xref: /netbsd-src/sys/kern/kern_descrip.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: kern_descrip.c,v 1.126 2004/05/31 15:30:55 pk 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 1.126 2004/05/31 15:30:55 pk 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 void	fd_used(struct filedesc *, int);
87 static __inline void	fd_unused(struct filedesc *, int);
88 static __inline int	find_next_zero(uint32_t *, int, u_int);
89 int			finishdup(struct proc *, int, int, register_t *);
90 int			find_last_set(struct filedesc *, int);
91 int			fcntl_forfs(int, struct proc *, int, void *);
92 
93 dev_type_open(filedescopen);
94 
95 const struct cdevsw filedesc_cdevsw = {
96 	filedescopen, noclose, noread, nowrite, noioctl,
97 	nostop, notty, nopoll, nommap, nokqfilter,
98 };
99 
100 static __inline int
101 find_next_zero(uint32_t *bitmap, int want, u_int bits)
102 {
103 	int i, off, maxoff;
104 	uint32_t sub;
105 
106 	if (want > bits)
107 		return -1;
108 
109 	off = want >> NDENTRYSHIFT;
110 	i = want & NDENTRYMASK;
111 	if (i) {
112 		sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i));
113 		if (sub != ~0)
114 			goto found;
115 		off++;
116 	}
117 
118 	maxoff = NDLOSLOTS(bits);
119 	while (off < maxoff) {
120 		if ((sub = bitmap[off]) != ~0)
121 			goto found;
122 		off++;
123 	}
124 
125 	return (-1);
126 
127  found:
128 	return (off << NDENTRYSHIFT) + ffs(~sub) - 1;
129 }
130 
131 int
132 find_last_set(struct filedesc *fd, int last)
133 {
134 	int off, i;
135 	struct file **ofiles = fd->fd_ofiles;
136 	uint32_t *bitmap = fd->fd_lomap;
137 
138 	off = (last - 1) >> NDENTRYSHIFT;
139 
140 	while (off >= 0 && !bitmap[off])
141 		off--;
142 
143 	if (off < 0)
144 		return (0);
145 
146 	i = ((off + 1) << NDENTRYSHIFT) - 1;
147 	if (i >= last)
148 		i = last - 1;
149 
150 	while (i > 0 && ofiles[i] == NULL)
151 		i--;
152 
153 	return (i);
154 }
155 
156 static __inline void
157 fd_used(struct filedesc *fdp, int fd)
158 {
159 	u_int off = fd >> NDENTRYSHIFT;
160 
161 	LOCK_ASSERT(simple_lock_held(&fdp->fd_slock));
162 	KDASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) == 0);
163 
164 	fdp->fd_lomap[off] |= 1 << (fd & NDENTRYMASK);
165 	if (fdp->fd_lomap[off] == ~0) {
166 		KDASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
167 		    (1 << (off & NDENTRYMASK))) == 0);
168 		fdp->fd_himap[off >> NDENTRYSHIFT] |= 1 << (off & NDENTRYMASK);
169 	}
170 
171 	if (fd > fdp->fd_lastfile)
172 		fdp->fd_lastfile = fd;
173 }
174 
175 static __inline void
176 fd_unused(struct filedesc *fdp, int fd)
177 {
178 	u_int off = fd >> NDENTRYSHIFT;
179 
180 	LOCK_ASSERT(simple_lock_held(&fdp->fd_slock));
181 	if (fd < fdp->fd_freefile)
182 		fdp->fd_freefile = fd;
183 
184 	if (fdp->fd_lomap[off] == ~0) {
185 		KDASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
186 		    (1 << (off & NDENTRYMASK))) != 0);
187 		fdp->fd_himap[off >> NDENTRYSHIFT] &=
188 		    ~(1 << (off & NDENTRYMASK));
189 	}
190 	KDASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0);
191 	fdp->fd_lomap[off] &= ~(1 << (fd & NDENTRYMASK));
192 
193 #ifdef DIAGNOSTIC
194 	if (fd > fdp->fd_lastfile)
195 		panic("fd_unused: fd_lastfile inconsistent");
196 #endif
197 	if (fd == fdp->fd_lastfile)
198 		fdp->fd_lastfile = find_last_set(fdp, fd);
199 }
200 
201 /*
202  * Lookup the file structure corresponding to a file descriptor
203  * and return it locked.
204  * Note: typical usage is: `fp = fd_getfile(..); FILE_USE(fp);'
205  * The locking strategy has been optimised for this case, i.e.
206  * fd_getfile() returns the file locked while FILE_USE() will increment
207  * the file's use count and unlock.
208  */
209 struct file *
210 fd_getfile(struct filedesc *fdp, int fd)
211 {
212 	struct file *fp;
213 
214 	if ((u_int) fd >= fdp->fd_nfiles || (fp = fdp->fd_ofiles[fd]) == NULL)
215 		return (NULL);
216 
217 	simple_lock(&fp->f_slock);
218 	if (FILE_IS_USABLE(fp) == 0) {
219 		simple_unlock(&fp->f_slock);
220 		return (NULL);
221 	}
222 
223 	return (fp);
224 }
225 
226 /*
227  * System calls on descriptors.
228  */
229 
230 /*
231  * Duplicate a file descriptor.
232  */
233 /* ARGSUSED */
234 int
235 sys_dup(struct lwp *l, void *v, register_t *retval)
236 {
237 	struct sys_dup_args /* {
238 		syscallarg(int)	fd;
239 	} */ *uap = v;
240 	struct file	*fp;
241 	struct filedesc	*fdp;
242 	struct proc	*p;
243 	int		old, new, error;
244 
245 	p = l->l_proc;
246 	fdp = p->p_fd;
247 	old = SCARG(uap, fd);
248 
249  restart:
250 	if ((fp = fd_getfile(fdp, old)) == NULL)
251 		return (EBADF);
252 
253 	FILE_USE(fp);
254 
255 	if ((error = fdalloc(p, 0, &new)) != 0) {
256 		if (error == ENOSPC) {
257 			fdexpand(p);
258 			FILE_UNUSE(fp, p);
259 			goto restart;
260 		}
261 		FILE_UNUSE(fp, p);
262 		return (error);
263 	}
264 
265 	/* finishdup() will unuse the descriptors for us */
266 	return (finishdup(p, old, new, retval));
267 }
268 
269 /*
270  * Duplicate a file descriptor to a particular value.
271  */
272 /* ARGSUSED */
273 int
274 sys_dup2(struct lwp *l, void *v, register_t *retval)
275 {
276 	struct sys_dup2_args /* {
277 		syscallarg(int)	from;
278 		syscallarg(int)	to;
279 	} */ *uap = v;
280 	struct file	*fp;
281 	struct filedesc	*fdp;
282 	struct proc	*p;
283 	int		old, new, i, error;
284 
285 	p = l->l_proc;
286 	fdp = p->p_fd;
287 	old = SCARG(uap, from);
288 	new = SCARG(uap, to);
289 
290  restart:
291 	if ((fp = fd_getfile(fdp, old)) == NULL)
292 		return (EBADF);
293 
294 	if ((u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
295 	    (u_int)new >= maxfiles) {
296 		simple_unlock(&fp->f_slock);
297 		return (EBADF);
298 	}
299 
300 	if (old == new) {
301 		simple_unlock(&fp->f_slock);
302 		*retval = new;
303 		return (0);
304 	}
305 
306 	FILE_USE(fp);
307 
308 	if (new >= fdp->fd_nfiles) {
309 		if ((error = fdalloc(p, new, &i)) != 0) {
310 			if (error == ENOSPC) {
311 				fdexpand(p);
312 				FILE_UNUSE(fp, p);
313 				goto restart;
314 			}
315 			FILE_UNUSE(fp, p);
316 			return (error);
317 		}
318 		if (new != i)
319 			panic("dup2: fdalloc");
320 	} else {
321 		simple_lock(&fdp->fd_slock);
322 		/*
323 		 * Mark `new' slot "used" only if it was empty.
