xref: /freebsd/sys/kern/kern_descrip.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_ddb.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 
46 #include <sys/conf.h>
47 #include <sys/domain.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/filedesc.h>
51 #include <sys/filio.h>
52 #include <sys/jail.h>
53 #include <sys/kernel.h>
54 #include <sys/limits.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mount.h>
58 #include <sys/mqueue.h>
59 #include <sys/mutex.h>
60 #include <sys/namei.h>
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/protosw.h>
64 #include <sys/resourcevar.h>
65 #include <sys/signalvar.h>
66 #include <sys/socketvar.h>
67 #include <sys/stat.h>
68 #include <sys/sx.h>
69 #include <sys/syscallsubr.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysproto.h>
72 #include <sys/unistd.h>
73 #include <sys/user.h>
74 #include <sys/vnode.h>
75 
76 #include <security/audit/audit.h>
77 
78 #include <vm/uma.h>
79 
80 #include <ddb/ddb.h>
81 
82 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
83 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
84 		     "file desc to leader structures");
85 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
86 
87 static uma_zone_t file_zone;
88 
89 
90 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
91 enum dup_type { DUP_VARIABLE, DUP_FIXED };
92 
93 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
94     register_t *retval);
95 static int	fd_first_free(struct filedesc *, int, int);
96 static int	fd_last_used(struct filedesc *, int, int);
97 static void	fdgrowtable(struct filedesc *, int);
98 static void	fdunused(struct filedesc *fdp, int fd);
99 static void	fdused(struct filedesc *fdp, int fd);
100 
101 /*
102  * A process is initially started out with NDFILE descriptors stored within
103  * this structure, selected to be enough for typical applications based on
104  * the historical limit of 20 open files (and the usage of descriptors by
105  * shells).  If these descriptors are exhausted, a larger descriptor table
106  * may be allocated, up to a process' resource limit; the internal arrays
107  * are then unused.
108  */
109 #define NDFILE		20
110 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
111 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
112 #define NDSLOT(x)	((x) / NDENTRIES)
113 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
114 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
115 
116 /*
117  * Storage required per open file descriptor.
118  */
119 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
120 
121 /*
122  * Basic allocation of descriptors:
123  * one of the above, plus arrays for NDFILE descriptors.
124  */
125 struct filedesc0 {
126 	struct	filedesc fd_fd;
127 	/*
128 	 * These arrays are used when the number of open files is
129 	 * <= NDFILE, and are then pointed to by the pointers above.
130 	 */
131 	struct	file *fd_dfiles[NDFILE];
132 	char	fd_dfileflags[NDFILE];
133 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
134 };
135 
136 /*
137  * Descriptor management.
138  */
139 volatile int openfiles;			/* actual number of open files */
140 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
141 void	(*mq_fdclose)(struct thread *td, int fd, struct file *fp);
142 
143 /* A mutex to protect the association between a proc and filedesc. */
144 static struct mtx	fdesc_mtx;
145 
146 /*
147  * Find the first zero bit in the given bitmap, starting at low and not
148  * exceeding size - 1.
149  */
150 static int
151 fd_first_free(struct filedesc *fdp, int low, int size)
152 {
153 	NDSLOTTYPE *map = fdp->fd_map;
154 	NDSLOTTYPE mask;
155 	int off, maxoff;
156 
157 	if (low >= size)
158 		return (low);
159 
160 	off = NDSLOT(low);
161 	if (low % NDENTRIES) {
162 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
163 		if ((mask &= ~map[off]) != 0UL)
164 			return (off * NDENTRIES + ffsl(mask) - 1);
165 		++off;
166 	}
167 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
168 		if (map[off] != ~0UL)
169 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
170 	return (size);
171 }
172 
173 /*
174  * Find the highest non-zero bit in the given bitmap, starting at low and
175  * not exceeding size - 1.
176  */
177 static int
178 fd_last_used(struct filedesc *fdp, int low, int size)
179 {
180 	NDSLOTTYPE *map = fdp->fd_map;
181 	NDSLOTTYPE mask;
182 	int off, minoff;
183 
184 	if (low >= size)
185 		return (-1);
186 
187 	off = NDSLOT(size);
188 	if (size % NDENTRIES) {
189 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
190 		if ((mask &= map[off]) != 0)
191 			return (off * NDENTRIES + flsl(mask) - 1);
192 		--off;
193 	}
194 	for (minoff = NDSLOT(low); off >= minoff; --off)
195 		if (map[off] != 0)
196 			return (off * NDENTRIES + flsl(map[off]) - 1);
197 	return (low - 1);
198 }
199 
200 static int
201 fdisused(struct filedesc *fdp, int fd)
202 {
203         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
204             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
205 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
206 }
207 
208 /*
209  * Mark a file descriptor as used.
210  */
211 static void
212 fdused(struct filedesc *fdp, int fd)
213 {
214 
215 	FILEDESC_XLOCK_ASSERT(fdp);
216 	KASSERT(!fdisused(fdp, fd),
217 	    ("fd already used"));
218 
219 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
220 	if (fd > fdp->fd_lastfile)
221 		fdp->fd_lastfile = fd;
222 	if (fd == fdp->fd_freefile)
223 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
224 }
225 
226 /*
227  * Mark a file descriptor as unused.
228  */
229 static void
230 fdunused(struct filedesc *fdp, int fd)
231 {
232 
233 	FILEDESC_XLOCK_ASSERT(fdp);
234 	KASSERT(fdisused(fdp, fd),
235 	    ("fd is already unused"));
236 	KASSERT(fdp->fd_ofiles[fd] == NULL,
237 	    ("fd is still in use"));
238 
239 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
240 	if (fd < fdp->fd_freefile)
241 		fdp->fd_freefile = fd;
242 	if (fd == fdp->fd_lastfile)
243 		fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
244 }
245 
246 /*
247  * System calls on descriptors.
248  */
249 #ifndef _SYS_SYSPROTO_H_
250 struct getdtablesize_args {
251 	int	dummy;
252 };
253 #endif
254 /* ARGSUSED */
255 int
256 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
257 {
258 	struct proc *p = td->td_proc;
259 
260 	PROC_LOCK(p);
261 	td->td_retval[0] =
262 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
263 	PROC_UNLOCK(p);
264 	return (0);
265 }
266 
267 /*
268  * Duplicate a file descriptor to a particular value.
269  *
270  * Note: keep in mind that a potential race condition exists when closing
271  * descriptors from a shared descriptor table (via rfork).
272  */
273 #ifndef _SYS_SYSPROTO_H_
274 struct dup2_args {
275 	u_int	from;
276 	u_int	to;
277 };
278 #endif
279 /* ARGSUSED */
280 int
281 dup2(struct thread *td, struct dup2_args *uap)
282 {
283 
284 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
285 		    td->td_retval));
286 }
287 
288 /*
289  * Duplicate a file descriptor.
290  */
291 #ifndef _SYS_SYSPROTO_H_
292 struct dup_args {
293 	u_int	fd;
294 };
295 #endif
296 /* ARGSUSED */
297 int
298 dup(struct thread *td, struct dup_args *uap)
299 {
300 
301 	return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
302 }
303 
304 /*
305  * The file control system call.
306  */
307 #ifndef _SYS_SYSPROTO_H_
308 struct fcntl_args {
309 	int	fd;
310 	int	cmd;
311 	long	arg;
312 };
313 #endif
314 /* ARGSUSED */
315 int
316 fcntl(struct thread *td, struct fcntl_args *uap)
317 {
318 	struct flock fl;
319 	intptr_t arg;
320 	int error;
321 
322 	error = 0;
323 	switch (uap->cmd) {
324 	case F_GETLK:
325 	case F_SETLK:
326 	case F_SETLKW:
327 		error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
328 		arg = (intptr_t)&fl;
329 		break;
330 	default:
331 		arg = uap->arg;
332 		break;
333 	}
334 	if (error)
335 		return (error);
336 	error = kern_fcntl(td, uap->fd, uap->cmd, arg);
337 	if (error)
338 		return (error);
339 	if (uap->cmd == F_GETLK)
340 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
341 	return (error);
342 }
343 
344 static inline struct file *
345 fdtofp(int fd, struct filedesc *fdp)
346 {
347 	struct file *fp;
348 
349 	FILEDESC_LOCK_ASSERT(fdp);
350 	if ((unsigned)fd >= fdp->fd_nfiles ||
351 	    (fp = fdp->fd_ofiles[fd]) == NULL)
352 		return (NULL);
353 	return (fp);
354 }
355 
356 int
357 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
358 {
359 	struct filedesc *fdp;
360 	struct flock *flp;
361 	struct file *fp;
362 	struct proc *p;
363 	char *pop;
364 	struct vnode *vp;
365 	u_int newmin;
366 	int error, flg, tmp;
367 	int vfslocked;
368 
369 	vfslocked = 0;
370 	error = 0;
371 	flg = F_POSIX;
372 	p = td->td_proc;
373 	fdp = p->p_fd;
374 
375 	switch (cmd) {
376 	case F_DUPFD:
377 		FILEDESC_SLOCK(fdp);
378 		if ((fp = fdtofp(fd, fdp)) == NULL) {
379 			FILEDESC_SUNLOCK(fdp);
380 			error = EBADF;
381 			break;
382 		}
383 		FILEDESC_SUNLOCK(fdp);
384 		newmin = arg;
385 		PROC_LOCK(p);
386 		if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
387 		    newmin >= maxfilesperproc) {
388 			PROC_UNLOCK(p);
389 			error = EINVAL;
390 			break;
391 		}
392 		PROC_UNLOCK(p);
393 		error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
394 		break;
395 
396 	case F_GETFD:
397 		FILEDESC_SLOCK(fdp);
398 		if ((fp = fdtofp(fd, fdp)) == NULL) {
399 			FILEDESC_SUNLOCK(fdp);
400 			error = EBADF;
401 			break;
402 		}
403 		pop = &fdp->fd_ofileflags[fd];
404 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
405 		FILEDESC_SUNLOCK(fdp);
406 		break;
407 
408 	case F_SETFD:
409 		FILEDESC_XLOCK(fdp);
410 		if ((fp = fdtofp(fd, fdp)) == NULL) {
411 			FILEDESC_XUNLOCK(fdp);
412 			error = EBADF;
413 			break;
414 		}
415 		pop = &fdp->fd_ofileflags[fd];
416 		*pop = (*pop &~ UF_EXCLOSE) |
417 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
418 		FILEDESC_XUNLOCK(fdp);
419 		break;
420 
421 	case F_GETFL:
422 		FILEDESC_SLOCK(fdp);
423 		if ((fp = fdtofp(fd, fdp)) == NULL) {
424 			FILEDESC_SUNLOCK(fdp);
425 			error = EBADF;
426 			break;
427 		}
428 		td->td_retval[0] = OFLAGS(fp->f_flag);
429 		FILEDESC_SUNLOCK(fdp);
430 		break;
431 
432 	case F_SETFL:
433 		FILEDESC_SLOCK(fdp);
434 		if ((fp = fdtofp(fd, fdp)) == NULL) {
435 			FILEDESC_SUNLOCK(fdp);
436 			error = EBADF;
437 			break;
438 		}
439 		fhold(fp);
440 		FILEDESC_SUNLOCK(fdp);
441 		do {
442 			tmp = flg = fp->f_flag;
443 			tmp &= ~FCNTLFLAGS;
444 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
445 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
446 		tmp = fp->f_flag & FNONBLOCK;
447 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
448 		if (error) {
449 			fdrop(fp, td);
450 			break;
451 		}
452 		tmp = fp->f_flag & FASYNC;
453 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
454 		if (error == 0) {
455 			fdrop(fp, td);
456 			break;
457 		}
458 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
459 		tmp = 0;
460 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
461 		fdrop(fp, td);
462 		break;
463 
464 	case F_GETOWN:
465 		FILEDESC_SLOCK(fdp);
466 		if ((fp = fdtofp(fd, fdp)) == NULL) {
467 			FILEDESC_SUNLOCK(fdp);
468 			error = EBADF;
469 			break;
470 		}
471 		fhold(fp);
472 		FILEDESC_SUNLOCK(fdp);
473 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
474 		if (error == 0)
475 			td->td_retval[0] = tmp;
476 		fdrop(fp, td);
477 		break;
478 
479 	case F_SETOWN:
480 		FILEDESC_SLOCK(fdp);
481 		if ((fp = fdtofp(fd, fdp)) == NULL) {
482 			FILEDESC_SUNLOCK(fdp);
483 			error = EBADF;
484 			break;
485 		}
486 		fhold(fp);
487 		FILEDESC_SUNLOCK(fdp);
488 		tmp = arg;
489 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
490 		fdrop(fp, td);
491 		break;
492 
493 	case F_SETLKW:
494 		flg |= F_WAIT;
495 		/* FALLTHROUGH F_SETLK */
496 
497 	case F_SETLK:
498 		FILEDESC_SLOCK(fdp);
499 		if ((fp = fdtofp(fd, fdp)) == NULL) {
500 			FILEDESC_SUNLOCK(fdp);
501 			error = EBADF;
502 			break;
503 		}
504 		if (fp->f_type != DTYPE_VNODE) {
505 			FILEDESC_SUNLOCK(fdp);
506 			error = EBADF;
507 			break;
508 		}
509 		flp = (struct flock *)arg;
510 		if (flp->l_whence == SEEK_CUR) {
511 			if (fp->f_offset < 0 ||
512 			    (flp->l_start > 0 &&
513 			     fp->f_offset > OFF_MAX - flp->l_start)) {
514 				FILEDESC_SUNLOCK(fdp);
515 				error = EOVERFLOW;
516 				break;
517 			}
518 			flp->l_start += fp->f_offset;
519 		}
520 
521 		/*
522 		 * VOP_ADVLOCK() may block.
