Home | History | Annotate | Line # | Download | only in kern
kern_descrip.c revision 1.48
      1 /*	$NetBSD: kern_descrip.c,v 1.48 1998/01/05 04:51:16 thorpej 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(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(int) from;
    147 		syscallarg(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 		case F_RDLCK:
    284 			if ((fp->f_flag & FREAD) == 0)
    285 				return (EBADF);
    286 			p->p_flag |= P_ADVLOCK;
    287 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    288 
    289 		case F_WRLCK:
    290 			if ((fp->f_flag & FWRITE) == 0)
    291 				return (EBADF);
    292 			p->p_flag |= P_ADVLOCK;
    293 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg));
    294 
    295 		case F_UNLCK:
    296 			return (VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &fl,
    297 				F_POSIX));
    298 
    299 		default:
    300 			return (EINVAL);
    301 		}
    302 
    303 	case F_GETLK:
    304 		if (fp->f_type != DTYPE_VNODE)
    305 			return (EBADF);
    306 		vp = (struct vnode *)fp->f_data;
    307 		/* Copy in the lock structure */
    308 		error = copyin((caddr_t)SCARG(uap, arg), (caddr_t)&fl,
    309 		    sizeof (fl));
    310 		if (error)
    311 			return (error);
    312 		if (fl.l_whence == SEEK_CUR)
    313 			fl.l_start += fp->f_offset;
    314 		if (fl.l_type != F_RDLCK &&
    315 		    fl.l_type != F_WRLCK &&
    316 		    fl.l_type != F_UNLCK)
    317 			return (EINVAL);
    318 		error = VOP_ADVLOCK(vp, (caddr_t)p, F_GETLK, &fl, F_POSIX);
    319 		if (error)
    320 			return (error);
    321 		return (copyout((caddr_t)&fl, (caddr_t)SCARG(uap, arg),
    322 		    sizeof (fl)));
    323 
    324 	default:
    325 		return (EINVAL);
    326 	}
    327 	/* NOTREACHED */
    328 }
    329 
    330 /*
    331  * Common code for dup, dup2, and fcntl(F_DUPFD).
    332  */
    333 int
    334 finishdup(fdp, old, new, retval)
    335 	register struct filedesc *fdp;
    336 	register int old, new;
    337 	register_t *retval;
    338 {
    339 	register struct file *fp;
    340 
    341 	fp = fdp->fd_ofiles[old];
    342 	fdp->fd_ofiles[new] = fp;
    343 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
    344 	fp->f_count++;
    345 	fd_used(fdp, new);
    346 	*retval = new;
    347 	return (0);
    348 }
    349 
    350 int
    351 fdrelease(p, fd)
    352 	struct proc *p;
    353 	int fd;
    354 {
    355 	register struct filedesc *fdp = p->p_fd;
    356 	register struct file **fpp, *fp;
    357 	register char *pf;
    358 
    359 	fpp = &fdp->fd_ofiles[fd];
    360 	fp = *fpp;
    361 	if (fp == NULL)
    362 		return (EBADF);
    363 	pf = &fdp->fd_ofileflags[fd];
    364 	if (*pf & UF_MAPPED)
    365 		(void) munmapfd(p, fd);
    366 	*fpp = NULL;
    367 	*pf = 0;
    368 	fd_unused(fdp, fd);
    369 	return (closef(fp, p));
    370 }
    371 
    372 /*
    373  * Close a file descriptor.
    374  */
    375 /* ARGSUSED */
    376 int
    377 sys_close(p, v, retval)
    378 	struct proc *p;
    379 	void *v;
    380 	register_t *retval;
    381 {
    382 	struct sys_close_args /* {
    383 		syscallarg(int) fd;
    384 	} */ *uap = v;
    385 	int fd = SCARG(uap, fd);
    386 	register struct filedesc *fdp = p->p_fd;
    387 
    388 	if ((u_int)fd >= fdp->fd_nfiles)
    389 		return (EBADF);
    390 	return (fdrelease(p, fd));
    391 }
    392 
    393 /*
    394  * Return status information about a file descriptor.
