ofiles, newofile,
    529 			(i = sizeof(struct file *) * fdp->fd_nfiles));
    530 		bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i);
    531 		bcopy(fdp->fd_ofileflags, newofileflags,
    532 			(i = sizeof(char) * fdp->fd_nfiles));
    533 		bzero(newofileflags + i, nfiles * sizeof(char) - i);
    534 		if (fdp->fd_nfiles > NDFILE)
    535 			FREE(fdp->fd_ofiles, M_FILEDESC);
    536 		fdp->fd_ofiles = newofile;
    537 		fdp->fd_ofileflags = newofileflags;
    538 		fdp->fd_nfiles = nfiles;
    539 		fdexpand++;
    540 	}
    541 }
    542 
    543 /*
    544  * Check to see whether n user file descriptors
    545  * are available to the process p.
    546  */
    547 int
    548 fdavail(p, n)
    549 	struct proc *p;
    550 	register int n;
    551 {
    552 	register struct filedesc *fdp = p->p_fd;
    553 	register struct file **fpp;
    554 	register int i, lim;
    555 
    556 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
    557 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
    558 		return (1);
    559 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
    560 	for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++)
    561 		if (*fpp == NULL && --n <= 0)
    562 			return (1);
    563 	return (0);
    564 }
    565 
    566 /*
    567  * Create a new open file structure and allocate
    568  * a file decriptor for the process that refers to it.
    569  */
    570 int
    571 falloc(p, resultfp, resultfd)
    572 	register struct proc *p;
    573 	struct file **resultfp;
    574 	int *resultfd;
    575 {
    576 	register struct file *fp, *fq;
    577 	int error, i;
    578 
    579 	if ((error = fdalloc(p, 0, &i)) != 0)
    580 		return (error);
    581 	if (nfiles >= maxfiles) {
    582 		tablefull("file");
    583 		return (ENFILE);
    584 	}
    585 	/*
    586 	 * Allocate a new file descriptor.
    587 	 * If the process has file descriptor zero open, add to the list
    588 	 * of open files at that point, otherwise put it at the front of
    589 	 * the list of open files.
    590 	 */
    591 	nfiles++;
    592 	MALLOC(fp, struct file *, sizeof(struct file), M_FILE, M_WAITOK);
    593 	bzero(fp, sizeof(struct file));
    594 	if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
    595 		LIST_INSERT_AFTER(fq, fp, f_list);
    596 	} else {
    597 		LIST_INSERT_HEAD(&filehead, fp, f_list);
    598 	}
    599 	p->p_fd->fd_ofiles[i] = fp;
    600 	fp->f_count = 1;
    601 	fp->f_cred = p->p_ucred;
    602 	crhold(fp->f_cred);
    603 	if (resultfp)
    604 		*resultfp = fp;
    605 	if (resultfd)
    606 		*resultfd = i;
    607 	return (0);
    608 }
    609 
    610 /*
    611  * Free a file descriptor.
    612  */
    613 void
    614 ffree(fp)
    615 	register struct file *fp;
    616 {
    617 	LIST_REMOVE(fp, f_list);
    618 	crfree(fp->f_cred);
    619 #ifdef DIAGNOSTIC
    620 	fp->f_count = 0;
    621 #endif
    622 	nfiles--;
    623 	FREE(fp, M_FILE);
    624 }
    625 
    626 /*
    627  * Copy a filedesc structure.
    628  */
    629 struct filedesc *
    630 fdcopy(p)
    631 	struct proc *p;
    632 {
    633 	register struct filedesc *newfdp, *fdp = p->p_fd;
    634 	register struct file **fpp;
    635 	register int i;
    636 
    637 	MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0),
    638 	    M_FILEDESC, M_WAITOK);
    639 	bcopy(fdp, newfdp, sizeof(struct filedesc));
    640 	VREF(newfdp->fd_cdir);
    641 	if (newfdp->fd_rdir)
    642 		VREF(newfdp->fd_rdir);
    643 	newfdp->fd_refcnt = 1;
    644 
    645 	/*
    646 	 * If the number of open files fits in the internal arrays
    647 	 * of the open file structure, use them, otherwise allocate
    648 	 * additional memory for the number of descriptors currently
    649 	 * in use.
    650 	 */
    651 	if (newfdp->fd_lastfile < NDFILE) {
    652 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
    653 		newfdp->fd_ofileflags =
    654 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
    655 		i = NDFILE;
    656 	} else {
    657 		/*
    658 		 * Compute the smallest multiple of NDEXTENT needed
    659 		 * for the file descriptors currently in use,
    660 		 * allowing the table to shrink.