523 		 */
524 		fhold(fp);
525 		FILEDESC_SUNLOCK(fdp);
526 		vp = fp->f_vnode;
527 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
528 		switch (flp->l_type) {
529 		case F_RDLCK:
530 			if ((fp->f_flag & FREAD) == 0) {
531 				error = EBADF;
532 				break;
533 			}
534 			PROC_LOCK(p->p_leader);
535 			p->p_leader->p_flag |= P_ADVLOCK;
536 			PROC_UNLOCK(p->p_leader);
537 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
538 			    flp, flg);
539 			break;
540 		case F_WRLCK:
541 			if ((fp->f_flag & FWRITE) == 0) {
542 				error = EBADF;
543 				break;
544 			}
545 			PROC_LOCK(p->p_leader);
546 			p->p_leader->p_flag |= P_ADVLOCK;
547 			PROC_UNLOCK(p->p_leader);
548 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
549 			    flp, flg);
550 			break;
551 		case F_UNLCK:
552 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
553 			    flp, F_POSIX);
554 			break;
555 		default:
556 			error = EINVAL;
557 			break;
558 		}
559 		VFS_UNLOCK_GIANT(vfslocked);
560 		vfslocked = 0;
561 		/* Check for race with close */
562 		FILEDESC_SLOCK(fdp);
563 		if ((unsigned) fd >= fdp->fd_nfiles ||
564 		    fp != fdp->fd_ofiles[fd]) {
565 			FILEDESC_SUNLOCK(fdp);
566 			flp->l_whence = SEEK_SET;
567 			flp->l_start = 0;
568 			flp->l_len = 0;
569 			flp->l_type = F_UNLCK;
570 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
571 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
572 					   F_UNLCK, flp, F_POSIX);
573 			VFS_UNLOCK_GIANT(vfslocked);
574 			vfslocked = 0;
575 		} else
576 			FILEDESC_SUNLOCK(fdp);
577 		fdrop(fp, td);
578 		break;
579 
580 	case F_GETLK:
581 		FILEDESC_SLOCK(fdp);
582 		if ((fp = fdtofp(fd, fdp)) == NULL) {
583 			FILEDESC_SUNLOCK(fdp);
584 			error = EBADF;
585 			break;
586 		}
587 		if (fp->f_type != DTYPE_VNODE) {
588 			FILEDESC_SUNLOCK(fdp);
589 			error = EBADF;
590 			break;
591 		}
592 		flp = (struct flock *)arg;
593 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
594 		    flp->l_type != F_UNLCK) {
595 			FILEDESC_SUNLOCK(fdp);
596 			error = EINVAL;
597 			break;
598 		}
599 		if (flp->l_whence == SEEK_CUR) {
600 			if ((flp->l_start > 0 &&
601 			    fp->f_offset > OFF_MAX - flp->l_start) ||
602 			    (flp->l_start < 0 &&
603 			     fp->f_offset < OFF_MIN - flp->l_start)) {
604 				FILEDESC_SUNLOCK(fdp);
605 				error = EOVERFLOW;
606 				break;
607 			}
608 			flp->l_start += fp->f_offset;
609 		}
610 		/*
611 		 * VOP_ADVLOCK() may block.
612 		 */
613 		fhold(fp);
614 		FILEDESC_SUNLOCK(fdp);
615 		vp = fp->f_vnode;
616 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
617 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
618 		    F_POSIX);
619 		VFS_UNLOCK_GIANT(vfslocked);
620 		vfslocked = 0;
621 		fdrop(fp, td);
622 		break;
623 	default:
624 		error = EINVAL;
625 		break;
626 	}
627 	VFS_UNLOCK_GIANT(vfslocked);
628 	return (error);
629 }
630 
631 /*
632  * Common code for dup, dup2, and fcntl(F_DUPFD).
633  */
634 static int
635 do_dup(struct thread *td, enum dup_type type, int old, int new,
636     register_t *retval)
637 {
638 	struct filedesc *fdp;
639 	struct proc *p;
640 	struct file *fp;
641 	struct file *delfp;
642 	int error, holdleaders, maxfd;
643 
644 	KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
645 	    ("invalid dup type %d", type));
646 
647 	p = td->td_proc;
648 	fdp = p->p_fd;
649 
650 	/*
651 	 * Verify we have a valid descriptor to dup from and possibly to
652 	 * dup to.
653 	 */
654 	if (old < 0 || new < 0)
655 		return (EBADF);
656 	PROC_LOCK(p);
657 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
658 	PROC_UNLOCK(p);
659 	if (new >= maxfd)
660 		return (EMFILE);
661 
662 	FILEDESC_XLOCK(fdp);
663 	if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
664 		FILEDESC_XUNLOCK(fdp);
665 		return (EBADF);
666 	}
667 	if (type == DUP_FIXED && old == new) {
668 		*retval = new;
669 		FILEDESC_XUNLOCK(fdp);
670 		return (0);
671 	}
672 	fp = fdp->fd_ofiles[old];
673 	fhold(fp);
674 
675 	/*
676 	 * If the caller specified a file descriptor, make sure the file
677 	 * table is large enough to hold it, and grab it.  Otherwise, just
678 	 * allocate a new descriptor the usual way.  Since the filedesc
679 	 * lock may be temporarily dropped in the process, we have to look
680 	 * out for a race.
681 	 */
682 	if (type == DUP_FIXED) {
683 		if (new >= fdp->fd_nfiles)
684 			fdgrowtable(fdp, new + 1);
685 		if (fdp->fd_ofiles[new] == NULL)
686 			fdused(fdp, new);
687 	} else {
688 		if ((error = fdalloc(td, new, &new)) != 0) {
689 			FILEDESC_XUNLOCK(fdp);
690 			fdrop(fp, td);
691 			return (error);
692 		}
693 	}
694 
695 	/*
696 	 * If the old file changed out from under us then treat it as a
697 	 * bad file descriptor.  Userland should do its own locking to
698 	 * avoid this case.
699 	 */
700 	if (fdp->fd_ofiles[old] != fp) {
701 		/* we've allocated a descriptor which we won't use */
702 		if (fdp->fd_ofiles[new] == NULL)
703 			fdunused(fdp, new);
704 		FILEDESC_XUNLOCK(fdp);
705 		fdrop(fp, td);
706 		return (EBADF);
707 	}
708 	KASSERT(old != new,
709 	    ("new fd is same as old"));
710 
711 	/*
712 	 * Save info on the descriptor being overwritten.  We cannot close
713 	 * it without introducing an ownership race for the slot, since we
714 	 * need to drop the filedesc lock to call closef().
715 	 *
716 	 * XXX this duplicates parts of close().
717 	 */
718 	delfp = fdp->fd_ofiles[new];
719 	holdleaders = 0;
720 	if (delfp != NULL) {
721 		if (td->td_proc->p_fdtol != NULL) {
722 			/*
723 			 * Ask fdfree() to sleep to ensure that all relevant
724 			 * process leaders can be traversed in closef().
725 			 */
726 			fdp->fd_holdleaderscount++;
727 			holdleaders = 1;
728 		}
729 	}
730 
731 	/*
732 	 * Duplicate the source descriptor
733 	 */
734 	fdp->fd_ofiles[new] = fp;
735 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
736 	if (new > fdp->fd_lastfile)
737 		fdp->fd_lastfile = new;
738 	*retval = new;
739 
740 	/*
741 	 * If we dup'd over a valid file, we now own the reference to it
742 	 * and must dispose of it using closef() semantics (as if a
743 	 * close() were performed on it).
744 	 *
745 	 * XXX this duplicates parts of close().
746 	 */
747 	if (delfp != NULL) {
748 		knote_fdclose(td, new);
749 		if (delfp->f_type == DTYPE_MQUEUE)
750 			mq_fdclose(td, new, delfp);
751 		FILEDESC_XUNLOCK(fdp);
752 		(void) closef(delfp, td);
753 		if (holdleaders) {
754 			FILEDESC_XLOCK(fdp);
755 			fdp->fd_holdleaderscount--;
756 			if (fdp->fd_holdleaderscount == 0 &&
757 			    fdp->fd_holdleaderswakeup != 0) {
758 				fdp->fd_holdleaderswakeup = 0;
759 				wakeup(&fdp->fd_holdleaderscount);
760 			}
761 			FILEDESC_XUNLOCK(fdp);
762 		}
763 	} else {
764 		FILEDESC_XUNLOCK(fdp);
765 	}
766 	return (0);
767 }
768 
769 /*
770  * If sigio is on the list associated with a process or process group,
771  * disable signalling from the device, remove sigio from the list and
772  * free sigio.
