Home | History | Annotate | Line # | Download | only in kern
kern_descrip.c revision 1.40
      1 /*	$NetBSD: kern_descrip.c,v 1.40 1996/03/14 19:01:10 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_conf.h>
     66 
     67 /*
     68  * Descriptor management.
     69  */
     70 struct filelist filehead;	/* head of list of open files */
     71 int nfiles;			/* actual number of open files */
     72 
     73 static __inline void fd_used __P((struct filedesc *, int));
     74 static __inline void fd_unused __P((struct filedesc *, int));
     75 int finishdup __P((struct filedesc *, int, int, register_t *));
     76 
     77 static __inline void
     78 fd_used(fdp, fd)
     79 	register struct filedesc *fdp;
     80 	register int fd;
     81 {
     82 
     83 	if (fd > fdp->fd_lastfile)
     84 		fdp->fd_lastfile = fd;
     85 }
     86 
     87 static __inline void
     88 fd_unused(fdp, fd)
     89 	register struct filedesc *fdp;
     90 	register int fd;
     91 {
     92 
     93 	if (fd < fdp->fd_freefile)
     94 		fdp->fd_freefile = fd;
     95 #ifdef DIAGNOSTIC
     96 	if (fd > fdp->fd_lastfile)
     97 		panic("fd_unused: fd_lastfile inconsistent");
     98 #endif
     99 	if (fd == fdp->fd_lastfile) {
    100 		do {
    101 			fd--;
    102 		} while (fd >= 0 && fdp->fd_ofiles[fd] == NULL);
    103 		fdp->fd_lastfile = fd;
    104 	}
    105 }
    106 
    107 /*
    108  * System calls on descriptors.
    109  */
    110 
    111 /*
    112  * Duplicate a file descriptor.
    113  */
    114 /* ARGSUSED */
    115 int
    116 sys_dup(p, v, retval)
    117 	struct proc *p;
    118 	void *v;
    119 	register_t *retval;
    120 {
    121 	struct sys_dup_args /* {
    122 		syscallarg(u_int) fd;
    123 	} */ *uap = v;
    124 	register struct filedesc *fdp = p->p_fd;
    125 	register int old = SCARG(uap, fd);
    126 	int new;
    127 	int error;
    128 
    129 	if ((u_int)old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL)
    130 		return (EBADF);
    131 	if ((error = fdalloc(p, 0, &new)) != 0)
    132 		return (error);
    133 	return (finishdup(fdp, old, new, retval));
    134 }
    135 
    136 /*
    137  * Duplicate a file descriptor to a particular value.
    138  */
    139 /* ARGSUSED */
    140 int
    141 sys_dup2(p, v, retval)
    142 	struct proc *p;
    143 	void *v;
    144 	register_t *retval;
    145 {
    146 	struct sys_dup2_args /* {
    147 		syscallarg(u_int) from;
    148 		syscallarg(u_int) to;
    149 	} */ *uap = v;
    150 	register struct filedesc *fdp = p->p_fd;
    151 	register int old = SCARG(uap, from), new = SCARG(uap, to);
    152 	int i, error;
    153 
    154 	if ((u_int)old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL ||
    155 	    (u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
    156 	    (u_int)new >= maxfiles)
    157 		return (EBADF);
    158 	if (old == new) {
    159 		*retval = new;
    160 		return (0);
    161 	}
    162 	if (new >= fdp->fd_nfiles) {
    163 		if ((error = fdalloc(p, new, &i)) != 0)
    164 			return (error);
    165 		if (new != i)
    166 			panic("dup2: fdalloc");
    167 	} else {
    168 		(void) fdrelease(p, new);
    169 	}
    170 	return (finishdup(fdp, old, new, retval));
    171 }
    172 
    173 /*
    174  * The file control system call.