324 		 */
325 		if (fdp->fd_ofiles[new] == NULL)
326 			fd_used(fdp, new);
327 		simple_unlock(&fdp->fd_slock);
328 	}
329 
330 	/*
331 	 * finishdup() will close the file that's in the `new'
332 	 * slot, if there's one there.
333 	 */
334 
335 	/* finishdup() will unuse the descriptors for us */
336 	return (finishdup(p, old, new, retval));
337 }
338 
339 /*
340  * The file control system call.
341  */
342 /* ARGSUSED */
343 int
344 sys_fcntl(struct lwp *l, void *v, register_t *retval)
345 {
346 	struct sys_fcntl_args /* {
347 		syscallarg(int)		fd;
348 		syscallarg(int)		cmd;
349 		syscallarg(void *)	arg;
350 	} */ *uap = v;
351 	struct filedesc *fdp;
352 	struct file	*fp;
353 	struct proc	*p;
354 	struct vnode	*vp;
355 	int		fd, i, tmp, error, flg, cmd, newmin;
356 	struct flock	fl;
357 
358 	p = l->l_proc;
359 	fd = SCARG(uap, fd);
360 	cmd = SCARG(uap, cmd);
361 	fdp = p->p_fd;
362 	error = 0;
363 	flg = F_POSIX;
364 
365 	switch (cmd) {
366 	case F_CLOSEM:
367 		if (fd < 0)
368 			return EBADF;
369 		while (fdp->fd_lastfile >= fd)
370 			fdrelease(p, fdp->fd_lastfile);
371 		return 0;
372 
373 	case F_MAXFD:
374 		*retval = fdp->fd_lastfile;
375 		return 0;
376 
377 	default:
378 		/* Handled below */
379 		break;
380 	}
381 
382  restart:
383 	if ((fp = fd_getfile(fdp, fd)) == NULL)
384 		return (EBADF);
385 
386 	FILE_USE(fp);
387 
388 	if ((cmd & F_FSCTL)) {
389 		error = fcntl_forfs(fd, p, cmd, SCARG(uap, arg));
390 		goto out;
391 	}
392 
393 	switch (cmd) {
394 
395 	case F_DUPFD:
396 		newmin = (long)SCARG(uap, arg);
397 		if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
398 		    (u_int)newmin >= maxfiles) {
399 			error = EINVAL;
400 			goto out;
401 		}
402 		if ((error = fdalloc(p, newmin, &i)) != 0) {
403 			if (error == ENOSPC) {
404 				fdexpand(p);
405 				FILE_UNUSE(fp, p);
406 				goto restart;
407 			}
408 			goto out;
409 		}
410 
411 		/* finishdup() will unuse the descriptors for us */
412 		return (finishdup(p, fd, i, retval));
413 
414 	case F_GETFD:
415 		*retval = fdp->fd_ofileflags[fd] & UF_EXCLOSE ? 1 : 0;
416 		break;
417 
418 	case F_SETFD:
419 		if ((long)SCARG(uap, arg) & 1)
420 			fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
421 		else
422 			fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
423 		break;
424 
425 	case F_GETFL:
426 		*retval = OFLAGS(fp->f_flag);
427 		break;
428 
429 	case F_SETFL:
430 		tmp = FFLAGS((long)SCARG(uap, arg)) & FCNTLFLAGS;
431 		error = (*fp->f_ops->fo_fcntl)(fp, F_SETFL, &tmp, p);
432 		if (error)
433 			break;
434 		i = tmp ^ fp->f_flag;
435 		if (i & FNONBLOCK) {
436 			int fl = tmp & FNONBLOCK;
437 			error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, &fl, p);
438 			if (error)
439 				goto reset_fcntl;
440 		}
441 		if (i & FASYNC) {
442 			int fl = tmp & FASYNC;
443 			error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, &fl, p);
444 			if (error) {
445 				if (i & FNONBLOCK) {
446 					tmp = fp->f_flag & FNONBLOCK;
447 					(void)(*fp->f_ops->fo_ioctl)(fp,
448 						FIONBIO, &tmp, p);
449 				}
450 				goto reset_fcntl;
451 			}
452 		}
453 		fp->f_flag = (fp->f_flag & ~FCNTLFLAGS) | tmp;
454 		break;
455 	    reset_fcntl:
456 		(void)(*fp->f_ops->fo_fcntl)(fp, F_SETFL, &fp->f_flag, p);
457 		break;
458 
459 	case F_GETOWN:
460 		error = (*fp->f_ops->fo_ioctl)(fp, FIOGETOWN, retval, p);
461 		break;
462 
463 	case F_SETOWN:
464 		tmp = (int)(intptr_t) SCARG(uap, arg);
465 		error = (*fp->f_ops->fo_ioctl)(fp, FIOSETOWN, &tmp, p);
466 		break;
467 
468 	case F_SETLKW:
469 		flg |= F_WAIT;
470 		/* Fall into F_SETLK */
471 
472 	case F_SETLK:
473 		if (fp->f_type != DTYPE_VNODE) {
474 			error = EINVAL;
475 			goto out;
476 		}
477 		vp = (struct vnode *)fp->f_data;
478 		/* Copy in the lock structure */
479 		error = copyin(SCARG(uap, arg), &fl, sizeof(fl));
480 		if (error)
481 			goto out;
482 		if (fl.l_whence == SEEK_CUR)
483 			fl.l_start += fp->f_offset;
484 		switch (fl.l_type) {
485 		case F_RDLCK:
486 			if ((fp->f_flag & FREAD) == 0) {
487 				error = EBADF;
488 				goto out;
489 			}
490 			p->p_flag |= P_ADVLOCK;
491 			error = VOP_ADVLOCK(vp, p, F_SETLK, &fl, flg);
492 			goto out;
493 
494 		case F_WRLCK:
495 			if ((fp->f_flag & FWRITE) == 0) {
496 				error = EBADF;
497 				goto out;
498 			}
499 			p->p_flag |= P_ADVLOCK;
500 			error = VOP_ADVLOCK(vp, p, F_SETLK, &fl, flg);
501 			goto out;
502 
503 		case F_UNLCK:
504 			error = VOP_ADVLOCK(vp, p, F_UNLCK, &fl, F_POSIX);
505 			goto out;
506 
507 		default:
508 			error = EINVAL;
509 			goto out;
510 		}
511 
512 	case F_GETLK:
513 		if (fp->f_type != DTYPE_VNODE) {
514 			error = EINVAL;
515 			goto out;
516 		}
517 		vp = (struct vnode *)fp->f_data;
518 		/* Copy in the lock structure */
519 		error = copyin(SCARG(uap, arg), &fl, sizeof(fl));
520 		if (error)
521 			goto out;
522 		if (fl.l_whence == SEEK_CUR)
523 			fl.l_start += fp->f_offset;
524 		if (fl.l_type != F_RDLCK &&
525 		    fl.l_type != F_WRLCK &&
526 		    fl.l_type != F_UNLCK) {
527 			error = EINVAL;
528 			goto out;
529 		}
530 		error = VOP_ADVLOCK(vp, p, F_GETLK, &fl, F_POSIX);
531 		if (error)
532 			goto out;
533 		error = copyout(&fl, SCARG(uap, arg), sizeof(fl));
534 		break;
535 
536 	default:
537 		error = EINVAL;
538 	}
539 
540  out:
541 	FILE_UNUSE(fp, p);
542 	return (error);
543 }
544 
545 /*
546  * Common code for dup, dup2, and fcntl(F_DUPFD).
