xref: /freebsd/sys/kern/kern_descrip.c (revision 7aa383846770374466b1dcb2cefd71bde9acf463)
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 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 
47 #include <sys/conf.h>
48 #include <sys/domain.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/filedesc.h>
52 #include <sys/filio.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/mqueue.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/priv.h>
63 #include <sys/proc.h>
64 #include <sys/protosw.h>
65 #include <sys/resourcevar.h>
66 #include <sys/signalvar.h>
67 #include <sys/socketvar.h>
68 #include <sys/stat.h>
69 #include <sys/sx.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysctl.h>
72 #include <sys/sysproto.h>
73 #include <sys/tty.h>
74 #include <sys/unistd.h>
75 #include <sys/user.h>
76 #include <sys/vnode.h>
77 #ifdef KTRACE
78 #include <sys/ktrace.h>
79 #endif
80 
81 #include <security/audit/audit.h>
82 
83 #include <vm/uma.h>
84 
85 #include <ddb/ddb.h>
86 
87 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
88 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
89 		     "file desc to leader structures");
90 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
91 
92 static uma_zone_t file_zone;
93 
94 
95 /* Flags for do_dup() */
96 #define DUP_FIXED	0x1	/* Force fixed allocation */
97 #define DUP_FCNTL	0x2	/* fcntl()-style errors */
98 
99 static int do_dup(struct thread *td, int flags, int old, int new,
100     register_t *retval);
101 static int	fd_first_free(struct filedesc *, int, int);
102 static int	fd_last_used(struct filedesc *, int, int);
103 static void	fdgrowtable(struct filedesc *, int);
104 static void	fdunused(struct filedesc *fdp, int fd);
105 static void	fdused(struct filedesc *fdp, int fd);
106 
107 /*
108  * A process is initially started out with NDFILE descriptors stored within
109  * this structure, selected to be enough for typical applications based on
110  * the historical limit of 20 open files (and the usage of descriptors by
111  * shells).  If these descriptors are exhausted, a larger descriptor table
112  * may be allocated, up to a process' resource limit; the internal arrays
113  * are then unused.
114  */
115 #define NDFILE		20
116 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
117 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
118 #define NDSLOT(x)	((x) / NDENTRIES)
119 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
120 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
121 
122 /*
123  * Storage required per open file descriptor.
124  */
125 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
126 
127 /*
128  * Storage to hold unused ofiles that need to be reclaimed.
129  */
130 struct freetable {
131 	struct file	**ft_table;
132 	SLIST_ENTRY(freetable) ft_next;
133 };
134 
135 /*
136  * Basic allocation of descriptors:
137  * one of the above, plus arrays for NDFILE descriptors.
138  */
139 struct filedesc0 {
140 	struct	filedesc fd_fd;
141 	/*
142 	 * ofiles which need to be reclaimed on free.
143 	 */
144 	SLIST_HEAD(,freetable) fd_free;
145 	/*
146 	 * These arrays are used when the number of open files is
147 	 * <= NDFILE, and are then pointed to by the pointers above.
148 	 */
149 	struct	file *fd_dfiles[NDFILE];
150 	char	fd_dfileflags[NDFILE];
151 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
152 };
153 
154 /*
155  * Descriptor management.
156  */
157 volatile int openfiles;			/* actual number of open files */
158 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
159 void	(*mq_fdclose)(struct thread *td, int fd, struct file *fp);
160 
161 /* A mutex to protect the association between a proc and filedesc. */
162 static struct mtx	fdesc_mtx;
163 
164 /*
165  * Find the first zero bit in the given bitmap, starting at low and not
166  * exceeding size - 1.
167  */
168 static int
169 fd_first_free(struct filedesc *fdp, int low, int size)
170 {
171 	NDSLOTTYPE *map = fdp->fd_map;
172 	NDSLOTTYPE mask;
173 	int off, maxoff;
174 
175 	if (low >= size)
176 		return (low);
177 
178 	off = NDSLOT(low);
179 	if (low % NDENTRIES) {
180 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
181 		if ((mask &= ~map[off]) != 0UL)
182 			return (off * NDENTRIES + ffsl(mask) - 1);
183 		++off;
184 	}
185 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
186 		if (map[off] != ~0UL)
187 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
188 	return (size);
189 }
190 
191 /*
192  * Find the highest non-zero bit in the given bitmap, starting at low and
193  * not exceeding size - 1.
194  */
195 static int
196 fd_last_used(struct filedesc *fdp, int low, int size)
197 {
198 	NDSLOTTYPE *map = fdp->fd_map;
199 	NDSLOTTYPE mask;
200 	int off, minoff;
201 
202 	if (low >= size)
203 		return (-1);
204 
205 	off = NDSLOT(size);
206 	if (size % NDENTRIES) {
207 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
208 		if ((mask &= map[off]) != 0)
209 			return (off * NDENTRIES + flsl(mask) - 1);
210 		--off;
211 	}
212 	for (minoff = NDSLOT(low); off >= minoff; --off)
213 		if (map[off] != 0)
214 			return (off * NDENTRIES + flsl(map[off]) - 1);
215 	return (low - 1);
216 }
217 
218 static int
219 fdisused(struct filedesc *fdp, int fd)
220 {
221         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
222             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
223 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
224 }
225 
226 /*
227  * Mark a file descriptor as used.
228  */
229 static void
230 fdused(struct filedesc *fdp, int fd)
231 {
232 
233 	FILEDESC_XLOCK_ASSERT(fdp);
234 	KASSERT(!fdisused(fdp, fd),
235 	    ("fd already used"));
236 
237 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
238 	if (fd > fdp->fd_lastfile)
239 		fdp->fd_lastfile = fd;
240 	if (fd == fdp->fd_freefile)
241 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
242 }
243 
244 /*
245  * Mark a file descriptor as unused.
246  */
247 static void
248 fdunused(struct filedesc *fdp, int fd)
249 {
250 
251 	FILEDESC_XLOCK_ASSERT(fdp);
252 	KASSERT(fdisused(fdp, fd),
253 	    ("fd is already unused"));
254 	KASSERT(fdp->fd_ofiles[fd] == NULL,
255 	    ("fd is still in use"));
256 
257 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
258 	if (fd < fdp->fd_freefile)
259 		fdp->fd_freefile = fd;
260 	if (fd == fdp->fd_lastfile)
261 		fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
262 }
263 
264 /*
265  * System calls on descriptors.
266  */
267 #ifndef _SYS_SYSPROTO_H_
268 struct getdtablesize_args {
269 	int	dummy;
270 };
271 #endif
272 /* ARGSUSED */
273 int
274 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
275 {
276 	struct proc *p = td->td_proc;
277 
278 	PROC_LOCK(p);
279 	td->td_retval[0] =
280 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
281 	PROC_UNLOCK(p);
282 	return (0);
283 }
284 
285 /*
286  * Duplicate a file descriptor to a particular value.
287  *
288  * Note: keep in mind that a potential race condition exists when closing
289  * descriptors from a shared descriptor table (via rfork).
290  */
291 #ifndef _SYS_SYSPROTO_H_
292 struct dup2_args {
293 	u_int	from;
294 	u_int	to;
295 };
296 #endif
297 /* ARGSUSED */
298 int
299 dup2(struct thread *td, struct dup2_args *uap)
300 {
301 
302 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
303 		    td->td_retval));
304 }
305 
306 /*
307  * Duplicate a file descriptor.
308  */
309 #ifndef _SYS_SYSPROTO_H_
310 struct dup_args {
311 	u_int	fd;
312 };
313 #endif
314 /* ARGSUSED */
315 int
316 dup(struct thread *td, struct dup_args *uap)
317 {
318 
319 	return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
320 }
321 
322 /*
323  * The file control system call.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct fcntl_args {
327 	int	fd;
328 	int	cmd;
329 	long	arg;
330 };
331 #endif
332 /* ARGSUSED */
333 int
334 fcntl(struct thread *td, struct fcntl_args *uap)
335 {
336 	struct flock fl;
337 	struct oflock ofl;
338 	intptr_t arg;
339 	int error;
340 	int cmd;
341 
342 	error = 0;
343 	cmd = uap->cmd;
344 	switch (uap->cmd) {
345 	case F_OGETLK:
346 	case F_OSETLK:
347 	case F_OSETLKW:
348 		/*
349 		 * Convert old flock structure to new.
350 		 */
351 		error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
352 		fl.l_start = ofl.l_start;
353 		fl.l_len = ofl.l_len;
354 		fl.l_pid = ofl.l_pid;
355 		fl.l_type = ofl.l_type;
356 		fl.l_whence = ofl.l_whence;
357 		fl.l_sysid = 0;
358 
359 		switch (uap->cmd) {
360 		case F_OGETLK:
361 		    cmd = F_GETLK;
362 		    break;
363 		case F_OSETLK:
364 		    cmd = F_SETLK;
365 		    break;
366 		case F_OSETLKW:
367 		    cmd = F_SETLKW;
368 		    break;
369 		}
370 		arg = (intptr_t)&fl;
371 		break;
372         case F_GETLK:
373         case F_SETLK:
374         case F_SETLKW:
375 	case F_SETLK_REMOTE:
376                 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
377                 arg = (intptr_t)&fl;
378                 break;
379 	default:
380 		arg = uap->arg;
381 		break;
382 	}
383 	if (error)
384 		return (error);
385 	error = kern_fcntl(td, uap->fd, cmd, arg);
386 	if (error)
387 		return (error);
388 	if (uap->cmd == F_OGETLK) {
389 		ofl.l_start = fl.l_start;
390 		ofl.l_len = fl.l_len;
391 		ofl.l_pid = fl.l_pid;
392 		ofl.l_type = fl.l_type;
393 		ofl.l_whence = fl.l_whence;
394 		error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
395 	} else if (uap->cmd == F_GETLK) {
396 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
397 	}
398 	return (error);
399 }
400 
401 static inline struct file *
402 fdtofp(int fd, struct filedesc *fdp)
403 {
404 	struct file *fp;
405 
406 	FILEDESC_LOCK_ASSERT(fdp);
407 	if ((unsigned)fd >= fdp->fd_nfiles ||
408 	    (fp = fdp->fd_ofiles[fd]) == NULL)
409 		return (NULL);
410 	return (fp);
411 }
412 
413 int
414 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
415 {
416 	struct filedesc *fdp;
417 	struct flock *flp;
418 	struct file *fp;
419 	struct proc *p;
420 	char *pop;
421 	struct vnode *vp;
422 	int error, flg, tmp;
423 	int vfslocked;
424 	u_int old, new;
425 	uint64_t bsize;
426 
427 	vfslocked = 0;
428 	error = 0;
429 	flg = F_POSIX;
430 	p = td->td_proc;
431 	fdp = p->p_fd;
432 
433 	switch (cmd) {
434 	case F_DUPFD:
435 		tmp = arg;
436 		error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
437 		break;
438 
439 	case F_DUP2FD:
440 		tmp = arg;
441 		error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
442 		break;
443 
444 	case F_GETFD:
445 		FILEDESC_SLOCK(fdp);
446 		if ((fp = fdtofp(fd, fdp)) == NULL) {
447 			FILEDESC_SUNLOCK(fdp);
448 			error = EBADF;
449 			break;
450 		}
451 		pop = &fdp->fd_ofileflags[fd];
452 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
453 		FILEDESC_SUNLOCK(fdp);
454 		break;
455 
456 	case F_SETFD:
457 		FILEDESC_XLOCK(fdp);
458 		if ((fp = fdtofp(fd, fdp)) == NULL) {
459 			FILEDESC_XUNLOCK(fdp);
460 			error = EBADF;
461 			break;
462 		}
463 		pop = &fdp->fd_ofileflags[fd];
464 		*pop = (*pop &~ UF_EXCLOSE) |
465 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
466 		FILEDESC_XUNLOCK(fdp);
467 		break;
468 
469 	case F_GETFL:
470 		FILEDESC_SLOCK(fdp);
471 		if ((fp = fdtofp(fd, fdp)) == NULL) {
472 			FILEDESC_SUNLOCK(fdp);
473 			error = EBADF;
474 			break;
475 		}
476 		td->td_retval[0] = OFLAGS(fp->f_flag);
477 		FILEDESC_SUNLOCK(fdp);
478 		break;
479 
480 	case F_SETFL:
481 		FILEDESC_SLOCK(fdp);
482 		if ((fp = fdtofp(fd, fdp)) == NULL) {
483 			FILEDESC_SUNLOCK(fdp);
484 			error = EBADF;
485 			break;
486 		}
487 		fhold(fp);
488 		FILEDESC_SUNLOCK(fdp);
489 		do {
490 			tmp = flg = fp->f_flag;
491 			tmp &= ~FCNTLFLAGS;
492 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
493 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
494 		tmp = fp->f_flag & FNONBLOCK;
495 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
496 		if (error) {
497 			fdrop(fp, td);
498 			break;
499 		}
500 		tmp = fp->f_flag & FASYNC;
501 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
502 		if (error == 0) {
503 			fdrop(fp, td);
504 			break;
505 		}
506 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
507 		tmp = 0;
508 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
509 		fdrop(fp, td);
510 		break;
511 
512 	case F_GETOWN:
513 		FILEDESC_SLOCK(fdp);
514 		if ((fp = fdtofp(fd, fdp)) == NULL) {
515 			FILEDESC_SUNLOCK(fdp);
516 			error = EBADF;
517 			break;
518 		}
519 		fhold(fp);
520 		FILEDESC_SUNLOCK(fdp);
521 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
522 		if (error == 0)
523 			td->td_retval[0] = tmp;
524 		fdrop(fp, td);
525 		break;
526 
527 	case F_SETOWN:
528 		FILEDESC_SLOCK(fdp);
529 		if ((fp = fdtofp(fd, fdp)) == NULL) {
530 			FILEDESC_SUNLOCK(fdp);
531 			error = EBADF;
532 			break;
533 		}
534 		fhold(fp);
535 		FILEDESC_SUNLOCK(fdp);
536 		tmp = arg;
537 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
538 		fdrop(fp, td);
539 		break;
540 
541 	case F_SETLK_REMOTE:
542 		error = priv_check(td, PRIV_NFS_LOCKD);
543 		if (error)
544 			return (error);
545 		flg = F_REMOTE;
546 		goto do_setlk;
547 
548 	case F_SETLKW:
549 		flg |= F_WAIT;
550 		/* FALLTHROUGH F_SETLK */
551 
552 	case F_SETLK:
553 	do_setlk:
554 		FILEDESC_SLOCK(fdp);
555 		if ((fp = fdtofp(fd, fdp)) == NULL) {
556 			FILEDESC_SUNLOCK(fdp);
557 			error = EBADF;
558 			break;
559 		}
560 		if (fp->f_type != DTYPE_VNODE) {
561 			FILEDESC_SUNLOCK(fdp);
562 			error = EBADF;
563 			break;
564 		}
565 		flp = (struct flock *)arg;
566 		if (flp->l_whence == SEEK_CUR) {
567 			if (fp->f_offset < 0 ||
568 			    (flp->l_start > 0 &&
569 			     fp->f_offset > OFF_MAX - flp->l_start)) {
570 				FILEDESC_SUNLOCK(fdp);
571 				error = EOVERFLOW;
572 				break;
573 			}
574 			flp->l_start += fp->f_offset;
575 		}
576 
577 		/*
578 		 * VOP_ADVLOCK() may block.