    175  */
    176 /* ARGSUSED */
    177 int
    178 sys_fcntl(p, v, retval)
    179 	struct proc *p;
    180 	void *v;
    181 	register_t *retval;
    182 {
    183 	register struct sys_fcntl_args /* {
    184 		syscallarg(int) fd;
    185 		syscallarg(int) cmd;
    186 		syscallarg(void *) arg;
    187 	} */ *uap = v;
    188 	int fd = SCARG(uap, fd);
    189 	register struct filedesc *fdp = p->p_fd;
    190 	register struct file *fp;
    191 	struct vnode *vp;
    192 	int i, tmp, error, flg = F_POSIX;
    193 	struct flock fl;
    194 	int newmin;
    195 
    196 	if ((u_int)fd >= fdp->fd_nfiles ||
    197 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    198 		return (EBADF);
    199 	switch (SCARG(uap, cmd)) {
    200 
    201 	case F_DUPFD:
    202 		newmin = (long)SCARG(uap, arg);
    203 		if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
    204 		    (u_int)newmin >= maxfiles)
    205 			return (EINVAL);
    206 		if ((error = fdalloc(p, newmin, &i)) != 0)
    207 			return (error);
    208 		return (finishdup(fdp, fd, i, retval));
    209 
    210 	case F_GETFD:
    211 		*retval = fdp->fd_ofileflags[fd] & UF_EXCLOSE ? 1 : 0;
    212 		return (0);
    213 
    214 	case F_SETFD:
    215 		if ((long)SCARG(uap, arg) & 1)
    216 			fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
    217 		else
    218 			fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
    219 		return (0);
    220 
    221 	case F_GETFL:
    222 		*retval = OFLAGS(fp->f_flag);
    223 		return (0);
    224 
    225 	case F_SETFL:
    226 		fp->f_flag &= ~FCNTLFLAGS;
    227 		fp->f_flag |= FFLAGS((long)SCARG(uap, arg)) & FCNTLFLAGS;
    228 		tmp = fp->f_flag & FNONBLOCK;
    229 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    230 		if (error)
    231 			return (error);
    232 		tmp = fp->f_flag & FASYNC;
    233 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
    234 		if (!error)
    235 			return (0);
    236 		fp->f_flag &= ~FNONBLOCK;
    237 		tmp = 0;
    238 		(void) (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    239 		return (error);
    240 
    241 	case F_GETOWN:
    242 		if (fp->f_type == DTYPE_SOCKET) {
    243 			*retval = ((struct socket *)fp->f_data)->so_pgid;
    244 			return (0);
    245 		}
    246 		error = (*fp->f_ops->fo_ioctl)
    247 			(fp, (int)TIOCGPGRP, (caddr_t)retval, p);
    248 		*retval = -*retval;
    249 		return (error);
    250 
    251 	case F_SETOWN:
    252 		if (fp->f_type == DTYPE_SOCKET) {
    253 			((struct socket *)fp->f_data)->so_pgid =
    254 			    (long)SCARG(uap, arg);
    255 			return (0);
    256 		}
    257 		if ((long)SCARG(uap, arg) <= 0) {
    258 			SCARG(uap, arg) = (void *)(-(long)SCARG(uap, arg));
    259 		} else {
    260 			struct proc *p1 = pfind((long)SCARG(uap, arg));
    261 			if (p1 == 0)
    262 				return (ESRCH);
    263 			SCARG(uap, arg) = (void *)(long)p1->p_pgrp->pg_id;
    264 		}
    265 		return ((*fp->f_ops->fo_ioctl)
    266 			(fp, (int)TIOCSPGRP, (caddr_t)&SCARG(uap, arg), p));
    267 
    268 	case F_SETLKW:
    269 		flg |= F_WAIT;
    270 		/* Fall into F_SETLK */
    271 
    272 	case F_SETLK:
    273 		if (fp->f_type != DTYPE_VNODE)
    274 			return (EBADF);
    275 		vp = (struct vnode *)fp->f_data;
    276 		/* Copy in the lock structure */
    277 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    278 		    sizeof (fl));
    279 		if (error)
    280 			return (error);
    281 		if (fl.l_whence == SEEK_CUR)
    282 			fl.l_start += fp->f_offset;
    283 		switch (fl.l_type) {
    284 
    285 		case F_RDLCK:
    286 			if ((fp->f_flag & FREAD) == 0)
    287 				return (EBADF);
    288 			p->p_flag |= P_ADVLOCK;
    289 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    290 
    291 		case F_WRLCK:
    292 			if ((fp->f_flag & FWRITE) == 0)
    293 				return (EBADF);
    294 			p->p_flag |= P_ADVLOCK;
    295 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    296 
    297 		case F_UNLCK:
    298 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &fl,
    299 				F_POSIX));
    300 
    301 		default:
    302 			return (EINVAL);
    303 		}
    304 
    305 	case F_GETLK:
    306 		if (fp->f_type != DTYPE_VNODE)
    307 			return (EBADF);
    308 		vp = (struct vnode *)fp->f_data;
    309 		/* Copy in the lock structure */
    310 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    311 		    sizeof (fl));
    312 		if (error)
    313 			return (error);
    314 		if (fl.l_whence == SEEK_CUR)
    315 			fl.l_start += fp->f_offset;
    316 		error = VOP_ADVLOCK(vp, (caddr_t)p, F_GETLK, &fl, F_POSIX);
    317 		if (error)
    318 			return (error);
    319 		return (copyout((caddr_t)&fl, (caddr_t)SCARG(uap, arg),
    320 		    sizeof (fl)));
    321 
    322 	default:
    323 		return (EINVAL);
    324 	}
    325 	/* NOTREACHED */
    326 }
    327 
    328 /*
    329  * Common code for dup, dup2, and fcntl(F_DUPFD).