547  */
548 int
549 finishdup(struct proc *p, int old, int new, register_t *retval)
550 {
551 	struct filedesc	*fdp;
552 	struct file	*fp, *delfp;
553 
554 	fdp = p->p_fd;
555 
556 	/*
557 	 * If there is a file in the new slot, remember it so we
558 	 * can close it after we've finished the dup.  We need
559 	 * to do it after the dup is finished, since closing
560 	 * the file may block.
561 	 *
562 	 * Note: `old' is already used for us.
563 	 * Note: Caller already marked `new' slot "used".
564 	 */
565 	simple_lock(&fdp->fd_slock);
566 	delfp = fdp->fd_ofiles[new];
567 
568 	fp = fdp->fd_ofiles[old];
569 	KDASSERT(fp != NULL);
570 	fdp->fd_ofiles[new] = fp;
571 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
572 	simple_unlock(&fdp->fd_slock);
573 
574 	*retval = new;
575 	simple_lock(&fp->f_slock);
576 	fp->f_count++;
577 	FILE_UNUSE_HAVELOCK(fp, p);
578 
579 	if (delfp != NULL) {
580 		simple_lock(&delfp->f_slock);
581 		FILE_USE(delfp);
582 		if (new < fdp->fd_knlistsize)
583 			knote_fdclose(p, new);
584 		(void) closef(delfp, p);
585 	}
586 	return (0);
587 }
588 
589 void
590 fdremove(struct filedesc *fdp, int fd)
591 {
592 
593 	simple_lock(&fdp->fd_slock);
594 	fdp->fd_ofiles[fd] = NULL;
595 	fd_unused(fdp, fd);
596 	simple_unlock(&fdp->fd_slock);
597 }
598 
599 int
600 fdrelease(struct proc *p, int fd)
601 {
602 	struct filedesc	*fdp;
603 	struct file	**fpp, *fp;
604 
605 	fdp = p->p_fd;
606 	simple_lock(&fdp->fd_slock);
607 	if ((u_int) fd > fdp->fd_lastfile)
608 		goto badf;
609 	fpp = &fdp->fd_ofiles[fd];
610 	fp = *fpp;
611 	if (fp == NULL)
612 		goto badf;
613 
614 	simple_lock(&fp->f_slock);
615 	if (!FILE_IS_USABLE(fp)) {
616 		simple_unlock(&fp->f_slock);
617 		goto badf;
618 	}
619 
620 	FILE_USE(fp);
621 
622 	*fpp = NULL;
623 	fdp->fd_ofileflags[fd] = 0;
624 	fd_unused(fdp, fd);
625 	simple_unlock(&fdp->fd_slock);
626 	if (fd < fdp->fd_knlistsize)
627 		knote_fdclose(p, fd);
628 	return (closef(fp, p));
629 
630 badf:
631 	simple_unlock(&fdp->fd_slock);
632 	return (EBADF);
633 }
634 
635 /*
636  * Close a file descriptor.
637  */
638 /* ARGSUSED */
639 int
640 sys_close(struct lwp *l, void *v, register_t *retval)
641 {
642 	struct sys_close_args /* {
643 		syscallarg(int)	fd;
644 	} */ *uap = v;
645 	int		fd;
646 	struct filedesc	*fdp;
647 	struct proc *p;
648 
649 	p = l->l_proc;
650 	fd = SCARG(uap, fd);
651 	fdp = p->p_fd;
652 
653 #if 0
654 	if (fd_getfile(fdp, fd) == NULL)
655 		return (EBADF);
656 #endif
657 
658 	return (fdrelease(p, fd));
659 }
660 
661 /*
662  * Return status information about a file descriptor.
663  */
664 /* ARGSUSED */
665 int
666 sys___fstat13(struct lwp *l, void *v, register_t *retval)
667 {
668 	struct sys___fstat13_args /* {
669 		syscallarg(int)			fd;
670 		syscallarg(struct stat *)	sb;
671 	} */ *uap = v;
672 	int		fd;
673 	struct filedesc	*fdp;
674 	struct file	*fp;
675 	struct proc	*p;
676 	struct stat	ub;
677 	int		error;
678 
679 	p = l->l_proc;
680 	fd = SCARG(uap, fd);
681 	fdp = p->p_fd;
682 
683 	if ((fp = fd_getfile(fdp, fd)) == NULL)
684 		return (EBADF);
685 
686 	FILE_USE(fp);
687 	error = (*fp->f_ops->fo_stat)(fp, &ub, p);
688 	FILE_UNUSE(fp, p);
689 
690 	if (error == 0)
691 		error = copyout(&ub, SCARG(uap, sb), sizeof(ub));
692 
693 	return (error);
694 }
695 
696 /*
697  * Return pathconf information about a file descriptor.
698  */
699 /* ARGSUSED */
700 int
701 sys_fpathconf(struct lwp *l, void *v, register_t *retval)
702 {
703 	struct sys_fpathconf_args /* {
704 		syscallarg(int)	fd;
705 		syscallarg(int)	name;
706 	} */ *uap = v;
707 	int		fd;
708 	struct filedesc	*fdp;
709 	struct file	*fp;
710 	struct proc 	*p;
711 	struct vnode	*vp;
712 	int		error;
713 
714 	p = l->l_proc;
715 	fd = SCARG(uap, fd);
716 	fdp = p->p_fd;
717 	error = 0;
718 
719 	if ((fp = fd_getfile(fdp, fd)) == NULL)
720 		return (EBADF);
721 
722 	FILE_USE(fp);
723 
724 	switch (fp->f_type) {
725 
726 	case DTYPE_SOCKET:
727 	case DTYPE_PIPE:
728 		if (SCARG(uap, name) != _PC_PIPE_BUF)
729 			error = EINVAL;
730 		else
731 			*retval = PIPE_BUF;
732 		break;
733 
734 	case DTYPE_VNODE:
735 		vp = (struct vnode *)fp->f_data;
736 		error = VOP_PATHCONF(vp, SCARG(uap, name), retval);
737 		break;
738 
739 	case DTYPE_KQUEUE:
740 		error = EINVAL;
741 		break;
742 
743 	default:
744 		error = EOPNOTSUPP;
745 		break;
746 	}
747 
748 	FILE_UNUSE(fp, p);
749 	return (error);
750 }
751 
752 /*
753  * Allocate a file descriptor for the process.
754  */
755 int	fdexpanded;		/* XXX: what else uses this? */
756 
757 int
758 fdalloc(struct proc *p, int want, int *result)
759 {
760 	struct filedesc	*fdp;
761 	int i, lim, last, error;
762 	u_int off, new;
763 
764 	fdp = p->p_fd;
765 	simple_lock(&fdp->fd_slock);
766 
767 	/*
768 	 * Search for a free descriptor starting at the higher
769 	 * of want or fd_freefile.  If that fails, consider
770 	 * expanding the ofile array.
771 	 */
772 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
773 	last = min(fdp->fd_nfiles, lim);
774  again:
775 	if ((i = want) < fdp->fd_freefile)
776 		i = fdp->fd_freefile;
777 	off = i >> NDENTRYSHIFT;
778 	new = find_next_zero(fdp->fd_himap, off,
779 	    (last + NDENTRIES - 1) >> NDENTRYSHIFT);
780 	if (new != -1) {
781 		i = find_next_zero(&fdp->fd_lomap[new],
782 		    new > off ? 0 : i & NDENTRYMASK, NDENTRIES);
783 		if (i == -1) {
784 			/*
785 			 * free file descriptor in this block was
786 			 * below want, try again with higher want.