579 		 */
580 		fhold(fp);
581 		FILEDESC_SUNLOCK(fdp);
582 		vp = fp->f_vnode;
583 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
584 		switch (flp->l_type) {
585 		case F_RDLCK:
586 			if ((fp->f_flag & FREAD) == 0) {
587 				error = EBADF;
588 				break;
589 			}
590 			PROC_LOCK(p->p_leader);
591 			p->p_leader->p_flag |= P_ADVLOCK;
592 			PROC_UNLOCK(p->p_leader);
593 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
594 			    flp, flg);
595 			break;
596 		case F_WRLCK:
597 			if ((fp->f_flag & FWRITE) == 0) {
598 				error = EBADF;
599 				break;
600 			}
601 			PROC_LOCK(p->p_leader);
602 			p->p_leader->p_flag |= P_ADVLOCK;
603 			PROC_UNLOCK(p->p_leader);
604 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
605 			    flp, flg);
606 			break;
607 		case F_UNLCK:
608 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
609 			    flp, flg);
610 			break;
611 		case F_UNLCKSYS:
612 			/*
613 			 * Temporary api for testing remote lock
614 			 * infrastructure.
615 			 */
616 			if (flg != F_REMOTE) {
617 				error = EINVAL;
618 				break;
619 			}
620 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
621 			    F_UNLCKSYS, flp, flg);
622 			break;
623 		default:
624 			error = EINVAL;
625 			break;
626 		}
627 		VFS_UNLOCK_GIANT(vfslocked);
628 		vfslocked = 0;
629 		/* Check for race with close */
630 		FILEDESC_SLOCK(fdp);
631 		if ((unsigned) fd >= fdp->fd_nfiles ||
632 		    fp != fdp->fd_ofiles[fd]) {
633 			FILEDESC_SUNLOCK(fdp);
634 			flp->l_whence = SEEK_SET;
635 			flp->l_start = 0;
636 			flp->l_len = 0;
637 			flp->l_type = F_UNLCK;
638 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
639 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
640 					   F_UNLCK, flp, F_POSIX);
641 			VFS_UNLOCK_GIANT(vfslocked);
642 			vfslocked = 0;
643 		} else
644 			FILEDESC_SUNLOCK(fdp);
645 		fdrop(fp, td);
646 		break;
647 
648 	case F_GETLK:
649 		FILEDESC_SLOCK(fdp);
650 		if ((fp = fdtofp(fd, fdp)) == NULL) {
651 			FILEDESC_SUNLOCK(fdp);
652 			error = EBADF;
653 			break;
654 		}
655 		if (fp->f_type != DTYPE_VNODE) {
656 			FILEDESC_SUNLOCK(fdp);
657 			error = EBADF;
658 			break;
659 		}
660 		flp = (struct flock *)arg;
661 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
662 		    flp->l_type != F_UNLCK) {
663 			FILEDESC_SUNLOCK(fdp);
664 			error = EINVAL;
665 			break;
666 		}
667 		if (flp->l_whence == SEEK_CUR) {
668 			if ((flp->l_start > 0 &&
669 			    fp->f_offset > OFF_MAX - flp->l_start) ||
670 			    (flp->l_start < 0 &&
671 			     fp->f_offset < OFF_MIN - flp->l_start)) {
672 				FILEDESC_SUNLOCK(fdp);
673 				error = EOVERFLOW;
674 				break;
675 			}
676 			flp->l_start += fp->f_offset;
677 		}
678 		/*
679 		 * VOP_ADVLOCK() may block.
680 		 */
681 		fhold(fp);
682 		FILEDESC_SUNLOCK(fdp);
683 		vp = fp->f_vnode;
684 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
685 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
686 		    F_POSIX);
687 		VFS_UNLOCK_GIANT(vfslocked);
688 		vfslocked = 0;
689 		fdrop(fp, td);
690 		break;
691 
692 	case F_RDAHEAD:
693 		arg = arg ? 128 * 1024: 0;
694 		/* FALLTHROUGH */
695 	case F_READAHEAD:
696 		FILEDESC_SLOCK(fdp);
697 		if ((fp = fdtofp(fd, fdp)) == NULL) {
698 			FILEDESC_SUNLOCK(fdp);
699 			error = EBADF;
700 			break;
701 		}
702 		if (fp->f_type != DTYPE_VNODE) {
703 			FILEDESC_SUNLOCK(fdp);
704 			error = EBADF;
705 			break;
706 		}
707 		fhold(fp);
708 		FILEDESC_SUNLOCK(fdp);
709 		if (arg != 0) {
710 			vp = fp->f_vnode;
711 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
712 			error = vn_lock(vp, LK_SHARED);
713 			if (error != 0)
714 				goto readahead_vnlock_fail;
715 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
716 			VOP_UNLOCK(vp, 0);
717 			fp->f_seqcount = (arg + bsize - 1) / bsize;
718 			do {
719 				new = old = fp->f_flag;
720 				new |= FRDAHEAD;
721 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
722 readahead_vnlock_fail:
723 			VFS_UNLOCK_GIANT(vfslocked);
724 			vfslocked = 0;
725 		} else {
726 			do {
727 				new = old = fp->f_flag;
728 				new &= ~FRDAHEAD;
729 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
730 		}
731 		fdrop(fp, td);
732 		break;
733 
734 	default:
735 		error = EINVAL;
736 		break;
737 	}
738 	VFS_UNLOCK_GIANT(vfslocked);
739 	return (error);
740 }
741 
742 /*
743  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
744  */
745 static int
746 do_dup(struct thread *td, int flags, int old, int new,
747     register_t *retval)
748 {
749 	struct filedesc *fdp;
750 	struct proc *p;
751 	struct file *fp;
752 	struct file *delfp;
753 	int error, holdleaders, maxfd;
754 
755 	p = td->td_proc;
756 	fdp = p->p_fd;
757 
758 	/*
759 	 * Verify we have a valid descriptor to dup from and possibly to
760 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
761 	 * return EINVAL when the new descriptor is out of bounds.
762 	 */
763 	if (old < 0)
764 		return (EBADF);
765 	if (new < 0)
766 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
767 	PROC_LOCK(p);
768 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
769 	PROC_UNLOCK(p);
770 	if (new >= maxfd)
771 		return (flags & DUP_FCNTL ? EINVAL : EMFILE);
772 
773 	FILEDESC_XLOCK(fdp);
774 	if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
775 		FILEDESC_XUNLOCK(fdp);
776 		return (EBADF);
777 	}
778 	if (flags & DUP_FIXED && old == new) {
779 		*retval = new;
780 		FILEDESC_XUNLOCK(fdp);
781 		return (0);
782 	}
783 	fp = fdp->fd_ofiles[old];
784 	fhold(fp);
785 
786 	/*
787 	 * If the caller specified a file descriptor, make sure the file
788 	 * table is large enough to hold it, and grab it.  Otherwise, just
789 	 * allocate a new descriptor the usual way.  Since the filedesc
790 	 * lock may be temporarily dropped in the process, we have to look
791 	 * out for a race.
792 	 */
793 	if (flags & DUP_FIXED) {
794 		if (new >= fdp->fd_nfiles)
795 			fdgrowtable(fdp, new + 1);
796 		if (fdp->fd_ofiles[new] == NULL)
797 			fdused(fdp, new);
798 	} else {
799 		if ((error = fdalloc(td, new, &new)) != 0) {
800 			FILEDESC_XUNLOCK(fdp);
801 			fdrop(fp, td);
802 			return (error);
803 		}
804 	}
805 
806 	/*
807 	 * If the old file changed out from under us then treat it as a
808 	 * bad file descriptor.  Userland should do its own locking to
809 	 * avoid this case.
810 	 */
811 	if (fdp->fd_ofiles[old] != fp) {
812 		/* we've allocated a descriptor which we won't use */
813 		if (fdp->fd_ofiles[new] == NULL)
814 			fdunused(fdp, new);
815 		FILEDESC_XUNLOCK(fdp);
816 		fdrop(fp, td);
817 		return (EBADF);
818 	}
819 	KASSERT(old != new,
820 	    ("new fd is same as old"));
821 
822 	/*
823 	 * Save info on the descriptor being overwritten.  We cannot close
824 	 * it without introducing an ownership race for the slot, since we
825 	 * need to drop the filedesc lock to call closef().
826 	 *
827 	 * XXX this duplicates parts of close().
828 	 */
829 	delfp = fdp->fd_ofiles[new];
830 	holdleaders = 0;
831 	if (delfp != NULL) {
832 		if (td->td_proc->p_fdtol != NULL) {
833 			/*
834 			 * Ask fdfree() to sleep to ensure that all relevant
835 			 * process leaders can be traversed in closef().
836 			 */
837 			fdp->fd_holdleaderscount++;
838 			holdleaders = 1;
839 		}
840 	}
841 
842 	/*
843 	 * Duplicate the source descriptor
844 	 */
845 	fdp->fd_ofiles[new] = fp;
846 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
847 	if (new > fdp->fd_lastfile)
848 		fdp->fd_lastfile = new;
849 	*retval = new;
850 
851 	/*
852 	 * If we dup'd over a valid file, we now own the reference to it
853 	 * and must dispose of it using closef() semantics (as if a
854 	 * close() were performed on it).
855 	 *
856 	 * XXX this duplicates parts of close().
857 	 */
858 	if (delfp != NULL) {
859 		knote_fdclose(td, new);
860 		if (delfp->f_type == DTYPE_MQUEUE)
861 			mq_fdclose(td, new, delfp);
862 		FILEDESC_XUNLOCK(fdp);
863 		(void) closef(delfp, td);
864 		if (holdleaders) {
865 			FILEDESC_XLOCK(fdp);
866 			fdp->fd_holdleaderscount--;
867 			if (fdp->fd_holdleaderscount == 0 &&
868 			    fdp->fd_holdleaderswakeup != 0) {
869 				fdp->fd_holdleaderswakeup = 0;
870 				wakeup(&fdp->fd_holdleaderscount);
871 			}
872 			FILEDESC_XUNLOCK(fdp);
873 		}
874 	} else {
875 		FILEDESC_XUNLOCK(fdp);
876 	}
877 	return (0);
878 }
879 
880 /*
881  * If sigio is on the list associated with a process or process group,
882  * disable signalling from the device, remove sigio from the list and
883  * free sigio.
884  */
885 void
886 funsetown(struct sigio **sigiop)
887 {
888 	struct sigio *sigio;
889 
890 	SIGIO_LOCK();
891 	sigio = *sigiop;
892 	if (sigio == NULL) {
893 		SIGIO_UNLOCK();
894 		return;
895 	}
896 	*(sigio->sio_myref) = NULL;
897 	if ((sigio)->sio_pgid < 0) {
898 		struct pgrp *pg = (sigio)->sio_pgrp;
899 		PGRP_LOCK(pg);
900 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
901 			     sigio, sio_pgsigio);
902 		PGRP_UNLOCK(pg);
903 	} else {
904 		struct proc *p = (sigio)->sio_proc;
905 		PROC_LOCK(p);
906 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
907 			     sigio, sio_pgsigio);
908 		PROC_UNLOCK(p);
909 	}
910 	SIGIO_UNLOCK();
911 	crfree(sigio->sio_ucred);
912 	free(sigio, M_SIGIO);
913 }
914 
915 /*
916  * Free a list of sigio structures.
917  * We only need to lock the SIGIO_LOCK because we have made ourselves
918  * inaccessible to callers of fsetown and therefore do not need to lock
919  * the proc or pgrp struct for the list manipulation.
920  */
921 void
922 funsetownlst(struct sigiolst *sigiolst)
923 {
924 	struct proc *p;
925 	struct pgrp *pg;
926 	struct sigio *sigio;
927 
928 	sigio = SLIST_FIRST(sigiolst);
929 	if (sigio == NULL)
930 		return;
931 	p = NULL;
932 	pg = NULL;
933 
934 	/*
935 	 * Every entry of the list should belong
936 	 * to a single proc or pgrp.
