Home | History | Annotate | Line # | Download | only in kern
kern_descrip.c revision 1.42
      1 /*	$NetBSD: kern_descrip.c,v 1.42 1996/03/30 22:24:38 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 #include <sys/conf.h>
     60 
     61 #include <sys/mount.h>
     62 #include <sys/syscallargs.h>
     63 
     64 #include <vm/vm.h>
     65 
     66 /*
     67  * Descriptor management.
     68  */
     69 struct filelist filehead;	/* head of list of open files */
     70 int nfiles;			/* actual number of open files */
     71 
     72 static __inline void fd_used __P((struct filedesc *, int));
     73 static __inline void fd_unused __P((struct filedesc *, int));
     74 int finishdup __P((struct filedesc *, int, int, register_t *));
     75 
     76 static __inline void
     77 fd_used(fdp, fd)
     78 	register struct filedesc *fdp;
     79 	register int fd;
     80 {
     81 
     82 	if (fd > fdp->fd_lastfile)
     83 		fdp->fd_lastfile = fd;
     84 }
     85 
     86 static __inline void
     87 fd_unused(fdp, fd)
     88 	register struct filedesc *fdp;
     89 	register int fd;
     90 {
     91 
     92 	if (fd < fdp->fd_freefile)
     93 		fdp->fd_freefile = fd;
     94 #ifdef DIAGNOSTIC
     95 	if (fd > fdp->fd_lastfile)
     96 		panic("fd_unused: fd_lastfile inconsistent");
     97 #endif
     98 	if (fd == fdp->fd_lastfile) {
     99 		do {
    100 			fd--;
    101 		} while (fd >= 0 && fdp->fd_ofiles[fd] == NULL);
    102 		fdp->fd_lastfile = fd;
    103 	}
    104 }
    105 
    106 /*
    107  * System calls on descriptors.
    108  */
    109 
    110 /*
    111  * Duplicate a file descriptor.
    112  */
    113 /* ARGSUSED */
    114 int
    115 sys_dup(p, v, retval)
    116 	struct proc *p;
    117 	void *v;
    118 	register_t *retval;
    119 {
    120 	struct sys_dup_args /* {
    121 		syscallarg(u_int) fd;
    122 	} */ *uap = v;
    123 	register struct filedesc *fdp = p->p_fd;
    124 	register int old = SCARG(uap, fd);
    125 	int new;
    126 	int error;
    127 
    128 	if ((u_int)old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL)
    129 		return (EBADF);
    130 	if ((error = fdalloc(p, 0, &new)) != 0)
    131 		return (error);
    132 	return (finishdup(fdp, old, new, retval));
    133 }
    134 
    135 /*
    136  * Duplicate a file descriptor to a particular value.
    137  */
    138 /* ARGSUSED */
    139 int
    140 sys_dup2(p, v, retval)
    141 	struct proc *p;
    142 	void *v;
    143 	register_t *retval;
    144 {
    145 	struct sys_dup2_args /* {
    146 		syscallarg(u_int) from;
    147 		syscallarg(u_int) to;
    148 	} */ *uap = v;
    149 	register struct filedesc *fdp = p->p_fd;
    150 	register int old = SCARG(uap, from), new = SCARG(uap, to);
    151 	int i, error;
    152 
    153 	if ((u_int)old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL ||
    154 	    (u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
    155 	    (u_int)new >= maxfiles)
    156 		return (EBADF);
    157 	if (old == new) {
    158 		*retval = new;
    159 		return (0);
    160 	}
    161 	if (new >= fdp->fd_nfiles) {
    162 		if ((error = fdalloc(p, new, &i)) != 0)
    163 			return (error);
    164 		if (new != i)
    165 			panic("dup2: fdalloc");
    166 	} else {
    167 		(void) fdrelease(p, new);
    168 	}
    169 	return (finishdup(fdp, old, new, retval));
    170 }
    171 
    172 /*
    173  * The file control system call.