773  */
774 void
775 funsetown(struct sigio **sigiop)
776 {
777 	struct sigio *sigio;
778 
779 	SIGIO_LOCK();
780 	sigio = *sigiop;
781 	if (sigio == NULL) {
782 		SIGIO_UNLOCK();
783 		return;
784 	}
785 	*(sigio->sio_myref) = NULL;
786 	if ((sigio)->sio_pgid < 0) {
787 		struct pgrp *pg = (sigio)->sio_pgrp;
788 		PGRP_LOCK(pg);
789 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
790 			     sigio, sio_pgsigio);
791 		PGRP_UNLOCK(pg);
792 	} else {
793 		struct proc *p = (sigio)->sio_proc;
794 		PROC_LOCK(p);
795 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
796 			     sigio, sio_pgsigio);
797 		PROC_UNLOCK(p);
798 	}
799 	SIGIO_UNLOCK();
800 	crfree(sigio->sio_ucred);
801 	FREE(sigio, M_SIGIO);
802 }
803 
804 /*
805  * Free a list of sigio structures.
806  * We only need to lock the SIGIO_LOCK because we have made ourselves
807  * inaccessible to callers of fsetown and therefore do not need to lock
808  * the proc or pgrp struct for the list manipulation.
809  */
810 void
811 funsetownlst(struct sigiolst *sigiolst)
812 {
813 	struct proc *p;
814 	struct pgrp *pg;
815 	struct sigio *sigio;
816 
817 	sigio = SLIST_FIRST(sigiolst);
818 	if (sigio == NULL)
819 		return;
820 	p = NULL;
821 	pg = NULL;
822 
823 	/*
824 	 * Every entry of the list should belong
825 	 * to a single proc or pgrp.
826 	 */
827 	if (sigio->sio_pgid < 0) {
828 		pg = sigio->sio_pgrp;
829 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
830 	} else /* if (sigio->sio_pgid > 0) */ {
831 		p = sigio->sio_proc;
832 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
833 	}
834 
835 	SIGIO_LOCK();
836 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
837 		*(sigio->sio_myref) = NULL;
838 		if (pg != NULL) {
839 			KASSERT(sigio->sio_pgid < 0,
840 			    ("Proc sigio in pgrp sigio list"));
841 			KASSERT(sigio->sio_pgrp == pg,
842 			    ("Bogus pgrp in sigio list"));
843 			PGRP_LOCK(pg);
844 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
845 			    sio_pgsigio);
846 			PGRP_UNLOCK(pg);
847 		} else /* if (p != NULL) */ {
848 			KASSERT(sigio->sio_pgid > 0,
849 			    ("Pgrp sigio in proc sigio list"));
850 			KASSERT(sigio->sio_proc == p,
851 			    ("Bogus proc in sigio list"));
852 			PROC_LOCK(p);
853 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
854 			    sio_pgsigio);
855 			PROC_UNLOCK(p);
856 		}
857 		SIGIO_UNLOCK();
858 		crfree(sigio->sio_ucred);
859 		FREE(sigio, M_SIGIO);
860 		SIGIO_LOCK();
861 	}
862 	SIGIO_UNLOCK();
863 }
864 
865 /*
866  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
867  *
868  * After permission checking, add a sigio structure to the sigio list for
869  * the process or process group.
870  */
871 int
872 fsetown(pid_t pgid, struct sigio **sigiop)
873 {
874 	struct proc *proc;
875 	struct pgrp *pgrp;
876 	struct sigio *sigio;
877 	int ret;
878 
879 	if (pgid == 0) {
880 		funsetown(sigiop);
881 		return (0);
882 	}
883 
884 	ret = 0;
885 
886 	/* Allocate and fill in the new sigio out of locks. */
887 	MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
888 	sigio->sio_pgid = pgid;
889 	sigio->sio_ucred = crhold(curthread->td_ucred);
890 	sigio->sio_myref = sigiop;
891 
892 	sx_slock(&proctree_lock);
893 	if (pgid > 0) {
894 		proc = pfind(pgid);
895 		if (proc == NULL) {
896 			ret = ESRCH;
897 			goto fail;
898 		}
899 
900 		/*
901 		 * Policy - Don't allow a process to FSETOWN a process
902 		 * in another session.
903 		 *
904 		 * Remove this test to allow maximum flexibility or
905 		 * restrict FSETOWN to the current process or process
906 		 * group for maximum safety.
907 		 */
908 		PROC_UNLOCK(proc);
909 		if (proc->p_session != curthread->td_proc->p_session) {
910 			ret = EPERM;
911 			goto fail;
912 		}
913 
914 		pgrp = NULL;
915 	} else /* if (pgid < 0) */ {
916 		pgrp = pgfind(-pgid);
917 		if (pgrp == NULL) {
918 			ret = ESRCH;
919 			goto fail;
920 		}
921 		PGRP_UNLOCK(pgrp);
922 
923 		/*
924 		 * Policy - Don't allow a process to FSETOWN a process
925 		 * in another session.
926 		 *
927 		 * Remove this test to allow maximum flexibility or
928 		 * restrict FSETOWN to the current process or process
929 		 * group for maximum safety.
930 		 */
931 		if (pgrp->pg_session != curthread->td_proc->p_session) {
932 			ret = EPERM;
933 			goto fail;
934 		}
935 
936 		proc = NULL;
937 	}
938 	funsetown(sigiop);
939 	if (pgid > 0) {
940 		PROC_LOCK(proc);
941 		/*
942 		 * Since funsetownlst() is called without the proctree
943 		 * locked, we need to check for P_WEXIT.
944 		 * XXX: is ESRCH correct?
945 		 */
946 		if ((proc->p_flag & P_WEXIT) != 0) {
947 			PROC_UNLOCK(proc);
948 			ret = ESRCH;
949 			goto fail;
950 		}
951 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
952 		sigio->sio_proc = proc;
953 		PROC_UNLOCK(proc);
954 	} else {
955 		PGRP_LOCK(pgrp);
956 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
957 		sigio->sio_pgrp = pgrp;
958 		PGRP_UNLOCK(pgrp);
959 	}
960 	sx_sunlock(&proctree_lock);
961 	SIGIO_LOCK();
962 	*sigiop = sigio;
963 	SIGIO_UNLOCK();
964 	return (0);
965 
966 fail:
967 	sx_sunlock(&proctree_lock);
968 	crfree(sigio->sio_ucred);
969 	FREE(sigio, M_SIGIO);
970 	return (ret);
971 }
972 
973 /*
974  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
975  */
976 pid_t
977 fgetown(sigiop)
978 	struct sigio **sigiop;
979 {
980 	pid_t pgid;
981 
982 	SIGIO_LOCK();
983 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
984 	SIGIO_UNLOCK();
985 	return (pgid);
986 }
987 
988 /*
989  * Close a file descriptor.
990  */
991 #ifndef _SYS_SYSPROTO_H_
992 struct close_args {
993 	int     fd;
994 };
995 #endif
996 /* ARGSUSED */
997 int
998 close(td, uap)
999 	struct thread *td;
1000 	struct close_args *uap;
1001 {
1002 
1003 	return (kern_close(td, uap->fd));
1004 }
1005 
1006 int
1007 kern_close(td, fd)
1008 	struct thread *td;
1009 	int fd;
1010 {
1011 	struct filedesc *fdp;
1012 	struct file *fp;
1013 	int error;
1014 	int holdleaders;
1015 
1016 	error = 0;
1017 	holdleaders = 0;
1018 	fdp = td->td_proc->p_fd;
1019 
1020 	AUDIT_SYSCLOSE(td, fd);
1021 
1022 	FILEDESC_XLOCK(fdp);
1023 	if ((unsigned)fd >= fdp->fd_nfiles ||
1024 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
1025 		FILEDESC_XUNLOCK(fdp);
1026 		return (EBADF);
1027 	}
1028 	fdp->fd_ofiles[fd] = NULL;
1029 	fdp->fd_ofileflags[fd] = 0;
1030 	fdunused(fdp, fd);
1031 	if (td->td_proc->p_fdtol != NULL) {
1032 		/*
1033 		 * Ask fdfree() to sleep to ensure that all relevant
1034 		 * process leaders can be traversed in closef().
1035 		 */
1036 		fdp->fd_holdleaderscount++;
1037 		holdleaders = 1;
1038 	}
1039 
1040 	/*
1041 	 * We now hold the fp reference that used to be owned by the
1042 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1043 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1044 	 * added, and deleteing a knote for the new fd.
1045 	 */
1046 	knote_fdclose(td, fd);
1047 	if (fp->f_type == DTYPE_MQUEUE)
1048 		mq_fdclose(td, fd, fp);
1049 	FILEDESC_XUNLOCK(fdp);
1050 
1051 	error = closef(fp, td);
1052 	if (holdleaders) {
1053 		FILEDESC_XLOCK(fdp);
1054 		fdp->fd_holdleaderscount--;
1055 		if (fdp->fd_holdleaderscount == 0 &&
1056 		    fdp->fd_holdleaderswakeup != 0) {
1057 			fdp->fd_holdleaderswakeup = 0;
1058 			wakeup(&fdp->fd_holdleaderscount);
1059 		}
1060 		FILEDESC_XUNLOCK(fdp);
1061 	}
1062 	return (error);
1063 }
1064 
1065 #if defined(COMPAT_43)
1066 /*
1067  * Return status information about a file descriptor.
1068  */
1069 #ifndef _SYS_SYSPROTO_H_
1070 struct ofstat_args {
1071 	int	fd;
1072 	struct	ostat *sb;
1073 };
1074 #endif
1075 /* ARGSUSED */
1076 int
1077 ofstat(struct thread *td, struct ofstat_args *uap)
1078 {
1079 	struct ostat oub;
1080 	struct stat ub;
1081 	int error;
1082 
1083 	error = kern_fstat(td, uap->fd, &ub);
1084 	if (error == 0) {
1085 		cvtstat(&ub, &oub);
1086 		error = copyout(&oub, uap->sb, sizeof(oub));
1087 	}
1088 	return (error);
1089 }
1090 #endif /* COMPAT_43 */
1091 
1092 /*
1093  * Return status information about a file descriptor.
1094  */
1095 #ifndef _SYS_SYSPROTO_H_
1096 struct fstat_args {
1097 	int	fd;
1098 	struct	stat *sb;
1099 };
1100 #endif
1101 /* ARGSUSED */
1102 int
1103 fstat(struct thread *td, struct fstat_args *uap)
1104 {
1105 	struct stat ub;
1106 	int error;
1107 
1108 	error = kern_fstat(td, uap->fd, &ub);
1109 	if (error == 0)
1110 		error = copyout(&ub, uap->sb, sizeof(ub));
1111 	return (error);
1112 }
1113 
1114 int
1115 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1116 {
1117 	struct file *fp;
1118 	int error;
1119 
1120 	AUDIT_ARG(fd, fd);
1121 
1122 	if ((error = fget(td, fd, &fp)) != 0)
1123 		return (error);
1124 
1125 	AUDIT_ARG(file, td->td_proc, fp);
1126 
1127 	error = fo_stat(fp, sbp, td->td_ucred, td);
1128 	fdrop(fp, td);
1129 	return (error);
1130 }
1131 
1132 /*
1133  * Return status information about a file descriptor.