937 	 */
938 	if (sigio->sio_pgid < 0) {
939 		pg = sigio->sio_pgrp;
940 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
941 	} else /* if (sigio->sio_pgid > 0) */ {
942 		p = sigio->sio_proc;
943 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
944 	}
945 
946 	SIGIO_LOCK();
947 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
948 		*(sigio->sio_myref) = NULL;
949 		if (pg != NULL) {
950 			KASSERT(sigio->sio_pgid < 0,
951 			    ("Proc sigio in pgrp sigio list"));
952 			KASSERT(sigio->sio_pgrp == pg,
953 			    ("Bogus pgrp in sigio list"));
954 			PGRP_LOCK(pg);
955 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
956 			    sio_pgsigio);
957 			PGRP_UNLOCK(pg);
958 		} else /* if (p != NULL) */ {
959 			KASSERT(sigio->sio_pgid > 0,
960 			    ("Pgrp sigio in proc sigio list"));
961 			KASSERT(sigio->sio_proc == p,
962 			    ("Bogus proc in sigio list"));
963 			PROC_LOCK(p);
964 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
965 			    sio_pgsigio);
966 			PROC_UNLOCK(p);
967 		}
968 		SIGIO_UNLOCK();
969 		crfree(sigio->sio_ucred);
970 		free(sigio, M_SIGIO);
971 		SIGIO_LOCK();
972 	}
973 	SIGIO_UNLOCK();
974 }
975 
976 /*
977  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
978  *
979  * After permission checking, add a sigio structure to the sigio list for
980  * the process or process group.
981  */
982 int
983 fsetown(pid_t pgid, struct sigio **sigiop)
984 {
985 	struct proc *proc;
986 	struct pgrp *pgrp;
987 	struct sigio *sigio;
988 	int ret;
989 
990 	if (pgid == 0) {
991 		funsetown(sigiop);
992 		return (0);
993 	}
994 
995 	ret = 0;
996 
997 	/* Allocate and fill in the new sigio out of locks. */
998 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
999 	sigio->sio_pgid = pgid;
1000 	sigio->sio_ucred = crhold(curthread->td_ucred);
1001 	sigio->sio_myref = sigiop;
1002 
1003 	sx_slock(&proctree_lock);
1004 	if (pgid > 0) {
1005 		proc = pfind(pgid);
1006 		if (proc == NULL) {
1007 			ret = ESRCH;
1008 			goto fail;
1009 		}
1010 
1011 		/*
1012 		 * Policy - Don't allow a process to FSETOWN a process
1013 		 * in another session.
1014 		 *
1015 		 * Remove this test to allow maximum flexibility or
1016 		 * restrict FSETOWN to the current process or process
1017 		 * group for maximum safety.
1018 		 */
1019 		PROC_UNLOCK(proc);
1020 		if (proc->p_session != curthread->td_proc->p_session) {
1021 			ret = EPERM;
1022 			goto fail;
1023 		}
1024 
1025 		pgrp = NULL;
1026 	} else /* if (pgid < 0) */ {
1027 		pgrp = pgfind(-pgid);
1028 		if (pgrp == NULL) {
1029 			ret = ESRCH;
1030 			goto fail;
1031 		}
1032 		PGRP_UNLOCK(pgrp);
1033 
1034 		/*
1035 		 * Policy - Don't allow a process to FSETOWN a process
1036 		 * in another session.
1037 		 *
1038 		 * Remove this test to allow maximum flexibility or
1039 		 * restrict FSETOWN to the current process or process
1040 		 * group for maximum safety.
1041 		 */
1042 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1043 			ret = EPERM;
1044 			goto fail;
1045 		}
1046 
1047 		proc = NULL;
1048 	}
1049 	funsetown(sigiop);
1050 	if (pgid > 0) {
1051 		PROC_LOCK(proc);
1052 		/*
1053 		 * Since funsetownlst() is called without the proctree
1054 		 * locked, we need to check for P_WEXIT.
1055 		 * XXX: is ESRCH correct?
1056 		 */
1057 		if ((proc->p_flag & P_WEXIT) != 0) {
1058 			PROC_UNLOCK(proc);
1059 			ret = ESRCH;
1060 			goto fail;
1061 		}
1062 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1063 		sigio->sio_proc = proc;
1064 		PROC_UNLOCK(proc);
1065 	} else {
1066 		PGRP_LOCK(pgrp);
1067 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1068 		sigio->sio_pgrp = pgrp;
1069 		PGRP_UNLOCK(pgrp);
1070 	}
1071 	sx_sunlock(&proctree_lock);
1072 	SIGIO_LOCK();
1073 	*sigiop = sigio;
1074 	SIGIO_UNLOCK();
1075 	return (0);
1076 
1077 fail:
1078 	sx_sunlock(&proctree_lock);
1079 	crfree(sigio->sio_ucred);
1080 	free(sigio, M_SIGIO);
1081 	return (ret);
1082 }
1083 
1084 /*
1085  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1086  */
1087 pid_t
1088 fgetown(sigiop)
1089 	struct sigio **sigiop;
1090 {
1091 	pid_t pgid;
1092 
1093 	SIGIO_LOCK();
1094 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1095 	SIGIO_UNLOCK();
1096 	return (pgid);
1097 }
1098 
1099 /*
1100  * Close a file descriptor.
1101  */
1102 #ifndef _SYS_SYSPROTO_H_
1103 struct close_args {
1104 	int     fd;
1105 };
1106 #endif
1107 /* ARGSUSED */
1108 int
1109 close(td, uap)
1110 	struct thread *td;
1111 	struct close_args *uap;
1112 {
1113 
1114 	return (kern_close(td, uap->fd));
1115 }
1116 
1117 int
1118 kern_close(td, fd)
1119 	struct thread *td;
1120 	int fd;
1121 {
1122 	struct filedesc *fdp;
1123 	struct file *fp;
1124 	int error;
1125 	int holdleaders;
1126 
1127 	error = 0;
1128 	holdleaders = 0;
1129 	fdp = td->td_proc->p_fd;
1130 
1131 	AUDIT_SYSCLOSE(td, fd);
1132 
1133 	FILEDESC_XLOCK(fdp);
1134 	if ((unsigned)fd >= fdp->fd_nfiles ||
1135 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
1136 		FILEDESC_XUNLOCK(fdp);
1137 		return (EBADF);
1138 	}
1139 	fdp->fd_ofiles[fd] = NULL;
1140 	fdp->fd_ofileflags[fd] = 0;
1141 	fdunused(fdp, fd);
1142 	if (td->td_proc->p_fdtol != NULL) {
1143 		/*
1144 		 * Ask fdfree() to sleep to ensure that all relevant
1145 		 * process leaders can be traversed in closef().
1146 		 */
1147 		fdp->fd_holdleaderscount++;
1148 		holdleaders = 1;
1149 	}
1150 
1151 	/*
1152 	 * We now hold the fp reference that used to be owned by the
1153 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1154 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1155 	 * added, and deleteing a knote for the new fd.
1156 	 */
1157 	knote_fdclose(td, fd);
1158 	if (fp->f_type == DTYPE_MQUEUE)
1159 		mq_fdclose(td, fd, fp);
1160 	FILEDESC_XUNLOCK(fdp);
1161 
1162 	error = closef(fp, td);
1163 	if (holdleaders) {
1164 		FILEDESC_XLOCK(fdp);
1165 		fdp->fd_holdleaderscount--;
1166 		if (fdp->fd_holdleaderscount == 0 &&
1167 		    fdp->fd_holdleaderswakeup != 0) {
1168 			fdp->fd_holdleaderswakeup = 0;
1169 			wakeup(&fdp->fd_holdleaderscount);
1170 		}
1171 		FILEDESC_XUNLOCK(fdp);
1172 	}
1173 	return (error);
1174 }
1175 
1176 /*
1177  * Close open file descriptors.
1178  */
1179 #ifndef _SYS_SYSPROTO_H_
1180 struct closefrom_args {
1181 	int	lowfd;
1182 };
1183 #endif
1184 /* ARGSUSED */
1185 int
1186 closefrom(struct thread *td, struct closefrom_args *uap)
1187 {
1188 	struct filedesc *fdp;
1189 	int fd;
1190 
1191 	fdp = td->td_proc->p_fd;
1192 	AUDIT_ARG_FD(uap->lowfd);
1193 
1194 	/*
1195 	 * Treat negative starting file descriptor values identical to
1196 	 * closefrom(0) which closes all files.
1197 	 */
1198 	if (uap->lowfd < 0)
1199 		uap->lowfd = 0;
1200 	FILEDESC_SLOCK(fdp);
1201 	for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1202 		if (fdp->fd_ofiles[fd] != NULL) {
1203 			FILEDESC_SUNLOCK(fdp);
1204 			(void)kern_close(td, fd);
1205 			FILEDESC_SLOCK(fdp);
1206 		}
1207 	}
1208 	FILEDESC_SUNLOCK(fdp);
1209 	return (0);
1210 }
1211 
1212 #if defined(COMPAT_43)
1213 /*
1214  * Return status information about a file descriptor.
1215  */
1216 #ifndef _SYS_SYSPROTO_H_
1217 struct ofstat_args {
1218 	int	fd;
1219 	struct	ostat *sb;
1220 };
1221 #endif
1222 /* ARGSUSED */
1223 int
1224 ofstat(struct thread *td, struct ofstat_args *uap)
1225 {
1226 	struct ostat oub;
1227 	struct stat ub;
1228 	int error;
1229 
1230 	error = kern_fstat(td, uap->fd, &ub);
1231 	if (error == 0) {
1232 		cvtstat(&ub, &oub);
1233 		error = copyout(&oub, uap->sb, sizeof(oub));
1234 	}
1235 	return (error);
1236 }
1237 #endif /* COMPAT_43 */
1238 
1239 /*
1240  * Return status information about a file descriptor.
1241  */
1242 #ifndef _SYS_SYSPROTO_H_
1243 struct fstat_args {
1244 	int	fd;
1245 	struct	stat *sb;
1246 };
1247 #endif
1248 /* ARGSUSED */
1249 int
1250 fstat(struct thread *td, struct fstat_args *uap)
1251 {
1252 	struct stat ub;
1253 	int error;
1254 
1255 	error = kern_fstat(td, uap->fd, &ub);
1256 	if (error == 0)
1257 		error = copyout(&ub, uap->sb, sizeof(ub));
1258 	return (error);
1259 }
1260 
1261 int
1262 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1263 {
1264 	struct file *fp;
1265 	int error;
1266 
1267 	AUDIT_ARG_FD(fd);
1268 
1269 	if ((error = fget(td, fd, &fp)) != 0)
1270 		return (error);
1271 
1272 	AUDIT_ARG_FILE(td->td_proc, fp);
1273 
1274 	error = fo_stat(fp, sbp, td->td_ucred, td);
1275 	fdrop(fp, td);
1276 #ifdef KTRACE
1277 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1278 		ktrstat(sbp);
1279 #endif
1280 	return (error);
1281 }
1282 
1283 /*
1284  * Return status information about a file descriptor.
1285  */
1286 #ifndef _SYS_SYSPROTO_H_
1287 struct nfstat_args {
1288 	int	fd;
1289 	struct	nstat *sb;
1290 };
1291 #endif
1292 /* ARGSUSED */
1293 int
1294 nfstat(struct thread *td, struct nfstat_args *uap)
1295 {
1296 	struct nstat nub;
1297 	struct stat ub;
1298 	int error;
1299 
1300 	error = kern_fstat(td, uap->fd, &ub);
1301 	if (error == 0) {
1302 		cvtnstat(&ub, &nub);
1303 		error = copyout(&nub, uap->sb, sizeof(nub));
1304 	}
1305 	return (error);
1306 }
1307 
1308 /*
1309  * Return pathconf information about a file descriptor.
1310  */
1311 #ifndef _SYS_SYSPROTO_H_
1312 struct fpathconf_args {
1313 	int	fd;
1314 	int	name;
1315 };
1316 #endif
1317 /* ARGSUSED */
1318 int
1319 fpathconf(struct thread *td, struct fpathconf_args *uap)
1320 {
1321 	struct file *fp;
1322 	struct vnode *vp;
1323 	int error;
1324 
1325 	if ((error = fget(td, uap->fd, &fp)) != 0)
1326 		return (error);
1327 
1328 	/* If asynchronous I/O is available, it works for all descriptors. */
1329 	if (uap->name == _PC_ASYNC_IO) {
1330 		td->td_retval[0] = async_io_version;
1331 		goto out;
1332 	}
1333 	vp = fp->f_vnode;
1334 	if (vp != NULL) {
1335 		int vfslocked;
1336 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1337 		vn_lock(vp, LK_SHARED | LK_RETRY);
1338 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1339 		VOP_UNLOCK(vp, 0);
1340 		VFS_UNLOCK_GIANT(vfslocked);
1341 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1342 		if (uap->name != _PC_PIPE_BUF) {
1343 			error = EINVAL;
1344 		} else {
1345 			td->td_retval[0] = PIPE_BUF;
1346 		error = 0;
1347 		}
1348 	} else {
1349 		error = EOPNOTSUPP;
1350 	}
1351 out:
1352 	fdrop(fp, td);
1353 	return (error);
1354 }
1355 
1356 /*
1357  * Grow the file table to accomodate (at least) nfd descriptors.  This may
1358  * block and drop the filedesc lock, but it will reacquire it before
1359  * returning.