    395  */
    396 /* ARGSUSED */
    397 int
    398 sys___fstat13(p, v, retval)
    399 	struct proc *p;
    400 	void *v;
    401 	register_t *retval;
    402 {
    403 	register struct sys___fstat13_args /* {
    404 		syscallarg(int) fd;
    405 		syscallarg(struct stat *) sb;
    406 	} */ *uap = v;
    407 	int fd = SCARG(uap, fd);
    408 	register struct filedesc *fdp = p->p_fd;
    409 	register struct file *fp;
    410 	struct stat ub;
    411 	int error;
    412 
    413 	if ((u_int)fd >= fdp->fd_nfiles ||
    414 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    415 		return (EBADF);
    416 	switch (fp->f_type) {
    417 
    418 	case DTYPE_VNODE:
    419 		error = vn_stat((struct vnode *)fp->f_data, &ub, p);
    420 		break;
    421 
    422 	case DTYPE_SOCKET:
    423 		error = soo_stat((struct socket *)fp->f_data, &ub);
    424 		break;
    425 
    426 	default:
    427 		panic("fstat");
    428 		/*NOTREACHED*/
    429 	}
    430 	if (error == 0)
    431 		error = copyout(&ub, SCARG(uap, sb), 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  * Create an initial filedesc structure, using the same current and root
    628  * directories as p.
    629  */
    630 struct filedesc *
    631 fdinit(p)
    632 	struct proc *p;
    633 {
    634 	struct filedesc0 *newfdp;
    635 	struct filedesc *fdp = p->p_fd;
    636 
    637 	MALLOC(newfdp, struct filedesc0 *, sizeof(struct filedesc0),
    638 	    M_FILEDESC, M_WAITOK);
    639 	bzero(newfdp, sizeof(struct filedesc0));
    640 	newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
    641 	VREF(newfdp->fd_fd.fd_cdir);
    642 	newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
    643 	if (newfdp->fd_fd.fd_rdir)
    644 		VREF(newfdp->fd_fd.fd_rdir);
    645 
    646 	fdinit1(newfdp);
    647 
    648 	return (&newfdp->fd_fd);
    649 }
    650 
    651 /*
    652  * Initialize a file descriptor table.
    653  */
    654 void
    655 fdinit1(newfdp)
    656 	struct filedesc0 *newfdp;
    657 {
    658 	extern int cmask;		/* init_main.c */
    659 
    660 	newfdp->fd_fd.fd_refcnt = 1;
    661 	newfdp->fd_fd.fd_cmask = cmask;
    662 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
    663 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
    664 	newfdp->fd_fd.fd_nfiles = NDFILE;
    665 }
    666 
    667 /*
    668  * Make p2 share p1's filedesc structure.
    669  */
    670 void
    671 fdshare(p1, p2)
    672 	struct proc *p1, *p2;
    673 {
    674 
    675 	p2->p_fd = p1->p_fd;
    676 	p1->p_fd->fd_refcnt++;
    677 }
    678 
    679 /*
    680  * Make this process not share its filedesc structure, maintaining
    681  * all file descriptor state.
    682  */
    683 void
    684 fdunshare(p)
    685 	struct proc *p;
    686 {
    687 	struct filedesc *newfd;
    688 
    689 	if (p->p_fd->fd_refcnt == 1)
    690 		return;
    691 
    692 	newfd = fdcopy(p);
    693 	fdfree(p);
    694 	p->p_fd = newfd;
    695 }
    696 
    697 /*
    698  * Clear a process's fd table.
    699  */
    700 void
    701 fdclear(p)
    702 	struct proc *p;
    703 {
    704 	struct filedesc *newfd;
    705 
    706 	newfd = fdinit(p);
    707 	fdfree(p);
    708 	p->p_fd = newfd;
    709 }
    710 
    711 /*
    712  * Copy a filedesc structure.
    713  */
    714 struct filedesc *
    715 fdcopy(p)
    716 	struct proc *p;
    717 {
    718 	register struct filedesc *newfdp, *fdp = p->p_fd;
    719 	register struct file **fpp;
    720 	register int i;
    721 
    722 	MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0),
    723 	    M_FILEDESC, M_WAITOK);
    724 	bcopy(fdp, newfdp, sizeof(struct filedesc));
    725 	VREF(newfdp->fd_cdir);
    726 	if (newfdp->fd_rdir)
    727 		VREF(newfdp->fd_rdir);
    728 	newfdp->fd_refcnt = 1;
    729 
    730 	/*
    731 	 * If the number of open files fits in the internal arrays
    732 	 * of the open file structure, use them, otherwise allocate
    733 	 * additional memory for the number of descriptors currently
    734 	 * in use.
    735 	 */
    736 	if (newfdp->fd_lastfile < NDFILE) {
    737 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
    738 		newfdp->fd_ofileflags =
    739 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
    740 		i = NDFILE;
    741 	} else {
    742 		/*
    743 		 * Compute the smallest multiple of NDEXTENT needed
    744 		 * for the file descriptors currently in use,
    745 		 * allowing the table to shrink.