1134  */
1135 #ifndef _SYS_SYSPROTO_H_
1136 struct nfstat_args {
1137 	int	fd;
1138 	struct	nstat *sb;
1139 };
1140 #endif
1141 /* ARGSUSED */
1142 int
1143 nfstat(struct thread *td, struct nfstat_args *uap)
1144 {
1145 	struct nstat nub;
1146 	struct stat ub;
1147 	int error;
1148 
1149 	error = kern_fstat(td, uap->fd, &ub);
1150 	if (error == 0) {
1151 		cvtnstat(&ub, &nub);
1152 		error = copyout(&nub, uap->sb, sizeof(nub));
1153 	}
1154 	return (error);
1155 }
1156 
1157 /*
1158  * Return pathconf information about a file descriptor.
1159  */
1160 #ifndef _SYS_SYSPROTO_H_
1161 struct fpathconf_args {
1162 	int	fd;
1163 	int	name;
1164 };
1165 #endif
1166 /* ARGSUSED */
1167 int
1168 fpathconf(struct thread *td, struct fpathconf_args *uap)
1169 {
1170 	struct file *fp;
1171 	struct vnode *vp;
1172 	int error;
1173 
1174 	if ((error = fget(td, uap->fd, &fp)) != 0)
1175 		return (error);
1176 
1177 	/* If asynchronous I/O is available, it works for all descriptors. */
1178 	if (uap->name == _PC_ASYNC_IO) {
1179 		td->td_retval[0] = async_io_version;
1180 		goto out;
1181 	}
1182 	vp = fp->f_vnode;
1183 	if (vp != NULL) {
1184 		int vfslocked;
1185 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1186 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1187 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1188 		VOP_UNLOCK(vp, 0);
1189 		VFS_UNLOCK_GIANT(vfslocked);
1190 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1191 		if (uap->name != _PC_PIPE_BUF) {
1192 			error = EINVAL;
1193 		} else {
1194 			td->td_retval[0] = PIPE_BUF;
1195 		error = 0;
1196 		}
1197 	} else {
1198 		error = EOPNOTSUPP;
1199 	}
1200 out:
1201 	fdrop(fp, td);
1202 	return (error);
1203 }
1204 
1205 /*
1206  * Grow the file table to accomodate (at least) nfd descriptors.  This may
1207  * block and drop the filedesc lock, but it will reacquire it before
1208  * returning.
1209  */
1210 static void
1211 fdgrowtable(struct filedesc *fdp, int nfd)
1212 {
1213 	struct file **ntable;
1214 	char *nfileflags;
1215 	int nnfiles, onfiles;
1216 	NDSLOTTYPE *nmap;
1217 
1218 	FILEDESC_XLOCK_ASSERT(fdp);
1219 
1220 	KASSERT(fdp->fd_nfiles > 0,
1221 	    ("zero-length file table"));
1222 
1223 	/* compute the size of the new table */
1224 	onfiles = fdp->fd_nfiles;
1225 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1226 	if (nnfiles <= onfiles)
1227 		/* the table is already large enough */
1228 		return;
1229 
1230 	/* allocate a new table and (if required) new bitmaps */
1231 	FILEDESC_XUNLOCK(fdp);
1232 	MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1233 	    M_FILEDESC, M_ZERO | M_WAITOK);
1234 	nfileflags = (char *)&ntable[nnfiles];
1235 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1236 		MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1237 		    M_FILEDESC, M_ZERO | M_WAITOK);
1238 	else
1239 		nmap = NULL;
1240 	FILEDESC_XLOCK(fdp);
1241 
1242 	/*
1243 	 * We now have new tables ready to go.  Since we dropped the
1244 	 * filedesc lock to call malloc(), watch out for a race.
1245 	 */
1246 	onfiles = fdp->fd_nfiles;
1247 	if (onfiles >= nnfiles) {
1248 		/* we lost the race, but that's OK */
1249 		free(ntable, M_FILEDESC);
1250 		if (nmap != NULL)
1251 			free(nmap, M_FILEDESC);
1252 		return;
1253 	}
1254 	bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1255 	bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1256 	if (onfiles > NDFILE)
1257 		free(fdp->fd_ofiles, M_FILEDESC);
1258 	fdp->fd_ofiles = ntable;
1259 	fdp->fd_ofileflags = nfileflags;
1260 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1261 		bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1262 		if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1263 			free(fdp->fd_map, M_FILEDESC);
1264 		fdp->fd_map = nmap;
1265 	}
1266 	fdp->fd_nfiles = nnfiles;
1267 }
1268 
1269 /*
1270  * Allocate a file descriptor for the process.
1271  */
1272 int
1273 fdalloc(struct thread *td, int minfd, int *result)
1274 {
1275 	struct proc *p = td->td_proc;
1276 	struct filedesc *fdp = p->p_fd;
1277 	int fd = -1, maxfd;
1278 
1279 	FILEDESC_XLOCK_ASSERT(fdp);
1280 
1281 	if (fdp->fd_freefile > minfd)
1282 		minfd = fdp->fd_freefile;
1283 
1284 	PROC_LOCK(p);
1285 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1286 	PROC_UNLOCK(p);
1287 
1288 	/*
1289 	 * Search the bitmap for a free descriptor.  If none is found, try
1290 	 * to grow the file table.  Keep at it until we either get a file
1291 	 * descriptor or run into process or system limits; fdgrowtable()
1292 	 * may drop the filedesc lock, so we're in a race.
1293 	 */
1294 	for (;;) {
1295 		fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1296 		if (fd >= maxfd)
1297 			return (EMFILE);
1298 		if (fd < fdp->fd_nfiles)
1299 			break;
1300 		fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1301 	}
1302 
1303 	/*
1304 	 * Perform some sanity checks, then mark the file descriptor as
1305 	 * used and return it to the caller.
1306 	 */
1307 	KASSERT(!fdisused(fdp, fd),
1308 	    ("fd_first_free() returned non-free descriptor"));
1309 	KASSERT(fdp->fd_ofiles[fd] == NULL,
1310 	    ("free descriptor isn't"));
1311 	fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1312 	fdused(fdp, fd);
1313 	*result = fd;
1314 	return (0);
1315 }
1316 
1317 /*
1318  * Check to see whether n user file descriptors are available to the process
1319  * p.
1320  */
1321 int
1322 fdavail(struct thread *td, int n)
1323 {
1324 	struct proc *p = td->td_proc;
1325 	struct filedesc *fdp = td->td_proc->p_fd;
1326 	struct file **fpp;
1327 	int i, lim, last;
1328 
1329 	FILEDESC_LOCK_ASSERT(fdp);
1330 
1331 	PROC_LOCK(p);
1332 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1333 	PROC_UNLOCK(p);
1334 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1335 		return (1);
1336 	last = min(fdp->fd_nfiles, lim);
1337 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1338 	for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1339 		if (*fpp == NULL && --n <= 0)
1340 			return (1);
1341 	}
1342 	return (0);
1343 }
1344 
1345 /*
1346  * Create a new open file structure and allocate a file decriptor for the
1347  * process that refers to it.  We add one reference to the file for the
1348  * descriptor table and one reference for resultfp. This is to prevent us
1349  * being preempted and the entry in the descriptor table closed after we
1350  * release the FILEDESC lock.
1351  */
1352 int
1353 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1354 {
1355 	struct proc *p = td->td_proc;
1356 	struct file *fp;
1357 	int error, i;
1358 	int maxuserfiles = maxfiles - (maxfiles / 20);
1359 	static struct timeval lastfail;
1360 	static int curfail;
1361 
1362 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1363 	if ((openfiles >= maxuserfiles &&
1364 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1365 	    openfiles >= maxfiles) {
1366 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1367 			printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1368 				td->td_ucred->cr_ruid);
1369 		}
1370 		uma_zfree(file_zone, fp);
1371 		return (ENFILE);
1372 	}
1373 	atomic_add_int(&openfiles, 1);
1374 
1375 	/*
1376 	 * If the process has file descriptor zero open, add the new file
1377 	 * descriptor to the list of open files at that point, otherwise
1378 	 * put it at the front of the list of open files.
1379 	 */
1380 	fp->f_count = 1;
1381 	if (resultfp)
1382 		fp->f_count++;
1383 	fp->f_cred = crhold(td->td_ucred);
1384 	fp->f_ops = &badfileops;
1385 	fp->f_data = NULL;
1386 	fp->f_vnode = NULL;
1387 	FILEDESC_XLOCK(p->p_fd);
1388 	if ((error = fdalloc(td, 0, &i))) {
1389 		FILEDESC_XUNLOCK(p->p_fd);
1390 		fdrop(fp, td);
1391 		if (resultfp)
1392 			fdrop(fp, td);
1393 		return (error);
1394 	}
1395 	p->p_fd->fd_ofiles[i] = fp;
1396 	FILEDESC_XUNLOCK(p->p_fd);
1397 	if (resultfp)
1398 		*resultfp = fp;
1399 	if (resultfd)
1400 		*resultfd = i;
1401 	return (0);
1402 }
1403 
1404 /*
1405  * Build a new filedesc structure from another.
1406  * Copy the current, root, and jail root vnode references.
1407  */
1408 struct filedesc *
1409 fdinit(struct filedesc *fdp)
1410 {
1411 	struct filedesc0 *newfdp;
1412 
1413 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1414 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1415 	if (fdp != NULL) {
1416 		FILEDESC_XLOCK(fdp);
1417 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1418 		if (newfdp->fd_fd.fd_cdir)
1419 			VREF(newfdp->fd_fd.fd_cdir);
1420 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1421 		if (newfdp->fd_fd.fd_rdir)
1422 			VREF(newfdp->fd_fd.fd_rdir);
1423 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1424 		if (newfdp->fd_fd.fd_jdir)
1425 			VREF(newfdp->fd_fd.fd_jdir);
1426 		FILEDESC_XUNLOCK(fdp);
1427 	}
1428 
1429 	/* Create the file descriptor table. */
1430 	newfdp->fd_fd.fd_refcnt = 1;
1431 	newfdp->fd_fd.fd_holdcnt = 1;
1432 	newfdp->fd_fd.fd_cmask = CMASK;
1433 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1434 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1435 	newfdp->fd_fd.fd_nfiles = NDFILE;
1436 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1437 	newfdp->fd_fd.fd_lastfile = -1;
1438 	return (&newfdp->fd_fd);
1439 }
1440 
1441 static struct filedesc *
1442 fdhold(struct proc *p)
1443 {
1444 	struct filedesc *fdp;
1445 
1446 	mtx_lock(&fdesc_mtx);
1447 	fdp = p->p_fd;
1448 	if (fdp != NULL)
1449 		fdp->fd_holdcnt++;
1450 	mtx_unlock(&fdesc_mtx);
1451 	return (fdp);
1452 }
1453 
1454 static void
1455 fddrop(struct filedesc *fdp)
1456 {
1457 	int i;
1458 
1459 	mtx_lock(&fdesc_mtx);
1460 	i = --fdp->fd_holdcnt;
1461 	mtx_unlock(&fdesc_mtx);
1462 	if (i > 0)
1463 		return;
1464 
1465 	FILEDESC_LOCK_DESTROY(fdp);
1466 	FREE(fdp, M_FILEDESC);
1467 }
1468 
1469 /*
1470  * Share a filedesc structure.
