xref: /netbsd-src/sys/kern/kern_descrip.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: kern_descrip.c,v 1.139 2005/12/24 19:12:23 perry 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. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.8 (Berkeley) 2/14/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: kern_descrip.c,v 1.139 2005/12/24 19:12:23 perry Exp $");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/filedesc.h>
45 #include <sys/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/proc.h>
48 #include <sys/file.h>
49 #include <sys/namei.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/pool.h>
57 #include <sys/syslog.h>
58 #include <sys/unistd.h>
59 #include <sys/resourcevar.h>
60 #include <sys/conf.h>
61 #include <sys/event.h>
62 
63 #include <sys/mount.h>
64 #include <sys/sa.h>
65 #include <sys/syscallargs.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 POOL_INIT(file_pool, sizeof(struct file), 0, 0, 0, "filepl",
73     &pool_allocator_nointr);
74 POOL_INIT(cwdi_pool, sizeof(struct cwdinfo), 0, 0, 0, "cwdipl",
75     &pool_allocator_nointr);
76 POOL_INIT(filedesc0_pool, sizeof(struct filedesc0), 0, 0, 0, "fdescpl",
77     &pool_allocator_nointr);
78 
79 /* Global file list lock */
80 static struct simplelock filelist_slock = SIMPLELOCK_INITIALIZER;
81 
82 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
83 MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
84 MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
85 
86 static inline int
87 find_next_zero(uint32_t *bitmap, int want, u_int bits)
88 {
89 	int i, off, maxoff;
90 	uint32_t sub;
91 
92 	if (want > bits)
93 		return -1;
94 
95 	off = want >> NDENTRYSHIFT;
96 	i = want & NDENTRYMASK;
97 	if (i) {
98 		sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i));
99 		if (sub != ~0)
100 			goto found;
101 		off++;
102 	}
103 
104 	maxoff = NDLOSLOTS(bits);
105 	while (off < maxoff) {
106 		if ((sub = bitmap[off]) != ~0)
107 			goto found;
108 		off++;
109 	}
110 
111 	return (-1);
112 
113  found:
114 	return (off << NDENTRYSHIFT) + ffs(~sub) - 1;
115 }
116 
117 static int
118 find_last_set(struct filedesc *fd, int last)
119 {
120 	int off, i;
121 	struct file **ofiles = fd->fd_ofiles;
122 	uint32_t *bitmap = fd->fd_lomap;
123 
124 	off = (last - 1) >> NDENTRYSHIFT;
125 
126 	while (off >= 0 && !bitmap[off])
127 		off--;
128 
129 	if (off < 0)
130 		return (-1);
131 
132 	i = ((off + 1) << NDENTRYSHIFT) - 1;
133 	if (i >= last)
134 		i = last - 1;
135 
136 	while (i > 0 && ofiles[i] == NULL)
137 		i--;
138 
139 	return (i);
140 }
141 
142 static inline void
143 fd_used(struct filedesc *fdp, int fd)
144 {
145 	u_int off = fd >> NDENTRYSHIFT;
146 
147 	LOCK_ASSERT(simple_lock_held(&fdp->fd_slock));
148 	KDASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) == 0);
149 
150 	fdp->fd_lomap[off] |= 1 << (fd & NDENTRYMASK);
151 	if (fdp->fd_lomap[off] == ~0) {
152 		KDASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
153 		    (1 << (off & NDENTRYMASK))) == 0);
154 		fdp->fd_himap[off >> NDENTRYSHIFT] |= 1 << (off & NDENTRYMASK);
155 	}
156 
157 	if (fd > fdp->fd_lastfile)
158 		fdp->fd_lastfile = fd;
159 }
160 
161 static inline void
162 fd_unused(struct filedesc *fdp, int fd)
163 {
164 	u_int off = fd >> NDENTRYSHIFT;
165 
166 	LOCK_ASSERT(simple_lock_held(&fdp->fd_slock));
167 	if (fd < fdp->fd_freefile)
168 		fdp->fd_freefile = fd;
169 
170 	if (fdp->fd_lomap[off] == ~0) {
171 		KDASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
172 		    (1 << (off & NDENTRYMASK))) != 0);
173 		fdp->fd_himap[off >> NDENTRYSHIFT] &=
174 		    ~(1 << (off & NDENTRYMASK));
175 	}
176 	KDASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0);
177 	fdp->fd_lomap[off] &= ~(1 << (fd & NDENTRYMASK));
178 
179 #ifdef DIAGNOSTIC
180 	if (fd > fdp->fd_lastfile)
181 		panic("fd_unused: fd_lastfile inconsistent");
182 #endif
183 	if (fd == fdp->fd_lastfile)
184 		fdp->fd_lastfile = find_last_set(fdp, fd);
185 }
186 
187 /*
188  * Lookup the file structure corresponding to a file descriptor
189  * and return it locked.
190  * Note: typical usage is: `fp = fd_getfile(..); FILE_USE(fp);'
191  * The locking strategy has been optimised for this case, i.e.
192  * fd_getfile() returns the file locked while FILE_USE() will increment
193  * the file's use count and unlock.
194  */
195 struct file *
196 fd_getfile(struct filedesc *fdp, int fd)
197 {
198 	struct file *fp;
199 
200 	if ((u_int) fd >= fdp->fd_nfiles || (fp = fdp->fd_ofiles[fd]) == NULL)
201 		return (NULL);
202 
203 	simple_lock(&fp->f_slock);
204 	if (FILE_IS_USABLE(fp) == 0) {
205 		simple_unlock(&fp->f_slock);
206 		return (NULL);
207 	}
208 
209 	return (fp);
210 }
211 
212 /*
213  * Common code for dup, dup2, and fcntl(F_DUPFD).
214  */
215 static int
216 finishdup(struct lwp *l, int old, int new, register_t *retval)
217 {
218 	struct filedesc	*fdp;
219 	struct file	*fp, *delfp;
220 
221 	fdp = l->l_proc->p_fd;
222 
223 	/*
224 	 * If there is a file in the new slot, remember it so we
225 	 * can close it after we've finished the dup.  We need
226 	 * to do it after the dup is finished, since closing
227 	 * the file may block.
228 	 *
229 	 * Note: `old' is already used for us.
230 	 * Note: Caller already marked `new' slot "used".
231 	 */
232 	simple_lock(&fdp->fd_slock);
233 	delfp = fdp->fd_ofiles[new];
234 
235 	fp = fdp->fd_ofiles[old];
236 	KDASSERT(fp != NULL);
237 	fdp->fd_ofiles[new] = fp;
238 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
239 	simple_unlock(&fdp->fd_slock);
240 
241 	*retval = new;
242 	simple_lock(&fp->f_slock);
243 	fp->f_count++;
244 	FILE_UNUSE_HAVELOCK(fp, l);
245 
246 	if (delfp != NULL) {
247 		simple_lock(&delfp->f_slock);
248 		FILE_USE(delfp);
249 		if (new < fdp->fd_knlistsize)
250 			knote_fdclose(l, new);
251 		(void) closef(delfp, l);
252 	}
253 	return (0);
254 }
255 
256 /*
257  * System calls on descriptors.
258  */
259 
260 /*
261  * Duplicate a file descriptor.
262  */
263 /* ARGSUSED */
264 int
265 sys_dup(struct lwp *l, void *v, register_t *retval)
266 {
267 	struct sys_dup_args /* {
268 		syscallarg(int)	fd;
269 	} */ *uap = v;
270 	struct file	*fp;
271 	struct filedesc	*fdp;
272 	struct proc	*p;
273 	int		old, new, error;
274 
275 	p = l->l_proc;
276 	fdp = p->p_fd;
277 	old = SCARG(uap, fd);
278 
279  restart:
280 	if ((fp = fd_getfile(fdp, old)) == NULL)
281 		return (EBADF);
282 
283 	FILE_USE(fp);
284 
285 	if ((error = fdalloc(p, 0, &new)) != 0) {
286 		if (error == ENOSPC) {
287 			fdexpand(p);
288 			FILE_UNUSE(fp, l);
289 			goto restart;
290 		}
291 		FILE_UNUSE(fp, l);
292 		return (error);
293 	}
294 
295 	/* finishdup() will unuse the descriptors for us */
296 	return (finishdup(l, old, new, retval));
297 }
298 
299 /*
300  * Duplicate a file descriptor to a particular value.