    174  */
    175 /* ARGSUSED */
    176 int
    177 sys_fcntl(p, v, retval)
    178 	struct proc *p;
    179 	void *v;
    180 	register_t *retval;
    181 {
    182 	register struct sys_fcntl_args /* {
    183 		syscallarg(int) fd;
    184 		syscallarg(int) cmd;
    185 		syscallarg(void *) arg;
    186 	} */ *uap = v;
    187 	int fd = SCARG(uap, fd);
    188 	register struct filedesc *fdp = p->p_fd;
    189 	register struct file *fp;
    190 	struct vnode *vp;
    191 	int i, tmp, error, flg = F_POSIX;
    192 	struct flock fl;
    193 	int newmin;
    194 
    195 	if ((u_int)fd >= fdp->fd_nfiles ||
    196 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    197 		return (EBADF);
    198 	switch (SCARG(uap, cmd)) {
    199 
    200 	case F_DUPFD:
    201 		newmin = (long)SCARG(uap, arg);
    202 		if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
    203 		    (u_int)newmin >= maxfiles)
    204 			return (EINVAL);
    205 		if ((error = fdalloc(p, newmin, &i)) != 0)
    206 			return (error);
    207 		return (finishdup(fdp, fd, i, retval));
    208 
    209 	case F_GETFD:
    210 		*retval = fdp->fd_ofileflags[fd] & UF_EXCLOSE ? 1 : 0;
    211 		return (0);
    212 
    213 	case F_SETFD:
    214 		if ((long)SCARG(uap, arg) & 1)
    215 			fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
    216 		else
    217 			fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
    218 		return (0);
    219 
    220 	case F_GETFL:
    221 		*retval = OFLAGS(fp->f_flag);
    222 		return (0);
    223 
    224 	case F_SETFL:
    225 		fp->f_flag &= ~FCNTLFLAGS;
    226 		fp->f_flag |= FFLAGS((long)SCARG(uap, arg)) & FCNTLFLAGS;
    227 		tmp = fp->f_flag & FNONBLOCK;
    228 		error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    229 		if (error)
    230 			return (error);
    231 		tmp = fp->f_flag & FASYNC;
    232 		error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p);
    233 		if (!error)
    234 			return (0);
    235 		fp->f_flag &= ~FNONBLOCK;
    236 		tmp = 0;
    237 		(void) (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p);
    238 		return (error);
    239 
    240 	case F_GETOWN:
    241 		if (fp->f_type == DTYPE_SOCKET) {
    242 			*retval = ((struct socket *)fp->f_data)->so_pgid;
    243 			return (0);
    244 		}
    245 		error = (*fp->f_ops->fo_ioctl)
    246 			(fp, TIOCGPGRP, (caddr_t)retval, p);
    247 		*retval = -*retval;
    248 		return (error);
    249 
    250 	case F_SETOWN:
    251 		if (fp->f_type == DTYPE_SOCKET) {
    252 			((struct socket *)fp->f_data)->so_pgid =
    253 			    (long)SCARG(uap, arg);
    254 			return (0);
    255 		}
    256 		if ((long)SCARG(uap, arg) <= 0) {
    257 			SCARG(uap, arg) = (void *)(-(long)SCARG(uap, arg));
    258 		} else {
    259 			struct proc *p1 = pfind((long)SCARG(uap, arg));
    260 			if (p1 == 0)
    261 				return (ESRCH);
    262 			SCARG(uap, arg) = (void *)(long)p1->p_pgrp->pg_id;
    263 		}
    264 		return ((*fp->f_ops->fo_ioctl)
    265 			(fp, TIOCSPGRP, (caddr_t)&SCARG(uap, arg), p));
    266 
    267 	case F_SETLKW:
    268 		flg |= F_WAIT;
    269 		/* Fall into F_SETLK */
    270 
    271 	case F_SETLK:
    272 		if (fp->f_type != DTYPE_VNODE)
    273 			return (EBADF);
    274 		vp = (struct vnode *)fp->f_data;
    275 		/* Copy in the lock structure */
    276 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    277 		    sizeof (fl));
    278 		if (error)
    279 			return (error);
    280 		if (fl.l_whence == SEEK_CUR)
    281 			fl.l_start += fp->f_offset;
    282 		switch (fl.l_type) {
    283 
    284 		case F_RDLCK:
    285 			if ((fp->f_flag & FREAD) == 0)
    286 				return (EBADF);
    287 			p->p_flag |= P_ADVLOCK;
    288 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    289 
    290 		case F_WRLCK:
    291 			if ((fp->f_flag & FWRITE) == 0)
    292 				return (EBADF);
    293 			p->p_flag |= P_ADVLOCK;
    294 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    295 
    296 		case F_UNLCK:
    297 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &fl,
    298 				F_POSIX));
    299 
    300 		default:
    301 			return (EINVAL);
    302 		}
    303 
    304 	case F_GETLK:
    305 		if (fp->f_type != DTYPE_VNODE)
    306 			return (EBADF);
    307 		vp = (struct vnode *)fp->f_data;
    308 		/* Copy in the lock structure */
    309 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    310 		    sizeof (fl));
    311 		if (error)
    312 			return (error);
    313 		if (fl.l_whence == SEEK_CUR)
    314 			fl.l_start += fp->f_offset;
    315 		error = VOP_ADVLOCK(vp, (caddr_t)p, F_GETLK, &fl, F_POSIX);
    316 		if (error)
    317 			return (error);
    318 		return (copyout((caddr_t)&fl, (caddr_t)SCARG(uap, arg),
    319 		    sizeof (fl)));
    320 
    321 	default:
    322 		return (EINVAL);
    323 	}
    324 	/* NOTREACHED */
    325 }
    326 
    327 /*
    328  * Common code for dup, dup2, and fcntl(F_DUPFD).