1471  */
1472 struct filedesc *
1473 fdshare(struct filedesc *fdp)
1474 {
1475 
1476 	FILEDESC_XLOCK(fdp);
1477 	fdp->fd_refcnt++;
1478 	FILEDESC_XUNLOCK(fdp);
1479 	return (fdp);
1480 }
1481 
1482 /*
1483  * Unshare a filedesc structure, if necessary by making a copy
1484  */
1485 void
1486 fdunshare(struct proc *p, struct thread *td)
1487 {
1488 
1489 	FILEDESC_XLOCK(p->p_fd);
1490 	if (p->p_fd->fd_refcnt > 1) {
1491 		struct filedesc *tmp;
1492 
1493 		FILEDESC_XUNLOCK(p->p_fd);
1494 		tmp = fdcopy(p->p_fd);
1495 		fdfree(td);
1496 		p->p_fd = tmp;
1497 	} else
1498 		FILEDESC_XUNLOCK(p->p_fd);
1499 }
1500 
1501 /*
1502  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1503  * this is to ease callers, not catch errors.
1504  */
1505 struct filedesc *
1506 fdcopy(struct filedesc *fdp)
1507 {
1508 	struct filedesc *newfdp;
1509 	int i;
1510 
1511 	/* Certain daemons might not have file descriptors. */
1512 	if (fdp == NULL)
1513 		return (NULL);
1514 
1515 	newfdp = fdinit(fdp);
1516 	FILEDESC_SLOCK(fdp);
1517 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1518 		FILEDESC_SUNLOCK(fdp);
1519 		FILEDESC_XLOCK(newfdp);
1520 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1521 		FILEDESC_XUNLOCK(newfdp);
1522 		FILEDESC_SLOCK(fdp);
1523 	}
1524 	/* copy everything except kqueue descriptors */
1525 	newfdp->fd_freefile = -1;
1526 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1527 		if (fdisused(fdp, i) &&
1528 		    fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1529 			newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1530 			newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1531 			fhold(newfdp->fd_ofiles[i]);
1532 			newfdp->fd_lastfile = i;
1533 		} else {
1534 			if (newfdp->fd_freefile == -1)
1535 				newfdp->fd_freefile = i;
1536 		}
1537 	}
1538 	FILEDESC_SUNLOCK(fdp);
1539 	FILEDESC_XLOCK(newfdp);
1540 	for (i = 0; i <= newfdp->fd_lastfile; ++i)
1541 		if (newfdp->fd_ofiles[i] != NULL)
1542 			fdused(newfdp, i);
1543 	FILEDESC_XUNLOCK(newfdp);
1544 	FILEDESC_SLOCK(fdp);
1545 	if (newfdp->fd_freefile == -1)
1546 		newfdp->fd_freefile = i;
1547 	newfdp->fd_cmask = fdp->fd_cmask;
1548 	FILEDESC_SUNLOCK(fdp);
1549 	return (newfdp);
1550 }
1551 
1552 /*
1553  * Release a filedesc structure.
1554  */
1555 void
1556 fdfree(struct thread *td)
1557 {
1558 	struct filedesc *fdp;
1559 	struct file **fpp;
1560 	int i, locked;
1561 	struct filedesc_to_leader *fdtol;
1562 	struct file *fp;
1563 	struct vnode *cdir, *jdir, *rdir, *vp;
1564 	struct flock lf;
1565 
1566 	/* Certain daemons might not have file descriptors. */
1567 	fdp = td->td_proc->p_fd;
1568 	if (fdp == NULL)
1569 		return;
1570 
1571 	/* Check for special need to clear POSIX style locks */
1572 	fdtol = td->td_proc->p_fdtol;
1573 	if (fdtol != NULL) {
1574 		FILEDESC_XLOCK(fdp);
1575 		KASSERT(fdtol->fdl_refcount > 0,
1576 			("filedesc_to_refcount botch: fdl_refcount=%d",
1577 			 fdtol->fdl_refcount));
1578 		if (fdtol->fdl_refcount == 1 &&
1579 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1580 			for (i = 0, fpp = fdp->fd_ofiles;
1581 			     i <= fdp->fd_lastfile;
1582 			     i++, fpp++) {
1583 				if (*fpp == NULL ||
1584 				    (*fpp)->f_type != DTYPE_VNODE)
1585 					continue;
1586 				fp = *fpp;
1587 				fhold(fp);
1588 				FILEDESC_XUNLOCK(fdp);
1589 				lf.l_whence = SEEK_SET;
1590 				lf.l_start = 0;
1591 				lf.l_len = 0;
1592 				lf.l_type = F_UNLCK;
1593 				vp = fp->f_vnode;
1594 				locked = VFS_LOCK_GIANT(vp->v_mount);
1595 				(void) VOP_ADVLOCK(vp,
1596 						   (caddr_t)td->td_proc->
1597 						   p_leader,
1598 						   F_UNLCK,
1599 						   &lf,
1600 						   F_POSIX);
1601 				VFS_UNLOCK_GIANT(locked);
1602 				FILEDESC_XLOCK(fdp);
1603 				fdrop(fp, td);
1604 				fpp = fdp->fd_ofiles + i;
1605 			}
1606 		}
1607 	retry:
1608 		if (fdtol->fdl_refcount == 1) {
1609 			if (fdp->fd_holdleaderscount > 0 &&
1610 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1611 				/*
1612 				 * close() or do_dup() has cleared a reference
1613 				 * in a shared file descriptor table.
1614 				 */
1615 				fdp->fd_holdleaderswakeup = 1;
1616 				sx_sleep(&fdp->fd_holdleaderscount,
1617 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1618 				goto retry;
1619 			}
1620 			if (fdtol->fdl_holdcount > 0) {
1621 				/*
1622 				 * Ensure that fdtol->fdl_leader remains
1623 				 * valid in closef().
1624 				 */
1625 				fdtol->fdl_wakeup = 1;
1626 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1627 				    "fdlhold", 0);
1628 				goto retry;
1629 			}
1630 		}
1631 		fdtol->fdl_refcount--;
1632 		if (fdtol->fdl_refcount == 0 &&
1633 		    fdtol->fdl_holdcount == 0) {
1634 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1635 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1636 		} else
1637 			fdtol = NULL;
1638 		td->td_proc->p_fdtol = NULL;
1639 		FILEDESC_XUNLOCK(fdp);
1640 		if (fdtol != NULL)
1641 			FREE(fdtol, M_FILEDESC_TO_LEADER);
1642 	}
1643 	FILEDESC_XLOCK(fdp);
1644 	i = --fdp->fd_refcnt;
1645 	FILEDESC_XUNLOCK(fdp);
1646 	if (i > 0)
1647 		return;
1648 	/*
1649 	 * We are the last reference to the structure, so we can
1650 	 * safely assume it will not change out from under us.
1651 	 */
1652 	fpp = fdp->fd_ofiles;
1653 	for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1654 		if (*fpp)
1655 			(void) closef(*fpp, td);
1656 	}
1657 	FILEDESC_XLOCK(fdp);
1658 
1659 	/* XXX This should happen earlier. */
1660 	mtx_lock(&fdesc_mtx);
1661 	td->td_proc->p_fd = NULL;
1662 	mtx_unlock(&fdesc_mtx);
1663 
1664 	if (fdp->fd_nfiles > NDFILE)
1665 		FREE(fdp->fd_ofiles, M_FILEDESC);
1666 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1667 		FREE(fdp->fd_map, M_FILEDESC);
1668 
1669 	fdp->fd_nfiles = 0;
1670 
1671 	cdir = fdp->fd_cdir;
1672 	fdp->fd_cdir = NULL;
1673 	rdir = fdp->fd_rdir;
1674 	fdp->fd_rdir = NULL;
1675 	jdir = fdp->fd_jdir;
1676 	fdp->fd_jdir = NULL;
1677 	FILEDESC_XUNLOCK(fdp);
1678 
1679 	if (cdir) {
1680 		locked = VFS_LOCK_GIANT(cdir->v_mount);
1681 		vrele(cdir);
1682 		VFS_UNLOCK_GIANT(locked);
1683 	}
1684 	if (rdir) {
1685 		locked = VFS_LOCK_GIANT(rdir->v_mount);
1686 		vrele(rdir);
1687 		VFS_UNLOCK_GIANT(locked);
1688 	}
1689 	if (jdir) {
1690 		locked = VFS_LOCK_GIANT(jdir->v_mount);
1691 		vrele(jdir);
1692 		VFS_UNLOCK_GIANT(locked);
1693 	}
1694 
1695 	fddrop(fdp);
1696 }
1697 
1698 /*
1699  * For setugid programs, we don't want to people to use that setugidness
1700  * to generate error messages which write to a file which otherwise would
1701  * otherwise be off-limits to the process.  We check for filesystems where
1702  * the vnode can change out from under us after execve (like [lin]procfs).
1703  *
1704  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1705  * sufficient.  We also don't check for setugidness since we know we are.
1706  */
1707 static int
1708 is_unsafe(struct file *fp)
1709 {
1710 	if (fp->f_type == DTYPE_VNODE) {
1711 		struct vnode *vp = fp->f_vnode;
1712 
1713 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1714 			return (1);
1715 	}
1716 	return (0);
1717 }
1718 
1719 /*
1720  * Make this setguid thing safe, if at all possible.
1721  */
1722 void
1723 setugidsafety(struct thread *td)
1724 {
1725 	struct filedesc *fdp;
1726 	int i;
1727 
1728 	/* Certain daemons might not have file descriptors. */
1729 	fdp = td->td_proc->p_fd;
1730 	if (fdp == NULL)
1731 		return;
1732 
1733 	/*
1734 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1735 	 * we are blocked in a close.  Be careful!
1736 	 */
1737 	FILEDESC_XLOCK(fdp);
1738 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1739 		if (i > 2)
1740 			break;
1741 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1742 			struct file *fp;
1743 
1744 			knote_fdclose(td, i);
1745 			/*
1746 			 * NULL-out descriptor prior to close to avoid
1747 			 * a race while close blocks.
1748 			 */
1749 			fp = fdp->fd_ofiles[i];
1750 			fdp->fd_ofiles[i] = NULL;
1751 			fdp->fd_ofileflags[i] = 0;
1752 			fdunused(fdp, i);
1753 			FILEDESC_XUNLOCK(fdp);
1754 			(void) closef(fp, td);
1755 			FILEDESC_XLOCK(fdp);
1756 		}
1757 	}
1758 	FILEDESC_XUNLOCK(fdp);
1759 }
1760 
1761 /*
1762  * If a specific file object occupies a specific file descriptor, close the
1763  * file descriptor entry and drop a reference on the file object.  This is a
1764  * convenience function to handle a subsequent error in a function that calls
1765  * falloc() that handles the race that another thread might have closed the
1766  * file descriptor out from under the thread creating the file object.
1767  */
1768 void
1769 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1770 {
1771 
1772 	FILEDESC_XLOCK(fdp);
1773 	if (fdp->fd_ofiles[idx] == fp) {
1774 		fdp->fd_ofiles[idx] = NULL;
1775 		fdunused(fdp, idx);
1776 		FILEDESC_XUNLOCK(fdp);
1777 		fdrop(fp, td);
1778 	} else
1779 		FILEDESC_XUNLOCK(fdp);
1780 }
1781 
1782 /*
1783  * Close any files on exec?
1784  */
1785 void
1786 fdcloseexec(struct thread *td)
1787 {
1788 	struct filedesc *fdp;
1789 	int i;
1790 
1791 	/* Certain daemons might not have file descriptors. */
1792 	fdp = td->td_proc->p_fd;
1793 	if (fdp == NULL)
1794 		return;
1795 
1796 	FILEDESC_XLOCK(fdp);
1797 
1798 	/*
1799 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
1800 	 * may block and rip them out from under us.