301  */
302 /* ARGSUSED */
303 int
304 sys_dup2(struct lwp *l, void *v, register_t *retval)
305 {
306 	struct sys_dup2_args /* {
307 		syscallarg(int)	from;
308 		syscallarg(int)	to;
309 	} */ *uap = v;
310 	struct file	*fp;
311 	struct filedesc	*fdp;
312 	struct proc	*p;
313 	int		old, new, i, error;
314 
315 	p = l->l_proc;
316 	fdp = p->p_fd;
317 	old = SCARG(uap, from);
318 	new = SCARG(uap, to);
319 
320  restart:
321 	if ((fp = fd_getfile(fdp, old)) == NULL)
322 		return (EBADF);
323 
324 	if ((u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
325 	    (u_int)new >= maxfiles) {
326 		simple_unlock(&fp->f_slock);
327 		return (EBADF);
328 	}
329 
330 	if (old == new) {
331 		simple_unlock(&fp->f_slock);
332 		*retval = new;
333 		return (0);
334 	}
335 
336 	FILE_USE(fp);
337 
338 	if (new >= fdp->fd_nfiles) {
339 		if ((error = fdalloc(p, new, &i)) != 0) {
340 			if (error == ENOSPC) {
341 				fdexpand(p);
342 				FILE_UNUSE(fp, l);
343 				goto restart;
344 			}
345 			FILE_UNUSE(fp, l);
346 			return (error);
347 		}
348 		if (new != i)
349 			panic("dup2: fdalloc");
350 	} else {
351 		simple_lock(&fdp->fd_slock);
352 		/*
353 		 * Mark `new' slot "used" only if it was empty.
354 		 */
355 		if (fdp->fd_ofiles[new] == NULL)
356 			fd_used(fdp, new);
357 		simple_unlock(&fdp->fd_slock);
358 	}
359 
360 	/*
361 	 * finishdup() will close the file that's in the `new'
362 	 * slot, if there's one there.
363 	 */
364 
365 	/* finishdup() will unuse the descriptors for us */
366 	return (finishdup(l, old, new, retval));
367 }
368 
369 /*
370  * fcntl call which is being passed to the file's fs.
371  */
372 static int
373 fcntl_forfs(int fd, struct lwp *l, int cmd, void *arg)
374 {
375 	struct file	*fp;
376 	struct filedesc	*fdp;
377 	int		error;
378 	u_int		size;
379 	void		*data, *memp;
380 #define STK_PARAMS	128
381 	char		stkbuf[STK_PARAMS];
382 
383 	/* fd's value was validated in sys_fcntl before calling this routine */
384 	fdp = l->l_proc->p_fd;
385 	fp = fdp->fd_ofiles[fd];
386 
387 	if ((fp->f_flag & (FREAD | FWRITE)) == 0)
388 		return (EBADF);
389 
390 	/*
391 	 * Interpret high order word to find amount of data to be
392 	 * copied to/from the user's address space.
393 	 */
394 	size = (size_t)F_PARAM_LEN(cmd);
395 	if (size > F_PARAM_MAX)
396 		return (EINVAL);
397 	memp = NULL;
398 	if (size > sizeof(stkbuf)) {
399 		memp = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
400 		data = memp;
401 	} else
402 		data = stkbuf;
403 	if (cmd & F_FSIN) {
404 		if (size) {
405 			error = copyin(arg, data, size);
406 			if (error) {
407 				if (memp)
408 					free(memp, M_IOCTLOPS);
409 				return (error);
410 			}
411 		} else
412 			*(void **)data = arg;
413 	} else if ((cmd & F_FSOUT) && size)
414 		/*
415 		 * Zero the buffer so the user always
416 		 * gets back something deterministic.
417 		 */
418 		memset(data, 0, size);
419 	else if (cmd & F_FSVOID)
420 		*(void **)data = arg;
421 
422 
423 	error = (*fp->f_ops->fo_fcntl)(fp, cmd, data, l);
424 
425 	/*
426 	 * Copy any data to user, size was
427 	 * already set and checked above.
428 	 */
429 	if (error == 0 && (cmd & F_FSOUT) && size)
430 		error = copyout(data, arg, size);
431 	if (memp)
432 		free(memp, M_IOCTLOPS);
433 	return (error);
434 }
435 
436 /*
437  * The file control system call.
438  */
439 /* ARGSUSED */
440 int
441 sys_fcntl(struct lwp *l, void *v, register_t *retval)
442 {
443 	struct sys_fcntl_args /* {
444 		syscallarg(int)		fd;
445 		syscallarg(int)		cmd;
446 		syscallarg(void *)	arg;
447 	} */ *uap = v;
448 	struct filedesc *fdp;
449 	struct file	*fp;
450 	struct proc	*p;
451 	struct vnode	*vp;
452 	int		fd, i, tmp, error, flg, cmd, newmin;
453 	struct flock	fl;
454 
455 	p = l->l_proc;
456 	fd = SCARG(uap, fd);
457 	cmd = SCARG(uap, cmd);
458 	fdp = p->p_fd;
459 	error = 0;
460 	flg = F_POSIX;
461 
462 	switch (cmd) {
463 	case F_CLOSEM:
464 		if (fd < 0)
465 			return EBADF;
466 		while (fdp->fd_lastfile >= fd)
467 			fdrelease(l, fdp->fd_lastfile);
468 		return 0;
469 
470 	case F_MAXFD:
471 		*retval = fdp->fd_lastfile;
472 		return 0;
473 
474 	default:
475 		/* Handled below */
476 		break;
477 	}
478 
479  restart:
480 	if ((fp = fd_getfile(fdp, fd)) == NULL)
481 		return (EBADF);
482 
483 	FILE_USE(fp);
484 
485 	if ((cmd & F_FSCTL)) {
486 		error = fcntl_forfs(fd, l, cmd, SCARG(uap, arg));
487 		goto out;
488 	}
489 
490 	switch (cmd) {
491 
492 	case F_DUPFD:
493 		newmin = (long)SCARG(uap, arg);
494 		if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
495 		    (u_int)newmin >= maxfiles) {
496 			error = EINVAL;
497 			goto out;
498 		}
499 		if ((error = fdalloc(p, newmin, &i)) != 0) {
500 			if (error == ENOSPC) {
501 				fdexpand(p);
502 				FILE_UNUSE(fp, l);
503 				goto restart;
504 			}
505 			goto out;
506 		}
507 
508 		/* finishdup() will unuse the descriptors for us */
509 		return (finishdup(l, fd, i, retval));
510 
511 	case F_GETFD:
512 		*retval = fdp->fd_ofileflags[fd] & UF_EXCLOSE ? 1 : 0;
513 		break;
514 
515 	case F_SETFD:
516 		if ((long)SCARG(uap, arg) & 1)
517 			fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
518 		else
519 			fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
520 		break;
521 
522 	case F_GETFL:
523 		*retval = OFLAGS(fp->f_flag);
524 		break;
525 
526 	case F_SETFL:
527 		tmp = FFLAGS((long)SCARG(uap, arg)) & FCNTLFLAGS;
528 		error = (*fp->f_ops->fo_fcntl)(fp, F_SETFL, &tmp, l);
529 		if (error)
530 			break;
531 		i = tmp ^ fp->f_flag;
532 		if (i & FNONBLOCK) {
533 			int flgs = tmp & FNONBLOCK;
534 			error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, &flgs, l);
535 			if (error)
536 				goto reset_fcntl;
537 		}
538 		if (i & FASYNC) {
539 			int flgs = tmp & FASYNC;
540 			error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, &flgs, l);
541 			if (error) {
542 				if (i & FNONBLOCK) {
543 					tmp = fp->f_flag & FNONBLOCK;
544 					(void)(*fp->f_ops->fo_ioctl)(fp,
545 						FIONBIO, &tmp, l);
546 				}
547 				goto reset_fcntl;
548 			}
549 		}
550 		fp->f_flag = (fp->f_flag & ~FCNTLFLAGS) | tmp;