787 			 */
788 			want = (new + 1) << NDENTRYSHIFT;
789 			goto again;
790 		}
791 		i += (new << NDENTRYSHIFT);
792 		if (i < last) {
793 			if (fdp->fd_ofiles[i] == NULL) {
794 				fd_used(fdp, i);
795 				if (want <= fdp->fd_freefile)
796 					fdp->fd_freefile = i;
797 				*result = i;
798 				error = 0;
799 				goto out;
800 			}
801 		}
802 	}
803 
804 	/* No space in current array.  Expand or let the caller do it. */
805 	error = (fdp->fd_nfiles >= lim) ? EMFILE : ENOSPC;
806 
807 out:
808 	simple_unlock(&fdp->fd_slock);
809 	return (error);
810 }
811 
812 void
813 fdexpand(struct proc *p)
814 {
815 	struct filedesc	*fdp;
816 	int		i, nfiles, oldnfiles;
817 	struct file	**newofile;
818 	char		*newofileflags;
819 	uint32_t	*newhimap = NULL, *newlomap = NULL;
820 
821 	fdp = p->p_fd;
822 
823 restart:
824 	oldnfiles = fdp->fd_nfiles;
825 
826 	if (oldnfiles < NDEXTENT)
827 		nfiles = NDEXTENT;
828 	else
829 		nfiles = 2 * oldnfiles;
830 
831 	newofile = malloc(nfiles * OFILESIZE, M_FILEDESC, M_WAITOK);
832 	if (NDHISLOTS(nfiles) > NDHISLOTS(oldnfiles)) {
833 		newhimap = malloc(NDHISLOTS(nfiles) * sizeof(uint32_t),
834 		    M_FILEDESC, M_WAITOK);
835 		newlomap = malloc(NDLOSLOTS(nfiles) * sizeof(uint32_t),
836 		    M_FILEDESC, M_WAITOK);
837 	}
838 
839 	simple_lock(&fdp->fd_slock);
840 	/* lock fdp */
841 	if (fdp->fd_nfiles != oldnfiles) {
842 		/* fdp changed; retry */
843 		simple_unlock(&fdp->fd_slock);
844 		free(newofile, M_FILEDESC);
845 		if (newhimap != NULL) free(newhimap, M_FILEDESC);
846 		if (newlomap != NULL) free(newlomap, M_FILEDESC);
847 		goto restart;
848 	}
849 
850 	newofileflags = (char *) &newofile[nfiles];
851 	/*
852 	 * Copy the existing ofile and ofileflags arrays
853 	 * and zero the new portion of each array.
854 	 */
855 	memcpy(newofile, fdp->fd_ofiles,
856 	    (i = sizeof(struct file *) * fdp->fd_nfiles));
857 	memset((char *)newofile + i, 0,
858 	    nfiles * sizeof(struct file *) - i);
859 	memcpy(newofileflags, fdp->fd_ofileflags,
860 	    (i = sizeof(char) * fdp->fd_nfiles));
861 	memset(newofileflags + i, 0, nfiles * sizeof(char) - i);
862 	if (oldnfiles > NDFILE)
863 		free(fdp->fd_ofiles, M_FILEDESC);
864 
865 	if (NDHISLOTS(nfiles) > NDHISLOTS(oldnfiles)) {
866 		memcpy(newhimap, fdp->fd_himap,
867 		    (i = NDHISLOTS(oldnfiles) * sizeof(uint32_t)));
868 		memset((char *)newhimap + i, 0,
869 		    NDHISLOTS(nfiles) * sizeof(uint32_t) - i);
870 
871 		memcpy(newlomap, fdp->fd_lomap,
872 		    (i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t)));
873 		memset((char *)newlomap + i, 0,
874 		    NDLOSLOTS(nfiles) * sizeof(uint32_t) - i);
875 
876 		if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) {
877 			free(fdp->fd_himap, M_FILEDESC);
878 			free(fdp->fd_lomap, M_FILEDESC);
879 		}
880 		fdp->fd_himap = newhimap;
881 		fdp->fd_lomap = newlomap;
882 	}
883 
884 	fdp->fd_ofiles = newofile;
885 	fdp->fd_ofileflags = newofileflags;
886 	fdp->fd_nfiles = nfiles;
887 
888 	simple_unlock(&fdp->fd_slock);
889 
890 	fdexpanded++;
891 }
892 
893 /*
894  * Check to see whether n user file descriptors
895  * are available to the process p.
896  */
897 int
898 fdavail(struct proc *p, int n)
899 {
900 	struct filedesc	*fdp;
901 	struct file	**fpp;
902 	int		i, lim;
903 
904 	fdp = p->p_fd;
905 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
906 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
907 		return (1);
908 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
909 	for (i = min(lim,fdp->fd_nfiles) - fdp->fd_freefile; --i >= 0; fpp++)
910 		if (*fpp == NULL && --n <= 0)
911 			return (1);
912 	return (0);
913 }
914 
915 /*
916  * Create a new open file structure and allocate
917  * a file descriptor for the process that refers to it.
918  */
919 int
920 falloc(struct proc *p, struct file **resultfp, int *resultfd)
921 {
922 	struct file	*fp, *fq;
923 	int		error, i;
924 
925  restart:
926 	if ((error = fdalloc(p, 0, &i)) != 0) {
927 		if (error == ENOSPC) {
928 			fdexpand(p);
929 			goto restart;
930 		}
931 		return (error);
932 	}
933 
934 	fp = pool_get(&file_pool, PR_WAITOK);
935 	simple_lock(&filelist_slock);
936 	if (nfiles >= maxfiles) {
937 		tablefull("file", "increase kern.maxfiles or MAXFILES");
938 		simple_unlock(&filelist_slock);
939 		fd_unused(p->p_fd, i);
940 		pool_put(&file_pool, fp);
941 		return (ENFILE);
942 	}
943 	/*
944 	 * Allocate a new file descriptor.
945 	 * If the process has file descriptor zero open, add to the list
946 	 * of open files at that point, otherwise put it at the front of
947 	 * the list of open files.
948 	 */
949 	nfiles++;
950 	memset(fp, 0, sizeof(struct file));
951 	fp->f_iflags = FIF_LARVAL;
952 	if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
953 		LIST_INSERT_AFTER(fq, fp, f_list);
954 	} else {
955 		LIST_INSERT_HEAD(&filehead, fp, f_list);
956 	}
957 	simple_unlock(&filelist_slock);
958 	KDASSERT(p->p_fd->fd_ofiles[i] == NULL);
959 	p->p_fd->fd_ofiles[i] = fp;
960 	simple_lock_init(&fp->f_slock);
961 	fp->f_count = 1;
962 	fp->f_cred = p->p_ucred;
963 	crhold(fp->f_cred);
964 	if (resultfp) {
965 		fp->f_usecount = 1;
966 		*resultfp = fp;
967 	}
968 	if (resultfd)
969 		*resultfd = i;
970 	return (0);
971 }
972 
973 /*
974  * Free a file descriptor.
975  */
976 void
977 ffree(struct file *fp)
978 {
979 
980 #ifdef DIAGNOSTIC
981 	if (fp->f_usecount)
982 		panic("ffree");
983 #endif
984 
985 	simple_lock(&filelist_slock);
986 	LIST_REMOVE(fp, f_list);
987 	crfree(fp->f_cred);
988 #ifdef DIAGNOSTIC
989 	fp->f_count = 0; /* What's the point? */
990 #endif
991 	nfiles--;
992 	simple_unlock(&filelist_slock);
993 	pool_put(&file_pool, fp);
994 }
995 
996 /*
997  * Create an initial cwdinfo structure, using the same current and root
998  * directories as p.