    330  */
    331 int
    332 finishdup(fdp, old, new, retval)
    333 	register struct filedesc *fdp;
    334 	register int old, new;
    335 	register_t *retval;
    336 {
    337 	register struct file *fp;
    338 
    339 	fp = fdp->fd_ofiles[old];
    340 	fdp->fd_ofiles[new] = fp;
    341 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
    342 	fp->f_count++;
    343 	fd_used(fdp, new);
    344 	*retval = new;
    345 	return (0);
    346 }
    347 
    348 int
    349 fdrelease(p, fd)
    350 	struct proc *p;
    351 	int fd;
    352 {
    353 	register struct filedesc *fdp = p->p_fd;
    354 	register struct file **fpp, *fp;
    355 	register char *pf;
    356 
    357 	fpp = &fdp->fd_ofiles[fd];
    358 	fp = *fpp;
    359 	if (fp == NULL)
    360 		return (EBADF);
    361 	pf = &fdp->fd_ofileflags[fd];
    362 	if (*pf & UF_MAPPED)
    363 		(void) munmapfd(p, fd);
    364 	*fpp = NULL;
    365 	*pf = 0;
    366 	fd_unused(fdp, fd);
    367 	return (closef(fp, p));
    368 }
    369 
    370 /*
    371  * Close a file descriptor.
    372  */
    373 /* ARGSUSED */
    374 int
    375 sys_close(p, v, retval)
    376 	struct proc *p;
    377 	void *v;
    378 	register_t *retval;
    379 {
    380 	struct sys_close_args /* {
    381 		syscallarg(int) fd;
    382 	} */ *uap = v;
    383 	int fd = SCARG(uap, fd);
    384 	register struct filedesc *fdp = p->p_fd;
    385 
    386 	if ((u_int)fd >= fdp->fd_nfiles)
    387 		return (EBADF);
    388 	return (fdrelease(p, fd));
    389 }
    390 
    391 /*
    392  * Return status information about a file descriptor.
    393  */
    394 /* ARGSUSED */
    395 int
    396 sys_fstat(p, v, retval)
    397 	struct proc *p;
    398 	void *v;
    399 	register_t *retval;
    400 {
    401 	register struct sys_fstat_args /* {
    402 		syscallarg(int) fd;
    403 		syscallarg(struct stat *) sb;
    404 	} */ *uap = v;
    405 	int fd = SCARG(uap, fd);
    406 	register struct filedesc *fdp = p->p_fd;
    407 	register struct file *fp;
    408 	struct stat ub;
    409 	int error;
    410 
    411 	if ((u_int)fd >= fdp->fd_nfiles ||
    412 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    413 		return (EBADF);
    414 	switch (fp->f_type) {
    415 
    416 	case DTYPE_VNODE:
    417 		error = vn_stat((struct vnode *)fp->f_data, &ub, p);
    418 		break;
    419 
    420 	case DTYPE_SOCKET:
    421 		error = soo_stat((struct socket *)fp->f_data, &ub);
    422 		break;
    423 
    424 	default:
    425 		panic("fstat");
    426 		/*NOTREACHED*/
    427 	}
    428 	if (error == 0)
    429 		error = copyout((caddr_t)&ub, (caddr_t)SCARG(uap, sb),
    430 		    sizeof (ub));
    431 	return (error);
    432 }
    433 
    434 /*
    435  * Return pathconf information about a file descriptor.