551 		break;
552 	    reset_fcntl:
553 		(void)(*fp->f_ops->fo_fcntl)(fp, F_SETFL, &fp->f_flag, l);
554 		break;
555 
556 	case F_GETOWN:
557 		error = (*fp->f_ops->fo_ioctl)(fp, FIOGETOWN, &tmp, l);
558 		*retval = tmp;
559 		break;
560 
561 	case F_SETOWN:
562 		tmp = (int)(intptr_t) SCARG(uap, arg);
563 		error = (*fp->f_ops->fo_ioctl)(fp, FIOSETOWN, &tmp, l);
564 		break;
565 
566 	case F_SETLKW:
567 		flg |= F_WAIT;
568 		/* Fall into F_SETLK */
569 
570 	case F_SETLK:
571 		if (fp->f_type != DTYPE_VNODE) {
572 			error = EINVAL;
573 			goto out;
574 		}
575 		vp = (struct vnode *)fp->f_data;
576 		/* Copy in the lock structure */
577 		error = copyin(SCARG(uap, arg), &fl, sizeof(fl));
578 		if (error)
579 			goto out;
580 		if (fl.l_whence == SEEK_CUR)
581 			fl.l_start += fp->f_offset;
582 		switch (fl.l_type) {
583 		case F_RDLCK:
584 			if ((fp->f_flag & FREAD) == 0) {
585 				error = EBADF;
586 				goto out;
587 			}
588 			p->p_flag |= P_ADVLOCK;
589 			error = VOP_ADVLOCK(vp, p, F_SETLK, &fl, flg);
590 			goto out;
591 
592 		case F_WRLCK:
593 			if ((fp->f_flag & FWRITE) == 0) {
594 				error = EBADF;
595 				goto out;
596 			}
597 			p->p_flag |= P_ADVLOCK;
598 			error = VOP_ADVLOCK(vp, p, F_SETLK, &fl, flg);
599 			goto out;
600 
601 		case F_UNLCK:
602 			error = VOP_ADVLOCK(vp, p, F_UNLCK, &fl, F_POSIX);
603 			goto out;
604 
605 		default:
606 			error = EINVAL;
607 			goto out;
608 		}
609 
610 	case F_GETLK:
611 		if (fp->f_type != DTYPE_VNODE) {
612 			error = EINVAL;
613 			goto out;
614 		}
615 		vp = (struct vnode *)fp->f_data;
616 		/* Copy in the lock structure */
617 		error = copyin(SCARG(uap, arg), &fl, sizeof(fl));
618 		if (error)
619 			goto out;
620 		if (fl.l_whence == SEEK_CUR)
621 			fl.l_start += fp->f_offset;
622 		if (fl.l_type != F_RDLCK &&
623 		    fl.l_type != F_WRLCK &&
624 		    fl.l_type != F_UNLCK) {
625 			error = EINVAL;
626 			goto out;
627 		}
628 		error = VOP_ADVLOCK(vp, p, F_GETLK, &fl, F_POSIX);
629 		if (error)
630 			goto out;
631 		error = copyout(&fl, SCARG(uap, arg), sizeof(fl));
632 		break;
633 
634 	default:
635 		error = EINVAL;
636 	}
637 
638  out:
639 	FILE_UNUSE(fp, l);
640 	return (error);
641 }
642 
643 void
644 fdremove(struct filedesc *fdp, int fd)
645 {
646 
647 	simple_lock(&fdp->fd_slock);
648 	fdp->fd_ofiles[fd] = NULL;
649 	fd_unused(fdp, fd);
650 	simple_unlock(&fdp->fd_slock);
651 }
652 
653 int
654 fdrelease(struct lwp *l, int fd)
655 {
656 	struct proc *p = l->l_proc;
657 	struct filedesc	*fdp;
658 	struct file	**fpp, *fp;
659 
660 	fdp = p->p_fd;
661 	simple_lock(&fdp->fd_slock);
662 	if (fd < 0 || fd > fdp->fd_lastfile)
663 		goto badf;
664 	fpp = &fdp->fd_ofiles[fd];
665 	fp = *fpp;
666 	if (fp == NULL)
667 		goto badf;
668 
669 	simple_lock(&fp->f_slock);
670 	if (!FILE_IS_USABLE(fp)) {
671 		simple_unlock(&fp->f_slock);
672 		goto badf;
673 	}
674 
675 	FILE_USE(fp);
676 
677 	*fpp = NULL;
678 	fdp->fd_ofileflags[fd] = 0;
679 	fd_unused(fdp, fd);
680 	simple_unlock(&fdp->fd_slock);
681 	if (fd < fdp->fd_knlistsize)
682 		knote_fdclose(l, fd);
683 	return (closef(fp, l));
684 
685 badf:
686 	simple_unlock(&fdp->fd_slock);
687 	return (EBADF);
688 }
689 
690 /*
691  * Close a file descriptor.
692  */
693 /* ARGSUSED */
694 int
695 sys_close(struct lwp *l, void *v, register_t *retval)
696 {
697 	struct sys_close_args /* {
698 		syscallarg(int)	fd;
699 	} */ *uap = v;
700 	int		fd;
701 	struct filedesc	*fdp;
702 	struct proc *p;
703 
704 	p = l->l_proc;
705 	fd = SCARG(uap, fd);
706 	fdp = p->p_fd;
707 
708 #if 0
709 	if (fd_getfile(fdp, fd) == NULL)
710 		return (EBADF);
711 #endif
712 
713 	return (fdrelease(l, fd));
714 }
715 
716 /*
717  * Return status information about a file descriptor.
718  */
719 /* ARGSUSED */
720 int
721 sys___fstat30(struct lwp *l, void *v, register_t *retval)
722 {
723 	struct sys___fstat30_args /* {
724 		syscallarg(int)			fd;
725 		syscallarg(struct stat *)	sb;
726 	} */ *uap = v;
727 	int		fd;
728 	struct filedesc	*fdp;
729 	struct file	*fp;
730 	struct proc	*p;
731 	struct stat	ub;
732 	int		error;
733 
734 	p = l->l_proc;
735 	fd = SCARG(uap, fd);
736 	fdp = p->p_fd;
737 
738 	if ((fp = fd_getfile(fdp, fd)) == NULL)
739 		return (EBADF);
740 
741 	FILE_USE(fp);
742 	error = (*fp->f_ops->fo_stat)(fp, &ub, l);
743 	FILE_UNUSE(fp, l);
744 
745 	if (error == 0)
746 		error = copyout(&ub, SCARG(uap, sb), sizeof(ub));
747 
748 	return (error);
749 }
750 
751 /*
752  * Return pathconf information about a file descriptor.
753  */
754 /* ARGSUSED */
755 int
756 sys_fpathconf(struct lwp *l, void *v, register_t *retval)
757 {
758 	struct sys_fpathconf_args /* {
759 		syscallarg(int)	fd;
760 		syscallarg(int)	name;
761 	} */ *uap = v;
762 	int		fd;
763 	struct filedesc	*fdp;
764 	struct file	*fp;
765 	struct proc 	*p;
766 	struct vnode	*vp;
767 	int		error;
768 
769 	p = l->l_proc;
770 	fd = SCARG(uap, fd);
771 	fdp = p->p_fd;
772 	error = 0;
773 
774 	if ((fp = fd_getfile(fdp, fd)) == NULL)
775 		return (EBADF);
776 
777 	FILE_USE(fp);
778 
779 	switch (fp->f_type) {
780 
781 	case DTYPE_SOCKET:
782 	case DTYPE_PIPE:
783 		if (SCARG(uap, name) != _PC_PIPE_BUF)
784 			error = EINVAL;
785 		else
786 			*retval = PIPE_BUF;
787 		break;
788 
789 	case DTYPE_VNODE:
790 		vp = (struct vnode *)fp->f_data;
791 		error = VOP_PATHCONF(vp, SCARG(uap, name), retval);
792 		break;
793 
794 	case DTYPE_KQUEUE:
795 		error = EINVAL;
796 		break;
797 
798 	default:
799 		error = EOPNOTSUPP;
800 		break;
801 	}
802 
803 	FILE_UNUSE(fp, l);
804 	return (error);
805 }
806 
807 /*
808  * Allocate a file descriptor for the process.
809  */
810 int	fdexpanded;		/* XXX: what else uses this? */
811 
812 int
813 fdalloc(struct proc *p, int want, int *result)
814 {
815 	struct filedesc	*fdp;
816 	int i, lim, last, error;
817 	u_int off, new;
818 
819 	fdp = p->p_fd;
820 	simple_lock(&fdp->fd_slock);
821 
822 	/*
823 	 * Search for a free descriptor starting at the higher
824 	 * of want or fd_freefile.  If that fails, consider
825 	 * expanding the ofile array.