1360  */
1361 static void
1362 fdgrowtable(struct filedesc *fdp, int nfd)
1363 {
1364 	struct filedesc0 *fdp0;
1365 	struct freetable *fo;
1366 	struct file **ntable;
1367 	struct file **otable;
1368 	char *nfileflags;
1369 	int nnfiles, onfiles;
1370 	NDSLOTTYPE *nmap;
1371 
1372 	FILEDESC_XLOCK_ASSERT(fdp);
1373 
1374 	KASSERT(fdp->fd_nfiles > 0,
1375 	    ("zero-length file table"));
1376 
1377 	/* compute the size of the new table */
1378 	onfiles = fdp->fd_nfiles;
1379 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1380 	if (nnfiles <= onfiles)
1381 		/* the table is already large enough */
1382 		return;
1383 
1384 	/* allocate a new table and (if required) new bitmaps */
1385 	FILEDESC_XUNLOCK(fdp);
1386 	ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable),
1387 	    M_FILEDESC, M_ZERO | M_WAITOK);
1388 	nfileflags = (char *)&ntable[nnfiles];
1389 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1390 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE,
1391 		    M_FILEDESC, M_ZERO | M_WAITOK);
1392 	else
1393 		nmap = NULL;
1394 	FILEDESC_XLOCK(fdp);
1395 
1396 	/*
1397 	 * We now have new tables ready to go.  Since we dropped the
1398 	 * filedesc lock to call malloc(), watch out for a race.
1399 	 */
1400 	onfiles = fdp->fd_nfiles;
1401 	if (onfiles >= nnfiles) {
1402 		/* we lost the race, but that's OK */
1403 		free(ntable, M_FILEDESC);
1404 		if (nmap != NULL)
1405 			free(nmap, M_FILEDESC);
1406 		return;
1407 	}
1408 	bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1409 	bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1410 	otable = fdp->fd_ofiles;
1411 	fdp->fd_ofileflags = nfileflags;
1412 	fdp->fd_ofiles = ntable;
1413 	/*
1414 	 * We must preserve ofiles until the process exits because we can't
1415 	 * be certain that no threads have references to the old table via
1416 	 * _fget().
1417 	 */
1418 	if (onfiles > NDFILE) {
1419 		fo = (struct freetable *)&otable[onfiles];
1420 		fdp0 = (struct filedesc0 *)fdp;
1421 		fo->ft_table = otable;
1422 		SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next);
1423 	}
1424 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1425 		bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1426 		if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1427 			free(fdp->fd_map, M_FILEDESC);
1428 		fdp->fd_map = nmap;
1429 	}
1430 	fdp->fd_nfiles = nnfiles;
1431 }
1432 
1433 /*
1434  * Allocate a file descriptor for the process.
1435  */
1436 int
1437 fdalloc(struct thread *td, int minfd, int *result)
1438 {
1439 	struct proc *p = td->td_proc;
1440 	struct filedesc *fdp = p->p_fd;
1441 	int fd = -1, maxfd;
1442 
1443 	FILEDESC_XLOCK_ASSERT(fdp);
1444 
1445 	if (fdp->fd_freefile > minfd)
1446 		minfd = fdp->fd_freefile;
1447 
1448 	PROC_LOCK(p);
1449 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1450 	PROC_UNLOCK(p);
1451 
1452 	/*
1453 	 * Search the bitmap for a free descriptor.  If none is found, try
1454 	 * to grow the file table.  Keep at it until we either get a file
1455 	 * descriptor or run into process or system limits; fdgrowtable()
1456 	 * may drop the filedesc lock, so we're in a race.
1457 	 */
1458 	for (;;) {
1459 		fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1460 		if (fd >= maxfd)
1461 			return (EMFILE);
1462 		if (fd < fdp->fd_nfiles)
1463 			break;
1464 		fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1465 	}
1466 
1467 	/*
1468 	 * Perform some sanity checks, then mark the file descriptor as
1469 	 * used and return it to the caller.
1470 	 */
1471 	KASSERT(!fdisused(fdp, fd),
1472 	    ("fd_first_free() returned non-free descriptor"));
1473 	KASSERT(fdp->fd_ofiles[fd] == NULL,
1474 	    ("free descriptor isn't"));
1475 	fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1476 	fdused(fdp, fd);
1477 	*result = fd;
1478 	return (0);
1479 }
1480 
1481 /*
1482  * Check to see whether n user file descriptors are available to the process
1483  * p.
1484  */
1485 int
1486 fdavail(struct thread *td, int n)
1487 {
1488 	struct proc *p = td->td_proc;
1489 	struct filedesc *fdp = td->td_proc->p_fd;
1490 	struct file **fpp;
1491 	int i, lim, last;
1492 
1493 	FILEDESC_LOCK_ASSERT(fdp);
1494 
1495 	PROC_LOCK(p);
1496 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1497 	PROC_UNLOCK(p);
1498 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1499 		return (1);
1500 	last = min(fdp->fd_nfiles, lim);
1501 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1502 	for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1503 		if (*fpp == NULL && --n <= 0)
1504 			return (1);
1505 	}
1506 	return (0);
1507 }
1508 
1509 /*
1510  * Create a new open file structure and allocate a file decriptor for the
1511  * process that refers to it.  We add one reference to the file for the
1512  * descriptor table and one reference for resultfp. This is to prevent us
1513  * being preempted and the entry in the descriptor table closed after we
1514  * release the FILEDESC lock.
1515  */
1516 int
1517 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1518 {
1519 	struct proc *p = td->td_proc;
1520 	struct file *fp;
1521 	int error, i;
1522 	int maxuserfiles = maxfiles - (maxfiles / 20);
1523 	static struct timeval lastfail;
1524 	static int curfail;
1525 
1526 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1527 	if ((openfiles >= maxuserfiles &&
1528 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1529 	    openfiles >= maxfiles) {
1530 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1531 			printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1532 				td->td_ucred->cr_ruid);
1533 		}
1534 		uma_zfree(file_zone, fp);
1535 		return (ENFILE);
1536 	}
1537 	atomic_add_int(&openfiles, 1);
1538 
1539 	/*
1540 	 * If the process has file descriptor zero open, add the new file
1541 	 * descriptor to the list of open files at that point, otherwise
1542 	 * put it at the front of the list of open files.
1543 	 */
1544 	refcount_init(&fp->f_count, 1);
1545 	if (resultfp)
1546 		fhold(fp);
1547 	fp->f_cred = crhold(td->td_ucred);
1548 	fp->f_ops = &badfileops;
1549 	fp->f_data = NULL;
1550 	fp->f_vnode = NULL;
1551 	FILEDESC_XLOCK(p->p_fd);
1552 	if ((error = fdalloc(td, 0, &i))) {
1553 		FILEDESC_XUNLOCK(p->p_fd);
1554 		fdrop(fp, td);
1555 		if (resultfp)
1556 			fdrop(fp, td);
1557 		return (error);
1558 	}
1559 	p->p_fd->fd_ofiles[i] = fp;
1560 	FILEDESC_XUNLOCK(p->p_fd);
1561 	if (resultfp)
1562 		*resultfp = fp;
1563 	if (resultfd)
1564 		*resultfd = i;
1565 	return (0);
1566 }
1567 
1568 /*
1569  * Build a new filedesc structure from another.
1570  * Copy the current, root, and jail root vnode references.
1571  */
1572 struct filedesc *
1573 fdinit(struct filedesc *fdp)
1574 {
1575 	struct filedesc0 *newfdp;
1576 
1577 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1578 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1579 	if (fdp != NULL) {
1580 		FILEDESC_XLOCK(fdp);
1581 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1582 		if (newfdp->fd_fd.fd_cdir)
1583 			VREF(newfdp->fd_fd.fd_cdir);
1584 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1585 		if (newfdp->fd_fd.fd_rdir)
1586 			VREF(newfdp->fd_fd.fd_rdir);
1587 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1588 		if (newfdp->fd_fd.fd_jdir)
1589 			VREF(newfdp->fd_fd.fd_jdir);
1590 		FILEDESC_XUNLOCK(fdp);
1591 	}
1592 
1593 	/* Create the file descriptor table. */
1594 	newfdp->fd_fd.fd_refcnt = 1;
1595 	newfdp->fd_fd.fd_holdcnt = 1;
1596 	newfdp->fd_fd.fd_cmask = CMASK;
1597 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1598 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1599 	newfdp->fd_fd.fd_nfiles = NDFILE;
1600 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1601 	newfdp->fd_fd.fd_lastfile = -1;
1602 	return (&newfdp->fd_fd);
1603 }
1604 
1605 static struct filedesc *
1606 fdhold(struct proc *p)
1607 {
1608 	struct filedesc *fdp;
1609 
1610 	mtx_lock(&fdesc_mtx);
1611 	fdp = p->p_fd;
1612 	if (fdp != NULL)
1613 		fdp->fd_holdcnt++;
1614 	mtx_unlock(&fdesc_mtx);
1615 	return (fdp);
1616 }
1617 
1618 static void
1619 fddrop(struct filedesc *fdp)
1620 {
1621 	struct filedesc0 *fdp0;
1622 	struct freetable *ft;
1623 	int i;
1624 
1625 	mtx_lock(&fdesc_mtx);
1626 	i = --fdp->fd_holdcnt;
1627 	mtx_unlock(&fdesc_mtx);
1628 	if (i > 0)
1629 		return;
1630 
1631 	FILEDESC_LOCK_DESTROY(fdp);
1632 	fdp0 = (struct filedesc0 *)fdp;
1633 	while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1634 		SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1635 		free(ft->ft_table, M_FILEDESC);
1636 	}
1637 	free(fdp, M_FILEDESC);
1638 }
1639 
1640 /*
1641  * Share a filedesc structure.
1642  */
1643 struct filedesc *
1644 fdshare(struct filedesc *fdp)
1645 {
1646 
1647 	FILEDESC_XLOCK(fdp);
1648 	fdp->fd_refcnt++;
1649 	FILEDESC_XUNLOCK(fdp);
1650 	return (fdp);
1651 }
1652 
1653 /*
1654  * Unshare a filedesc structure, if necessary by making a copy
1655  */
1656 void
1657 fdunshare(struct proc *p, struct thread *td)
1658 {
1659 
1660 	FILEDESC_XLOCK(p->p_fd);
1661 	if (p->p_fd->fd_refcnt > 1) {
1662 		struct filedesc *tmp;
1663 
1664 		FILEDESC_XUNLOCK(p->p_fd);
1665 		tmp = fdcopy(p->p_fd);
1666 		fdfree(td);
1667 		p->p_fd = tmp;
1668 	} else
1669 		FILEDESC_XUNLOCK(p->p_fd);
1670 }
1671 
1672 /*
1673  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1674  * this is to ease callers, not catch errors.
1675  */
1676 struct filedesc *
1677 fdcopy(struct filedesc *fdp)
1678 {
1679 	struct filedesc *newfdp;
1680 	int i;
1681 
1682 	/* Certain daemons might not have file descriptors. */
1683 	if (fdp == NULL)
1684 		return (NULL);
1685 
1686 	newfdp = fdinit(fdp);
1687 	FILEDESC_SLOCK(fdp);
1688 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1689 		FILEDESC_SUNLOCK(fdp);
1690 		FILEDESC_XLOCK(newfdp);
1691 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1692 		FILEDESC_XUNLOCK(newfdp);
1693 		FILEDESC_SLOCK(fdp);
1694 	}
1695 	/* copy everything except kqueue descriptors */
1696 	newfdp->fd_freefile = -1;
1697 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1698 		if (fdisused(fdp, i) &&
1699 		    fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE &&
1700 		    fdp->fd_ofiles[i]->f_ops != &badfileops) {
1701 			newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1702 			newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1703 			fhold(newfdp->fd_ofiles[i]);
1704 			newfdp->fd_lastfile = i;
1705 		} else {
1706 			if (newfdp->fd_freefile == -1)
1707 				newfdp->fd_freefile = i;
1708 		}
1709 	}
1710 	newfdp->fd_cmask = fdp->fd_cmask;
1711 	FILEDESC_SUNLOCK(fdp);
1712 	FILEDESC_XLOCK(newfdp);
1713 	for (i = 0; i <= newfdp->fd_lastfile; ++i)
1714 		if (newfdp->fd_ofiles[i] != NULL)
1715 			fdused(newfdp, i);
1716 	if (newfdp->fd_freefile == -1)
1717 		newfdp->fd_freefile = i;
1718 	FILEDESC_XUNLOCK(newfdp);
1719 	return (newfdp);
1720 }
1721 
1722 /*
1723  * Release a filedesc structure.
1724  */
1725 void
1726 fdfree(struct thread *td)
1727 {
1728 	struct filedesc *fdp;
1729 	struct file **fpp;
1730 	int i, locked;
1731 	struct filedesc_to_leader *fdtol;
1732 	struct file *fp;
1733 	struct vnode *cdir, *jdir, *rdir, *vp;
1734 	struct flock lf;
1735 
1736 	/* Certain daemons might not have file descriptors. */
1737 	fdp = td->td_proc->p_fd;
1738 	if (fdp == NULL)
1739 		return;
1740 
1741 	/* Check for special need to clear POSIX style locks */
1742 	fdtol = td->td_proc->p_fdtol;
1743 	if (fdtol != NULL) {
1744 		FILEDESC_XLOCK(fdp);
1745 		KASSERT(fdtol->fdl_refcount > 0,
1746 			("filedesc_to_refcount botch: fdl_refcount=%d",
1747 			 fdtol->fdl_refcount));
1748 		if (fdtol->fdl_refcount == 1 &&
1749 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1750 			for (i = 0, fpp = fdp->fd_ofiles;
1751 			     i <= fdp->fd_lastfile;
1752 			     i++, fpp++) {
1753 				if (*fpp == NULL ||
1754 				    (*fpp)->f_type != DTYPE_VNODE)
1755 					continue;
1756 				fp = *fpp;
1757 				fhold(fp);
1758 				FILEDESC_XUNLOCK(fdp);
1759 				lf.l_whence = SEEK_SET;
1760 				lf.l_start = 0;
1761 				lf.l_len = 0;
1762 				lf.l_type = F_UNLCK;
1763 				vp = fp->f_vnode;
1764 				locked = VFS_LOCK_GIANT(vp->v_mount);
1765 				(void) VOP_ADVLOCK(vp,
1766 						   (caddr_t)td->td_proc->
1767 						   p_leader,
1768 						   F_UNLCK,
1769 						   &lf,
1770 						   F_POSIX);
1771 				VFS_UNLOCK_GIANT(locked);
1772 				FILEDESC_XLOCK(fdp);
1773 				fdrop(fp, td);
1774 				fpp = fdp->fd_ofiles + i;
1775 			}
1776 		}
1777 	retry:
1778 		if (fdtol->fdl_refcount == 1) {
1779 			if (fdp->fd_holdleaderscount > 0 &&
1780 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1781 				/*
1782 				 * close() or do_dup() has cleared a reference
1783 				 * in a shared file descriptor table.