    329  */
    330 int
    331 finishdup(fdp, old, new, retval)
    332 	register struct filedesc *fdp;
    333 	register int old, new;
    334 	register_t *retval;
    335 {
    336 	register struct file *fp;
    337 
    338 	fp = fdp->fd_ofiles[old];
    339 	fdp->fd_ofiles[new] = fp;
    340 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
    341 	fp->f_count++;
    342 	fd_used(fdp, new);
    343 	*retval = new;
    344 	return (0);
    345 }
    346 
    347 int
    348 fdrelease(p, fd)
    349 	struct proc *p;
    350 	int fd;
    351 {
    352 	register struct filedesc *fdp = p->p_fd;
    353 	register struct file **fpp, *fp;
    354 	register char *pf;
    355 
    356 	fpp = &fdp->fd_ofiles[fd];
    357 	fp = *fpp;
    358 	if (fp == NULL)
    359 		return (EBADF);
    360 	pf = &fdp->fd_ofileflags[fd];
    361 	if (*pf & UF_MAPPED)
    362 		(void) munmapfd(p, fd);
    363 	*fpp = NULL;
    364 	*pf = 0;
    365 	fd_unused(fdp, fd);
    366 	return (closef(fp, p));
    367 }
    368 
    369 /*
    370  * Close a file descriptor.
    371  */
    372 /* ARGSUSED */
    373 int
    374 sys_close(p, v, retval)
    375 	struct proc *p;
    376 	void *v;
    377 	register_t *retval;
    378 {
    379 	struct sys_close_args /* {
    380 		syscallarg(int) fd;
    381 	} */ *uap = v;
    382 	int fd = SCARG(uap, fd);
    383 	register struct filedesc *fdp = p->p_fd;
    384 
    385 	if ((u_int)fd >= fdp->fd_nfiles)
    386 		return (EBADF);
    387 	return (fdrelease(p, fd));
    388 }
    389 
    390 /*
    391  * Return status information about a file descriptor.
    392  */
    393 /* ARGSUSED */
    394 int
    395 sys_fstat(p, v, retval)
    396 	struct proc *p;
    397 	void *v;
    398 	register_t *retval;
    399 {
    400 	register struct sys_fstat_args /* {
    401 		syscallarg(int) fd;
    402 		syscallarg(struct stat *) sb;
    403 	} */ *uap = v;
    404 	int fd = SCARG(uap, fd);
    405 	register struct filedesc *fdp = p->p_fd;
    406 	register struct file *fp;
    407 	struct stat ub;
    408 	int error;
    409 
    410 	if ((u_int)fd >= fdp->fd_nfiles ||
    411 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    412 		return (EBADF);
    413 	switch (fp->f_type) {
    414 
    415 	case DTYPE_VNODE:
    416 		error = vn_stat((struct vnode *)fp->f_data, &ub, p);
    417 		break;
    418 
    419 	case DTYPE_SOCKET:
    420 		error = soo_stat((struct socket *)fp->f_data, &ub);
    421 		break;
    422 
    423 	default:
    424 		panic("fstat");
    425 		/*NOTREACHED*/
    426 	}
    427 	if (error == 0)
    428 		error = copyout((caddr_t)&ub, (caddr_t)SCARG(uap, sb),
    429 		    sizeof (ub));
    430 	return (error);
    431 }
    432 
    433 /*
    434  * Return pathconf information about a file descriptor.