999  */
1000 struct cwdinfo *
1001 cwdinit(struct proc *p)
1002 {
1003 	struct cwdinfo *cwdi;
1004 
1005 	cwdi = pool_get(&cwdi_pool, PR_WAITOK);
1006 
1007 	simple_lock_init(&cwdi->cwdi_slock);
1008 	cwdi->cwdi_cdir = p->p_cwdi->cwdi_cdir;
1009 	if (cwdi->cwdi_cdir)
1010 		VREF(cwdi->cwdi_cdir);
1011 	cwdi->cwdi_rdir = p->p_cwdi->cwdi_rdir;
1012 	if (cwdi->cwdi_rdir)
1013 		VREF(cwdi->cwdi_rdir);
1014 	cwdi->cwdi_cmask =  p->p_cwdi->cwdi_cmask;
1015 	cwdi->cwdi_refcnt = 1;
1016 
1017 	return (cwdi);
1018 }
1019 
1020 /*
1021  * Make p2 share p1's cwdinfo.
1022  */
1023 void
1024 cwdshare(struct proc *p1, struct proc *p2)
1025 {
1026 	struct cwdinfo *cwdi = p1->p_cwdi;
1027 
1028 	simple_lock(&cwdi->cwdi_slock);
1029 	cwdi->cwdi_refcnt++;
1030 	simple_unlock(&cwdi->cwdi_slock);
1031 	p2->p_cwdi = cwdi;
1032 }
1033 
1034 /*
1035  * Make this process not share its cwdinfo structure, maintaining
1036  * all cwdinfo state.
1037  */
1038 void
1039 cwdunshare(struct proc *p)
1040 {
1041 	struct cwdinfo *oldcwdi, *newcwdi;
1042 
1043 	if (p->p_cwdi->cwdi_refcnt == 1)
1044 		return;
1045 
1046 	newcwdi = cwdinit(p);
1047 	oldcwdi = p->p_cwdi;
1048 	p->p_cwdi = newcwdi;
1049 	cwdfree(oldcwdi);
1050 }
1051 
1052 /*
1053  * Release a cwdinfo structure.
1054  */
1055 void
1056 cwdfree(struct cwdinfo *cwdi)
1057 {
1058 	int n;
1059 
1060 	simple_lock(&cwdi->cwdi_slock);
1061 	n = --cwdi->cwdi_refcnt;
1062 	simple_unlock(&cwdi->cwdi_slock);
1063 	if (n > 0)
1064 		return;
1065 
1066 	vrele(cwdi->cwdi_cdir);
1067 	if (cwdi->cwdi_rdir)
1068 		vrele(cwdi->cwdi_rdir);
1069 	pool_put(&cwdi_pool, cwdi);
1070 }
1071 
1072 /*
1073  * Create an initial filedesc structure, using the same current and root
1074  * directories as p.
1075  */
1076 struct filedesc *
1077 fdinit(struct proc *p)
1078 {
1079 	struct filedesc0 *newfdp;
1080 
1081 	newfdp = pool_get(&filedesc0_pool, PR_WAITOK);
1082 	memset(newfdp, 0, sizeof(struct filedesc0));
1083 
1084 	fdinit1(newfdp);
1085 
1086 	return (&newfdp->fd_fd);
1087 }
1088 
1089 /*
1090  * Initialize a file descriptor table.
1091  */
1092 void
1093 fdinit1(struct filedesc0 *newfdp)
1094 {
1095 
1096 	newfdp->fd_fd.fd_refcnt = 1;
1097 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1098 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1099 	newfdp->fd_fd.fd_nfiles = NDFILE;
1100 	newfdp->fd_fd.fd_knlistsize = -1;
1101 	newfdp->fd_fd.fd_himap = newfdp->fd_dhimap;
1102 	newfdp->fd_fd.fd_lomap = newfdp->fd_dlomap;
1103 	simple_lock_init(&newfdp->fd_fd.fd_slock);
1104 }
1105 
1106 /*
1107  * Make p2 share p1's filedesc structure.
1108  */
1109 void
1110 fdshare(struct proc *p1, struct proc *p2)
1111 {
1112 	struct filedesc *fdp = p1->p_fd;
1113 
1114 	simple_lock(&fdp->fd_slock);
1115 	p2->p_fd = fdp;
1116 	fdp->fd_refcnt++;
1117 	simple_unlock(&fdp->fd_slock);
1118 }
1119 
1120 /*
1121  * Make this process not share its filedesc structure, maintaining
1122  * all file descriptor state.
1123  */
1124 void
1125 fdunshare(struct proc *p)
1126 {
1127 	struct filedesc *newfd;
1128 
1129 	if (p->p_fd->fd_refcnt == 1)
1130 		return;
1131 
1132 	newfd = fdcopy(p);
1133 	fdfree(p);
1134 	p->p_fd = newfd;
1135 }
1136 
1137 /*
1138  * Clear a process's fd table.
1139  */
1140 void
1141 fdclear(struct proc *p)
1142 {
1143 	struct filedesc *newfd;
1144 
1145 	newfd = fdinit(p);
1146 	fdfree(p);
1147 	p->p_fd = newfd;
1148 }
1149 
1150 /*
1151  * Copy a filedesc structure.
1152  */
1153 struct filedesc *
1154 fdcopy(struct proc *p)
1155 {
1156 	struct filedesc	*newfdp, *fdp;
1157 	struct file	**fpp, **nfpp;
1158 	int		i, nfiles, lastfile;
1159 
1160 	fdp = p->p_fd;
1161 	newfdp = pool_get(&filedesc0_pool, PR_WAITOK);
1162 	newfdp->fd_refcnt = 1;
1163 	simple_lock_init(&newfdp->fd_slock);
1164 
1165 restart:
1166 	nfiles = fdp->fd_nfiles;
1167 	lastfile = fdp->fd_lastfile;
1168 
1169 	/*
1170 	 * If the number of open files fits in the internal arrays
1171 	 * of the open file structure, use them, otherwise allocate
1172 	 * additional memory for the number of descriptors currently
1173 	 * in use.
1174 	 */
1175 	if (lastfile < NDFILE) {
1176 		i = NDFILE;
1177 	} else {
1178 		/*
1179 		 * Compute the smallest multiple of NDEXTENT needed
1180 		 * for the file descriptors currently in use,
1181 		 * allowing the table to shrink.