1801 	 */
1802 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1803 		if (fdp->fd_ofiles[i] != NULL &&
1804 		    (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1805 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1806 			struct file *fp;
1807 
1808 			knote_fdclose(td, i);
1809 			/*
1810 			 * NULL-out descriptor prior to close to avoid
1811 			 * a race while close blocks.
1812 			 */
1813 			fp = fdp->fd_ofiles[i];
1814 			fdp->fd_ofiles[i] = NULL;
1815 			fdp->fd_ofileflags[i] = 0;
1816 			fdunused(fdp, i);
1817 			if (fp->f_type == DTYPE_MQUEUE)
1818 				mq_fdclose(td, i, fp);
1819 			FILEDESC_XUNLOCK(fdp);
1820 			(void) closef(fp, td);
1821 			FILEDESC_XLOCK(fdp);
1822 		}
1823 	}
1824 	FILEDESC_XUNLOCK(fdp);
1825 }
1826 
1827 /*
1828  * It is unsafe for set[ug]id processes to be started with file
1829  * descriptors 0..2 closed, as these descriptors are given implicit
1830  * significance in the Standard C library.  fdcheckstd() will create a
1831  * descriptor referencing /dev/null for each of stdin, stdout, and
1832  * stderr that is not already open.
1833  */
1834 int
1835 fdcheckstd(struct thread *td)
1836 {
1837 	struct filedesc *fdp;
1838 	register_t retval, save;
1839 	int i, error, devnull;
1840 
1841 	fdp = td->td_proc->p_fd;
1842 	if (fdp == NULL)
1843 		return (0);
1844 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1845 	devnull = -1;
1846 	error = 0;
1847 	for (i = 0; i < 3; i++) {
1848 		if (fdp->fd_ofiles[i] != NULL)
1849 			continue;
1850 		if (devnull < 0) {
1851 			save = td->td_retval[0];
1852 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1853 			    O_RDWR, 0);
1854 			devnull = td->td_retval[0];
1855 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
1856 			td->td_retval[0] = save;
1857 			if (error)
1858 				break;
1859 		} else {
1860 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1861 			if (error != 0)
1862 				break;
1863 		}
1864 	}
1865 	return (error);
1866 }
1867 
1868 /*
1869  * Internal form of close.  Decrement reference count on file structure.
1870  * Note: td may be NULL when closing a file that was being passed in a
1871  * message.
1872  *
1873  * XXXRW: Giant is not required for the caller, but often will be held; this
1874  * makes it moderately likely the Giant will be recursed in the VFS case.
1875  */
1876 int
1877 closef(struct file *fp, struct thread *td)
1878 {
1879 	struct vnode *vp;
1880 	struct flock lf;
1881 	struct filedesc_to_leader *fdtol;
1882 	struct filedesc *fdp;
1883 
1884 	/*
1885 	 * POSIX record locking dictates that any close releases ALL
1886 	 * locks owned by this process.  This is handled by setting
1887 	 * a flag in the unlock to free ONLY locks obeying POSIX
1888 	 * semantics, and not to free BSD-style file locks.
1889 	 * If the descriptor was in a message, POSIX-style locks
1890 	 * aren't passed with the descriptor, and the thread pointer
1891 	 * will be NULL.  Callers should be careful only to pass a
1892 	 * NULL thread pointer when there really is no owning
1893 	 * context that might have locks, or the locks will be
1894 	 * leaked.
1895 	 */
1896 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
1897 		int vfslocked;
1898 
1899 		vp = fp->f_vnode;
1900 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1901 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1902 			lf.l_whence = SEEK_SET;
1903 			lf.l_start = 0;
1904 			lf.l_len = 0;
1905 			lf.l_type = F_UNLCK;
1906 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1907 					   F_UNLCK, &lf, F_POSIX);
1908 		}
1909 		fdtol = td->td_proc->p_fdtol;
1910 		if (fdtol != NULL) {
1911 			/*
1912 			 * Handle special case where file descriptor table is
1913 			 * shared between multiple process leaders.
1914 			 */
1915 			fdp = td->td_proc->p_fd;
1916 			FILEDESC_XLOCK(fdp);
1917 			for (fdtol = fdtol->fdl_next;
1918 			     fdtol != td->td_proc->p_fdtol;
1919 			     fdtol = fdtol->fdl_next) {
1920 				if ((fdtol->fdl_leader->p_flag &
1921 				     P_ADVLOCK) == 0)
1922 					continue;
1923 				fdtol->fdl_holdcount++;
1924 				FILEDESC_XUNLOCK(fdp);
1925 				lf.l_whence = SEEK_SET;
1926 				lf.l_start = 0;
1927 				lf.l_len = 0;
1928 				lf.l_type = F_UNLCK;
1929 				vp = fp->f_vnode;
1930 				(void) VOP_ADVLOCK(vp,
1931 						   (caddr_t)fdtol->fdl_leader,
1932 						   F_UNLCK, &lf, F_POSIX);
1933 				FILEDESC_XLOCK(fdp);
1934 				fdtol->fdl_holdcount--;
1935 				if (fdtol->fdl_holdcount == 0 &&
1936 				    fdtol->fdl_wakeup != 0) {
1937 					fdtol->fdl_wakeup = 0;
1938 					wakeup(fdtol);
1939 				}
1940 			}
1941 			FILEDESC_XUNLOCK(fdp);
1942 		}
1943 		VFS_UNLOCK_GIANT(vfslocked);
1944 	}
1945 	return (fdrop(fp, td));
1946 }
1947 
1948 /*
1949  * Initialize the file pointer with the specified properties.
1950  *
1951  * The ops are set with release semantics to be certain that the flags, type,
1952  * and data are visible when ops is.  This is to prevent ops methods from being
1953  * called with bad data.
1954  */
1955 void
1956 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
1957 {
1958 	fp->f_data = data;
1959 	fp->f_flag = flag;
1960 	fp->f_type = type;
1961 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
1962 }
1963 
1964 
1965 /*
1966  * Extract the file pointer associated with the specified descriptor for the
1967  * current user process.
1968  *
1969  * If the descriptor doesn't exist, EBADF is returned.
1970  *
1971  * If the descriptor exists but doesn't match 'flags' then return EBADF for
1972  * read attempts and EINVAL for write attempts.
1973  *
1974  * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1975  * It should be dropped with fdrop().  If it is not set, then the refcount
1976  * will not be bumped however the thread's filedesc struct will be returned
1977  * locked (for fgetsock).
1978  *
1979  * If an error occured the non-zero error is returned and *fpp is set to
1980  * NULL.  Otherwise *fpp is set and zero is returned.
1981  */
1982 static __inline int
1983 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1984 {
1985 	struct filedesc *fdp;
1986 	struct file *fp;
1987 
1988 	*fpp = NULL;
1989 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1990 		return (EBADF);
1991 	FILEDESC_SLOCK(fdp);
1992 	if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1993 		FILEDESC_SUNLOCK(fdp);
1994 		return (EBADF);
1995 	}
1996 
1997 	/*
1998 	 * FREAD and FWRITE failure return EBADF as per POSIX.
1999 	 *
2000 	 * Only one flag, or 0, may be specified.
2001 	 */
2002 	if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2003 		FILEDESC_SUNLOCK(fdp);
2004 		return (EBADF);
2005 	}
2006 	if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2007 		FILEDESC_SUNLOCK(fdp);
2008 		return (EBADF);
2009 	}
2010 	if (hold) {
2011 		fhold(fp);
2012 		FILEDESC_SUNLOCK(fdp);
2013 	}
2014 	*fpp = fp;
2015 	return (0);
2016 }
2017 
2018 int
2019 fget(struct thread *td, int fd, struct file **fpp)
2020 {
2021 
2022 	return(_fget(td, fd, fpp, 0, 1));
2023 }
2024 
2025 int
2026 fget_read(struct thread *td, int fd, struct file **fpp)
2027 {
2028 
2029 	return(_fget(td, fd, fpp, FREAD, 1));
2030 }
2031 
2032 int
2033 fget_write(struct thread *td, int fd, struct file **fpp)
2034 {
2035 
2036 	return(_fget(td, fd, fpp, FWRITE, 1));
2037 }
2038 
2039 /*
2040  * Like fget() but loads the underlying vnode, or returns an error if the
2041  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2042  * never have VM objects.  The returned vnode will be vref()'d.
2043  *
2044  * XXX: what about the unused flags ?
2045  */
2046 static __inline int
2047 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2048 {
2049 	struct file *fp;
2050 	int error;
2051 
2052 	*vpp = NULL;
2053 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2054 		return (error);
2055 	if (fp->f_vnode == NULL) {
2056 		error = EINVAL;
2057 	} else {
2058 		*vpp = fp->f_vnode;
2059 		vref(*vpp);
2060 	}
2061 	FILEDESC_SUNLOCK(td->td_proc->p_fd);
2062 	return (error);
2063 }
2064 
2065 int
2066 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2067 {
2068 
2069 	return (_fgetvp(td, fd, vpp, 0));
2070 }
2071 
2072 int
2073 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2074 {
2075 
2076 	return (_fgetvp(td, fd, vpp, FREAD));
2077 }
2078 
2079 #ifdef notyet
2080 int
2081 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2082 {
2083 
2084 	return (_fgetvp(td, fd, vpp, FWRITE));
2085 }
2086 #endif
2087 
2088 /*
2089  * Like fget() but loads the underlying socket, or returns an error if the
2090  * descriptor does not represent a socket.
2091  *
2092  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2093  * in the future.
2094  *
2095  * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely
2096  * on their file descriptor reference to prevent the socket from being free'd
2097  * during use.
2098  */
2099 int
2100 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2101 {
2102 	struct file *fp;
2103 	int error;
2104 
2105 	*spp = NULL;
2106 	if (fflagp != NULL)
2107 		*fflagp = 0;
2108 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2109 		return (error);
2110 	if (fp->f_type != DTYPE_SOCKET) {
2111 		error = ENOTSOCK;
2112 	} else {
2113 		*spp = fp->f_data;
2114 		if (fflagp)
2115 			*fflagp = fp->f_flag;
2116 		SOCK_LOCK(*spp);
2117 		soref(*spp);
2118 		SOCK_UNLOCK(*spp);
2119 	}
2120 	FILEDESC_SUNLOCK(td->td_proc->p_fd);
2121 	return (error);
2122 }
2123 
2124 /*
2125  * Drop the reference count on the socket and XXX release the SX lock in the
2126  * future.  The last reference closes the socket.
2127  *
2128  * XXXRW: fputsock() is deprecated, see comment for fgetsock().
2129  */
2130 void
2131 fputsock(struct socket *so)
2132 {
2133 
2134 	ACCEPT_LOCK();
2135 	SOCK_LOCK(so);
2136 	sorele(so);
2137 }
2138 
2139 /*
2140  * Handle the last reference to a file being closed.
2141  */
2142 int
2143 _fdrop(struct file *fp, struct thread *td)
2144 {
2145 	int error;
2146 
2147 	error = 0;
2148 	if (fp->f_count != 0)
2149 		panic("fdrop: count %d", fp->f_count);
2150 	if (fp->f_ops != &badfileops)
2151 		error = fo_close(fp, td);
2152 	atomic_subtract_int(&openfiles, 1);
2153 	crfree(fp->f_cred);
2154 	uma_zfree(file_zone, fp);
2155 
2156 	return (error);
2157 }
2158 
2159 /*
2160  * Apply an advisory lock on a file descriptor.