826 	 */
827 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
828 	last = min(fdp->fd_nfiles, lim);
829  again:
830 	if ((i = want) < fdp->fd_freefile)
831 		i = fdp->fd_freefile;
832 	off = i >> NDENTRYSHIFT;
833 	new = find_next_zero(fdp->fd_himap, off,
834 	    (last + NDENTRIES - 1) >> NDENTRYSHIFT);
835 	if (new != -1) {
836 		i = find_next_zero(&fdp->fd_lomap[new],
837 		    new > off ? 0 : i & NDENTRYMASK, NDENTRIES);
838 		if (i == -1) {
839 			/*
840 			 * free file descriptor in this block was
841 			 * below want, try again with higher want.
842 			 */
843 			want = (new + 1) << NDENTRYSHIFT;
844 			goto again;
845 		}
846 		i += (new << NDENTRYSHIFT);
847 		if (i < last) {
848 			if (fdp->fd_ofiles[i] == NULL) {
849 				fd_used(fdp, i);
850 				if (want <= fdp->fd_freefile)
851 					fdp->fd_freefile = i;
852 				*result = i;
853 				error = 0;
854 				goto out;
855 			}
856 		}
857 	}
858 
859 	/* No space in current array.  Expand or let the caller do it. */
860 	error = (fdp->fd_nfiles >= lim) ? EMFILE : ENOSPC;
861 
862 out:
863 	simple_unlock(&fdp->fd_slock);
864 	return (error);
865 }
866 
867 void
868 fdexpand(struct proc *p)
869 {
870 	struct filedesc	*fdp;
871 	int		i, numfiles, oldnfiles;
872 	struct file	**newofile;
873 	char		*newofileflags;
874 	uint32_t	*newhimap = NULL, *newlomap = NULL;
875 
876 	fdp = p->p_fd;
877 
878 restart:
879 	oldnfiles = fdp->fd_nfiles;
880 
881 	if (oldnfiles < NDEXTENT)
882 		numfiles = NDEXTENT;
883 	else
884 		numfiles = 2 * oldnfiles;
885 
886 	newofile = malloc(numfiles * OFILESIZE, M_FILEDESC, M_WAITOK);
887 	if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
888 		newhimap = malloc(NDHISLOTS(numfiles) * sizeof(uint32_t),
889 		    M_FILEDESC, M_WAITOK);
890 		newlomap = malloc(NDLOSLOTS(numfiles) * sizeof(uint32_t),
891 		    M_FILEDESC, M_WAITOK);
892 	}
893 
894 	simple_lock(&fdp->fd_slock);
895 	/* lock fdp */
896 	if (fdp->fd_nfiles != oldnfiles) {
897 		/* fdp changed; retry */
898 		simple_unlock(&fdp->fd_slock);
899 		free(newofile, M_FILEDESC);
900 		if (newhimap != NULL) free(newhimap, M_FILEDESC);
901 		if (newlomap != NULL) free(newlomap, M_FILEDESC);
902 		goto restart;
903 	}
904 
905 	newofileflags = (char *) &newofile[numfiles];
906 	/*
907 	 * Copy the existing ofile and ofileflags arrays
908 	 * and zero the new portion of each array.
909 	 */
910 	memcpy(newofile, fdp->fd_ofiles,
911 	    (i = sizeof(struct file *) * fdp->fd_nfiles));
912 	memset((char *)newofile + i, 0,
913 	    numfiles * sizeof(struct file *) - i);
914 	memcpy(newofileflags, fdp->fd_ofileflags,
915 	    (i = sizeof(char) * fdp->fd_nfiles));
916 	memset(newofileflags + i, 0, numfiles * sizeof(char) - i);
917 	if (oldnfiles > NDFILE)
918 		free(fdp->fd_ofiles, M_FILEDESC);
919 
920 	if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
921 		memcpy(newhimap, fdp->fd_himap,
922 		    (i = NDHISLOTS(oldnfiles) * sizeof(uint32_t)));
923 		memset((char *)newhimap + i, 0,
924 		    NDHISLOTS(numfiles) * sizeof(uint32_t) - i);
925 
926 		memcpy(newlomap, fdp->fd_lomap,
927 		    (i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t)));
928 		memset((char *)newlomap + i, 0,
929 		    NDLOSLOTS(numfiles) * sizeof(uint32_t) - i);
930 
931 		if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) {
932 			free(fdp->fd_himap, M_FILEDESC);
933 			free(fdp->fd_lomap, M_FILEDESC);
934 		}
935 		fdp->fd_himap = newhimap;
936 		fdp->fd_lomap = newlomap;
937 	}
938 
939 	fdp->fd_ofiles = newofile;
940 	fdp->fd_ofileflags = newofileflags;
941 	fdp->fd_nfiles = numfiles;
942 
943 	simple_unlock(&fdp->fd_slock);
944 
945 	fdexpanded++;
946 }
947 
948 /*
949  * Check to see whether n user file descriptors
950  * are available to the process p.
951  */
952 int
953 fdavail(struct proc *p, int n)
954 {
955 	struct filedesc	*fdp;
956 	struct file	**fpp;
957 	int		i, lim;
958 
959 	fdp = p->p_fd;
960 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
961 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
962 		return (1);
963 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
964 	for (i = min(lim,fdp->fd_nfiles) - fdp->fd_freefile; --i >= 0; fpp++)
965 		if (*fpp == NULL && --n <= 0)
966 			return (1);
967 	return (0);
968 }
969 
970 /*
971  * Create a new open file structure and allocate
972  * a file descriptor for the process that refers to it.
973  */
974 int
975 falloc(struct proc *p, struct file **resultfp, int *resultfd)
976 {
977 	struct file	*fp, *fq;
978 	int		error, i;
979 
980  restart:
981 	if ((error = fdalloc(p, 0, &i)) != 0) {
982 		if (error == ENOSPC) {
983 			fdexpand(p);
984 			goto restart;
985 		}
986 		return (error);
987 	}
988 
989 	fp = pool_get(&file_pool, PR_WAITOK);
990 	simple_lock(&filelist_slock);
991 	if (nfiles >= maxfiles) {
992 		tablefull("file", "increase kern.maxfiles or MAXFILES");
993 		simple_unlock(&filelist_slock);
994 		fd_unused(p->p_fd, i);
995 		pool_put(&file_pool, fp);
996 		return (ENFILE);
997 	}
998 	/*
999 	 * Allocate a new file descriptor.
1000 	 * If the process has file descriptor zero open, add to the list
1001 	 * of open files at that point, otherwise put it at the front of
1002 	 * the list of open files.
1003 	 */
1004 	nfiles++;
1005 	memset(fp, 0, sizeof(struct file));
1006 	fp->f_iflags = FIF_LARVAL;
1007 	if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
1008 		LIST_INSERT_AFTER(fq, fp, f_list);
1009 	} else {
1010 		LIST_INSERT_HEAD(&filehead, fp, f_list);
1011 	}
1012 	simple_unlock(&filelist_slock);
1013 	KDASSERT(p->p_fd->fd_ofiles[i] == NULL);
1014 	p->p_fd->fd_ofiles[i] = fp;
1015 	simple_lock_init(&fp->f_slock);
1016 	fp->f_count = 1;
1017 	fp->f_cred = p->p_ucred;
1018 	crhold(fp->f_cred);
1019 	if (resultfp) {
1020 		fp->f_usecount = 1;
1021 		*resultfp = fp;
1022 	}
1023 	if (resultfd)
1024 		*resultfd = i;
1025 	return (0);
1026 }
1027 
1028 /*
1029  * Free a file descriptor.
1030  */
1031 void
1032 ffree(struct file *fp)
1033 {
1034 
1035 #ifdef DIAGNOSTIC
1036 	if (fp->f_usecount)
1037 		panic("ffree");
1038 #endif
1039 
1040 	simple_lock(&filelist_slock);
1041 	LIST_REMOVE(fp, f_list);
1042 	crfree(fp->f_cred);
1043 #ifdef DIAGNOSTIC
1044 	fp->f_count = 0; /* What's the point? */
1045 #endif
1046 	nfiles--;
1047 	simple_unlock(&filelist_slock);
1048 	pool_put(&file_pool, fp);
1049 }
1050 
1051 /*
1052  * Create an initial cwdinfo structure, using the same current and root
1053  * directories as p.