1182 		 */
1183 		i = nfiles;
1184 		while (i >= 2 * NDEXTENT && i > lastfile * 2)
1185 			i /= 2;
1186 		newfdp->fd_ofiles = malloc(i * OFILESIZE, M_FILEDESC, M_WAITOK);
1187 	}
1188 	if (NDHISLOTS(i) > NDHISLOTS(NDFILE)) {
1189 		newfdp->fd_himap = malloc(NDHISLOTS(i) * sizeof(uint32_t),
1190 		    M_FILEDESC, M_WAITOK);
1191 		newfdp->fd_lomap = malloc(NDLOSLOTS(i) * sizeof(uint32_t),
1192 		    M_FILEDESC, M_WAITOK);
1193 	}
1194 
1195 	simple_lock(&fdp->fd_slock);
1196 	if (nfiles != fdp->fd_nfiles || lastfile != fdp->fd_lastfile) {
1197 		simple_unlock(&fdp->fd_slock);
1198 		if (i > NDFILE)
1199 			free(newfdp->fd_ofiles, M_FILEDESC);
1200 		if (NDHISLOTS(i) > NDHISLOTS(NDFILE)) {
1201 			free(newfdp->fd_himap, M_FILEDESC);
1202 			free(newfdp->fd_lomap, M_FILEDESC);
1203 		}
1204 		goto restart;
1205 	}
1206 
1207 	if (lastfile < NDFILE) {
1208 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
1209 		newfdp->fd_ofileflags =
1210 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
1211 	} else {
1212 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
1213 	}
1214 	if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) {
1215 		newfdp->fd_himap =
1216 		    ((struct filedesc0 *) newfdp)->fd_dhimap;
1217 		newfdp->fd_lomap =
1218 		    ((struct filedesc0 *) newfdp)->fd_dlomap;
1219 	}
1220 
1221 	newfdp->fd_nfiles = i;
1222 	newfdp->fd_lastfile = lastfile;
1223 	newfdp->fd_freefile = fdp->fd_freefile;
1224 
1225 	memset(newfdp->fd_ofiles + lastfile, 0,
1226 		(i - lastfile) * sizeof(struct file **));
1227 	memcpy(newfdp->fd_ofileflags, fdp->fd_ofileflags, i * sizeof(char));
1228 	if (i < NDENTRIES * NDENTRIES)
1229 		i = NDENTRIES * NDENTRIES; /* size of inlined bitmaps */
1230 	memcpy(newfdp->fd_himap, fdp->fd_himap, NDHISLOTS(i)*sizeof(uint32_t));
1231 	memcpy(newfdp->fd_lomap, fdp->fd_lomap, NDLOSLOTS(i)*sizeof(uint32_t));
1232 
1233 	fpp = fdp->fd_ofiles;
1234 	nfpp = newfdp->fd_ofiles;
1235 	for (i = 0; i <= lastfile; i++, fpp++, nfpp++) {
1236 		if ((*nfpp = *fpp) == NULL)
1237 			continue;
1238 
1239 		if ((*fpp)->f_type == DTYPE_KQUEUE)
1240 			/* kq descriptors cannot be copied. */
1241 			fdremove(newfdp, i);
1242 		else {
1243 			simple_lock(&(*fpp)->f_slock);
1244 			(*fpp)->f_count++;
1245 			simple_unlock(&(*fpp)->f_slock);
1246 		}
1247 	}
1248 
1249 	simple_unlock(&fdp->fd_slock);
1250 
1251 	newfdp->fd_knlist = NULL;
1252 	newfdp->fd_knlistsize = -1;
1253 	newfdp->fd_knhash = NULL;
1254 	newfdp->fd_knhashmask = 0;
1255 
1256 	return (newfdp);
1257 }
1258 
1259 /*
1260  * Release a filedesc structure.
1261  */
1262 void
1263 fdfree(struct proc *p)
1264 {
1265 	struct filedesc	*fdp;
1266 	struct file	**fpp, *fp;
1267 	int		i;
1268 
1269 	fdp = p->p_fd;
1270 	simple_lock(&fdp->fd_slock);
1271 	i = --fdp->fd_refcnt;
1272 	simple_unlock(&fdp->fd_slock);
1273 	if (i > 0)
1274 		return;
1275 
1276 	fpp = fdp->fd_ofiles;
1277 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
1278 		fp = *fpp;
1279 		if (fp != NULL) {
1280 			*fpp = NULL;
1281 			simple_lock(&fp->f_slock);
1282 			FILE_USE(fp);
1283 			if (i < fdp->fd_knlistsize)
1284 				knote_fdclose(p, fdp->fd_lastfile - i);
1285 			(void) closef(fp, p);
1286 		}
1287 	}
1288 	p->p_fd = NULL;
1289 	if (fdp->fd_nfiles > NDFILE)
1290 		free(fdp->fd_ofiles, M_FILEDESC);
1291 	if (NDHISLOTS(fdp->fd_nfiles) > NDHISLOTS(NDFILE)) {
1292 		free(fdp->fd_himap, M_FILEDESC);
1293 		free(fdp->fd_lomap, M_FILEDESC);
1294 	}
1295 	if (fdp->fd_knlist)
1296 		free(fdp->fd_knlist, M_KEVENT);
1297 	if (fdp->fd_knhash)
1298 		hashdone(fdp->fd_knhash, M_KEVENT);
1299 	pool_put(&filedesc0_pool, fdp);
1300 }
1301 
1302 /*
1303  * Internal form of close.
1304  * Decrement reference count on file structure.
1305  * Note: p may be NULL when closing a file
1306  * that was being passed in a message.
1307  *
1308  * Note: we expect the caller is holding a usecount, and expects us
1309  * to drop it (the caller thinks the file is going away forever).
1310  */
1311 int
1312 closef(struct file *fp, struct proc *p)
1313 {
1314 	struct vnode	*vp;
1315 	struct flock	lf;
1316 	int		error;
1317 
1318 	if (fp == NULL)
1319 		return (0);
1320 
1321 	/*
1322 	 * POSIX record locking dictates that any close releases ALL
1323 	 * locks owned by this process.  This is handled by setting
1324 	 * a flag in the unlock to free ONLY locks obeying POSIX
1325 	 * semantics, and not to free BSD-style file locks.
1326 	 * If the descriptor was in a message, POSIX-style locks
1327 	 * aren't passed with the descriptor.
1328 	 */
1329 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
1330 		lf.l_whence = SEEK_SET;
1331 		lf.l_start = 0;
1332 		lf.l_len = 0;
1333 		lf.l_type = F_UNLCK;
1334 		vp = (struct vnode *)fp->f_data;
1335 		(void) VOP_ADVLOCK(vp, p, F_UNLCK, &lf, F_POSIX);
1336 	}
1337 
1338 	/*
1339 	 * If WANTCLOSE is set, then the reference count on the file
1340 	 * is 0, but there were multiple users of the file.  This can
1341 	 * happen if a filedesc structure is shared by multiple
1342 	 * processes.
1343 	 */
1344 	simple_lock(&fp->f_slock);
1345 	if (fp->f_iflags & FIF_WANTCLOSE) {
1346 		/*
1347 		 * Another user of the file is already closing, and is
1348 		 * simply waiting for other users of the file to drain.
1349 		 * Release our usecount, and wake up the closer if it
1350 		 * is the only remaining use.
1351 		 */
1352 #ifdef DIAGNOSTIC
1353 		if (fp->f_count != 0)
1354 			panic("closef: wantclose and count != 0");
1355 		if (fp->f_usecount < 2)
1356 			panic("closef: wantclose and usecount < 2");
1357 #endif
1358 		if (--fp->f_usecount == 1)
1359 			wakeup(&fp->f_usecount);
1360 		simple_unlock(&fp->f_slock);
1361 		return (0);
1362 	} else {
1363 		/*
1364 		 * Decrement the reference count.  If we were not the
1365 		 * last reference, then release our use and just
1366 		 * return.
1367 		 */
1368 		if (--fp->f_count > 0) {
1369 #ifdef DIAGNOSTIC
1370 			if (fp->f_usecount < 1)
1371 				panic("closef: no wantclose and usecount < 1");
1372 #endif
1373 			fp->f_usecount--;
1374 			simple_unlock(&fp->f_slock);
1375 			return (0);
1376 		}
1377 	}
1378 
1379 	/*
1380 	 * The reference count is now 0.  However, there may be
1381 	 * multiple potential users of this file.  This can happen
1382 	 * if multiple processes shared a single filedesc structure.
1383 	 *
1384 	 * Notify these potential users that the file is closing.
1385 	 * This will prevent them from adding additional uses to
1386 	 * the file.
1387 	 */
1388 	fp->f_iflags |= FIF_WANTCLOSE;
1389 
1390 	/*
1391 	 * We expect the caller to add a use to the file.  So, if we
1392 	 * are the last user, usecount will be 1.  If it is not, we
1393 	 * must wait for the usecount to drain.  When it drains back
1394 	 * to 1, we will be awakened so that we may proceed with the
1395 	 * close.