1784 				 */
1785 				fdp->fd_holdleaderswakeup = 1;
1786 				sx_sleep(&fdp->fd_holdleaderscount,
1787 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1788 				goto retry;
1789 			}
1790 			if (fdtol->fdl_holdcount > 0) {
1791 				/*
1792 				 * Ensure that fdtol->fdl_leader remains
1793 				 * valid in closef().
1794 				 */
1795 				fdtol->fdl_wakeup = 1;
1796 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1797 				    "fdlhold", 0);
1798 				goto retry;
1799 			}
1800 		}
1801 		fdtol->fdl_refcount--;
1802 		if (fdtol->fdl_refcount == 0 &&
1803 		    fdtol->fdl_holdcount == 0) {
1804 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1805 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1806 		} else
1807 			fdtol = NULL;
1808 		td->td_proc->p_fdtol = NULL;
1809 		FILEDESC_XUNLOCK(fdp);
1810 		if (fdtol != NULL)
1811 			free(fdtol, M_FILEDESC_TO_LEADER);
1812 	}
1813 	FILEDESC_XLOCK(fdp);
1814 	i = --fdp->fd_refcnt;
1815 	FILEDESC_XUNLOCK(fdp);
1816 	if (i > 0)
1817 		return;
1818 
1819 	fpp = fdp->fd_ofiles;
1820 	for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1821 		if (*fpp) {
1822 			FILEDESC_XLOCK(fdp);
1823 			fp = *fpp;
1824 			*fpp = NULL;
1825 			FILEDESC_XUNLOCK(fdp);
1826 			(void) closef(fp, td);
1827 		}
1828 	}
1829 	FILEDESC_XLOCK(fdp);
1830 
1831 	/* XXX This should happen earlier. */
1832 	mtx_lock(&fdesc_mtx);
1833 	td->td_proc->p_fd = NULL;
1834 	mtx_unlock(&fdesc_mtx);
1835 
1836 	if (fdp->fd_nfiles > NDFILE)
1837 		free(fdp->fd_ofiles, M_FILEDESC);
1838 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1839 		free(fdp->fd_map, M_FILEDESC);
1840 
1841 	fdp->fd_nfiles = 0;
1842 
1843 	cdir = fdp->fd_cdir;
1844 	fdp->fd_cdir = NULL;
1845 	rdir = fdp->fd_rdir;
1846 	fdp->fd_rdir = NULL;
1847 	jdir = fdp->fd_jdir;
1848 	fdp->fd_jdir = NULL;
1849 	FILEDESC_XUNLOCK(fdp);
1850 
1851 	if (cdir) {
1852 		locked = VFS_LOCK_GIANT(cdir->v_mount);
1853 		vrele(cdir);
1854 		VFS_UNLOCK_GIANT(locked);
1855 	}
1856 	if (rdir) {
1857 		locked = VFS_LOCK_GIANT(rdir->v_mount);
1858 		vrele(rdir);
1859 		VFS_UNLOCK_GIANT(locked);
1860 	}
1861 	if (jdir) {
1862 		locked = VFS_LOCK_GIANT(jdir->v_mount);
1863 		vrele(jdir);
1864 		VFS_UNLOCK_GIANT(locked);
1865 	}
1866 
1867 	fddrop(fdp);
1868 }
1869 
1870 /*
1871  * For setugid programs, we don't want to people to use that setugidness
1872  * to generate error messages which write to a file which otherwise would
1873  * otherwise be off-limits to the process.  We check for filesystems where
1874  * the vnode can change out from under us after execve (like [lin]procfs).
1875  *
1876  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1877  * sufficient.  We also don't check for setugidness since we know we are.
1878  */
1879 static int
1880 is_unsafe(struct file *fp)
1881 {
1882 	if (fp->f_type == DTYPE_VNODE) {
1883 		struct vnode *vp = fp->f_vnode;
1884 
1885 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1886 			return (1);
1887 	}
1888 	return (0);
1889 }
1890 
1891 /*
1892  * Make this setguid thing safe, if at all possible.
1893  */
1894 void
1895 setugidsafety(struct thread *td)
1896 {
1897 	struct filedesc *fdp;
1898 	int i;
1899 
1900 	/* Certain daemons might not have file descriptors. */
1901 	fdp = td->td_proc->p_fd;
1902 	if (fdp == NULL)
1903 		return;
1904 
1905 	/*
1906 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1907 	 * we are blocked in a close.  Be careful!
1908 	 */
1909 	FILEDESC_XLOCK(fdp);
1910 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1911 		if (i > 2)
1912 			break;
1913 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1914 			struct file *fp;
1915 
1916 			knote_fdclose(td, i);
1917 			/*
1918 			 * NULL-out descriptor prior to close to avoid
1919 			 * a race while close blocks.
1920 			 */
1921 			fp = fdp->fd_ofiles[i];
1922 			fdp->fd_ofiles[i] = NULL;
1923 			fdp->fd_ofileflags[i] = 0;
1924 			fdunused(fdp, i);
1925 			FILEDESC_XUNLOCK(fdp);
1926 			(void) closef(fp, td);
1927 			FILEDESC_XLOCK(fdp);
1928 		}
1929 	}
1930 	FILEDESC_XUNLOCK(fdp);
1931 }
1932 
1933 /*
1934  * If a specific file object occupies a specific file descriptor, close the
1935  * file descriptor entry and drop a reference on the file object.  This is a
1936  * convenience function to handle a subsequent error in a function that calls
1937  * falloc() that handles the race that another thread might have closed the
1938  * file descriptor out from under the thread creating the file object.
1939  */
1940 void
1941 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1942 {
1943 
1944 	FILEDESC_XLOCK(fdp);
1945 	if (fdp->fd_ofiles[idx] == fp) {
1946 		fdp->fd_ofiles[idx] = NULL;
1947 		fdunused(fdp, idx);
1948 		FILEDESC_XUNLOCK(fdp);
1949 		fdrop(fp, td);
1950 	} else
1951 		FILEDESC_XUNLOCK(fdp);
1952 }
1953 
1954 /*
1955  * Close any files on exec?
1956  */
1957 void
1958 fdcloseexec(struct thread *td)
1959 {
1960 	struct filedesc *fdp;
1961 	int i;
1962 
1963 	/* Certain daemons might not have file descriptors. */
1964 	fdp = td->td_proc->p_fd;
1965 	if (fdp == NULL)
1966 		return;
1967 
1968 	FILEDESC_XLOCK(fdp);
1969 
1970 	/*
1971 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
1972 	 * may block and rip them out from under us.
1973 	 */
1974 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1975 		if (fdp->fd_ofiles[i] != NULL &&
1976 		    (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1977 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1978 			struct file *fp;
1979 
1980 			knote_fdclose(td, i);
1981 			/*
1982 			 * NULL-out descriptor prior to close to avoid
1983 			 * a race while close blocks.
1984 			 */
1985 			fp = fdp->fd_ofiles[i];
1986 			fdp->fd_ofiles[i] = NULL;
1987 			fdp->fd_ofileflags[i] = 0;
1988 			fdunused(fdp, i);
1989 			if (fp->f_type == DTYPE_MQUEUE)
1990 				mq_fdclose(td, i, fp);
1991 			FILEDESC_XUNLOCK(fdp);
1992 			(void) closef(fp, td);
1993 			FILEDESC_XLOCK(fdp);
1994 		}
1995 	}
1996 	FILEDESC_XUNLOCK(fdp);
1997 }
1998 
1999 /*
2000  * It is unsafe for set[ug]id processes to be started with file
2001  * descriptors 0..2 closed, as these descriptors are given implicit
2002  * significance in the Standard C library.  fdcheckstd() will create a
2003  * descriptor referencing /dev/null for each of stdin, stdout, and
2004  * stderr that is not already open.
2005  */
2006 int
2007 fdcheckstd(struct thread *td)
2008 {
2009 	struct filedesc *fdp;
2010 	register_t retval, save;
2011 	int i, error, devnull;
2012 
2013 	fdp = td->td_proc->p_fd;
2014 	if (fdp == NULL)
2015 		return (0);
2016 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2017 	devnull = -1;
2018 	error = 0;
2019 	for (i = 0; i < 3; i++) {
2020 		if (fdp->fd_ofiles[i] != NULL)
2021 			continue;
2022 		if (devnull < 0) {
2023 			save = td->td_retval[0];
2024 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2025 			    O_RDWR, 0);
2026 			devnull = td->td_retval[0];
2027 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
2028 			td->td_retval[0] = save;
2029 			if (error)
2030 				break;
2031 		} else {
2032 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2033 			if (error != 0)
2034 				break;
2035 		}
2036 	}
2037 	return (error);
2038 }
2039 
2040 /*
2041  * Internal form of close.  Decrement reference count on file structure.
2042  * Note: td may be NULL when closing a file that was being passed in a
2043  * message.
2044  *
2045  * XXXRW: Giant is not required for the caller, but often will be held; this
2046  * makes it moderately likely the Giant will be recursed in the VFS case.
2047  */
2048 int
2049 closef(struct file *fp, struct thread *td)
2050 {
2051 	struct vnode *vp;
2052 	struct flock lf;
2053 	struct filedesc_to_leader *fdtol;
2054 	struct filedesc *fdp;
2055 
2056 	/*
2057 	 * POSIX record locking dictates that any close releases ALL
2058 	 * locks owned by this process.  This is handled by setting
2059 	 * a flag in the unlock to free ONLY locks obeying POSIX
2060 	 * semantics, and not to free BSD-style file locks.
2061 	 * If the descriptor was in a message, POSIX-style locks
2062 	 * aren't passed with the descriptor, and the thread pointer
2063 	 * will be NULL.  Callers should be careful only to pass a
2064 	 * NULL thread pointer when there really is no owning
2065 	 * context that might have locks, or the locks will be
2066 	 * leaked.
2067 	 */
2068 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2069 		int vfslocked;
2070 
2071 		vp = fp->f_vnode;
2072 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2073 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2074 			lf.l_whence = SEEK_SET;
2075 			lf.l_start = 0;
2076 			lf.l_len = 0;
2077 			lf.l_type = F_UNLCK;
2078 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2079 					   F_UNLCK, &lf, F_POSIX);
2080 		}
2081 		fdtol = td->td_proc->p_fdtol;
2082 		if (fdtol != NULL) {
2083 			/*
2084 			 * Handle special case where file descriptor table is
2085 			 * shared between multiple process leaders.
2086 			 */
2087 			fdp = td->td_proc->p_fd;
2088 			FILEDESC_XLOCK(fdp);
2089 			for (fdtol = fdtol->fdl_next;
2090 			     fdtol != td->td_proc->p_fdtol;
2091 			     fdtol = fdtol->fdl_next) {
2092 				if ((fdtol->fdl_leader->p_flag &
2093 				     P_ADVLOCK) == 0)
2094 					continue;
2095 				fdtol->fdl_holdcount++;
2096 				FILEDESC_XUNLOCK(fdp);
2097 				lf.l_whence = SEEK_SET;
2098 				lf.l_start = 0;
2099 				lf.l_len = 0;
2100 				lf.l_type = F_UNLCK;
2101 				vp = fp->f_vnode;
2102 				(void) VOP_ADVLOCK(vp,
2103 						   (caddr_t)fdtol->fdl_leader,
2104 						   F_UNLCK, &lf, F_POSIX);
2105 				FILEDESC_XLOCK(fdp);
2106 				fdtol->fdl_holdcount--;
2107 				if (fdtol->fdl_holdcount == 0 &&
2108 				    fdtol->fdl_wakeup != 0) {
2109 					fdtol->fdl_wakeup = 0;
2110 					wakeup(fdtol);
2111 				}
2112 			}
2113 			FILEDESC_XUNLOCK(fdp);
2114 		}
2115 		VFS_UNLOCK_GIANT(vfslocked);
2116 	}
2117 	return (fdrop(fp, td));
2118 }
2119 
2120 /*
2121  * Initialize the file pointer with the specified properties.
2122  *
2123  * The ops are set with release semantics to be certain that the flags, type,
2124  * and data are visible when ops is.  This is to prevent ops methods from being
2125  * called with bad data.
2126  */
2127 void
2128 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2129 {
2130 	fp->f_data = data;
2131 	fp->f_flag = flag;
2132 	fp->f_type = type;
2133 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2134 }
2135 
2136 struct file *
2137 fget_unlocked(struct filedesc *fdp, int fd)
2138 {
2139 	struct file *fp;
2140 	u_int count;
2141 
2142 	if (fd < 0 || fd >= fdp->fd_nfiles)
2143 		return (NULL);
2144 	/*
2145 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2146 	 * never raising a refcount above 0.  To accomplish this we have
2147 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2148 	 * must be re-verified once we acquire a reference to be certain
2149 	 * that the identity is still correct and we did not lose a race
2150 	 * due to preemption.