    436  */
    437 /* ARGSUSED */
    438 int
    439 sys_fpathconf(p, v, retval)
    440 	struct proc *p;
    441 	void *v;
    442 	register_t *retval;
    443 {
    444 	register struct sys_fpathconf_args /* {
    445 		syscallarg(int) fd;
    446 		syscallarg(int) name;
    447 	} */ *uap = v;
    448 	int fd = SCARG(uap, fd);
    449 	struct filedesc *fdp = p->p_fd;
    450 	struct file *fp;
    451 	struct vnode *vp;
    452 
    453 	if ((u_int)fd >= fdp->fd_nfiles ||
    454 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    455 		return (EBADF);
    456 	switch (fp->f_type) {
    457 
    458 	case DTYPE_SOCKET:
    459 		if (SCARG(uap, name) != _PC_PIPE_BUF)
    460 			return (EINVAL);
    461 		*retval = PIPE_BUF;
    462 		return (0);
    463 
    464 	case DTYPE_VNODE:
    465 		vp = (struct vnode *)fp->f_data;
    466 		return (VOP_PATHCONF(vp, SCARG(uap, name), retval));
    467 
    468 	default:
    469 		panic("fpathconf");
    470 	}
    471 	/*NOTREACHED*/
    472 }
    473 
    474 /*
    475  * Allocate a file descriptor for the process.
    476  */
    477 int fdexpand;
    478 
    479 int
    480 fdalloc(p, want, result)
    481 	struct proc *p;
    482 	int want;
    483 	int *result;
    484 {
    485 	register struct filedesc *fdp = p->p_fd;
    486 	register int i;
    487 	int lim, last, nfiles;
    488 	struct file **newofile;
    489 	char *newofileflags;
    490 
    491 	/*
    492 	 * Search for a free descriptor starting at the higher
    493 	 * of want or fd_freefile.  If that fails, consider
    494 	 * expanding the ofile array.
    495 	 */
    496 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
    497 	for (;;) {
    498 		last = min(fdp->fd_nfiles, lim);
    499 		if ((i = want) < fdp->fd_freefile)
    500 			i = fdp->fd_freefile;
    501 		for (; i < last; i++) {
    502 			if (fdp->fd_ofiles[i] == NULL) {
    503 				fd_used(fdp, i);
    504 				if (want <= fdp->fd_freefile)
    505 					fdp->fd_freefile = i;
    506 				*result = i;
    507 				return (0);
    508 			}
    509 		}
    510 
    511 		/*
    512 		 * No space in current array.  Expand?
    513 		 */
    514 		if (fdp->fd_nfiles >= lim)
    515 			return (EMFILE);
    516 		if (fdp->fd_nfiles < NDEXTENT)
    517 			nfiles = NDEXTENT;
    518 		else
    519 			nfiles = 2 * fdp->fd_nfiles;
    520 		MALLOC(newofile, struct file **, nfiles * OFILESIZE,
    521 		    M_FILEDESC, M_WAITOK);
    522 		newofileflags = (char *) &newofile[nfiles];
    523 		/*
    524 		 * Copy the existing ofile and ofileflags arrays
    525 		 * and zero the new portion of each array.
    526 		 */
    527 		bcopy(fdp->fd_ofiles, newofile,
    528 			(i = sizeof(struct file *) * fdp->fd_nfiles));
    529 		bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i);
    530 		bcopy(fdp->fd_ofileflags, newofileflags,
    531 			(i = sizeof(char) * fdp->fd_nfiles));
    532 		bzero(newofileflags + i, nfiles * sizeof(char) - i);
    533 		if (fdp->fd_nfiles > NDFILE)
    534 			FREE(fdp->fd_ofiles, M_FILEDESC);
    535 		fdp->fd_ofiles = newofile;
    536 		fdp->fd_ofileflags = newofileflags;
    537 		fdp->fd_nfiles = nfiles;
    538 		fdexpand++;
    539 	}
    540 }
    541 
    542 /*
    543  * Check to see whether n user file descriptors
    544  * are available to the process p.