    661 		 */
    662 		i = newfdp->fd_nfiles;
    663 		while (i >= 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
    664 			i /= 2;
    665 		MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE,
    666 		    M_FILEDESC, M_WAITOK);
    667 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
    668 	}
    669 	newfdp->fd_nfiles = i;
    670 	bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
    671 	bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
    672 	fpp = newfdp->fd_ofiles;
    673 	for (i = newfdp->fd_lastfile; i >= 0; i--, fpp++)
    674 		if (*fpp != NULL)
    675 			(*fpp)->f_count++;
    676 	return (newfdp);
    677 }
    678 
    679 /*
    680  * Release a filedesc structure.
    681  */
    682 void
    683 fdfree(p)
    684 	struct proc *p;
    685 {
    686 	register struct filedesc *fdp = p->p_fd;
    687 	register struct file **fpp, *fp;
    688 	register int i;
    689 
    690 	if (--fdp->fd_refcnt > 0)
    691 		return;
    692 	fpp = fdp->fd_ofiles;
    693 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
    694 		fp = *fpp;
    695 		if (fp != NULL) {
    696 			*fpp = NULL;
    697 			(void) closef(fp, p);
    698 		}
    699 	}
    700 	p->p_fd = NULL;
    701 	if (fdp->fd_nfiles > NDFILE)
    702 		FREE(fdp->fd_ofiles, M_FILEDESC);
    703 	vrele(fdp->fd_cdir);
    704 	if (fdp->fd_rdir)
    705 		vrele(fdp->fd_rdir);
    706 	FREE(fdp, M_FILEDESC);
    707 }
    708 
    709 /*
    710  * Internal form of close.
    711  * Decrement reference count on file structure.
    712  * Note: p may be NULL when closing a file
    713  * that was being passed in a message.
    714  */
    715 int
    716 closef(fp, p)
    717 	register struct file *fp;
    718 	register struct proc *p;
    719 {
    720 	struct vnode *vp;
    721 	struct flock lf;
    722 	int error;
    723 
    724 	if (fp == NULL)
    725 		return (0);
    726 	/*
    727 	 * POSIX record locking dictates that any close releases ALL
    728 	 * locks owned by this process.  This is handled by setting
    729 	 * a flag in the unlock to free ONLY locks obeying POSIX
    730 	 * semantics, and not to free BSD-style file locks.
    731 	 * If the descriptor was in a message, POSIX-style locks
    732 	 * aren't passed with the descriptor.
    733 	 */
    734 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
    735 		lf.l_whence = SEEK_SET;
    736 		lf.l_start = 0;
    737 		lf.l_len = 0;
    738 		lf.l_type = F_UNLCK;
    739 		vp = (struct vnode *)fp->f_data;
    740 		(void) VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX);
    741 	}
    742 	if (--fp->f_count > 0)
    743 		return (0);
    744 	if (fp->f_count < 0)
    745 		panic("closef: count < 0");
    746 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
    747 		lf.l_whence = SEEK_SET;
    748 		lf.l_start = 0;
    749 		lf.l_len = 0;
    750 		lf.l_type = F_UNLCK;
    751 		vp = (struct vnode *)fp->f_data;
    752 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
    753 	}
    754 	if (fp->f_ops)
    755 		error = (*fp->f_ops->fo_close)(fp, p);
    756 	else
    757 		error = 0;
    758 	ffree(fp);
    759 	return (error);
    760 }
    761 
    762 /*
    763  * Apply an advisory lock on a file descriptor.
    764  *
    765  * Just attempt to get a record lock of the requested type on
    766  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
    767  */
    768 /* ARGSUSED */
    769 int
    770 sys_flock(p, v, retval)
    771 	struct proc *p;
    772 	void *v;
    773 	register_t *retval;
    774 {
    775 	register struct sys_flock_args /* {
    776 		syscallarg(int) fd;
    777 		syscallarg(int) how;
    778 	} */ *uap = v;
    779 	int fd = SCARG(uap, fd);
    780 	int how = SCARG(uap, how);
    781 	register struct filedesc *fdp = p->p_fd;
    782 	register struct file *fp;
    783 	struct vnode *vp;
    784 	struct flock lf;
    785 
    786 	if ((u_int)fd >= fdp->fd_nfiles ||
    787 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    788 		return (EBADF);
    789 	if (fp->f_type != DTYPE_VNODE)
    790 		return (EOPNOTSUPP);
    791 	vp = (struct vnode *)fp->f_data;
    792 	lf.l_whence = SEEK_SET;
    793 	lf.l_start = 0;
    794 	lf.l_len = 0;
    795 	if (how & LOCK_UN) {
    796 		lf.l_type = F_UNLCK;
    797 		fp->f_flag &= ~FHASLOCK;
    798 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK));
    799 	}
    800 	if (how & LOCK_EX)
    801 		lf.l_type = F_WRLCK;
    802 	else if (how & LOCK_SH)
    803 		lf.l_type = F_RDLCK;
    804 	else
    805 		return (EBADF);
    806 	fp->f_flag |= FHASLOCK;
    807 	if (how & LOCK_NB)
    808 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK));
    809 	return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT));
    810 }
    811 
    812 /*
    813  * File Descriptor pseudo-device driver (/dev/fd/).