2151 	 */
2152 	for (;;) {
2153 		fp = fdp->fd_ofiles[fd];
2154 		if (fp == NULL)
2155 			break;
2156 		count = fp->f_count;
2157 		if (count == 0)
2158 			continue;
2159 		/*
2160 		 * Use an acquire barrier to prevent caching of fd_ofiles
2161 		 * so it is refreshed for verification.
2162 		 */
2163 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2164 			continue;
2165 		if (fp == fdp->fd_ofiles[fd])
2166 			break;
2167 		fdrop(fp, curthread);
2168 	}
2169 
2170 	return (fp);
2171 }
2172 
2173 /*
2174  * Extract the file pointer associated with the specified descriptor for the
2175  * current user process.
2176  *
2177  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2178  * returned.
2179  *
2180  * If an error occured the non-zero error is returned and *fpp is set to
2181  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2182  * responsible for fdrop().
2183  */
2184 static __inline int
2185 _fget(struct thread *td, int fd, struct file **fpp, int flags)
2186 {
2187 	struct filedesc *fdp;
2188 	struct file *fp;
2189 
2190 	*fpp = NULL;
2191 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2192 		return (EBADF);
2193 	if ((fp = fget_unlocked(fdp, fd)) == NULL)
2194 		return (EBADF);
2195 	if (fp->f_ops == &badfileops) {
2196 		fdrop(fp, td);
2197 		return (EBADF);
2198 	}
2199 	/*
2200 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2201 	 *
2202 	 * Only one flag, or 0, may be specified.
2203 	 */
2204 	if ((flags == FREAD && (fp->f_flag & FREAD) == 0) ||
2205 	    (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) {
2206 		fdrop(fp, td);
2207 		return (EBADF);
2208 	}
2209 	*fpp = fp;
2210 	return (0);
2211 }
2212 
2213 int
2214 fget(struct thread *td, int fd, struct file **fpp)
2215 {
2216 
2217 	return(_fget(td, fd, fpp, 0));
2218 }
2219 
2220 int
2221 fget_read(struct thread *td, int fd, struct file **fpp)
2222 {
2223 
2224 	return(_fget(td, fd, fpp, FREAD));
2225 }
2226 
2227 int
2228 fget_write(struct thread *td, int fd, struct file **fpp)
2229 {
2230 
2231 	return(_fget(td, fd, fpp, FWRITE));
2232 }
2233 
2234 /*
2235  * Like fget() but loads the underlying vnode, or returns an error if the
2236  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2237  * never have VM objects.  The returned vnode will be vref()'d.
2238  *
2239  * XXX: what about the unused flags ?
2240  */
2241 static __inline int
2242 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2243 {
2244 	struct file *fp;
2245 	int error;
2246 
2247 	*vpp = NULL;
2248 	if ((error = _fget(td, fd, &fp, flags)) != 0)
2249 		return (error);
2250 	if (fp->f_vnode == NULL) {
2251 		error = EINVAL;
2252 	} else {
2253 		*vpp = fp->f_vnode;
2254 		vref(*vpp);
2255 	}
2256 	fdrop(fp, td);
2257 
2258 	return (error);
2259 }
2260 
2261 int
2262 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2263 {
2264 
2265 	return (_fgetvp(td, fd, vpp, 0));
2266 }
2267 
2268 int
2269 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2270 {
2271 
2272 	return (_fgetvp(td, fd, vpp, FREAD));
2273 }
2274 
2275 #ifdef notyet
2276 int
2277 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2278 {
2279 
2280 	return (_fgetvp(td, fd, vpp, FWRITE));
2281 }
2282 #endif
2283 
2284 /*
2285  * Like fget() but loads the underlying socket, or returns an error if the
2286  * descriptor does not represent a socket.
2287  *
2288  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2289  * in the future.
2290  *
2291  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2292  * on their file descriptor reference to prevent the socket from being free'd
2293  * during use.
2294  */
2295 int
2296 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2297 {
2298 	struct file *fp;
2299 	int error;
2300 
2301 	*spp = NULL;
2302 	if (fflagp != NULL)
2303 		*fflagp = 0;
2304 	if ((error = _fget(td, fd, &fp, 0)) != 0)
2305 		return (error);
2306 	if (fp->f_type != DTYPE_SOCKET) {
2307 		error = ENOTSOCK;
2308 	} else {
2309 		*spp = fp->f_data;
2310 		if (fflagp)
2311 			*fflagp = fp->f_flag;
2312 		SOCK_LOCK(*spp);
2313 		soref(*spp);
2314 		SOCK_UNLOCK(*spp);
2315 	}
2316 	fdrop(fp, td);
2317 
2318 	return (error);
2319 }
2320 
2321 /*
2322  * Drop the reference count on the socket and XXX release the SX lock in the
2323  * future.  The last reference closes the socket.
2324  *
2325  * Note: fputsock() is deprecated, see comment for fgetsock().
2326  */
2327 void
2328 fputsock(struct socket *so)
2329 {
2330 
2331 	ACCEPT_LOCK();
2332 	SOCK_LOCK(so);
2333 	sorele(so);
2334 }
2335 
2336 /*
2337  * Handle the last reference to a file being closed.
2338  */
2339 int
2340 _fdrop(struct file *fp, struct thread *td)
2341 {
2342 	int error;
2343 
2344 	error = 0;
2345 	if (fp->f_count != 0)
2346 		panic("fdrop: count %d", fp->f_count);
2347 	if (fp->f_ops != &badfileops)
2348 		error = fo_close(fp, td);
2349 	/*
2350 	 * The f_cdevpriv cannot be assigned non-NULL value while we
2351 	 * are destroying the file.
2352 	 */
2353 	if (fp->f_cdevpriv != NULL)
2354 		devfs_fpdrop(fp);
2355 	atomic_subtract_int(&openfiles, 1);
2356 	crfree(fp->f_cred);
2357 	uma_zfree(file_zone, fp);
2358 
2359 	return (error);
2360 }
2361 
2362 /*
2363  * Apply an advisory lock on a file descriptor.
2364  *
2365  * Just attempt to get a record lock of the requested type on the entire file
2366  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2367  */
2368 #ifndef _SYS_SYSPROTO_H_
2369 struct flock_args {
2370 	int	fd;
2371 	int	how;
2372 };
2373 #endif
2374 /* ARGSUSED */
2375 int
2376 flock(struct thread *td, struct flock_args *uap)
2377 {
2378 	struct file *fp;
2379 	struct vnode *vp;
2380 	struct flock lf;
2381 	int vfslocked;
2382 	int error;
2383 
2384 	if ((error = fget(td, uap->fd, &fp)) != 0)
2385 		return (error);
2386 	if (fp->f_type != DTYPE_VNODE) {
2387 		fdrop(fp, td);
2388 		return (EOPNOTSUPP);
2389 	}
2390 
2391 	vp = fp->f_vnode;
2392 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2393 	lf.l_whence = SEEK_SET;
2394 	lf.l_start = 0;
2395 	lf.l_len = 0;
2396 	if (uap->how & LOCK_UN) {
2397 		lf.l_type = F_UNLCK;
2398 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2399 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2400 		goto done2;
2401 	}
2402 	if (uap->how & LOCK_EX)
2403 		lf.l_type = F_WRLCK;
2404 	else if (uap->how & LOCK_SH)
2405 		lf.l_type = F_RDLCK;
2406 	else {
2407 		error = EBADF;
2408 		goto done2;
2409 	}
2410 	atomic_set_int(&fp->f_flag, FHASLOCK);
2411 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2412 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2413 done2:
2414 	fdrop(fp, td);
2415 	VFS_UNLOCK_GIANT(vfslocked);
2416 	return (error);
2417 }
2418 /*
2419  * Duplicate the specified descriptor to a free descriptor.
2420  */
2421 int
2422 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2423 {
2424 	struct file *wfp;
2425 	struct file *fp;
2426 
2427 	/*
2428 	 * If the to-be-dup'd fd number is greater than the allowed number
2429 	 * of file descriptors, or the fd to be dup'd has already been
2430 	 * closed, then reject.
2431 	 */
2432 	FILEDESC_XLOCK(fdp);
2433 	if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2434 	    (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2435 		FILEDESC_XUNLOCK(fdp);
2436 		return (EBADF);
2437 	}
2438 
2439 	/*
2440 	 * There are two cases of interest here.
2441 	 *
2442 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2443 	 *
2444 	 * For ENXIO steal away the file structure from (dfd) and store it in
2445 	 * (indx).  (dfd) is effectively closed by this operation.
2446 	 *
2447 	 * Any other error code is just returned.
2448 	 */
2449 	switch (error) {
2450 	case ENODEV:
2451 		/*
2452 		 * Check that the mode the file is being opened for is a
2453 		 * subset of the mode of the existing descriptor.
2454 		 */
2455 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2456 			FILEDESC_XUNLOCK(fdp);
2457 			return (EACCES);
2458 		}
2459 		fp = fdp->fd_ofiles[indx];
2460 		fdp->fd_ofiles[indx] = wfp;
2461 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2462 		if (fp == NULL)
2463 			fdused(fdp, indx);
2464 		fhold(wfp);
2465 		FILEDESC_XUNLOCK(fdp);
2466 		if (fp != NULL)
2467 			/*
2468 			 * We now own the reference to fp that the ofiles[]
2469 			 * array used to own.  Release it.
2470 			 */
2471 			fdrop(fp, td);
2472 		return (0);
2473 
2474 	case ENXIO:
2475 		/*
2476 		 * Steal away the file pointer from dfd and stuff it into indx.
2477 		 */
2478 		fp = fdp->fd_ofiles[indx];
2479 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2480 		fdp->fd_ofiles[dfd] = NULL;
2481 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2482 		fdp->fd_ofileflags[dfd] = 0;
2483 		fdunused(fdp, dfd);
2484 		if (fp == NULL)
2485 			fdused(fdp, indx);
2486 		FILEDESC_XUNLOCK(fdp);
2487 
2488 		/*
2489 		 * We now own the reference to fp that the ofiles[] array
2490 		 * used to own.  Release it.
2491 		 */
2492 		if (fp != NULL)
2493 			fdrop(fp, td);
2494 		return (0);
2495 
2496 	default:
2497 		FILEDESC_XUNLOCK(fdp);
2498 		return (error);
2499 	}
2500 	/* NOTREACHED */
2501 }
2502 
2503 /*
2504  * Scan all active processes and prisons to see if any of them have a current
2505  * or root directory of `olddp'. If so, replace them with the new mount point.
2506  */
2507 void
2508 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2509 {
2510 	struct filedesc *fdp;
2511 	struct prison *pr;
2512 	struct proc *p;
2513 	int nrele;
2514 
2515 	if (vrefcnt(olddp) == 1)
2516 		return;
2517 	nrele = 0;
2518 	sx_slock(&allproc_lock);
2519 	FOREACH_PROC_IN_SYSTEM(p) {
2520 		fdp = fdhold(p);
2521 		if (fdp == NULL)
2522 			continue;
2523 		FILEDESC_XLOCK(fdp);
2524 		if (fdp->fd_cdir == olddp) {
2525 			vref(newdp);
2526 			fdp->fd_cdir = newdp;
2527 			nrele++;
2528 		}
2529 		if (fdp->fd_rdir == olddp) {
2530 			vref(newdp);
2531 			fdp->fd_rdir = newdp;
2532 			nrele++;
2533 		}
2534 		if (fdp->fd_jdir == olddp) {
2535 			vref(newdp);
2536 			fdp->fd_jdir = newdp;
2537 			nrele++;
2538 		}
2539 		FILEDESC_XUNLOCK(fdp);
2540 		fddrop(fdp);
2541 	}
2542 	sx_sunlock(&allproc_lock);
2543 	if (rootvnode == olddp) {
2544 		vref(newdp);
2545 		rootvnode = newdp;
2546 		nrele++;
2547 	}
2548 	mtx_lock(&prison0.pr_mtx);
2549 	if (prison0.pr_root == olddp) {
2550 		vref(newdp);
2551 		prison0.pr_root = newdp;
2552 		nrele++;
2553 	}
2554 	mtx_unlock(&prison0.pr_mtx);
2555 	sx_slock(&allprison_lock);
2556 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2557 		mtx_lock(&pr->pr_mtx);
2558 		if (pr->pr_root == olddp) {
2559 			vref(newdp);
2560 			pr->pr_root = newdp;
2561 			nrele++;
2562 		}
2563 		mtx_unlock(&pr->pr_mtx);
2564 	}
2565 	sx_sunlock(&allprison_lock);
2566 	while (nrele--)
2567 		vrele(olddp);
2568 }
2569 
2570 struct filedesc_to_leader *
2571 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2572 {
2573 	struct filedesc_to_leader *fdtol;
2574 
2575 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2576 	       M_FILEDESC_TO_LEADER,
2577 	       M_WAITOK);
2578 	fdtol->fdl_refcount = 1;
2579 	fdtol->fdl_holdcount = 0;
2580 	fdtol->fdl_wakeup = 0;
2581 	fdtol->fdl_leader = leader;
2582 	if (old != NULL) {
2583 		FILEDESC_XLOCK(fdp);
2584 		fdtol->fdl_next = old->fdl_next;
2585 		fdtol->fdl_prev = old;
2586 		old->fdl_next = fdtol;
2587 		fdtol->fdl_next->fdl_prev = fdtol;
2588 		FILEDESC_XUNLOCK(fdp);
2589 	} else {
2590 		fdtol->fdl_next = fdtol;
2591 		fdtol->fdl_prev = fdtol;
2592 	}
2593 	return (fdtol);
2594 }
2595 
2596 /*
2597  * Get file structures globally.