1396 	 */
1397 #ifdef DIAGNOSTIC
1398 	if (fp->f_usecount < 1)
1399 		panic("closef: usecount < 1");
1400 #endif
1401 	while (fp->f_usecount > 1)
1402 		(void) ltsleep(&fp->f_usecount, PRIBIO, "closef", 0,
1403 				&fp->f_slock);
1404 #ifdef DIAGNOSTIC
1405 	if (fp->f_usecount != 1)
1406 		panic("closef: usecount != 1");
1407 #endif
1408 
1409 	simple_unlock(&fp->f_slock);
1410 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
1411 		lf.l_whence = SEEK_SET;
1412 		lf.l_start = 0;
1413 		lf.l_len = 0;
1414 		lf.l_type = F_UNLCK;
1415 		vp = (struct vnode *)fp->f_data;
1416 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1417 	}
1418 	if (fp->f_ops)
1419 		error = (*fp->f_ops->fo_close)(fp, p);
1420 	else
1421 		error = 0;
1422 
1423 	/* Nothing references the file now, drop the final use (us). */
1424 	fp->f_usecount--;
1425 
1426 	ffree(fp);
1427 	return (error);
1428 }
1429 
1430 /*
1431  * Apply an advisory lock on a file descriptor.
1432  *
1433  * Just attempt to get a record lock of the requested type on
1434  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
1435  */
1436 /* ARGSUSED */
1437 int
1438 sys_flock(struct lwp *l, void *v, register_t *retval)
1439 {
1440 	struct sys_flock_args /* {
1441 		syscallarg(int)	fd;
1442 		syscallarg(int)	how;
1443 	} */ *uap = v;
1444 	int		fd, how, error;
1445 	struct proc	*p;
1446 	struct filedesc	*fdp;
1447 	struct file	*fp;
1448 	struct vnode	*vp;
1449 	struct flock	lf;
1450 
1451 	p = l->l_proc;
1452 	fd = SCARG(uap, fd);
1453 	how = SCARG(uap, how);
1454 	fdp = p->p_fd;
1455 	error = 0;
1456 
1457 	if ((fp = fd_getfile(fdp, fd)) == NULL)
1458 		return (EBADF);
1459 
1460 	FILE_USE(fp);
1461 
1462 	if (fp->f_type != DTYPE_VNODE) {
1463 		error = EOPNOTSUPP;
1464 		goto out;
1465 	}
1466 
1467 	vp = (struct vnode *)fp->f_data;
1468 	lf.l_whence = SEEK_SET;
1469 	lf.l_start = 0;
1470 	lf.l_len = 0;
1471 	if (how & LOCK_UN) {
1472 		lf.l_type = F_UNLCK;
1473 		fp->f_flag &= ~FHASLOCK;
1474 		error = VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1475 		goto out;
1476 	}
1477 	if (how & LOCK_EX)
1478 		lf.l_type = F_WRLCK;
1479 	else if (how & LOCK_SH)
1480 		lf.l_type = F_RDLCK;
1481 	else {
1482 		error = EINVAL;
1483 		goto out;
1484 	}
1485 	fp->f_flag |= FHASLOCK;
1486 	if (how & LOCK_NB)
1487 		error = VOP_ADVLOCK(vp, fp, F_SETLK, &lf, F_FLOCK);
1488 	else
1489 		error = VOP_ADVLOCK(vp, fp, F_SETLK, &lf,
1490 		    F_FLOCK|F_WAIT);
1491  out:
1492 	FILE_UNUSE(fp, p);
1493 	return (error);
1494 }
1495 
1496 /*
1497  * File Descriptor pseudo-device driver (/dev/fd/).
1498  *
1499  * Opening minor device N dup()s the file (if any) connected to file
1500  * descriptor N belonging to the calling process.  Note that this driver
1501  * consists of only the ``open()'' routine, because all subsequent
1502  * references to this file will be direct to the other driver.
1503  */
1504 /* ARGSUSED */
1505 int
1506 filedescopen(dev_t dev, int mode, int type, struct proc *p)
1507 {
1508 
1509 	/*
1510 	 * XXX Kludge: set dupfd to contain the value of the
1511 	 * the file descriptor being sought for duplication. The error
1512 	 * return ensures that the vnode for this device will be released
1513 	 * by vn_open. Open will detect this special error and take the
1514 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
1515 	 * will simply report the error.
1516 	 */
1517 	curlwp->l_dupfd = minor(dev);	/* XXX */
1518 	return (ENODEV);
1519 }
1520 
1521 /*
1522  * Duplicate the specified descriptor to a free descriptor.
1523  *
1524  * 'indx' has been fdalloc'ed (and will be fdremove'ed on error) by the caller.
1525  */
1526 int
1527 dupfdopen(struct proc *p, int indx, int dfd, int mode, int error)
1528 {
1529 	struct filedesc	*fdp;
1530 	struct file	*wfp;
1531 
1532 	fdp = p->p_fd;
1533 
1534 	/* should be cleared by the caller */
1535 	KASSERT(fdp->fd_ofiles[indx] == NULL);
1536 
1537 	/*
1538 	 * If the to-be-dup'd fd number is greater than the allowed number
1539 	 * of file descriptors, or the fd to be dup'd has already been
1540 	 * closed, reject.
1541 	 */
1542 
1543 	/*
1544 	 * Note, in the case of indx == dfd, fd_getfile below returns NULL.
1545 	 */
1546 	if ((wfp = fd_getfile(fdp, dfd)) == NULL)
1547 		return (EBADF);
1548 
1549 	FILE_USE(wfp);
1550 
1551 	/*
1552 	 * There are two cases of interest here.
1553 	 *
1554 	 * For ENODEV simply dup (dfd) to file descriptor
1555 	 * (indx) and return.
1556 	 *
1557 	 * For ENXIO steal away the file structure from (dfd) and
1558 	 * store it in (indx).  (dfd) is effectively closed by
1559 	 * this operation.
1560 	 *
1561 	 * Any other error code is just returned.
1562 	 */
1563 	switch (error) {
1564 	case ENODEV:
1565 		/*
1566 		 * Check that the mode the file is being opened for is a
1567 		 * subset of the mode of the existing descriptor.
1568 		 */
1569 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
1570 			FILE_UNUSE(wfp, p);
1571 			return (EACCES);
1572 		}
1573 		simple_lock(&fdp->fd_slock);
1574 		fdp->fd_ofiles[indx] = wfp;
1575 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1576 		simple_unlock(&fdp->fd_slock);
1577 		simple_lock(&wfp->f_slock);
1578 		wfp->f_count++;
1579 		/* 'indx' has been fd_used'ed by caller */
1580 		FILE_UNUSE_HAVELOCK(wfp, p);
1581 		return (0);
1582 
1583 	case ENXIO:
1584 		/*
1585 		 * Steal away the file pointer from dfd, and stuff it into indx.
1586 		 */
1587 		simple_lock(&fdp->fd_slock);
1588 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
1589 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1590 		fdp->fd_ofiles[dfd] = NULL;
1591 		fdp->fd_ofileflags[dfd] = 0;
1592 		/*
1593 		 * Complete the clean up of the filedesc structure by
1594 		 * recomputing the various hints.
1595 		 */
1596 		/* 'indx' has been fd_used'ed by caller */
1597 		fd_unused(fdp, dfd);
1598 		simple_unlock(&fdp->fd_slock);
1599 		FILE_UNUSE(wfp, p);
1600 		return (0);
1601 
1602 	default:
1603 		FILE_UNUSE(wfp, p);
1604 		return (error);
1605 	}
1606 	/* NOTREACHED */
1607 }
1608 
1609 /*
1610  * fcntl call which is being passed to the file's fs.