1054  */
1055 struct cwdinfo *
1056 cwdinit(struct proc *p)
1057 {
1058 	struct cwdinfo *cwdi;
1059 
1060 	cwdi = pool_get(&cwdi_pool, PR_WAITOK);
1061 
1062 	simple_lock_init(&cwdi->cwdi_slock);
1063 	cwdi->cwdi_cdir = p->p_cwdi->cwdi_cdir;
1064 	if (cwdi->cwdi_cdir)
1065 		VREF(cwdi->cwdi_cdir);
1066 	cwdi->cwdi_rdir = p->p_cwdi->cwdi_rdir;
1067 	if (cwdi->cwdi_rdir)
1068 		VREF(cwdi->cwdi_rdir);
1069 	cwdi->cwdi_cmask =  p->p_cwdi->cwdi_cmask;
1070 	cwdi->cwdi_refcnt = 1;
1071 
1072 	return (cwdi);
1073 }
1074 
1075 /*
1076  * Make p2 share p1's cwdinfo.
1077  */
1078 void
1079 cwdshare(struct proc *p1, struct proc *p2)
1080 {
1081 	struct cwdinfo *cwdi = p1->p_cwdi;
1082 
1083 	simple_lock(&cwdi->cwdi_slock);
1084 	cwdi->cwdi_refcnt++;
1085 	simple_unlock(&cwdi->cwdi_slock);
1086 	p2->p_cwdi = cwdi;
1087 }
1088 
1089 /*
1090  * Make this process not share its cwdinfo structure, maintaining
1091  * all cwdinfo state.
1092  */
1093 void
1094 cwdunshare(struct proc *p)
1095 {
1096 	struct cwdinfo *oldcwdi, *newcwdi;
1097 
1098 	if (p->p_cwdi->cwdi_refcnt == 1)
1099 		return;
1100 
1101 	newcwdi = cwdinit(p);
1102 	oldcwdi = p->p_cwdi;
1103 	p->p_cwdi = newcwdi;
1104 	cwdfree(oldcwdi);
1105 }
1106 
1107 /*
1108  * Release a cwdinfo structure.
1109  */
1110 void
1111 cwdfree(struct cwdinfo *cwdi)
1112 {
1113 	int n;
1114 
1115 	simple_lock(&cwdi->cwdi_slock);
1116 	n = --cwdi->cwdi_refcnt;
1117 	simple_unlock(&cwdi->cwdi_slock);
1118 	if (n > 0)
1119 		return;
1120 
1121 	vrele(cwdi->cwdi_cdir);
1122 	if (cwdi->cwdi_rdir)
1123 		vrele(cwdi->cwdi_rdir);
1124 	pool_put(&cwdi_pool, cwdi);
1125 }
1126 
1127 /*
1128  * Create an initial filedesc structure, using the same current and root
1129  * directories as p.
1130  */
1131 struct filedesc *
1132 fdinit(struct proc *p)
1133 {
1134 	struct filedesc0 *newfdp;
1135 
1136 	newfdp = pool_get(&filedesc0_pool, PR_WAITOK);
1137 	memset(newfdp, 0, sizeof(struct filedesc0));
1138 
1139 	fdinit1(newfdp);
1140 
1141 	return (&newfdp->fd_fd);
1142 }
1143 
1144 /*
1145  * Initialize a file descriptor table.
1146  */
1147 void
1148 fdinit1(struct filedesc0 *newfdp)
1149 {
1150 
1151 	newfdp->fd_fd.fd_refcnt = 1;
1152 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1153 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1154 	newfdp->fd_fd.fd_nfiles = NDFILE;
1155 	newfdp->fd_fd.fd_knlistsize = -1;
1156 	newfdp->fd_fd.fd_himap = newfdp->fd_dhimap;
1157 	newfdp->fd_fd.fd_lomap = newfdp->fd_dlomap;
1158 	newfdp->fd_fd.fd_lastfile = -1;
1159 	simple_lock_init(&newfdp->fd_fd.fd_slock);
1160 }
1161 
1162 /*
1163  * Make p2 share p1's filedesc structure.
1164  */
1165 void
1166 fdshare(struct proc *p1, struct proc *p2)
1167 {
1168 	struct filedesc *fdp = p1->p_fd;
1169 
1170 	simple_lock(&fdp->fd_slock);
1171 	p2->p_fd = fdp;
1172 	fdp->fd_refcnt++;
1173 	simple_unlock(&fdp->fd_slock);
1174 }
1175 
1176 /*
1177  * Make this process not share its filedesc structure, maintaining
1178  * all file descriptor state.
1179  */
1180 void
1181 fdunshare(struct lwp *l)
1182 {
1183 	struct proc *p = l->l_proc;
1184 	struct filedesc *newfd;
1185 
1186 	if (p->p_fd->fd_refcnt == 1)
1187 		return;
1188 
1189 	newfd = fdcopy(p);
1190 	fdfree(l);
1191 	p->p_fd = newfd;
1192 }
1193 
1194 /*
1195  * Clear a process's fd table.
1196  */
1197 void
1198 fdclear(struct lwp *l)
1199 {
1200 	struct proc *p = l->l_proc;
1201 	struct filedesc *newfd;
1202 
1203 	newfd = fdinit(p);
1204 	fdfree(l);
1205 	p->p_fd = newfd;
1206 }
1207 
1208 /*
1209  * Copy a filedesc structure.
1210  */
1211 struct filedesc *
1212 fdcopy(struct proc *p)
1213 {
1214 	struct filedesc	*newfdp, *fdp;
1215 	struct file	**fpp, **nfpp;
1216 	int		i, numfiles, lastfile;
1217 
1218 	fdp = p->p_fd;
1219 	newfdp = pool_get(&filedesc0_pool, PR_WAITOK);
1220 	newfdp->fd_refcnt = 1;
1221 	simple_lock_init(&newfdp->fd_slock);
1222 
1223 restart:
1224 	numfiles = fdp->fd_nfiles;
1225 	lastfile = fdp->fd_lastfile;
1226 
1227 	/*
1228 	 * If the number of open files fits in the internal arrays
1229 	 * of the open file structure, use them, otherwise allocate
1230 	 * additional memory for the number of descriptors currently
1231 	 * in use.
1232 	 */
1233 	if (lastfile < NDFILE) {
1234 		i = NDFILE;
1235 	} else {
1236 		/*
1237 		 * Compute the smallest multiple of NDEXTENT needed
1238 		 * for the file descriptors currently in use,
1239 		 * allowing the table to shrink.
1240 		 */
1241 		i = numfiles;
1242 		while (i >= 2 * NDEXTENT && i > lastfile * 2)
1243 			i /= 2;
1244 		newfdp->fd_ofiles = malloc(i * OFILESIZE, M_FILEDESC, M_WAITOK);
1245 	}
1246 	if (NDHISLOTS(i) > NDHISLOTS(NDFILE)) {
1247 		newfdp->fd_himap = malloc(NDHISLOTS(i) * sizeof(uint32_t),
1248 		    M_FILEDESC, M_WAITOK);
1249 		newfdp->fd_lomap = malloc(NDLOSLOTS(i) * sizeof(uint32_t),
1250 		    M_FILEDESC, M_WAITOK);
1251 	}
1252 
1253 	simple_lock(&fdp->fd_slock);
1254 	if (numfiles != fdp->fd_nfiles || lastfile != fdp->fd_lastfile) {
1255 		simple_unlock(&fdp->fd_slock);
1256 		if (i > NDFILE)
1257 			free(newfdp->fd_ofiles, M_FILEDESC);
1258 		if (NDHISLOTS(i) > NDHISLOTS(NDFILE)) {
1259 			free(newfdp->fd_himap, M_FILEDESC);
1260 			free(newfdp->fd_lomap, M_FILEDESC);
1261 		}
1262 		goto restart;
1263 	}
1264 
1265 	if (lastfile < NDFILE) {
1266 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
1267 		newfdp->fd_ofileflags =
1268 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
1269 	} else {
1270 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
1271 	}
1272 	if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) {
1273 		newfdp->fd_himap =
1274 		    ((struct filedesc0 *) newfdp)->fd_dhimap;
1275 		newfdp->fd_lomap =
1276 		    ((struct filedesc0 *) newfdp)->fd_dlomap;
1277 	}
1278 
1279 	newfdp->fd_nfiles = i;
1280 	newfdp->fd_lastfile = lastfile;
1281 	newfdp->fd_freefile = fdp->fd_freefile;
1282 
1283 	/* Clear the entries that will not be copied over.