2598  */
2599 static int
2600 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2601 {
2602 	struct xfile xf;
2603 	struct filedesc *fdp;
2604 	struct file *fp;
2605 	struct proc *p;
2606 	int error, n;
2607 
2608 	error = sysctl_wire_old_buffer(req, 0);
2609 	if (error != 0)
2610 		return (error);
2611 	if (req->oldptr == NULL) {
2612 		n = 0;
2613 		sx_slock(&allproc_lock);
2614 		FOREACH_PROC_IN_SYSTEM(p) {
2615 			if (p->p_state == PRS_NEW)
2616 				continue;
2617 			fdp = fdhold(p);
2618 			if (fdp == NULL)
2619 				continue;
2620 			/* overestimates sparse tables. */
2621 			if (fdp->fd_lastfile > 0)
2622 				n += fdp->fd_lastfile;
2623 			fddrop(fdp);
2624 		}
2625 		sx_sunlock(&allproc_lock);
2626 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2627 	}
2628 	error = 0;
2629 	bzero(&xf, sizeof(xf));
2630 	xf.xf_size = sizeof(xf);
2631 	sx_slock(&allproc_lock);
2632 	FOREACH_PROC_IN_SYSTEM(p) {
2633 		if (p->p_state == PRS_NEW)
2634 			continue;
2635 		PROC_LOCK(p);
2636 		if (p_cansee(req->td, p) != 0) {
2637 			PROC_UNLOCK(p);
2638 			continue;
2639 		}
2640 		xf.xf_pid = p->p_pid;
2641 		xf.xf_uid = p->p_ucred->cr_uid;
2642 		PROC_UNLOCK(p);
2643 		fdp = fdhold(p);
2644 		if (fdp == NULL)
2645 			continue;
2646 		FILEDESC_SLOCK(fdp);
2647 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2648 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2649 				continue;
2650 			xf.xf_fd = n;
2651 			xf.xf_file = fp;
2652 			xf.xf_data = fp->f_data;
2653 			xf.xf_vnode = fp->f_vnode;
2654 			xf.xf_type = fp->f_type;
2655 			xf.xf_count = fp->f_count;
2656 			xf.xf_msgcount = 0;
2657 			xf.xf_offset = fp->f_offset;
2658 			xf.xf_flag = fp->f_flag;
2659 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2660 			if (error)
2661 				break;
2662 		}
2663 		FILEDESC_SUNLOCK(fdp);
2664 		fddrop(fdp);
2665 		if (error)
2666 			break;
2667 	}
2668 	sx_sunlock(&allproc_lock);
2669 	return (error);
2670 }
2671 
2672 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2673     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2674 
2675 #ifdef KINFO_OFILE_SIZE
2676 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2677 #endif
2678 
2679 #ifdef COMPAT_FREEBSD7
2680 static int
2681 export_vnode_for_osysctl(struct vnode *vp, int type,
2682     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2683 {
2684 	int error;
2685 	char *fullpath, *freepath;
2686 	int vfslocked;
2687 
2688 	bzero(kif, sizeof(*kif));
2689 	kif->kf_structsize = sizeof(*kif);
2690 
2691 	vref(vp);
2692 	kif->kf_fd = type;
2693 	kif->kf_type = KF_TYPE_VNODE;
2694 	/* This function only handles directories. */
2695 	if (vp->v_type != VDIR) {
2696 		vrele(vp);
2697 		return (ENOTDIR);
2698 	}
2699 	kif->kf_vnode_type = KF_VTYPE_VDIR;
2700 
2701 	/*
2702 	 * This is not a true file descriptor, so we set a bogus refcount
2703 	 * and offset to indicate these fields should be ignored.
2704 	 */
2705 	kif->kf_ref_count = -1;
2706 	kif->kf_offset = -1;
2707 
2708 	freepath = NULL;
2709 	fullpath = "-";
2710 	FILEDESC_SUNLOCK(fdp);
2711 	vn_fullpath(curthread, vp, &fullpath, &freepath);
2712 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2713 	vrele(vp);
2714 	VFS_UNLOCK_GIANT(vfslocked);
2715 	strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2716 	if (freepath != NULL)
2717 		free(freepath, M_TEMP);
2718 	error = SYSCTL_OUT(req, kif, sizeof(*kif));
2719 	FILEDESC_SLOCK(fdp);
2720 	return (error);
2721 }
2722 
2723 /*
2724  * Get per-process file descriptors for use by procstat(1), et al.
2725  */
2726 static int
2727 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2728 {
2729 	char *fullpath, *freepath;
2730 	struct kinfo_ofile *kif;
2731 	struct filedesc *fdp;
2732 	int error, i, *name;
2733 	struct socket *so;
2734 	struct vnode *vp;
2735 	struct file *fp;
2736 	struct proc *p;
2737 	struct tty *tp;
2738 	int vfslocked;
2739 
2740 	name = (int *)arg1;
2741 	if ((p = pfind((pid_t)name[0])) == NULL)
2742 		return (ESRCH);
2743 	if ((error = p_candebug(curthread, p))) {
2744 		PROC_UNLOCK(p);
2745 		return (error);
2746 	}
2747 	fdp = fdhold(p);
2748 	PROC_UNLOCK(p);
2749 	if (fdp == NULL)
2750 		return (ENOENT);
2751 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2752 	FILEDESC_SLOCK(fdp);
2753 	if (fdp->fd_cdir != NULL)
2754 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2755 				fdp, req);
2756 	if (fdp->fd_rdir != NULL)
2757 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2758 				fdp, req);
2759 	if (fdp->fd_jdir != NULL)
2760 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2761 				fdp, req);
2762 	for (i = 0; i < fdp->fd_nfiles; i++) {
2763 		if ((fp = fdp->fd_ofiles[i]) == NULL)
2764 			continue;
2765 		bzero(kif, sizeof(*kif));
2766 		kif->kf_structsize = sizeof(*kif);
2767 		vp = NULL;
2768 		so = NULL;
2769 		tp = NULL;
2770 		kif->kf_fd = i;
2771 		switch (fp->f_type) {
2772 		case DTYPE_VNODE:
2773 			kif->kf_type = KF_TYPE_VNODE;
2774 			vp = fp->f_vnode;
2775 			break;
2776 
2777 		case DTYPE_SOCKET:
2778 			kif->kf_type = KF_TYPE_SOCKET;
2779 			so = fp->f_data;
2780 			break;
2781 
2782 		case DTYPE_PIPE:
2783 			kif->kf_type = KF_TYPE_PIPE;
2784 			break;
2785 
2786 		case DTYPE_FIFO:
2787 			kif->kf_type = KF_TYPE_FIFO;
2788 			vp = fp->f_vnode;
2789 			break;
2790 
2791 		case DTYPE_KQUEUE:
2792 			kif->kf_type = KF_TYPE_KQUEUE;
2793 			break;
2794 
2795 		case DTYPE_CRYPTO:
2796 			kif->kf_type = KF_TYPE_CRYPTO;
2797 			break;
2798 
2799 		case DTYPE_MQUEUE:
2800 			kif->kf_type = KF_TYPE_MQUEUE;
2801 			break;
2802 
2803 		case DTYPE_SHM:
2804 			kif->kf_type = KF_TYPE_SHM;
2805 			break;
2806 
2807 		case DTYPE_SEM:
2808 			kif->kf_type = KF_TYPE_SEM;
2809 			break;
2810 
2811 		case DTYPE_PTS:
2812 			kif->kf_type = KF_TYPE_PTS;
2813 			tp = fp->f_data;
2814 			break;
2815 
2816 		default:
2817 			kif->kf_type = KF_TYPE_UNKNOWN;
2818 			break;
2819 		}
2820 		kif->kf_ref_count = fp->f_count;
2821 		if (fp->f_flag & FREAD)
2822 			kif->kf_flags |= KF_FLAG_READ;
2823 		if (fp->f_flag & FWRITE)
2824 			kif->kf_flags |= KF_FLAG_WRITE;
2825 		if (fp->f_flag & FAPPEND)
2826 			kif->kf_flags |= KF_FLAG_APPEND;
2827 		if (fp->f_flag & FASYNC)
2828 			kif->kf_flags |= KF_FLAG_ASYNC;
2829 		if (fp->f_flag & FFSYNC)
2830 			kif->kf_flags |= KF_FLAG_FSYNC;
2831 		if (fp->f_flag & FNONBLOCK)
2832 			kif->kf_flags |= KF_FLAG_NONBLOCK;
2833 		if (fp->f_flag & O_DIRECT)
2834 			kif->kf_flags |= KF_FLAG_DIRECT;
2835 		if (fp->f_flag & FHASLOCK)
2836 			kif->kf_flags |= KF_FLAG_HASLOCK;
2837 		kif->kf_offset = fp->f_offset;
2838 		if (vp != NULL) {
2839 			vref(vp);
2840 			switch (vp->v_type) {
2841 			case VNON:
2842 				kif->kf_vnode_type = KF_VTYPE_VNON;
2843 				break;
2844 			case VREG:
2845 				kif->kf_vnode_type = KF_VTYPE_VREG;
2846 				break;
2847 			case VDIR:
2848 				kif->kf_vnode_type = KF_VTYPE_VDIR;
2849 				break;
2850 			case VBLK:
2851 				kif->kf_vnode_type = KF_VTYPE_VBLK;
2852 				break;
2853 			case VCHR:
2854 				kif->kf_vnode_type = KF_VTYPE_VCHR;
2855 				break;
2856 			case VLNK:
2857 				kif->kf_vnode_type = KF_VTYPE_VLNK;
2858 				break;
2859 			case VSOCK:
2860 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
2861 				break;
2862 			case VFIFO:
2863 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
2864 				break;
2865 			case VBAD:
2866 				kif->kf_vnode_type = KF_VTYPE_VBAD;
2867 				break;
2868 			default:
2869 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2870 				break;
2871 			}
2872 			/*
2873 			 * It is OK to drop the filedesc lock here as we will
2874 			 * re-validate and re-evaluate its properties when
2875 			 * the loop continues.
2876 			 */
2877 			freepath = NULL;
2878 			fullpath = "-";
2879 			FILEDESC_SUNLOCK(fdp);
2880 			vn_fullpath(curthread, vp, &fullpath, &freepath);
2881 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2882 			vrele(vp);
2883 			VFS_UNLOCK_GIANT(vfslocked);
2884 			strlcpy(kif->kf_path, fullpath,
2885 			    sizeof(kif->kf_path));
2886 			if (freepath != NULL)
2887 				free(freepath, M_TEMP);
2888 			FILEDESC_SLOCK(fdp);
2889 		}
2890 		if (so != NULL) {
2891 			struct sockaddr *sa;
2892 
2893 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2894 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2895 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2896 				free(sa, M_SONAME);
2897 			}
2898 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2899 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2900 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2901 				free(sa, M_SONAME);
2902 			}
2903 			kif->kf_sock_domain =
2904 			    so->so_proto->pr_domain->dom_family;
2905 			kif->kf_sock_type = so->so_type;
2906 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
2907 		}
2908 		if (tp != NULL) {
2909 			strlcpy(kif->kf_path, tty_devname(tp),
2910 			    sizeof(kif->kf_path));
2911 		}
2912 		error = SYSCTL_OUT(req, kif, sizeof(*kif));
2913 		if (error)
2914 			break;
2915 	}
2916 	FILEDESC_SUNLOCK(fdp);
2917 	fddrop(fdp);
2918 	free(kif, M_TEMP);
2919 	return (0);
2920 }
2921 
2922 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
2923     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
2924 #endif	/* COMPAT_FREEBSD7 */
2925 
2926 #ifdef KINFO_FILE_SIZE
2927 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2928 #endif
2929 
2930 static int
2931 export_vnode_for_sysctl(struct vnode *vp, int type,
2932     struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2933 {
2934 	int error;
2935 	char *fullpath, *freepath;
2936 	int vfslocked;
2937 
2938 	bzero(kif, sizeof(*kif));
2939 
2940 	vref(vp);
2941 	kif->kf_fd = type;
2942 	kif->kf_type = KF_TYPE_VNODE;
2943 	/* This function only handles directories. */
2944 	if (vp->v_type != VDIR) {
2945 		vrele(vp);
2946 		return (ENOTDIR);
2947 	}
2948 	kif->kf_vnode_type = KF_VTYPE_VDIR;
2949 
2950 	/*
2951 	 * This is not a true file descriptor, so we set a bogus refcount
2952 	 * and offset to indicate these fields should be ignored.
2953 	 */
2954 	kif->kf_ref_count = -1;
2955 	kif->kf_offset = -1;
2956 
2957 	freepath = NULL;
2958 	fullpath = "-";
2959 	FILEDESC_SUNLOCK(fdp);
2960 	vn_fullpath(curthread, vp, &fullpath, &freepath);
2961 	vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2962 	vrele(vp);
2963 	VFS_UNLOCK_GIANT(vfslocked);
2964 	strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2965 	if (freepath != NULL)
2966 		free(freepath, M_TEMP);
2967 	/* Pack record size down */
2968 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
2969 	    strlen(kif->kf_path) + 1;
2970 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
2971 	error = SYSCTL_OUT(req, kif, kif->kf_structsize);
2972 	FILEDESC_SLOCK(fdp);
2973 	return (error);
2974 }
2975 
2976 /*
2977  * Get per-process file descriptors for use by procstat(1), et al.