    435  */
    436 /* ARGSUSED */
    437 int
    438 sys_fpathconf(p, v, retval)
    439 	struct proc *p;
    440 	void *v;
    441 	register_t *retval;
    442 {
    443 	register struct sys_fpathconf_args /* {
    444 		syscallarg(int) fd;
    445 		syscallarg(int) name;
    446 	} */ *uap = v;
    447 	int fd = SCARG(uap, fd);
    448 	struct filedesc *fdp = p->p_fd;
    449 	struct file *fp;
    450 	struct vnode *vp;
    451 
    452 	if ((u_int)fd >= fdp->fd_nfiles ||
    453 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    454 		return (EBADF);
    455 	switch (fp->f_type) {
    456 
    457 	case DTYPE_SOCKET:
    458 		if (SCARG(uap, name) != _PC_PIPE_BUF)
    459 			return (EINVAL);
    460 		*retval = PIPE_BUF;
    461 		return (0);
    462 
    463 	case DTYPE_VNODE:
    464 		vp = (struct vnode *)fp->f_data;
    465 		return (VOP_PATHCONF(vp, SCARG(uap, name), retval));
    466 
    467 	default:
    468 		panic("fpathconf");
    469 	}
    470 	/*NOTREACHED*/
    471 }
    472 
    473 /*
    474  * Allocate a file descriptor for the process.
    475  */
    476 int fdexpand;
    477 
    478 int
    479 fdalloc(p, want, result)
    480 	struct proc *p;
    481 	int want;
    482 	int *result;
    483 {
    484 	register struct filedesc *fdp = p->p_fd;
    485 	register int i;
    486 	int lim, last, nfiles;
    487 	struct file **newofile;
    488 	char *newofileflags;
    489 
    490 	/*
    491 	 * Search for a free descriptor starting at the higher
    492 	 * of want or fd_freefile.  If that fails, consider
    493 	 * expanding the ofile array.
    494 	 */
    495 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
    496 	for (;;) {
    497 		last = min(fdp->fd_nfiles, lim);
    498 		if ((i = want) < fdp->fd_freefile)
    499 			i = fdp->fd_freefile;
    500 		for (; i < last; i++) {
    501 			if (fdp->fd_ofiles[i] == NULL) {
    502 				fd_used(fdp, i);
    503 				if (want <= fdp->fd_freefile)
    504 					fdp->fd_freefile = i;
    505 				*result = i;
    506 				return (0);
    507 			}
    508 		}
    509 
    510 		/*
    511 		 * No space in current array.  Expand?
    512 		 */
    513 		if (fdp->fd_nfiles >= lim)
    514 			return (EMFILE);
    515 		if (fdp->fd_nfiles < NDEXTENT)
    516 			nfiles = NDEXTENT;
    517 		else
    518 			nfiles = 2 * fdp->fd_nfiles;
    519 		MALLOC(newofile, struct file **, nfiles * OFILESIZE,
    520 		    M_FILEDESC, M_WAITOK);
    521 		newofileflags = (char *) &newofile[nfiles];
    522 		/*
    523 		 * Copy the existing ofile and ofileflags arrays
    524 		 * and zero the new portion of each array.
    525 		 */
    526 		bcopy(fdp->fd_ofiles, newofile,
    527 			(i = sizeof(struct file *) * fdp->fd_nfiles));
    528 		bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i);
    529 		bcopy(fdp->fd_ofileflags, newofileflags,
    530 			(i = sizeof(char) * fdp->fd_nfiles));
    531 		bzero(newofileflags + i, nfiles * sizeof(char) - i);
    532 		if (fdp->fd_nfiles > NDFILE)
    533 			FREE(fdp->fd_ofiles, M_FILEDESC);
    534 		fdp->fd_ofiles = newofile;
    535 		fdp->fd_ofileflags = newofileflags;
    536 		fdp->fd_nfiles = nfiles;
    537 		fdexpand++;
    538 	}
    539 }
    540 
    541 /*
    542  * Check to see whether n user file descriptors
    543  * are available to the process p.