    545  */
    546 int
    547 fdavail(p, n)
    548 	struct proc *p;
    549 	register int n;
    550 {
    551 	register struct filedesc *fdp = p->p_fd;
    552 	register struct file **fpp;
    553 	register int i, lim;
    554 
    555 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
    556 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
    557 		return (1);
    558 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
    559 	for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++)
    560 		if (*fpp == NULL && --n <= 0)
    561 			return (1);
    562 	return (0);
    563 }
    564 
    565 /*
    566  * Create a new open file structure and allocate
    567  * a file decriptor for the process that refers to it.
    568  */
    569 int
    570 falloc(p, resultfp, resultfd)
    571 	register struct proc *p;
    572 	struct file **resultfp;
    573 	int *resultfd;
    574 {
    575 	register struct file *fp, *fq;
    576 	int error, i;
    577 
    578 	if ((error = fdalloc(p, 0, &i)) != 0)
    579 		return (error);
    580 	if (nfiles >= maxfiles) {
    581 		tablefull("file");
    582 		return (ENFILE);
    583 	}
    584 	/*
    585 	 * Allocate a new file descriptor.
    586 	 * If the process has file descriptor zero open, add to the list
    587 	 * of open files at that point, otherwise put it at the front of
    588 	 * the list of open files.
    589 	 */
    590 	nfiles++;
    591 	MALLOC(fp, struct file *, sizeof(struct file), M_FILE, M_WAITOK);
    592 	bzero(fp, sizeof(struct file));
    593 	if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
    594 		LIST_INSERT_AFTER(fq, fp, f_list);
    595 	} else {
    596 		LIST_INSERT_HEAD(&filehead, fp, f_list);
    597 	}
    598 	p->p_fd->fd_ofiles[i] = fp;
    599 	fp->f_count = 1;
    600 	fp->f_cred = p->p_ucred;
    601 	crhold(fp->f_cred);
    602 	if (resultfp)
    603 		*resultfp = fp;
    604 	if (resultfd)
    605 		*resultfd = i;
    606 	return (0);
    607 }
    608 
    609 /*
    610  * Free a file descriptor.
    611  */
    612 void
    613 ffree(fp)
    614 	register struct file *fp;
    615 {
    616 	LIST_REMOVE(fp, f_list);
    617 	crfree(fp->f_cred);
    618 #ifdef DIAGNOSTIC
    619 	fp->f_count = 0;
    620 #endif
    621 	nfiles--;
    622 	FREE(fp, M_FILE);
    623 }
    624 
    625 /*
    626  * Copy a filedesc structure.
    627  */
    628 struct filedesc *
    629 fdcopy(p)
    630 	struct proc *p;
    631 {
    632 	register struct filedesc *newfdp, *fdp = p->p_fd;
    633 	register struct file **fpp;
    634 	register int i;
    635 
    636 	MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0),
    637 	    M_FILEDESC, M_WAITOK);
    638 	bcopy(fdp, newfdp, sizeof(struct filedesc));
    639 	VREF(newfdp->fd_cdir);
    640 	if (newfdp->fd_rdir)
    641 		VREF(newfdp->fd_rdir);
    642 	newfdp->fd_refcnt = 1;
    643 
    644 	/*
    645 	 * If the number of open files fits in the internal arrays
    646 	 * of the open file structure, use them, otherwise allocate
    647 	 * additional memory for the number of descriptors currently
    648 	 * in use.
    649 	 */
    650 	if (newfdp->fd_lastfile < NDFILE) {
    651 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
    652 		newfdp->fd_ofileflags =
    653 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
    654 		i = NDFILE;
    655 	} else {
    656 		/*
    657 		 * Compute the smallest multiple of NDEXTENT needed
    658 		 * for the file descriptors currently in use,
    659 		 * allowing the table to shrink.
    660 		 */
    661 		i = newfdp->fd_nfiles;
    662 		while (i >= 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
    663 			i /= 2;
    664 		MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE,
    665 		    M_FILEDESC, M_WAITOK);
    666 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
    667 	}
    668 	newfdp->fd_nfiles = i;
    669 	bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
    670 	bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
    671 	fpp = newfdp->fd_ofiles;
    672 	for (i = newfdp->fd_lastfile; i >= 0; i--, fpp++)
    673 		if (*fpp != NULL)
    674 			(*fpp)->f_count++;
    675 	return (newfdp);
    676 }
    677 
    678 /*
    679  * Release a filedesc structure.