2978  */
2979 static int
2980 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2981 {
2982 	char *fullpath, *freepath;
2983 	struct kinfo_file *kif;
2984 	struct filedesc *fdp;
2985 	int error, i, *name;
2986 	struct socket *so;
2987 	struct vnode *vp;
2988 	struct file *fp;
2989 	struct proc *p;
2990 	struct tty *tp;
2991 	int vfslocked;
2992 	size_t oldidx;
2993 
2994 	name = (int *)arg1;
2995 	if ((p = pfind((pid_t)name[0])) == NULL)
2996 		return (ESRCH);
2997 	if ((error = p_candebug(curthread, p))) {
2998 		PROC_UNLOCK(p);
2999 		return (error);
3000 	}
3001 	fdp = fdhold(p);
3002 	PROC_UNLOCK(p);
3003 	if (fdp == NULL)
3004 		return (ENOENT);
3005 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3006 	FILEDESC_SLOCK(fdp);
3007 	if (fdp->fd_cdir != NULL)
3008 		export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3009 				fdp, req);
3010 	if (fdp->fd_rdir != NULL)
3011 		export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3012 				fdp, req);
3013 	if (fdp->fd_jdir != NULL)
3014 		export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3015 				fdp, req);
3016 	for (i = 0; i < fdp->fd_nfiles; i++) {
3017 		if ((fp = fdp->fd_ofiles[i]) == NULL)
3018 			continue;
3019 		bzero(kif, sizeof(*kif));
3020 		vp = NULL;
3021 		so = NULL;
3022 		tp = NULL;
3023 		kif->kf_fd = i;
3024 		switch (fp->f_type) {
3025 		case DTYPE_VNODE:
3026 			kif->kf_type = KF_TYPE_VNODE;
3027 			vp = fp->f_vnode;
3028 			break;
3029 
3030 		case DTYPE_SOCKET:
3031 			kif->kf_type = KF_TYPE_SOCKET;
3032 			so = fp->f_data;
3033 			break;
3034 
3035 		case DTYPE_PIPE:
3036 			kif->kf_type = KF_TYPE_PIPE;
3037 			break;
3038 
3039 		case DTYPE_FIFO:
3040 			kif->kf_type = KF_TYPE_FIFO;
3041 			vp = fp->f_vnode;
3042 			break;
3043 
3044 		case DTYPE_KQUEUE:
3045 			kif->kf_type = KF_TYPE_KQUEUE;
3046 			break;
3047 
3048 		case DTYPE_CRYPTO:
3049 			kif->kf_type = KF_TYPE_CRYPTO;
3050 			break;
3051 
3052 		case DTYPE_MQUEUE:
3053 			kif->kf_type = KF_TYPE_MQUEUE;
3054 			break;
3055 
3056 		case DTYPE_SHM:
3057 			kif->kf_type = KF_TYPE_SHM;
3058 			break;
3059 
3060 		case DTYPE_SEM:
3061 			kif->kf_type = KF_TYPE_SEM;
3062 			break;
3063 
3064 		case DTYPE_PTS:
3065 			kif->kf_type = KF_TYPE_PTS;
3066 			tp = fp->f_data;
3067 			break;
3068 
3069 		default:
3070 			kif->kf_type = KF_TYPE_UNKNOWN;
3071 			break;
3072 		}
3073 		kif->kf_ref_count = fp->f_count;
3074 		if (fp->f_flag & FREAD)
3075 			kif->kf_flags |= KF_FLAG_READ;
3076 		if (fp->f_flag & FWRITE)
3077 			kif->kf_flags |= KF_FLAG_WRITE;
3078 		if (fp->f_flag & FAPPEND)
3079 			kif->kf_flags |= KF_FLAG_APPEND;
3080 		if (fp->f_flag & FASYNC)
3081 			kif->kf_flags |= KF_FLAG_ASYNC;
3082 		if (fp->f_flag & FFSYNC)
3083 			kif->kf_flags |= KF_FLAG_FSYNC;
3084 		if (fp->f_flag & FNONBLOCK)
3085 			kif->kf_flags |= KF_FLAG_NONBLOCK;
3086 		if (fp->f_flag & O_DIRECT)
3087 			kif->kf_flags |= KF_FLAG_DIRECT;
3088 		if (fp->f_flag & FHASLOCK)
3089 			kif->kf_flags |= KF_FLAG_HASLOCK;
3090 		kif->kf_offset = fp->f_offset;
3091 		if (vp != NULL) {
3092 			vref(vp);
3093 			switch (vp->v_type) {
3094 			case VNON:
3095 				kif->kf_vnode_type = KF_VTYPE_VNON;
3096 				break;
3097 			case VREG:
3098 				kif->kf_vnode_type = KF_VTYPE_VREG;
3099 				break;
3100 			case VDIR:
3101 				kif->kf_vnode_type = KF_VTYPE_VDIR;
3102 				break;
3103 			case VBLK:
3104 				kif->kf_vnode_type = KF_VTYPE_VBLK;
3105 				break;
3106 			case VCHR:
3107 				kif->kf_vnode_type = KF_VTYPE_VCHR;
3108 				break;
3109 			case VLNK:
3110 				kif->kf_vnode_type = KF_VTYPE_VLNK;
3111 				break;
3112 			case VSOCK:
3113 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
3114 				break;
3115 			case VFIFO:
3116 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
3117 				break;
3118 			case VBAD:
3119 				kif->kf_vnode_type = KF_VTYPE_VBAD;
3120 				break;
3121 			default:
3122 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3123 				break;
3124 			}
3125 			/*
3126 			 * It is OK to drop the filedesc lock here as we will
3127 			 * re-validate and re-evaluate its properties when
3128 			 * the loop continues.
3129 			 */
3130 			freepath = NULL;
3131 			fullpath = "-";
3132 			FILEDESC_SUNLOCK(fdp);
3133 			vn_fullpath(curthread, vp, &fullpath, &freepath);
3134 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3135 			vrele(vp);
3136 			VFS_UNLOCK_GIANT(vfslocked);
3137 			strlcpy(kif->kf_path, fullpath,
3138 			    sizeof(kif->kf_path));
3139 			if (freepath != NULL)
3140 				free(freepath, M_TEMP);
3141 			FILEDESC_SLOCK(fdp);
3142 		}
3143 		if (so != NULL) {
3144 			struct sockaddr *sa;
3145 
3146 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3147 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3148 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3149 				free(sa, M_SONAME);
3150 			}
3151 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3152 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3153 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3154 				free(sa, M_SONAME);
3155 			}
3156 			kif->kf_sock_domain =
3157 			    so->so_proto->pr_domain->dom_family;
3158 			kif->kf_sock_type = so->so_type;
3159 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
3160 		}
3161 		if (tp != NULL) {
3162 			strlcpy(kif->kf_path, tty_devname(tp),
3163 			    sizeof(kif->kf_path));
3164 		}
3165 		/* Pack record size down */
3166 		kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3167 		    strlen(kif->kf_path) + 1;
3168 		kif->kf_structsize = roundup(kif->kf_structsize,
3169 		    sizeof(uint64_t));
3170 		oldidx = req->oldidx;
3171 		error = SYSCTL_OUT(req, kif, kif->kf_structsize);
3172 		if (error) {
3173 			if (error == ENOMEM) {
3174 				/*
3175 				 * The hack to keep the ABI of sysctl
3176 				 * kern.proc.filedesc intact, but not
3177 				 * to account a partially copied
3178 				 * kinfo_file into the oldidx.
3179 				 */
3180 				req->oldidx = oldidx;
3181 				error = 0;
3182 			}
3183 			break;
3184 		}
3185 	}
3186 	FILEDESC_SUNLOCK(fdp);
3187 	fddrop(fdp);
3188 	free(kif, M_TEMP);
3189 	return (error);
3190 }
3191 
3192 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3193     sysctl_kern_proc_filedesc, "Process filedesc entries");
3194 
3195 #ifdef DDB
3196 /*
3197  * For the purposes of debugging, generate a human-readable string for the
3198  * file type.
3199  */
3200 static const char *
3201 file_type_to_name(short type)
3202 {
3203 
3204 	switch (type) {
3205 	case 0:
3206 		return ("zero");
3207 	case DTYPE_VNODE:
3208 		return ("vnod");
3209 	case DTYPE_SOCKET:
3210 		return ("sock");
3211 	case DTYPE_PIPE:
3212 		return ("pipe");
3213 	case DTYPE_FIFO:
3214 		return ("fifo");
3215 	case DTYPE_KQUEUE:
3216 		return ("kque");
3217 	case DTYPE_CRYPTO:
3218 		return ("crpt");
3219 	case DTYPE_MQUEUE:
3220 		return ("mque");
3221 	case DTYPE_SHM:
3222 		return ("shm");
3223 	case DTYPE_SEM:
3224 		return ("ksem");
3225 	default:
3226 		return ("unkn");
3227 	}
3228 }
3229 
3230 /*
3231  * For the purposes of debugging, identify a process (if any, perhaps one of
3232  * many) that references the passed file in its file descriptor array. Return
3233  * NULL if none.
3234  */
3235 static struct proc *
3236 file_to_first_proc(struct file *fp)
3237 {
3238 	struct filedesc *fdp;
3239 	struct proc *p;
3240 	int n;
3241 
3242 	FOREACH_PROC_IN_SYSTEM(p) {
3243 		if (p->p_state == PRS_NEW)
3244 			continue;
3245 		fdp = p->p_fd;
3246 		if (fdp == NULL)
3247 			continue;
3248 		for (n = 0; n < fdp->fd_nfiles; n++) {
3249 			if (fp == fdp->fd_ofiles[n])
3250 				return (p);
3251 		}
3252 	}
3253 	return (NULL);
3254 }
3255 
3256 static void
3257 db_print_file(struct file *fp, int header)
3258 {
3259 	struct proc *p;
3260 
3261 	if (header)
3262 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3263 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3264 		    "MCount", "Vnode", "FPID", "FCmd");
3265 	p = file_to_first_proc(fp);
3266 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3267 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3268 	    0, fp->f_count, 0, fp->f_vnode,
3269 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3270 }
3271 
3272 DB_SHOW_COMMAND(file, db_show_file)
3273 {
3274 	struct file *fp;
3275 
3276 	if (!have_addr) {
3277 		db_printf("usage: show file <addr>\n");
3278 		return;
3279 	}
3280 	fp = (struct file *)addr;
3281 	db_print_file(fp, 1);
3282 }
3283 
3284 DB_SHOW_COMMAND(files, db_show_files)
3285 {
3286 	struct filedesc *fdp;
3287 	struct file *fp;
3288 	struct proc *p;
3289 	int header;
3290 	int n;
3291 
3292 	header = 1;
3293 	FOREACH_PROC_IN_SYSTEM(p) {
3294 		if (p->p_state == PRS_NEW)
3295 			continue;
3296 		if ((fdp = p->p_fd) == NULL)
3297 			continue;
3298 		for (n = 0; n < fdp->fd_nfiles; ++n) {
3299 			if ((fp = fdp->fd_ofiles[n]) == NULL)
3300 				continue;
3301 			db_print_file(fp, header);
3302 			header = 0;
3303 		}
3304 	}
3305 }
3306 #endif
3307 
3308 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3309     &maxfilesperproc, 0, "Maximum files allowed open per process");
3310 
3311 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3312     &maxfiles, 0, "Maximum number of files");
3313 
3314 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3315     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3316 
3317 /* ARGSUSED*/
3318 static void
3319 filelistinit(void *dummy)
3320 {
3321 
3322 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3323 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3324 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3325 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3326 }
3327 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3328 
3329 /*-------------------------------------------------------------------*/
3330 
3331 static int
3332 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
3333 {
3334 
3335 	return (EBADF);
3336 }
3337 
3338 static int
3339 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
3340 {
3341 
3342 	return (EINVAL);
3343 }
3344 
3345 static int
3346 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
3347 {
3348 
3349 	return (EBADF);
3350 }
3351 
3352 static int
3353 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
3354 {
3355 
3356 	return (0);
3357 }
3358 
3359 static int
3360 badfo_kqfilter(struct file *fp, struct knote *kn)
3361 {
3362 
3363 	return (EBADF);
3364 }
3365 
3366 static int
3367 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
3368 {
3369 
3370 	return (EBADF);
3371 }
3372 
3373 static int
3374 badfo_close(struct file *fp, struct thread *td)
3375 {
3376 
3377 	return (EBADF);
3378 }
3379 
3380 struct fileops badfileops = {
3381 	.fo_read = badfo_readwrite,
3382 	.fo_write = badfo_readwrite,
3383 	.fo_truncate = badfo_truncate,
3384 	.fo_ioctl = badfo_ioctl,
3385 	.fo_poll = badfo_poll,
3386 	.fo_kqfilter = badfo_kqfilter,
3387 	.fo_stat = badfo_stat,
3388 	.fo_close = badfo_close,
3389 };
3390 
3391 
3392 /*-------------------------------------------------------------------*/
3393 
3394 /*
3395  * File Descriptor pseudo-device driver (/dev/fd/).
3396  *
3397  * Opening minor device N dup()s the file (if any) connected to file
3398  * descriptor N belonging to the calling process.  Note that this driver
3399  * consists of only the ``open()'' routine, because all subsequent
3400  * references to this file will be direct to the other driver.
3401  *
3402  * XXX: we could give this one a cloning event handler if necessary.
3403  */
3404 
3405 /* ARGSUSED */
3406 static int
3407 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3408 {
3409 
3410 	/*
3411 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3412 	 * the file descriptor being sought for duplication. The error
3413 	 * return ensures that the vnode for this device will be released
3414 	 * by vn_open. Open will detect this special error and take the
3415 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3416 	 * will simply report the error.
3417 	 */
3418 	td->td_dupfd = dev2unit(dev);
3419 	return (ENODEV);
3420 }
3421 
3422 static struct cdevsw fildesc_cdevsw = {
3423 	.d_version =	D_VERSION,
3424 	.d_open =	fdopen,
3425 	.d_name =	"FD",
3426 };
3427 
3428 static void
3429 fildesc_drvinit(void *unused)
3430 {
3431 	struct cdev *dev;
3432 
3433 	dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
3434 	make_dev_alias(dev, "stdin");
3435 	dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
3436 	make_dev_alias(dev, "stdout");
3437 	dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
3438 	make_dev_alias(dev, "stderr");
3439 }
3440 
3441 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
3442