    746 		 */
    747 		i = newfdp->fd_nfiles;
    748 		while (i >= 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
    749 			i /= 2;
    750 		MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE,
    751 		    M_FILEDESC, M_WAITOK);
    752 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
    753 	}
    754 	newfdp->fd_nfiles = i;
    755 	bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
    756 	bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
    757 	fpp = newfdp->fd_ofiles;
    758 	for (i = newfdp->fd_lastfile; i >= 0; i--, fpp++)
    759 		if (*fpp != NULL)
    760 			(*fpp)->f_count++;
    761 	return (newfdp);
    762 }
    763 
    764 /*
    765  * Release a filedesc structure.
    766  */
    767 void
    768 fdfree(p)
    769 	struct proc *p;
    770 {
    771 	register struct filedesc *fdp = p->p_fd;
    772 	register struct file **fpp, *fp;
    773 	register int i;
    774 
    775 	if (--fdp->fd_refcnt > 0)
    776 		return;
    777 	fpp = fdp->fd_ofiles;
    778 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
    779 		fp = *fpp;
    780 		if (fp != NULL) {
    781 			*fpp = NULL;
    782 			(void) closef(fp, p);
    783 		}
    784 	}
    785 	p->p_fd = NULL;
    786 	if (fdp->fd_nfiles > NDFILE)
    787 		FREE(fdp->fd_ofiles, M_FILEDESC);
    788 	vrele(fdp->fd_cdir);
    789 	if (fdp->fd_rdir)
    790 		vrele(fdp->fd_rdir);
    791 	FREE(fdp, M_FILEDESC);
    792 }
    793 
    794 /*
    795  * Internal form of close.
    796  * Decrement reference count on file structure.
    797  * Note: p may be NULL when closing a file
    798  * that was being passed in a message.
    799  */
    800 int
    801 closef(fp, p)
    802 	register struct file *fp;
    803 	register struct proc *p;
    804 {
    805 	struct vnode *vp;
    806 	struct flock lf;
    807 	int error;
    808 
    809 	if (fp == NULL)
    810 		return (0);
    811 	/*
    812 	 * POSIX record locking dictates that any close releases ALL
    813 	 * locks owned by this process.  This is handled by setting
    814 	 * a flag in the unlock to free ONLY locks obeying POSIX
    815 	 * semantics, and not to free BSD-style file locks.
    816 	 * If the descriptor was in a message, POSIX-style locks
    817 	 * aren't passed with the descriptor.
    818 	 */
    819 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
    820 		lf.l_whence = SEEK_SET;
    821 		lf.l_start = 0;
    822 		lf.l_len = 0;
    823 		lf.l_type = F_UNLCK;
    824 		vp = (struct vnode *)fp->f_data;
    825 		(void) VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX);
    826 	}
    827 	if (--fp->f_count > 0)
    828 		return (0);
    829 	if (fp->f_count < 0)
    830 		panic("closef: count < 0");
    831 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
    832 		lf.l_whence = SEEK_SET;
    833 		lf.l_start = 0;
    834 		lf.l_len = 0;
    835 		lf.l_type = F_UNLCK;
    836 		vp = (struct vnode *)fp->f_data;
    837 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
    838 	}
    839 	if (fp->f_ops)
    840 		error = (*fp->f_ops->fo_close)(fp, p);
    841 	else
    842 		error = 0;
    843 	ffree(fp);
    844 	return (error);
    845 }
    846 
    847 /*
    848  * Apply an advisory lock on a file descriptor.
    849  *
    850  * Just attempt to get a record lock of the requested type on
    851  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
    852  */
    853 /* ARGSUSED */
    854 int
    855 sys_flock(p, v, retval)
    856 	struct proc *p;
    857 	void *v;
    858 	register_t *retval;
    859 {
    860 	register struct sys_flock_args /* {
    861 		syscallarg(int) fd;
    862 		syscallarg(int) how;
    863 	} */ *uap = v;
    864 	int fd = SCARG(uap, fd);
    865 	int how = SCARG(uap, how);
    866 	register struct filedesc *fdp = p->p_fd;
    867 	register struct file *fp;
    868 	struct vnode *vp;
    869 	struct flock lf;
    870 
    871 	if ((u_int)fd >= fdp->fd_nfiles ||
    872 	    (fp = fdp->fd_ofiles[fd]) == NULL)
    873 		return (EBADF);
    874 	if (fp->f_type != DTYPE_VNODE)
    875 		return (EOPNOTSUPP);
    876 	vp = (struct vnode *)fp->f_data;
    877 	lf.l_whence = SEEK_SET;
    878 	lf.l_start = 0;
    879 	lf.l_len = 0;
    880 	if (how & LOCK_UN) {
    881 		lf.l_type = F_UNLCK;
    882 		fp->f_flag &= ~FHASLOCK;
    883 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK));
    884 	}
    885 	if (how & LOCK_EX)
    886 		lf.l_type = F_WRLCK;
    887 	else if (how & LOCK_SH)
    888 		lf.l_type = F_RDLCK;
    889 	else
    890 		return (EINVAL);
    891 	fp->f_flag |= FHASLOCK;
    892 	if (how & LOCK_NB)
    893 		return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK));
    894 	return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT));
    895 }
    896 
    897 /*
    898  * File Descriptor pseudo-device driver (/dev/fd/).