2161  *
2162  * Just attempt to get a record lock of the requested type on the entire file
2163  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2164  */
2165 #ifndef _SYS_SYSPROTO_H_
2166 struct flock_args {
2167 	int	fd;
2168 	int	how;
2169 };
2170 #endif
2171 /* ARGSUSED */
2172 int
2173 flock(struct thread *td, struct flock_args *uap)
2174 {
2175 	struct file *fp;
2176 	struct vnode *vp;
2177 	struct flock lf;
2178 	int vfslocked;
2179 	int error;
2180 
2181 	if ((error = fget(td, uap->fd, &fp)) != 0)
2182 		return (error);
2183 	if (fp->f_type != DTYPE_VNODE) {
2184 		fdrop(fp, td);
2185 		return (EOPNOTSUPP);
2186 	}
2187 
2188 	vp = fp->f_vnode;
2189 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2190 	lf.l_whence = SEEK_SET;
2191 	lf.l_start = 0;
2192 	lf.l_len = 0;
2193 	if (uap->how & LOCK_UN) {
2194 		lf.l_type = F_UNLCK;
2195 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2196 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2197 		goto done2;
2198 	}
2199 	if (uap->how & LOCK_EX)
2200 		lf.l_type = F_WRLCK;
2201 	else if (uap->how & LOCK_SH)
2202 		lf.l_type = F_RDLCK;
2203 	else {
2204 		error = EBADF;
2205 		goto done2;
2206 	}
2207 	atomic_set_int(&fp->f_flag, FHASLOCK);
2208 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2209 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2210 done2:
2211 	fdrop(fp, td);
2212 	VFS_UNLOCK_GIANT(vfslocked);
2213 	return (error);
2214 }
2215 /*
2216  * Duplicate the specified descriptor to a free descriptor.
2217  */
2218 int
2219 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2220 {
2221 	struct file *wfp;
2222 	struct file *fp;
2223 
2224 	/*
2225 	 * If the to-be-dup'd fd number is greater than the allowed number
2226 	 * of file descriptors, or the fd to be dup'd has already been
2227 	 * closed, then reject.
2228 	 */
2229 	FILEDESC_XLOCK(fdp);
2230 	if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2231 	    (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2232 		FILEDESC_XUNLOCK(fdp);
2233 		return (EBADF);
2234 	}
2235 
2236 	/*
2237 	 * There are two cases of interest here.
2238 	 *
2239 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2240 	 *
2241 	 * For ENXIO steal away the file structure from (dfd) and store it in
2242 	 * (indx).  (dfd) is effectively closed by this operation.
2243 	 *
2244 	 * Any other error code is just returned.
2245 	 */
2246 	switch (error) {
2247 	case ENODEV:
2248 		/*
2249 		 * Check that the mode the file is being opened for is a
2250 		 * subset of the mode of the existing descriptor.
2251 		 */
2252 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2253 			FILEDESC_XUNLOCK(fdp);
2254 			return (EACCES);
2255 		}
2256 		fp = fdp->fd_ofiles[indx];
2257 		fdp->fd_ofiles[indx] = wfp;
2258 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2259 		if (fp == NULL)
2260 			fdused(fdp, indx);
2261 		fhold(wfp);
2262 		FILEDESC_XUNLOCK(fdp);
2263 		if (fp != NULL)
2264 			/*
2265 			 * We now own the reference to fp that the ofiles[]
2266 			 * array used to own.  Release it.
2267 			 */
2268 			fdrop(fp, td);
2269 		return (0);
2270 
2271 	case ENXIO:
2272 		/*
2273 		 * Steal away the file pointer from dfd and stuff it into indx.
2274 		 */
2275 		fp = fdp->fd_ofiles[indx];
2276 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2277 		fdp->fd_ofiles[dfd] = NULL;
2278 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2279 		fdp->fd_ofileflags[dfd] = 0;
2280 		fdunused(fdp, dfd);
2281 		if (fp == NULL)
2282 			fdused(fdp, indx);
2283 		FILEDESC_XUNLOCK(fdp);
2284 
2285 		/*
2286 		 * We now own the reference to fp that the ofiles[] array
2287 		 * used to own.  Release it.
2288 		 */
2289 		if (fp != NULL)
2290 			fdrop(fp, td);
2291 		return (0);
2292 
2293 	default:
2294 		FILEDESC_XUNLOCK(fdp);
2295 		return (error);
2296 	}
2297 	/* NOTREACHED */
2298 }
2299 
2300 /*
2301  * Scan all active processes to see if any of them have a current or root
2302  * directory of `olddp'. If so, replace them with the new mount point.
2303  */
2304 void
2305 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2306 {
2307 	struct filedesc *fdp;
2308 	struct proc *p;
2309 	int nrele;
2310 
2311 	if (vrefcnt(olddp) == 1)
2312 		return;
2313 	sx_slock(&allproc_lock);
2314 	FOREACH_PROC_IN_SYSTEM(p) {
2315 		fdp = fdhold(p);
2316 		if (fdp == NULL)
2317 			continue;
2318 		nrele = 0;
2319 		FILEDESC_XLOCK(fdp);
2320 		if (fdp->fd_cdir == olddp) {
2321 			vref(newdp);
2322 			fdp->fd_cdir = newdp;
2323 			nrele++;
2324 		}
2325 		if (fdp->fd_rdir == olddp) {
2326 			vref(newdp);
2327 			fdp->fd_rdir = newdp;
2328 			nrele++;
2329 		}
2330 		FILEDESC_XUNLOCK(fdp);
2331 		fddrop(fdp);
2332 		while (nrele--)
2333 			vrele(olddp);
2334 	}
2335 	sx_sunlock(&allproc_lock);
2336 	if (rootvnode == olddp) {
2337 		vrele(rootvnode);
2338 		vref(newdp);
2339 		rootvnode = newdp;
2340 	}
2341 }
2342 
2343 struct filedesc_to_leader *
2344 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2345 {
2346 	struct filedesc_to_leader *fdtol;
2347 
2348 	MALLOC(fdtol, struct filedesc_to_leader *,
2349 	       sizeof(struct filedesc_to_leader),
2350 	       M_FILEDESC_TO_LEADER,
2351 	       M_WAITOK);
2352 	fdtol->fdl_refcount = 1;
2353 	fdtol->fdl_holdcount = 0;
2354 	fdtol->fdl_wakeup = 0;
2355 	fdtol->fdl_leader = leader;
2356 	if (old != NULL) {
2357 		FILEDESC_XLOCK(fdp);
2358 		fdtol->fdl_next = old->fdl_next;
2359 		fdtol->fdl_prev = old;
2360 		old->fdl_next = fdtol;
2361 		fdtol->fdl_next->fdl_prev = fdtol;
2362 		FILEDESC_XUNLOCK(fdp);
2363 	} else {
2364 		fdtol->fdl_next = fdtol;
2365 		fdtol->fdl_prev = fdtol;
2366 	}
2367 	return (fdtol);
2368 }
2369 
2370 /*
2371  * Get file structures globally.
2372  */
2373 static int
2374 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2375 {
2376 	struct xfile xf;
2377 	struct filedesc *fdp;
2378 	struct file *fp;
2379 	struct proc *p;
2380 	int error, n;
2381 
2382 	error = sysctl_wire_old_buffer(req, 0);
2383 	if (error != 0)
2384 		return (error);
2385 	if (req->oldptr == NULL) {
2386 		n = 0;
2387 		sx_slock(&allproc_lock);
2388 		FOREACH_PROC_IN_SYSTEM(p) {
2389 			if (p->p_state == PRS_NEW)
2390 				continue;
2391 			fdp = fdhold(p);
2392 			if (fdp == NULL)
2393 				continue;
2394 			/* overestimates sparse tables. */
2395 			if (fdp->fd_lastfile > 0)
2396 				n += fdp->fd_lastfile;
2397 			fddrop(fdp);
2398 		}
2399 		sx_sunlock(&allproc_lock);
2400 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2401 	}
2402 	error = 0;
2403 	bzero(&xf, sizeof(xf));
2404 	xf.xf_size = sizeof(xf);
2405 	sx_slock(&allproc_lock);
2406 	FOREACH_PROC_IN_SYSTEM(p) {
2407 		if (p->p_state == PRS_NEW)
2408 			continue;
2409 		PROC_LOCK(p);
2410 		if (p_cansee(req->td, p) != 0) {
2411 			PROC_UNLOCK(p);
2412 			continue;
2413 		}
2414 		xf.xf_pid = p->p_pid;
2415 		xf.xf_uid = p->p_ucred->cr_uid;
2416 		PROC_UNLOCK(p);
2417 		fdp = fdhold(p);
2418 		if (fdp == NULL)
2419 			continue;
2420 		FILEDESC_SLOCK(fdp);
2421 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2422 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2423 				continue;
2424 			xf.xf_fd = n;
2425 			xf.xf_file = fp;
2426 			xf.xf_data = fp->f_data;
2427 			xf.xf_vnode = fp->f_vnode;
2428 			xf.xf_type = fp->f_type;
2429 			xf.xf_count = fp->f_count;
2430 			xf.xf_msgcount = 0;
2431 			xf.xf_offset = fp->f_offset;
2432 			xf.xf_flag = fp->f_flag;
2433 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2434 			if (error)
2435 				break;
2436 		}
2437 		FILEDESC_SUNLOCK(fdp);
2438 		fddrop(fdp);
2439 		if (error)
2440 			break;
2441 	}
2442 	sx_sunlock(&allproc_lock);
2443 	return (error);
2444 }
2445 
2446 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2447     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2448 
2449 /*
2450  * Get per-process file descriptors for use by procstat(1), et al.