1284 	 * Avoid calling memset with 0 size (i.e. when
1285 	 * lastfile == i-1 */
1286 	if (lastfile < (i-1))
1287 		memset(newfdp->fd_ofiles + lastfile + 1, 0,
1288 		    (i - lastfile - 1) * sizeof(struct file **));
1289 	memcpy(newfdp->fd_ofileflags, fdp->fd_ofileflags, i * sizeof(char));
1290 	if (i < NDENTRIES * NDENTRIES)
1291 		i = NDENTRIES * NDENTRIES; /* size of inlined bitmaps */
1292 	memcpy(newfdp->fd_himap, fdp->fd_himap, NDHISLOTS(i)*sizeof(uint32_t));
1293 	memcpy(newfdp->fd_lomap, fdp->fd_lomap, NDLOSLOTS(i)*sizeof(uint32_t));
1294 
1295 	fpp = fdp->fd_ofiles;
1296 	nfpp = newfdp->fd_ofiles;
1297 	for (i = 0; i <= lastfile; i++, fpp++, nfpp++) {
1298 		if ((*nfpp = *fpp) == NULL)
1299 			continue;
1300 
1301 		if ((*fpp)->f_type == DTYPE_KQUEUE)
1302 			/* kq descriptors cannot be copied. */
1303 			fdremove(newfdp, i);
1304 		else {
1305 			simple_lock(&(*fpp)->f_slock);
1306 			(*fpp)->f_count++;
1307 			simple_unlock(&(*fpp)->f_slock);
1308 		}
1309 	}
1310 
1311 	simple_unlock(&fdp->fd_slock);
1312 
1313 	newfdp->fd_knlist = NULL;
1314 	newfdp->fd_knlistsize = -1;
1315 	newfdp->fd_knhash = NULL;
1316 	newfdp->fd_knhashmask = 0;
1317 
1318 	return (newfdp);
1319 }
1320 
1321 /*
1322  * Release a filedesc structure.
1323  */
1324 void
1325 fdfree(struct lwp *l)
1326 {
1327 	struct proc	*p = l->l_proc;
1328 	struct filedesc	*fdp;
1329 	struct file	**fpp, *fp;
1330 	int		i;
1331 
1332 	fdp = p->p_fd;
1333 	simple_lock(&fdp->fd_slock);
1334 	i = --fdp->fd_refcnt;
1335 	simple_unlock(&fdp->fd_slock);
1336 	if (i > 0)
1337 		return;
1338 
1339 	fpp = fdp->fd_ofiles;
1340 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
1341 		fp = *fpp;
1342 		if (fp != NULL) {
1343 			*fpp = NULL;
1344 			simple_lock(&fp->f_slock);
1345 			FILE_USE(fp);
1346 			if ((fdp->fd_lastfile - i) < fdp->fd_knlistsize)
1347 				knote_fdclose(l, fdp->fd_lastfile - i);
1348 			(void) closef(fp, l);
1349 		}
1350 	}
1351 	p->p_fd = NULL;
1352 	if (fdp->fd_nfiles > NDFILE)
1353 		free(fdp->fd_ofiles, M_FILEDESC);
1354 	if (NDHISLOTS(fdp->fd_nfiles) > NDHISLOTS(NDFILE)) {
1355 		free(fdp->fd_himap, M_FILEDESC);
1356 		free(fdp->fd_lomap, M_FILEDESC);
1357 	}
1358 	if (fdp->fd_knlist)
1359 		free(fdp->fd_knlist, M_KEVENT);
1360 	if (fdp->fd_knhash)
1361 		hashdone(fdp->fd_knhash, M_KEVENT);
1362 	pool_put(&filedesc0_pool, fdp);
1363 }
1364 
1365 /*
1366  * Internal form of close.
1367  * Decrement reference count on file structure.
1368  * Note: p may be NULL when closing a file
1369  * that was being passed in a message.
1370  *
1371  * Note: we expect the caller is holding a usecount, and expects us
1372  * to drop it (the caller thinks the file is going away forever).
1373  */
1374 int
1375 closef(struct file *fp, struct lwp *l)
1376 {
1377 	struct proc	*p = l ? l->l_proc : NULL;
1378 	struct vnode	*vp;
1379 	struct flock	lf;
1380 	int		error;
1381 
1382 	if (fp == NULL)
1383 		return (0);
1384 
1385 	/*
1386 	 * POSIX record locking dictates that any close releases ALL
1387 	 * locks owned by this process.  This is handled by setting
1388 	 * a flag in the unlock to free ONLY locks obeying POSIX
1389 	 * semantics, and not to free BSD-style file locks.
1390 	 * If the descriptor was in a message, POSIX-style locks
1391 	 * aren't passed with the descriptor.
1392 	 */
1393 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
1394 		lf.l_whence = SEEK_SET;
1395 		lf.l_start = 0;
1396 		lf.l_len = 0;
1397 		lf.l_type = F_UNLCK;
1398 		vp = (struct vnode *)fp->f_data;
1399 		(void) VOP_ADVLOCK(vp, p, F_UNLCK, &lf, F_POSIX);
1400 	}
1401 
1402 	/*
1403 	 * If WANTCLOSE is set, then the reference count on the file
1404 	 * is 0, but there were multiple users of the file.  This can
1405 	 * happen if a filedesc structure is shared by multiple
1406 	 * processes.
1407 	 */
1408 	simple_lock(&fp->f_slock);
1409 	if (fp->f_iflags & FIF_WANTCLOSE) {
1410 		/*
1411 		 * Another user of the file is already closing, and is
1412 		 * simply waiting for other users of the file to drain.
1413 		 * Release our usecount, and wake up the closer if it
1414 		 * is the only remaining use.
1415 		 */
1416 #ifdef DIAGNOSTIC
1417 		if (fp->f_count != 0)
1418 			panic("closef: wantclose and count != 0");
1419 		if (fp->f_usecount < 2)
1420 			panic("closef: wantclose and usecount < 2");
1421 #endif
1422 		if (--fp->f_usecount == 1)
1423 			wakeup(&fp->f_usecount);
1424 		simple_unlock(&fp->f_slock);
1425 		return (0);
1426 	} else {
1427 		/*
1428 		 * Decrement the reference count.  If we were not the
1429 		 * last reference, then release our use and just
1430 		 * return.
1431 		 */
1432 		if (--fp->f_count > 0) {
1433 #ifdef DIAGNOSTIC
1434 			if (fp->f_usecount < 1)
1435 				panic("closef: no wantclose and usecount < 1");
1436 #endif
1437 			fp->f_usecount--;
1438 			simple_unlock(&fp->f_slock);
1439 			return (0);
1440 		}
1441 	}
1442 
1443 	/*
1444 	 * The reference count is now 0.  However, there may be
1445 	 * multiple potential users of this file.  This can happen
1446 	 * if multiple processes shared a single filedesc structure.
1447 	 *
1448 	 * Notify these potential users that the file is closing.
1449 	 * This will prevent them from adding additional uses to
1450 	 * the file.
1451 	 */
1452 	fp->f_iflags |= FIF_WANTCLOSE;
1453 
1454 	/*
1455 	 * We expect the caller to add a use to the file.  So, if we
1456 	 * are the last user, usecount will be 1.  If it is not, we
1457 	 * must wait for the usecount to drain.  When it drains back
1458 	 * to 1, we will be awakened so that we may proceed with the
1459 	 * close.
1460 	 */
1461 #ifdef DIAGNOSTIC
1462 	if (fp->f_usecount < 1)
1463 		panic("closef: usecount < 1");
1464 #endif
1465 	while (fp->f_usecount > 1)
1466 		(void) ltsleep(&fp->f_usecount, PRIBIO, "closef", 0,
1467 				&fp->f_slock);
1468 #ifdef DIAGNOSTIC
1469 	if (fp->f_usecount != 1)
1470 		panic("closef: usecount != 1");
1471 #endif
1472 
1473 	simple_unlock(&fp->f_slock);
1474 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
1475 		lf.l_whence = SEEK_SET;
1476 		lf.l_start = 0;
1477 		lf.l_len = 0;
1478 		lf.l_type = F_UNLCK;
1479 		vp = (struct vnode *)fp->f_data;
1480 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1481 	}
1482 	if (fp->f_ops)
1483 		error = (*fp->f_ops->fo_close)(fp, l);
1484 	else
1485 		error = 0;
1486 
1487 	/* Nothing references the file now, drop the final use (us). */
1488 	fp->f_usecount--;
1489 
1490 	ffree(fp);
1491 	return (error);
1492 }
1493 
1494 /*
1495  * Apply an advisory lock on a file descriptor.