    680  */
    681 void
    682 fdfree(p)
    683 	struct proc *p;
    684 {
    685 	register struct filedesc *fdp = p->p_fd;
    686 	register struct file **fpp, *fp;
    687 	register int i;
    688 
    689 	if (--fdp->fd_refcnt > 0)
    690 		return;
    691 	fpp = fdp->fd_ofiles;
    692 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
    693 		fp = *fpp;
    694 		if (fp != NULL) {
    695 			*fpp = NULL;
    696 			(void) closef(fp, p);
    697 		}
    698 	}
    699 	p->p_fd = NULL;
    700 	if (fdp->fd_nfiles > NDFILE)
    701 		FREE(fdp->fd_ofiles, M_FILEDESC);
    702 	vrele(fdp->fd_cdir);
    703 	if (fdp->fd_rdir)
    704 		vrele(fdp->fd_rdir);
    705 	FREE(fdp, M_FILEDESC);
    706 }
    707 
    708 /*
    709  * Internal form of close.
    710  * Decrement reference count on file structure.
    711  * Note: p may be NULL when closing a file
    712  * that was being passed in a message.
    713  */
    714 int
    715 closef(fp, p)
    716 	register struct file *fp;
    717 	register struct proc *p;
    718 {
    719 	struct vnode *vp;
    720 	struct flock lf;
    721 	int error;
    722 
    723 	if (fp == NULL)
    724 		return (0);
    725 	/*
    726 	 * POSIX record locking dictates that any close releases ALL
    727 	 * locks owned by this process.  This is handled by setting
    728 	 * a flag in the unlock to free ONLY locks obeying POSIX
    729 	 * semantics, and not to free BSD-style file locks.
    730 	 * If the descriptor was in a message, POSIX-style locks
    731 	 * aren't passed with the descriptor.
    732 	 */
    733 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
    734 		lf.l_whence = SEEK_SET;
    735 		lf.l_start = 0;
    736 		lf.l_len = 0;
    737 		lf.l_type = F_UNLCK;
    738 		vp = (struct vnode *)fp->f_data;
    739 		(void) VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX);
    740 	}
    741 	if (--fp->f_count > 0)
    742 		return (0);
    743 	if (fp->f_count < 0)
    744 		panic("closef: count < 0");
    745 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
    746 		lf.l_whence = SEEK_SET;
    747 		lf.l_start = 0;
    748 		lf.l_len = 0;
    749 		lf.l_type = F_UNLCK;
    750 		vp = (struct vnode *)fp->f_data;
    751 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
    752 	}
    753 	if (fp->f_ops)
    754 		error = (*fp->f_ops->fo_close)(fp, p);
    755 	else
    756 		error = 0;
    757 	ffree(fp);
    758 	return (error);
    759 }
    760 
    761 /*
    762  * Apply an advisory lock on a file descriptor.
    763  *
    764  * Just attempt to get a record lock of the requested type on
    765  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
    766  */
    767 /* ARGSUSED */
    768 int
    769 sys_flock(p, v, retval)
    770 	struct proc *p;
    771 	void *v;
    772 	register_t *retval;
    773 {
    774 	register struct sys_flock_args /* {
    775 		syscallarg(int) fd;
    776 		syscallarg(int) how;
    777 	} */ *uap = v;
    778 	int fd = SCARG(uap, fd);
    779 	int how = SCARG(uap, how);
    780 	register struct filedesc *fdp = p->p_fd;
    781 	register struct file *fp;
    782 	struct vnode *vp;
    783 	struct flock lf;
    784 
    785 	if ((u_int)fd >= fdp->fd_nfiles ||
    786 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    787 		return (EBADF);
    788 	if (fp->f_type != DTYPE_VNODE)
    789 		return (EOPNOTSUPP);
    790 	vp = (struct vnode *)fp->f_data;
    791 	lf.l_whence = SEEK_SET;
    792 	lf.l_start = 0;
    793 	lf.l_len = 0;
    794 	if (how & LOCK_UN) {
    795 		lf.l_type = F_UNLCK;
    796 		fp->f_flag &= ~FHASLOCK;
    797 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK));
    798 	}
    799 	if (how & LOCK_EX)
    800 		lf.l_type = F_WRLCK;
    801 	else if (how & LOCK_SH)
    802 		lf.l_type = F_RDLCK;
    803 	else
    804 		return (EBADF);
    805 	fp->f_flag |= FHASLOCK;
    806 	if (how & LOCK_NB)
    807 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK));
    808 	return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT));
    809 }
    810 
    811 /*
    812  * File Descriptor pseudo-device driver (/dev/fd/).