    544  */
    545 int
    546 fdavail(p, n)
    547 	struct proc *p;
    548 	register int n;
    549 {
    550 	register struct filedesc *fdp = p->p_fd;
    551 	register struct file **fpp;
    552 	register int i, lim;
    553 
    554 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
    555 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
    556 		return (1);
    557 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
    558 	for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++)
    559 		if (*fpp == NULL && --n <= 0)
    560 			return (1);
    561 	return (0);
    562 }
    563 
    564 /*
    565  * Create a new open file structure and allocate
    566  * a file decriptor for the process that refers to it.
    567  */
    568 int
    569 falloc(p, resultfp, resultfd)
    570 	register struct proc *p;
    571 	struct file **resultfp;
    572 	int *resultfd;
    573 {
    574 	register struct file *fp, *fq;
    575 	int error, i;
    576 
    577 	if ((error = fdalloc(p, 0, &i)) != 0)
    578 		return (error);
    579 	if (nfiles >= maxfiles) {
    580 		tablefull("file");
    581 		return (ENFILE);
    582 	}
    583 	/*
    584 	 * Allocate a new file descriptor.
    585 	 * If the process has file descriptor zero open, add to the list
    586 	 * of open files at that point, otherwise put it at the front of
    587 	 * the list of open files.
    588 	 */
    589 	nfiles++;
    590 	MALLOC(fp, struct file *, sizeof(struct file), M_FILE, M_WAITOK);
    591 	bzero(fp, sizeof(struct file));
    592 	if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
    593 		LIST_INSERT_AFTER(fq, fp, f_list);
    594 	} else {
    595 		LIST_INSERT_HEAD(&filehead, fp, f_list);
    596 	}
    597 	p->p_fd->fd_ofiles[i] = fp;
    598 	fp->f_count = 1;
    599 	fp->f_cred = p->p_ucred;
    600 	crhold(fp->f_cred);
    601 	if (resultfp)
    602 		*resultfp = fp;
    603 	if (resultfd)
    604 		*resultfd = i;
    605 	return (0);
    606 }
    607 
    608 /*
    609  * Free a file descriptor.
    610  */
    611 void
    612 ffree(fp)
    613 	register struct file *fp;
    614 {
    615 	LIST_REMOVE(fp, f_list);
    616 	crfree(fp->f_cred);
    617 #ifdef DIAGNOSTIC
    618 	fp->f_count = 0;
    619 #endif
    620 	nfiles--;
    621 	FREE(fp, M_FILE);
    622 }
    623 
    624 /*
    625  * Copy a filedesc structure.
    626  */
    627 struct filedesc *
    628 fdcopy(p)
    629 	struct proc *p;
    630 {
    631 	register struct filedesc *newfdp, *fdp = p->p_fd;
    632 	register struct file **fpp;
    633 	register int i;
    634 
    635 	MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0),
    636 	    M_FILEDESC, M_WAITOK);
    637 	bcopy(fdp, newfdp, sizeof(struct filedesc));
    638 	VREF(newfdp->fd_cdir);
    639 	if (newfdp->fd_rdir)
    640 		VREF(newfdp->fd_rdir);
    641 	newfdp->fd_refcnt = 1;
    642 
    643 	/*
    644 	 * If the number of open files fits in the internal arrays
    645 	 * of the open file structure, use them, otherwise allocate
    646 	 * additional memory for the number of descriptors currently
    647 	 * in use.
    648 	 */
    649 	if (newfdp->fd_lastfile < NDFILE) {
    650 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
    651 		newfdp->fd_ofileflags =
    652 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
    653 		i = NDFILE;
    654 	} else {
    655 		/*
    656 		 * Compute the smallest multiple of NDEXTENT needed
    657 		 * for the file descriptors currently in use,
    658 		 * allowing the table to shrink.
    659 		 */
    660 		i = newfdp->fd_nfiles;
    661 		while (i >= 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
    662 			i /= 2;
    663 		MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE,
    664 		    M_FILEDESC, M_WAITOK);
    665 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
    666 	}
    667 	newfdp->fd_nfiles = i;
    668 	bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
    669 	bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
    670 	fpp = newfdp->fd_ofiles;
    671 	for (i = newfdp->fd_lastfile; i >= 0; i--, fpp++)
    672 		if (*fpp != NULL)
    673 			(*fpp)->f_count++;
    674 	return (newfdp);
    675 }
    676 
    677 /*
    678  * Release a filedesc structure.