1496  *
1497  * Just attempt to get a record lock of the requested type on
1498  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
1499  */
1500 /* ARGSUSED */
1501 int
1502 sys_flock(struct lwp *l, void *v, register_t *retval)
1503 {
1504 	struct sys_flock_args /* {
1505 		syscallarg(int)	fd;
1506 		syscallarg(int)	how;
1507 	} */ *uap = v;
1508 	int		fd, how, error;
1509 	struct proc	*p;
1510 	struct filedesc	*fdp;
1511 	struct file	*fp;
1512 	struct vnode	*vp;
1513 	struct flock	lf;
1514 
1515 	p = l->l_proc;
1516 	fd = SCARG(uap, fd);
1517 	how = SCARG(uap, how);
1518 	fdp = p->p_fd;
1519 	error = 0;
1520 
1521 	if ((fp = fd_getfile(fdp, fd)) == NULL)
1522 		return (EBADF);
1523 
1524 	FILE_USE(fp);
1525 
1526 	if (fp->f_type != DTYPE_VNODE) {
1527 		error = EOPNOTSUPP;
1528 		goto out;
1529 	}
1530 
1531 	vp = (struct vnode *)fp->f_data;
1532 	lf.l_whence = SEEK_SET;
1533 	lf.l_start = 0;
1534 	lf.l_len = 0;
1535 	if (how & LOCK_UN) {
1536 		lf.l_type = F_UNLCK;
1537 		fp->f_flag &= ~FHASLOCK;
1538 		error = VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1539 		goto out;
1540 	}
1541 	if (how & LOCK_EX)
1542 		lf.l_type = F_WRLCK;
1543 	else if (how & LOCK_SH)
1544 		lf.l_type = F_RDLCK;
1545 	else {
1546 		error = EINVAL;
1547 		goto out;
1548 	}
1549 	fp->f_flag |= FHASLOCK;
1550 	if (how & LOCK_NB)
1551 		error = VOP_ADVLOCK(vp, fp, F_SETLK, &lf, F_FLOCK);
1552 	else
1553 		error = VOP_ADVLOCK(vp, fp, F_SETLK, &lf,
1554 		    F_FLOCK|F_WAIT);
1555  out:
1556 	FILE_UNUSE(fp, l);
1557 	return (error);
1558 }
1559 
1560 /* ARGSUSED */
1561 int
1562 sys_posix_fadvise(struct lwp *l, void *v, register_t *retval)
1563 {
1564 	const struct sys_posix_fadvise_args /* {
1565 		syscallarg(int) fd;
1566 		syscallarg(off_t) offset;
1567 		syscallarg(off_t) len;
1568 		syscallarg(int) advice;
1569 	} */ *uap = v;
1570 	const int fd = SCARG(uap, fd);
1571 	const int advice = SCARG(uap, advice);
1572 	struct proc *p = l->l_proc;
1573 	struct file *fp;
1574 	int error = 0;
1575 
1576 	fp = fd_getfile(p->p_fd, fd);
1577 	if (fp == NULL) {
1578 		error = EBADF;
1579 		goto out;
1580 	}
1581 	FILE_USE(fp);
1582 
1583 	if (fp->f_type != DTYPE_VNODE) {
1584 		if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1585 			error = ESPIPE;
1586 		} else {
1587 			error = EOPNOTSUPP;
1588 		}
1589 		goto out;
1590 	}
1591 
1592 	switch (advice) {
1593 	case POSIX_FADV_NORMAL:
1594 	case POSIX_FADV_RANDOM:
1595 	case POSIX_FADV_SEQUENTIAL:
1596 		KASSERT(POSIX_FADV_NORMAL == UVM_ADV_NORMAL);
1597 		KASSERT(POSIX_FADV_RANDOM == UVM_ADV_RANDOM);
1598 		KASSERT(POSIX_FADV_SEQUENTIAL == UVM_ADV_SEQUENTIAL);
1599 
1600 		/*
1601 		 * we ignore offset and size.
1602 		 */
1603 
1604 		fp->f_advice = advice;
1605 		break;
1606 
1607 	case POSIX_FADV_WILLNEED:
1608 	case POSIX_FADV_DONTNEED:
1609 	case POSIX_FADV_NOREUSE:
1610 
1611 		/*
1612 		 * not implemented yet.
1613 		 */
1614 
1615 		break;
1616 	default:
1617 		error = EINVAL;
1618 		break;
1619 	}
1620 out:
1621 	if (fp != NULL) {
1622 		FILE_UNUSE(fp, l);
1623 	}
1624 	*retval = error;
1625 	return 0;
1626 }
1627 
1628 /*
1629  * File Descriptor pseudo-device driver (/dev/fd/).
1630  *
1631  * Opening minor device N dup()s the file (if any) connected to file
1632  * descriptor N belonging to the calling process.  Note that this driver
1633  * consists of only the ``open()'' routine, because all subsequent
1634  * references to this file will be direct to the other driver.
1635  */
1636 /* ARGSUSED */
1637 static int
1638 filedescopen(dev_t dev, int mode, int type, struct lwp *l)
1639 {
1640 
1641 	/*
1642 	 * XXX Kludge: set dupfd to contain the value of the
1643 	 * the file descriptor being sought for duplication. The error
1644 	 * return ensures that the vnode for this device will be released
1645 	 * by vn_open. Open will detect this special error and take the
1646 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
1647 	 * will simply report the error.
1648 	 */
1649 	l->l_dupfd = minor(dev);	/* XXX */
1650 	return EDUPFD;
1651 }
1652 
1653 const struct cdevsw filedesc_cdevsw = {
1654 	filedescopen, noclose, noread, nowrite, noioctl,
1655 	    nostop, notty, nopoll, nommap, nokqfilter,
1656 };
1657 
1658 /*
1659  * Duplicate the specified descriptor to a free descriptor.
1660  *
1661  * 'indx' has been fdalloc'ed (and will be fdremove'ed on error) by the caller.
1662  */
1663 int
1664 dupfdopen(struct lwp *l, int indx, int dfd, int mode, int error)
1665 {
1666 	struct proc	*p = l->l_proc;
1667 	struct filedesc *fdp;
1668 	struct file	*wfp;
1669 
1670 	fdp = p->p_fd;
1671 
1672 	/* should be cleared by the caller */
1673 	KASSERT(fdp->fd_ofiles[indx] == NULL);
1674 
1675 	/*
1676 	 * If the to-be-dup'd fd number is greater than the allowed number
1677 	 * of file descriptors, or the fd to be dup'd has already been
1678 	 * closed, reject.
1679 	 */
1680 
1681 	/*
1682 	 * Note, in the case of indx == dfd, fd_getfile below returns NULL.
1683 	 */
1684 	if ((wfp = fd_getfile(fdp, dfd)) == NULL)
1685 		return (EBADF);
1686 
1687 	FILE_USE(wfp);
1688 
1689 	/*
1690 	 * There are two cases of interest here.
1691 	 *
1692 	 * For EDUPFD simply dup (dfd) to file descriptor
1693 	 * (indx) and return.
1694 	 *
1695 	 * For EMOVEFD steal away the file structure from (dfd) and
1696 	 * store it in (indx).  (dfd) is effectively closed by
1697 	 * this operation.
1698 	 *
1699 	 * Any other error code is just returned.
1700 	 */
1701 	switch (error) {
1702 	case EDUPFD:
1703 		/*
1704 		 * Check that the mode the file is being opened for is a
1705 		 * subset of the mode of the existing descriptor.
1706 		 */
1707 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
1708 			FILE_UNUSE(wfp, l);
1709 			return (EACCES);
1710 		}
1711 		simple_lock(&fdp->fd_slock);
1712 		fdp->fd_ofiles[indx] = wfp;
1713 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1714 		simple_unlock(&fdp->fd_slock);
1715 		simple_lock(&wfp->f_slock);
1716 		wfp->f_count++;
1717 		/* 'indx' has been fd_used'ed by caller */
1718 		FILE_UNUSE_HAVELOCK(wfp, l);
1719 		return (0);
1720 
1721 	case EMOVEFD:
1722 		/*
1723 		 * Steal away the file pointer from dfd, and stuff it into indx.