2451  */
2452 static int
2453 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2454 {
2455 	char *fullpath, *freepath;
2456 	struct kinfo_file *kif;
2457 	struct filedesc *fdp;
2458 	int error, i, *name;
2459 	struct socket *so;
2460 	struct vnode *vp;
2461 	struct file *fp;
2462 	struct proc *p;
2463 	int vfslocked;
2464 
2465 	name = (int *)arg1;
2466 	if ((p = pfind((pid_t)name[0])) == NULL)
2467 		return (ESRCH);
2468 	if ((error = p_candebug(curthread, p))) {
2469 		PROC_UNLOCK(p);
2470 		return (error);
2471 	}
2472 	fdp = fdhold(p);
2473 	PROC_UNLOCK(p);
2474 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2475 	FILEDESC_SLOCK(fdp);
2476 	for (i = 0; i < fdp->fd_nfiles; i++) {
2477 		if ((fp = fdp->fd_ofiles[i]) == NULL)
2478 			continue;
2479 		bzero(kif, sizeof(*kif));
2480 		kif->kf_structsize = sizeof(*kif);
2481 		vp = NULL;
2482 		so = NULL;
2483 		kif->kf_fd = i;
2484 		switch (fp->f_type) {
2485 		case DTYPE_VNODE:
2486 			kif->kf_type = KF_TYPE_VNODE;
2487 			vp = fp->f_vnode;
2488 			break;
2489 
2490 		case DTYPE_SOCKET:
2491 			kif->kf_type = KF_TYPE_SOCKET;
2492 			so = fp->f_data;
2493 			break;
2494 
2495 		case DTYPE_PIPE:
2496 			kif->kf_type = KF_TYPE_PIPE;
2497 			break;
2498 
2499 		case DTYPE_FIFO:
2500 			kif->kf_type = KF_TYPE_FIFO;
2501 			vp = fp->f_vnode;
2502 			vref(vp);
2503 			break;
2504 
2505 		case DTYPE_KQUEUE:
2506 			kif->kf_type = KF_TYPE_KQUEUE;
2507 			break;
2508 
2509 		case DTYPE_CRYPTO:
2510 			kif->kf_type = KF_TYPE_CRYPTO;
2511 			break;
2512 
2513 		case DTYPE_MQUEUE:
2514 			kif->kf_type = KF_TYPE_MQUEUE;
2515 			break;
2516 
2517 		default:
2518 			kif->kf_type = KF_TYPE_UNKNOWN;
2519 			break;
2520 		}
2521 		kif->kf_ref_count = fp->f_count;
2522 		if (fp->f_flag & FREAD)
2523 			kif->kf_flags |= KF_FLAG_READ;
2524 		if (fp->f_flag & FWRITE)
2525 			kif->kf_flags |= KF_FLAG_WRITE;
2526 		if (fp->f_flag & FAPPEND)
2527 			kif->kf_flags |= KF_FLAG_APPEND;
2528 		if (fp->f_flag & FASYNC)
2529 			kif->kf_flags |= KF_FLAG_ASYNC;
2530 		if (fp->f_flag & FFSYNC)
2531 			kif->kf_flags |= KF_FLAG_FSYNC;
2532 		if (fp->f_flag & FNONBLOCK)
2533 			kif->kf_flags |= KF_FLAG_NONBLOCK;
2534 		if (fp->f_flag & O_DIRECT)
2535 			kif->kf_flags |= KF_FLAG_DIRECT;
2536 		if (fp->f_flag & FHASLOCK)
2537 			kif->kf_flags |= KF_FLAG_HASLOCK;
2538 		kif->kf_offset = fp->f_offset;
2539 		if (vp != NULL) {
2540 			vref(vp);
2541 			switch (vp->v_type) {
2542 			case VNON:
2543 				kif->kf_vnode_type = KF_VTYPE_VNON;
2544 				break;
2545 			case VREG:
2546 				kif->kf_vnode_type = KF_VTYPE_VREG;
2547 				break;
2548 			case VDIR:
2549 				kif->kf_vnode_type = KF_VTYPE_VDIR;
2550 				break;
2551 			case VBLK:
2552 				kif->kf_vnode_type = KF_VTYPE_VBLK;
2553 				break;
2554 			case VCHR:
2555 				kif->kf_vnode_type = KF_VTYPE_VCHR;
2556 				break;
2557 			case VLNK:
2558 				kif->kf_vnode_type = KF_VTYPE_VLNK;
2559 				break;
2560 			case VSOCK:
2561 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
2562 				break;
2563 			case VFIFO:
2564 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
2565 				break;
2566 			case VBAD:
2567 				kif->kf_vnode_type = KF_VTYPE_VBAD;
2568 				break;
2569 			default:
2570 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2571 				break;
2572 			}
2573 			/*
2574 			 * It is OK to drop the filedesc lock here as we will
2575 			 * re-validate and re-evaluate its properties when
2576 			 * the loop continues.
2577 			 */
2578 			freepath = NULL;
2579 			fullpath = "-";
2580 			FILEDESC_SUNLOCK(fdp);
2581 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2582 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2583 			vn_fullpath(curthread, vp, &fullpath, &freepath);
2584 			vput(vp);
2585 			VFS_UNLOCK_GIANT(vfslocked);
2586 			strlcpy(kif->kf_path, fullpath,
2587 			    sizeof(kif->kf_path));
2588 			if (freepath != NULL)
2589 				free(freepath, M_TEMP);
2590 			FILEDESC_SLOCK(fdp);
2591 		}
2592 		if (so != NULL) {
2593 			struct sockaddr *sa;
2594 
2595 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2596 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2597 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2598 				free(sa, M_SONAME);
2599 			}
2600 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2601 			    == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2602 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2603 				free(sa, M_SONAME);
2604 			}
2605 			kif->kf_sock_domain =
2606 			    so->so_proto->pr_domain->dom_family;
2607 			kif->kf_sock_type = so->so_type;
2608 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
2609 		}
2610 		error = SYSCTL_OUT(req, kif, sizeof(*kif));
2611 		if (error)
2612 			break;
2613 	}
2614 	FILEDESC_SUNLOCK(fdp);
2615 	fddrop(fdp);
2616 	free(kif, M_TEMP);
2617 	return (0);
2618 }
2619 
2620 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
2621     sysctl_kern_proc_filedesc, "Process filedesc entries");
2622 
2623 #ifdef DDB
2624 /*
2625  * For the purposes of debugging, generate a human-readable string for the
2626  * file type.
2627  */
2628 static const char *
2629 file_type_to_name(short type)
2630 {
2631 
2632 	switch (type) {
2633 	case 0:
2634 		return ("zero");
2635 	case DTYPE_VNODE:
2636 		return ("vnod");
2637 	case DTYPE_SOCKET:
2638 		return ("sock");
2639 	case DTYPE_PIPE:
2640 		return ("pipe");
2641 	case DTYPE_FIFO:
2642 		return ("fifo");
2643 	case DTYPE_KQUEUE:
2644 		return ("kque");
2645 	case DTYPE_CRYPTO:
2646 		return ("crpt");
2647 	case DTYPE_MQUEUE:
2648 		return ("mque");
2649 	case DTYPE_SHM:
2650 		return ("shm");
2651 	default:
2652 		return ("unkn");
2653 	}
2654 }
2655 
2656 /*
2657  * For the purposes of debugging, identify a process (if any, perhaps one of
2658  * many) that references the passed file in its file descriptor array. Return
2659  * NULL if none.
2660  */
2661 static struct proc *
2662 file_to_first_proc(struct file *fp)
2663 {
2664 	struct filedesc *fdp;
2665 	struct proc *p;
2666 	int n;
2667 
2668 	FOREACH_PROC_IN_SYSTEM(p) {
2669 		if (p->p_state == PRS_NEW)
2670 			continue;
2671 		fdp = p->p_fd;
2672 		if (fdp == NULL)
2673 			continue;
2674 		for (n = 0; n < fdp->fd_nfiles; n++) {
2675 			if (fp == fdp->fd_ofiles[n])
2676 				return (p);
2677 		}
2678 	}
2679 	return (NULL);
2680 }
2681 
2682 static void
2683 db_print_file(struct file *fp, int header)
2684 {
2685 	struct proc *p;
2686 
2687 	if (header)
2688 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
2689 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
2690 		    "MCount", "Vnode", "FPID", "FCmd");
2691 	p = file_to_first_proc(fp);
2692 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2693 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2694 	    0, fp->f_count, 0, fp->f_vnode,
2695 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2696 }
2697 
2698 DB_SHOW_COMMAND(file, db_show_file)
2699 {
2700 	struct file *fp;
2701 
2702 	if (!have_addr) {
2703 		db_printf("usage: show file <addr>\n");
2704 		return;
2705 	}
2706 	fp = (struct file *)addr;
2707 	db_print_file(fp, 1);
2708 }
2709 
2710 DB_SHOW_COMMAND(files, db_show_files)
2711 {
2712 	struct filedesc *fdp;
2713 	struct file *fp;
2714 	struct proc *p;
2715 	int header;
2716 	int n;
2717 
2718 	header = 1;
2719 	FOREACH_PROC_IN_SYSTEM(p) {
2720 		if (p->p_state == PRS_NEW)
2721 			continue;
2722 		if ((fdp = p->p_fd) == NULL)
2723 			continue;
2724 		for (n = 0; n < fdp->fd_nfiles; ++n) {
2725 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2726 				continue;
2727 			db_print_file(fp, header);
2728 			header = 0;
2729 		}
2730 	}
2731 }
2732 #endif
2733 
2734 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2735     &maxfilesperproc, 0, "Maximum files allowed open per process");
2736 
2737 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2738     &maxfiles, 0, "Maximum number of files");
2739 
2740 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2741     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
2742 
2743 /* ARGSUSED*/
2744 static void
2745 filelistinit(void *dummy)
2746 {
2747 
2748 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2749 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2750 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2751 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2752 }
2753 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2754 
2755 /*-------------------------------------------------------------------*/
2756 
2757 static int
2758 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2759 {
2760 
2761 	return (EBADF);
2762 }
2763 
2764 static int
2765 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
2766 {
2767 
2768 	return (EINVAL);
2769 }
2770 
2771 static int
2772 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2773 {
2774 
2775 	return (EBADF);
2776 }
2777 
2778 static int
2779 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2780 {
2781 
2782 	return (0);
2783 }
2784 
2785 static int
2786 badfo_kqfilter(struct file *fp, struct knote *kn)
2787 {
2788 
2789 	return (EBADF);
2790 }
2791 
2792 static int
2793 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2794 {
2795 
2796 	return (EBADF);
2797 }
2798 
2799 static int
2800 badfo_close(struct file *fp, struct thread *td)
2801 {
2802 
2803 	return (EBADF);
2804 }
2805 
2806 struct fileops badfileops = {
2807 	.fo_read = badfo_readwrite,
2808 	.fo_write = badfo_readwrite,
2809 	.fo_truncate = badfo_truncate,
2810 	.fo_ioctl = badfo_ioctl,
2811 	.fo_poll = badfo_poll,
2812 	.fo_kqfilter = badfo_kqfilter,
2813 	.fo_stat = badfo_stat,
2814 	.fo_close = badfo_close,
2815 };
2816 
2817 
2818 /*-------------------------------------------------------------------*/
2819 
2820 /*
2821  * File Descriptor pseudo-device driver (/dev/fd/).
2822  *
2823  * Opening minor device N dup()s the file (if any) connected to file
2824  * descriptor N belonging to the calling process.  Note that this driver
2825  * consists of only the ``open()'' routine, because all subsequent
2826  * references to this file will be direct to the other driver.
2827  *
2828  * XXX: we could give this one a cloning event handler if necessary.
2829  */
2830 
2831 /* ARGSUSED */
2832 static int
2833 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2834 {
2835 
2836 	/*
2837 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2838 	 * the file descriptor being sought for duplication. The error
2839 	 * return ensures that the vnode for this device will be released
2840 	 * by vn_open. Open will detect this special error and take the
2841 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2842 	 * will simply report the error.
2843 	 */
2844 	td->td_dupfd = dev2unit(dev);
2845 	return (ENODEV);
2846 }
2847 
2848 static struct cdevsw fildesc_cdevsw = {
2849 	.d_version =	D_VERSION,
2850 	.d_flags =	D_NEEDGIANT,
2851 	.d_open =	fdopen,
2852 	.d_name =	"FD",
2853 };
2854 
2855 static void
2856 fildesc_drvinit(void *unused)
2857 {
2858 	struct cdev *dev;
2859 
2860 	dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2861 	make_dev_alias(dev, "stdin");
2862 	dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2863 	make_dev_alias(dev, "stdout");
2864 	dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2865 	make_dev_alias(dev, "stderr");
2866 }
2867 
2868 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
2869