1611  */
1612 int
1613 fcntl_forfs(int fd, struct proc *p, int cmd, void *arg)
1614 {
1615 	struct file	*fp;
1616 	struct filedesc	*fdp;
1617 	int		error;
1618 	u_int		size;
1619 	void		*data, *memp;
1620 #define STK_PARAMS	128
1621 	char		stkbuf[STK_PARAMS];
1622 
1623 	/* fd's value was validated in sys_fcntl before calling this routine */
1624 	fdp = p->p_fd;
1625 	fp = fdp->fd_ofiles[fd];
1626 
1627 	if ((fp->f_flag & (FREAD | FWRITE)) == 0)
1628 		return (EBADF);
1629 
1630 	/*
1631 	 * Interpret high order word to find amount of data to be
1632 	 * copied to/from the user's address space.
1633 	 */
1634 	size = (size_t)F_PARAM_LEN(cmd);
1635 	if (size > F_PARAM_MAX)
1636 		return (EINVAL);
1637 	memp = NULL;
1638 	if (size > sizeof(stkbuf)) {
1639 		memp = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
1640 		data = memp;
1641 	} else
1642 		data = stkbuf;
1643 	if (cmd & F_FSIN) {
1644 		if (size) {
1645 			error = copyin(arg, data, size);
1646 			if (error) {
1647 				if (memp)
1648 					free(memp, M_IOCTLOPS);
1649 				return (error);
1650 			}
1651 		} else
1652 			*(void **)data = arg;
1653 	} else if ((cmd & F_FSOUT) && size)
1654 		/*
1655 		 * Zero the buffer so the user always
1656 		 * gets back something deterministic.
1657 		 */
1658 		memset(data, 0, size);
1659 	else if (cmd & F_FSVOID)
1660 		*(void **)data = arg;
1661 
1662 
1663 	error = (*fp->f_ops->fo_fcntl)(fp, cmd, data, p);
1664 
1665 	/*
1666 	 * Copy any data to user, size was
1667 	 * already set and checked above.
1668 	 */
1669 	if (error == 0 && (cmd & F_FSOUT) && size)
1670 		error = copyout(data, arg, size);
1671 	if (memp)
1672 		free(memp, M_IOCTLOPS);
1673 	return (error);
1674 }
1675 
1676 /*
1677  * Close any files on exec?
1678  */
1679 void
1680 fdcloseexec(struct proc *p)
1681 {
1682 	struct filedesc	*fdp;
1683 	int		fd;
1684 
1685 	fdunshare(p);
1686 	cwdunshare(p);
1687 
1688 	fdp = p->p_fd;
1689 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
1690 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
1691 			(void) fdrelease(p, fd);
1692 }
1693 
1694 /*
1695  * It is unsafe for set[ug]id processes to be started with file
1696  * descriptors 0..2 closed, as these descriptors are given implicit
1697  * significance in the Standard C library.  fdcheckstd() will create a
1698  * descriptor referencing /dev/null for each of stdin, stdout, and
1699  * stderr that is not already open.
1700  */
1701 #define CHECK_UPTO 3
1702 int
1703 fdcheckstd(p)
1704 	struct proc *p;
1705 {
1706 	struct nameidata nd;
1707 	struct filedesc *fdp;
1708 	struct file *fp;
1709 	struct file *devnullfp = NULL;	/* Quell compiler warning */
1710 	struct proc *pp;
1711 	register_t retval;
1712 	int fd, i, error, flags = FREAD|FWRITE, devnull = -1;
1713 	char closed[CHECK_UPTO * 3 + 1], which[3 + 1];
1714 
1715 	closed[0] = '\0';
1716 	if ((fdp = p->p_fd) == NULL)
1717 		return (0);
1718 	for (i = 0; i < CHECK_UPTO; i++) {
1719 		if (fdp->fd_ofiles[i] != NULL)
1720 			continue;
1721 		snprintf(which, sizeof(which), ",%d", i);
1722 		strlcat(closed, which, sizeof(closed));
1723 		if (devnull < 0) {
1724 			if ((error = falloc(p, &fp, &fd)) != 0)
1725 				return (error);
1726 			NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1727 			    p);
1728 			if ((error = vn_open(&nd, flags, 0)) != 0) {
1729 				FILE_UNUSE(fp, p);
1730 				ffree(fp);
1731 				fdremove(p->p_fd, fd);
1732 				return (error);
1733 			}
1734 			fp->f_data = nd.ni_vp;
1735 			fp->f_flag = flags;
1736 			fp->f_ops = &vnops;
1737 			fp->f_type = DTYPE_VNODE;
1738 			VOP_UNLOCK(nd.ni_vp, 0);
1739 			devnull = fd;
1740 			devnullfp = fp;
1741 			FILE_SET_MATURE(fp);
1742 		} else {
1743 restart:
1744 			if ((error = fdalloc(p, 0, &fd)) != 0) {
1745 				if (error == ENOSPC) {
1746 					fdexpand(p);
1747 					goto restart;
1748 				}
1749 				return (error);
1750 			}
1751 
1752 			simple_lock(&devnullfp->f_slock);
1753 			FILE_USE(devnullfp);
1754 			/* finishdup() will unuse the descriptors for us */
1755 			if ((error = finishdup(p, devnull, fd, &retval)) != 0)
1756 				return (error);
1757 		}
1758 	}
1759 	if (devnullfp)
1760 		FILE_UNUSE(devnullfp, p);
1761 	if (closed[0] != '\0') {
1762 		pp = p->p_pptr;
1763 		log(LOG_WARNING, "set{u,g}id pid %d (%s) "
1764 		    "was invoked by uid %d ppid %d (%s) "
1765 		    "with fd %s closed\n",
1766 		    p->p_pid, p->p_comm, pp->p_ucred->cr_uid,
1767 		    pp->p_pid, pp->p_comm, &closed[1]);
1768 	}
1769 	return (0);
1770 }
1771 #undef CHECK_UPTO
1772 
1773 /*
1774  * Sets descriptor owner. If the owner is a process, 'pgid'
1775  * is set to positive value, process ID. If the owner is process group,
1776  * 'pgid' is set to -pg_id.
1777  */
1778 int
1779 fsetown(struct proc *p, pid_t *pgid, int cmd, const void *data)
1780 {
1781 	int id = *(int *)data;
1782 	int error;
1783 
1784 	switch (cmd) {
1785 	case TIOCSPGRP:
1786 		if (id < 0)
1787 			return (EINVAL);
1788 		id = -id;
1789 		break;
1790 	default:
1791 		break;
1792 	}
1793 
1794 	if (id > 0 && !pfind(id))
1795 		return (ESRCH);
1796 	else if (id < 0 && (error = pgid_in_session(p, -id)))
1797 		return (error);
1798 
1799 	*pgid = id;
1800 	return (0);
1801 }
1802 
1803 /*
1804  * Return descriptor owner information. If the value is positive,
1805  * it's process ID. If it's negative, it's process group ID and
1806  * needs the sign removed before use.
1807  */
1808 int
1809 fgetown(struct proc *p, pid_t pgid, int cmd, void *data)
1810 {
1811 	switch (cmd) {
1812 	case TIOCGPGRP:
1813 		*(int *)data = -pgid;
1814 		break;
1815 	default:
1816 		*(int *)data = pgid;
1817 		break;
1818 	}
1819 	return (0);
1820 }
1821 
1822 /*
1823  * Send signal to descriptor owner, either process or process group.
1824  */
1825 void
1826 fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata)
1827 {
1828 	struct proc *p1;
1829 	ksiginfo_t ksi;
1830 
1831 	memset(&ksi, 0, sizeof(ksi));
1832 	ksi.ksi_signo = signo;
1833 	ksi.ksi_code = code;
1834 	ksi.ksi_band = band;
1835 
1836 	if (pgid > 0 && (p1 = pfind(pgid)))
1837 		kpsignal(p1, &ksi, fdescdata);
1838 	else if (pgid < 0)
1839 		kgsignal(-pgid, &ksi, fdescdata);
1840 }
1841