1724 		 */
1725 		simple_lock(&fdp->fd_slock);
1726 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
1727 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1728 		fdp->fd_ofiles[dfd] = NULL;
1729 		fdp->fd_ofileflags[dfd] = 0;
1730 		/*
1731 		 * Complete the clean up of the filedesc structure by
1732 		 * recomputing the various hints.
1733 		 */
1734 		/* 'indx' has been fd_used'ed by caller */
1735 		fd_unused(fdp, dfd);
1736 		simple_unlock(&fdp->fd_slock);
1737 		FILE_UNUSE(wfp, l);
1738 		return (0);
1739 
1740 	default:
1741 		FILE_UNUSE(wfp, l);
1742 		return (error);
1743 	}
1744 	/* NOTREACHED */
1745 }
1746 
1747 /*
1748  * Close any files on exec?
1749  */
1750 void
1751 fdcloseexec(struct lwp *l)
1752 {
1753 	struct proc	*p = l->l_proc;
1754 	struct filedesc *fdp;
1755 	int		fd;
1756 
1757 	fdunshare(l);
1758 	cwdunshare(p);
1759 
1760 	fdp = p->p_fd;
1761 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
1762 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
1763 			(void) fdrelease(l, fd);
1764 }
1765 
1766 /*
1767  * It is unsafe for set[ug]id processes to be started with file
1768  * descriptors 0..2 closed, as these descriptors are given implicit
1769  * significance in the Standard C library.  fdcheckstd() will create a
1770  * descriptor referencing /dev/null for each of stdin, stdout, and
1771  * stderr that is not already open.
1772  */
1773 #define CHECK_UPTO 3
1774 int
1775 fdcheckstd(l)
1776 	struct lwp *l;
1777 {
1778 	struct proc *p;
1779 	struct nameidata nd;
1780 	struct filedesc *fdp;
1781 	struct file *fp;
1782 	struct file *devnullfp = NULL;	/* Quell compiler warning */
1783 	struct proc *pp;
1784 	register_t retval;
1785 	int fd, i, error, flags = FREAD|FWRITE, devnull = -1;
1786 	char closed[CHECK_UPTO * 3 + 1], which[3 + 1];
1787 
1788 	p = l->l_proc;
1789 	closed[0] = '\0';
1790 	if ((fdp = p->p_fd) == NULL)
1791 		return (0);
1792 	for (i = 0; i < CHECK_UPTO; i++) {
1793 		if (fdp->fd_ofiles[i] != NULL)
1794 			continue;
1795 		snprintf(which, sizeof(which), ",%d", i);
1796 		strlcat(closed, which, sizeof(closed));
1797 		if (devnull < 0) {
1798 			if ((error = falloc(p, &fp, &fd)) != 0)
1799 				return (error);
1800 			NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1801 			    l);
1802 			if ((error = vn_open(&nd, flags, 0)) != 0) {
1803 				FILE_UNUSE(fp, l);
1804 				ffree(fp);
1805 				fdremove(p->p_fd, fd);
1806 				return (error);
1807 			}
1808 			fp->f_data = nd.ni_vp;
1809 			fp->f_flag = flags;
1810 			fp->f_ops = &vnops;
1811 			fp->f_type = DTYPE_VNODE;
1812 			VOP_UNLOCK(nd.ni_vp, 0);
1813 			devnull = fd;
1814 			devnullfp = fp;
1815 			FILE_SET_MATURE(fp);
1816 		} else {
1817 restart:
1818 			if ((error = fdalloc(p, 0, &fd)) != 0) {
1819 				if (error == ENOSPC) {
1820 					fdexpand(p);
1821 					goto restart;
1822 				}
1823 				return (error);
1824 			}
1825 
1826 			simple_lock(&devnullfp->f_slock);
1827 			FILE_USE(devnullfp);
1828 			/* finishdup() will unuse the descriptors for us */
1829 			if ((error = finishdup(l, devnull, fd, &retval)) != 0)
1830 				return (error);
1831 		}
1832 	}
1833 	if (devnullfp)
1834 		FILE_UNUSE(devnullfp, l);
1835 	if (closed[0] != '\0') {
1836 		pp = p->p_pptr;
1837 		log(LOG_WARNING, "set{u,g}id pid %d (%s) "
1838 		    "was invoked by uid %d ppid %d (%s) "
1839 		    "with fd %s closed\n",
1840 		    p->p_pid, p->p_comm, pp->p_ucred->cr_uid,
1841 		    pp->p_pid, pp->p_comm, &closed[1]);
1842 	}
1843 	return (0);
1844 }
1845 #undef CHECK_UPTO
1846 
1847 /*
1848  * Sets descriptor owner. If the owner is a process, 'pgid'
1849  * is set to positive value, process ID. If the owner is process group,
1850  * 'pgid' is set to -pg_id.
1851  */
1852 int
1853 fsetown(struct proc *p, pid_t *pgid, int cmd, const void *data)
1854 {
1855 	int id = *(const int *)data;
1856 	int error;
1857 
1858 	switch (cmd) {
1859 	case TIOCSPGRP:
1860 		if (id < 0)
1861 			return (EINVAL);
1862 		id = -id;
1863 		break;
1864 	default:
1865 		break;
1866 	}
1867 
1868 	if (id > 0 && !pfind(id))
1869 		return (ESRCH);
1870 	else if (id < 0 && (error = pgid_in_session(p, -id)))
1871 		return (error);
1872 
1873 	*pgid = id;
1874 	return (0);
1875 }
1876 
1877 /*
1878  * Return descriptor owner information. If the value is positive,
1879  * it's process ID. If it's negative, it's process group ID and
1880  * needs the sign removed before use.
1881  */
1882 int
1883 fgetown(struct proc *p, pid_t pgid, int cmd, void *data)
1884 {
1885 	switch (cmd) {
1886 	case TIOCGPGRP:
1887 		*(int *)data = -pgid;
1888 		break;
1889 	default:
1890 		*(int *)data = pgid;
1891 		break;
1892 	}
1893 	return (0);
1894 }
1895 
1896 /*
1897  * Send signal to descriptor owner, either process or process group.
1898  */
1899 void
1900 fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata)
1901 {
1902 	struct proc *p1;
1903 	ksiginfo_t ksi;
1904 
1905 	memset(&ksi, 0, sizeof(ksi));
1906 	ksi.ksi_signo = signo;
1907 	ksi.ksi_code = code;
1908 	ksi.ksi_band = band;
1909 
1910 	if (pgid > 0 && (p1 = pfind(pgid)))
1911 		kpsignal(p1, &ksi, fdescdata);
1912 	else if (pgid < 0)
1913 		kgsignal(-pgid, &ksi, fdescdata);
1914 }
1915 
1916 int
1917 fdclone(struct lwp *l, struct file *fp, int fd, int flag,
1918     const struct fileops *fops, void *data)
1919 {
1920 	fp->f_flag = flag;
1921 	fp->f_type = DTYPE_MISC;
1922 	fp->f_ops = fops;
1923 	fp->f_data = data;
1924 
1925 	l->l_dupfd = fd;
1926 
1927 	FILE_SET_MATURE(fp);
1928 	FILE_UNUSE(fp, l);
1929 	return EMOVEFD;
1930 }
1931 
1932 /* ARGSUSED */
1933 int
1934 fnullop_fcntl(struct file *fp, u_int cmd, void *data, struct lwp *l)
1935 {
1936 	if (cmd == F_SETFL)
1937 		return 0;
1938 
1939 	return EOPNOTSUPP;
1940 }
1941 
1942 /* ARGSUSED */
1943 int
1944 fnullop_poll(struct file *fp, int which, struct lwp *l)
1945 {
1946 	return 0;
1947 }
1948 
1949 
1950 /* ARGSUSED */
1951 int
1952 fnullop_kqfilter(struct file *fp, struct knote *kn)
1953 {
1954 
1955 	return 0;
1956 }
1957 
1958 /* ARGSUSED */
1959 int
1960 fbadop_stat(struct file *fp, struct stat *sb, struct lwp *l)
1961 {
1962 	return EOPNOTSUPP;
1963 }
1964