    679  */
    680 void
    681 fdfree(p)
    682 	struct proc *p;
    683 {
    684 	register struct filedesc *fdp = p->p_fd;
    685 	register struct file **fpp, *fp;
    686 	register int i;
    687 
    688 	if (--fdp->fd_refcnt > 0)
    689 		return;
    690 	fpp = fdp->fd_ofiles;
    691 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
    692 		fp = *fpp;
    693 		if (fp != NULL) {
    694 			*fpp = NULL;
    695 			(void) closef(fp, p);
    696 		}
    697 	}
    698 	p->p_fd = NULL;
    699 	if (fdp->fd_nfiles > NDFILE)
    700 		FREE(fdp->fd_ofiles, M_FILEDESC);
    701 	vrele(fdp->fd_cdir);
    702 	if (fdp->fd_rdir)
    703 		vrele(fdp->fd_rdir);
    704 	FREE(fdp, M_FILEDESC);
    705 }
    706 
    707 /*
    708  * Internal form of close.
    709  * Decrement reference count on file structure.
    710  * Note: p may be NULL when closing a file
    711  * that was being passed in a message.
    712  */
    713 int
    714 closef(fp, p)
    715 	register struct file *fp;
    716 	register struct proc *p;
    717 {
    718 	struct vnode *vp;
    719 	struct flock lf;
    720 	int error;
    721 
    722 	if (fp == NULL)
    723 		return (0);
    724 	/*
    725 	 * POSIX record locking dictates that any close releases ALL
    726 	 * locks owned by this process.  This is handled by setting
    727 	 * a flag in the unlock to free ONLY locks obeying POSIX
    728 	 * semantics, and not to free BSD-style file locks.
    729 	 * If the descriptor was in a message, POSIX-style locks
    730 	 * aren't passed with the descriptor.
    731 	 */
    732 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
    733 		lf.l_whence = SEEK_SET;
    734 		lf.l_start = 0;
    735 		lf.l_len = 0;
    736 		lf.l_type = F_UNLCK;
    737 		vp = (struct vnode *)fp->f_data;
    738 		(void) VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX);
    739 	}
    740 	if (--fp->f_count > 0)
    741 		return (0);
    742 	if (fp->f_count < 0)
    743 		panic("closef: count < 0");
    744 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
    745 		lf.l_whence = SEEK_SET;
    746 		lf.l_start = 0;
    747 		lf.l_len = 0;
    748 		lf.l_type = F_UNLCK;
    749 		vp = (struct vnode *)fp->f_data;
    750 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
    751 	}
    752 	if (fp->f_ops)
    753 		error = (*fp->f_ops->fo_close)(fp, p);
    754 	else
    755 		error = 0;
    756 	ffree(fp);
    757 	return (error);
    758 }
    759 
    760 /*
    761  * Apply an advisory lock on a file descriptor.
    762  *
    763  * Just attempt to get a record lock of the requested type on
    764  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
    765  */
    766 /* ARGSUSED */
    767 int
    768 sys_flock(p, v, retval)
    769 	struct proc *p;
    770 	void *v;
    771 	register_t *retval;
    772 {
    773 	register struct sys_flock_args /* {
    774 		syscallarg(int) fd;
    775 		syscallarg(int) how;
    776 	} */ *uap = v;
    777 	int fd = SCARG(uap, fd);
    778 	int how = SCARG(uap, how);
    779 	register struct filedesc *fdp = p->p_fd;
    780 	register struct file *fp;
    781 	struct vnode *vp;
    782 	struct flock lf;
    783 
    784 	if ((u_int)fd >= fdp->fd_nfiles ||
    785 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    786 		return (EBADF);
    787 	if (fp->f_type != DTYPE_VNODE)
    788 		return (EOPNOTSUPP);
    789 	vp = (struct vnode *)fp->f_data;
    790 	lf.l_whence = SEEK_SET;
    791 	lf.l_start = 0;
    792 	lf.l_len = 0;
    793 	if (how & LOCK_UN) {
    794 		lf.l_type = F_UNLCK;
    795 		fp->f_flag &= ~FHASLOCK;
    796 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK));
    797 	}
    798 	if (how & LOCK_EX)
    799 		lf.l_type = F_WRLCK;
    800 	else if (how & LOCK_SH)
    801 		lf.l_type = F_RDLCK;
    802 	else
    803 		return (EBADF);
    804 	fp->f_flag |= FHASLOCK;
    805 	if (how & LOCK_NB)
    806 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK));
    807 	return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT));
    808 }
    809 
    810 /*
    811  * File Descriptor pseudo-device driver (/dev/fd/).