    899  *
    900  * Opening minor device N dup()s the file (if any) connected to file
    901  * descriptor N belonging to the calling process.  Note that this driver
    902  * consists of only the ``open()'' routine, because all subsequent
    903  * references to this file will be direct to the other driver.
    904  */
    905 /* ARGSUSED */
    906 int
    907 filedescopen(dev, mode, type, p)
    908 	dev_t dev;
    909 	int mode, type;
    910 	struct proc *p;
    911 {
    912 
    913 	/*
    914 	 * XXX Kludge: set curproc->p_dupfd to contain the value of the
    915 	 * the file descriptor being sought for duplication. The error
    916 	 * return ensures that the vnode for this device will be released
    917 	 * by vn_open. Open will detect this special error and take the
    918 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
    919 	 * will simply report the error.
    920 	 */
    921 	p->p_dupfd = minor(dev);
    922 	return (ENODEV);
    923 }
    924 
    925 /*
    926  * Duplicate the specified descriptor to a free descriptor.
    927  */
    928 int
    929 dupfdopen(fdp, indx, dfd, mode, error)
    930 	register struct filedesc *fdp;
    931 	register int indx, dfd;
    932 	int mode;
    933 	int error;
    934 {
    935 	register struct file *wfp;
    936 	struct file *fp;
    937 
    938 	/*
    939 	 * If the to-be-dup'd fd number is greater than the allowed number
    940 	 * of file descriptors, or the fd to be dup'd has already been
    941 	 * closed, reject.  Note, check for new == old is necessary as
    942 	 * falloc could allocate an already closed to-be-dup'd descriptor
    943 	 * as the new descriptor.
    944 	 */
    945 	fp = fdp->fd_ofiles[indx];
    946 	if ((u_int)dfd >= fdp->fd_nfiles ||
    947 	    (wfp = fdp->fd_ofiles[dfd]) == NULL || fp == wfp)
    948 		return (EBADF);
    949 
    950 	/*
    951 	 * There are two cases of interest here.
    952 	 *
    953 	 * For ENODEV simply dup (dfd) to file descriptor
    954 	 * (indx) and return.
    955 	 *
    956 	 * For ENXIO steal away the file structure from (dfd) and
    957 	 * store it in (indx).  (dfd) is effectively closed by
    958 	 * this operation.
    959 	 *
    960 	 * Any other error code is just returned.
    961 	 */
    962 	switch (error) {
    963 	case ENODEV:
    964 		/*
    965 		 * Check that the mode the file is being opened for is a
    966 		 * subset of the mode of the existing descriptor.
    967 		 */
    968 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag)
    969 			return (EACCES);
    970 		fdp->fd_ofiles[indx] = wfp;
    971 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    972 		wfp->f_count++;
    973 		fd_used(fdp, indx);
    974 		return (0);
    975 
    976 	case ENXIO:
    977 		/*
    978 		 * Steal away the file pointer from dfd, and stuff it into indx.
    979 		 */
    980 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
    981 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
    982 		fdp->fd_ofiles[dfd] = NULL;
    983 		fdp->fd_ofileflags[dfd] = 0;
    984 		/*
    985 		 * Complete the clean up of the filedesc structure by
    986 		 * recomputing the various hints.
    987 		 */
    988 		fd_used(fdp, indx);
    989 		fd_unused(fdp, dfd);
    990 		return (0);
    991 
    992 	default:
    993 		return (error);
    994 	}
    995 	/* NOTREACHED */
    996 }
    997 
    998 /*
    999  * Close any files on exec?
   1000  */
   1001 void
   1002 fdcloseexec(p)
   1003 	struct proc *p;
   1004 {
   1005 	register struct filedesc *fdp = p->p_fd;
   1006 	register int fd;
   1007 
   1008 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
   1009 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
   1010 			(void) fdrelease(p, fd);
   1011 }
   1012