    813  *
    814  * Opening minor device N dup()s the file (if any) connected to file
    815  * descriptor N belonging to the calling process.  Note that this driver
    816  * consists of only the ``open()'' routine, because all subsequent
    817  * references to this file will be direct to the other driver.
    818  */
    819 /* ARGSUSED */
    820 int
    821 filedescopen(dev, mode, type, p)
    822 	dev_t dev;
    823 	int mode, type;
    824 	struct proc *p;
    825 {
    826 
    827 	/*
    828 	 * XXX Kludge: set curproc->p_dupfd to contain the value of the
    829 	 * the file descriptor being sought for duplication. The error
    830 	 * return ensures that the vnode for this device will be released
    831 	 * by vn_open. Open will detect this special error and take the
    832 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
    833 	 * will simply report the error.
    834 	 */
    835 	p->p_dupfd = minor(dev);
    836 	return (ENODEV);
    837 }
    838 
    839 /*
    840  * Duplicate the specified descriptor to a free descriptor.
    841  */
    842 int
    843 dupfdopen(fdp, indx, dfd, mode, error)
    844 	register struct filedesc *fdp;
    845 	register int indx, dfd;
    846 	int mode;
    847 	int error;
    848 {
    849 	register struct file *wfp;
    850 	struct file *fp;
    851 
    852 	/*
    853 	 * If the to-be-dup'd fd number is greater than the allowed number
    854 	 * of file descriptors, or the fd to be dup'd has already been
    855 	 * closed, reject.  Note, check for new == old is necessary as
    856 	 * falloc could allocate an already closed to-be-dup'd descriptor
    857 	 * as the new descriptor.
    858 	 */
    859 	fp = fdp->fd_ofiles[indx];
    860 	if ((u_int)dfd >= fdp->fd_nfiles ||
    861 	    (wfp = fdp->fd_ofiles[dfd]) == NULL || fp == wfp)
    862 		return (EBADF);
    863 
    864 	/*
    865 	 * There are two cases of interest here.
    866 	 *
    867 	 * For ENODEV simply dup (dfd) to file descriptor
    868 	 * (indx) and return.
    869 	 *
    870 	 * For ENXIO steal away the file structure from (dfd) and
    871 	 * store it in (indx).  (dfd) is effectively closed by
    872 	 * this operation.
    873 	 *
    874 	 * Any other error code is just returned.
    875 	 */
    876 	switch (error) {
    877 	case ENODEV:
    878 		/*
    879 		 * Check that the mode the file is being opened for is a
    880 		 * subset of the mode of the existing descriptor.
    881 		 */
    882 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag)
    883 			return (EACCES);
    884 		fdp->fd_ofiles[indx] = wfp;
    885 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    886 		wfp->f_count++;
    887 		fd_used(fdp, indx);
    888 		return (0);
    889 
    890 	case ENXIO:
    891 		/*
    892 		 * Steal away the file pointer from dfd, and stuff it into indx.
    893 		 */
    894 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
    895 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    896 		fdp->fd_ofiles[dfd] = NULL;
    897 		fdp->fd_ofileflags[dfd] = 0;
    898 		/*
    899 		 * Complete the clean up of the filedesc structure by
    900 		 * recomputing the various hints.
    901 		 */
    902 		fd_used(fdp, indx);
    903 		fd_unused(fdp, dfd);
    904 		return (0);
    905 
    906 	default:
    907 		return (error);
    908 	}
    909 	/* NOTREACHED */
    910 }
    911 
    912 /*
    913  * Close any files on exec?
    914  */
    915 void
    916 fdcloseexec(p)
    917 	struct proc *p;
    918 {
    919 	register struct filedesc *fdp = p->p_fd;
    920 	register int fd;
    921 
    922 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
    923 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
    924 			(void) fdrelease(p, fd);
    925 }
    926