    814  *
    815  * Opening minor device N dup()s the file (if any) connected to file
    816  * descriptor N belonging to the calling process.  Note that this driver
    817  * consists of only the ``open()'' routine, because all subsequent
    818  * references to this file will be direct to the other driver.
    819  */
    820 /* ARGSUSED */
    821 int
    822 fdopen(dev, mode, type, p)
    823 	dev_t dev;
    824 	int mode, type;
    825 	struct proc *p;
    826 {
    827 
    828 	/*
    829 	 * XXX Kludge: set curproc->p_dupfd to contain the value of the
    830 	 * the file descriptor being sought for duplication. The error
    831 	 * return ensures that the vnode for this device will be released
    832 	 * by vn_open. Open will detect this special error and take the
    833 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
    834 	 * will simply report the error.
    835 	 */
    836 	p->p_dupfd = minor(dev);
    837 	return (ENODEV);
    838 }
    839 
    840 /*
    841  * Duplicate the specified descriptor to a free descriptor.
    842  */
    843 int
    844 dupfdopen(fdp, indx, dfd, mode, error)
    845 	register struct filedesc *fdp;
    846 	register int indx, dfd;
    847 	int mode;
    848 	int error;
    849 {
    850 	register struct file *wfp;
    851 	struct file *fp;
    852 
    853 	/*
    854 	 * If the to-be-dup'd fd number is greater than the allowed number
    855 	 * of file descriptors, or the fd to be dup'd has already been
    856 	 * closed, reject.  Note, check for new == old is necessary as
    857 	 * falloc could allocate an already closed to-be-dup'd descriptor
    858 	 * as the new descriptor.
    859 	 */
    860 	fp = fdp->fd_ofiles[indx];
    861 	if ((u_int)dfd >= fdp->fd_nfiles ||
    862 	    (wfp = fdp->fd_ofiles[dfd]) == NULL || fp == wfp)
    863 		return (EBADF);
    864 
    865 	/*
    866 	 * There are two cases of interest here.
    867 	 *
    868 	 * For ENODEV simply dup (dfd) to file descriptor
    869 	 * (indx) and return.
    870 	 *
    871 	 * For ENXIO steal away the file structure from (dfd) and
    872 	 * store it in (indx).  (dfd) is effectively closed by
    873 	 * this operation.
    874 	 *
    875 	 * Any other error code is just returned.
    876 	 */
    877 	switch (error) {
    878 	case ENODEV:
    879 		/*
    880 		 * Check that the mode the file is being opened for is a
    881 		 * subset of the mode of the existing descriptor.
    882 		 */
    883 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag)
    884 			return (EACCES);
    885 		fdp->fd_ofiles[indx] = wfp;
    886 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    887 		wfp->f_count++;
    888 		fd_used(fdp, indx);
    889 		return (0);
    890 
    891 	case ENXIO:
    892 		/*
    893 		 * Steal away the file pointer from dfd, and stuff it into indx.
    894 		 */
    895 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
    896 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    897 		fdp->fd_ofiles[dfd] = NULL;
    898 		fdp->fd_ofileflags[dfd] = 0;
    899 		/*
    900 		 * Complete the clean up of the filedesc structure by
    901 		 * recomputing the various hints.
    902 		 */
    903 		fd_used(fdp, indx);
    904 		fd_unused(fdp, dfd);
    905 		return (0);
    906 
    907 	default:
    908 		return (error);
    909 	}
    910 	/* NOTREACHED */
    911 }
    912 
    913 /*
    914  * Close any files on exec?
    915  */
    916 void
    917 fdcloseexec(p)
    918 	struct proc *p;
    919 {
    920 	register struct filedesc *fdp = p->p_fd;
    921 	register int fd;
    922 
    923 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
    924 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
    925 			(void) fdrelease(p, fd);
    926 }
    927