    812  *
    813  * Opening minor device N dup()s the file (if any) connected to file
    814  * descriptor N belonging to the calling process.  Note that this driver
    815  * consists of only the ``open()'' routine, because all subsequent
    816  * references to this file will be direct to the other driver.
    817  */
    818 /* ARGSUSED */
    819 int
    820 filedescopen(dev, mode, type, p)
    821 	dev_t dev;
    822 	int mode, type;
    823 	struct proc *p;
    824 {
    825 
    826 	/*
    827 	 * XXX Kludge: set curproc->p_dupfd to contain the value of the
    828 	 * the file descriptor being sought for duplication. The error
    829 	 * return ensures that the vnode for this device will be released
    830 	 * by vn_open. Open will detect this special error and take the
    831 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
    832 	 * will simply report the error.
    833 	 */
    834 	p->p_dupfd = minor(dev);
    835 	return (ENODEV);
    836 }
    837 
    838 /*
    839  * Duplicate the specified descriptor to a free descriptor.
    840  */
    841 int
    842 dupfdopen(fdp, indx, dfd, mode, error)
    843 	register struct filedesc *fdp;
    844 	register int indx, dfd;
    845 	int mode;
    846 	int error;
    847 {
    848 	register struct file *wfp;
    849 	struct file *fp;
    850 
    851 	/*
    852 	 * If the to-be-dup'd fd number is greater than the allowed number
    853 	 * of file descriptors, or the fd to be dup'd has already been
    854 	 * closed, reject.  Note, check for new == old is necessary as
    855 	 * falloc could allocate an already closed to-be-dup'd descriptor
    856 	 * as the new descriptor.
    857 	 */
    858 	fp = fdp->fd_ofiles[indx];
    859 	if ((u_int)dfd >= fdp->fd_nfiles ||
    860 	    (wfp = fdp->fd_ofiles[dfd]) == NULL || fp == wfp)
    861 		return (EBADF);
    862 
    863 	/*
    864 	 * There are two cases of interest here.
    865 	 *
    866 	 * For ENODEV simply dup (dfd) to file descriptor
    867 	 * (indx) and return.
    868 	 *
    869 	 * For ENXIO steal away the file structure from (dfd) and
    870 	 * store it in (indx).  (dfd) is effectively closed by
    871 	 * this operation.
    872 	 *
    873 	 * Any other error code is just returned.
    874 	 */
    875 	switch (error) {
    876 	case ENODEV:
    877 		/*
    878 		 * Check that the mode the file is being opened for is a
    879 		 * subset of the mode of the existing descriptor.
    880 		 */
    881 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag)
    882 			return (EACCES);
    883 		fdp->fd_ofiles[indx] = wfp;
    884 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    885 		wfp->f_count++;
    886 		fd_used(fdp, indx);
    887 		return (0);
    888 
    889 	case ENXIO:
    890 		/*
    891 		 * Steal away the file pointer from dfd, and stuff it into indx.
    892 		 */
    893 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
    894 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    895 		fdp->fd_ofiles[dfd] = NULL;
    896 		fdp->fd_ofileflags[dfd] = 0;
    897 		/*
    898 		 * Complete the clean up of the filedesc structure by
    899 		 * recomputing the various hints.
    900 		 */
    901 		fd_used(fdp, indx);
    902 		fd_unused(fdp, dfd);
    903 		return (0);
    904 
    905 	default:
    906 		return (error);
    907 	}
    908 	/* NOTREACHED */
    909 }
    910 
    911 /*
    912  * Close any files on exec?
    913  */
    914 void
    915 fdcloseexec(p)
    916 	struct proc *p;
    917 {
    918 	register struct filedesc *fdp = p->p_fd;
    919 	register int fd;
    920 
    921 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
    922 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
    923 			(void) fdrelease(p, fd);
    924 }
    925