xref: /freebsd/sys/kern/kern_descrip.c (revision d8462732ce74ccaf077dac625168cfa75f4a3b7e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, int holdleaders);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seqc_write_begin(&fde->fde_seqc);
313 #endif
314 	fde->fde_file = NULL;
315 #ifdef CAPABILITIES
316 	seqc_write_end(&fde->fde_seqc);
317 #endif
318 	fdefree_last(fde);
319 	fdunused(fdp, fd);
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 #ifdef	RACCT
335 	uint64_t lim;
336 #endif
337 
338 	td->td_retval[0] = getmaxfd(td);
339 #ifdef	RACCT
340 	PROC_LOCK(td->td_proc);
341 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 	PROC_UNLOCK(td->td_proc);
343 	if (lim < td->td_retval[0])
344 		td->td_retval[0] = lim;
345 #endif
346 	return (0);
347 }
348 
349 /*
350  * Duplicate a file descriptor to a particular value.
351  *
352  * Note: keep in mind that a potential race condition exists when closing
353  * descriptors from a shared descriptor table (via rfork).
354  */
355 #ifndef _SYS_SYSPROTO_H_
356 struct dup2_args {
357 	u_int	from;
358 	u_int	to;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_dup2(struct thread *td, struct dup2_args *uap)
364 {
365 
366 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
367 }
368 
369 /*
370  * Duplicate a file descriptor.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup_args {
374 	u_int	fd;
375 };
376 #endif
377 /* ARGSUSED */
378 int
379 sys_dup(struct thread *td, struct dup_args *uap)
380 {
381 
382 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
383 }
384 
385 /*
386  * The file control system call.
387  */
388 #ifndef _SYS_SYSPROTO_H_
389 struct fcntl_args {
390 	int	fd;
391 	int	cmd;
392 	long	arg;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 {
399 
400 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
401 }
402 
403 int
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
405 {
406 	struct flock fl;
407 	struct __oflock ofl;
408 	intptr_t arg1;
409 	int error, newcmd;
410 
411 	error = 0;
412 	newcmd = cmd;
413 	switch (cmd) {
414 	case F_OGETLK:
415 	case F_OSETLK:
416 	case F_OSETLKW:
417 		/*
418 		 * Convert old flock structure to new.
419 		 */
420 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 		fl.l_start = ofl.l_start;
422 		fl.l_len = ofl.l_len;
423 		fl.l_pid = ofl.l_pid;
424 		fl.l_type = ofl.l_type;
425 		fl.l_whence = ofl.l_whence;
426 		fl.l_sysid = 0;
427 
428 		switch (cmd) {
429 		case F_OGETLK:
430 			newcmd = F_GETLK;
431 			break;
432 		case F_OSETLK:
433 			newcmd = F_SETLK;
434 			break;
435 		case F_OSETLKW:
436 			newcmd = F_SETLKW;
437 			break;
438 		}
439 		arg1 = (intptr_t)&fl;
440 		break;
441 	case F_GETLK:
442 	case F_SETLK:
443 	case F_SETLKW:
444 	case F_SETLK_REMOTE:
445 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 		arg1 = (intptr_t)&fl;
447 		break;
448 	default:
449 		arg1 = arg;
450 		break;
451 	}
452 	if (error)
453 		return (error);
454 	error = kern_fcntl(td, fd, newcmd, arg1);
455 	if (error)
456 		return (error);
457 	if (cmd == F_OGETLK) {
458 		ofl.l_start = fl.l_start;
459 		ofl.l_len = fl.l_len;
460 		ofl.l_pid = fl.l_pid;
461 		ofl.l_type = fl.l_type;
462 		ofl.l_whence = fl.l_whence;
463 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 	} else if (cmd == F_GETLK) {
465 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
466 	}
467 	return (error);
468 }
469 
470 int
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
472 {
473 	struct filedesc *fdp;
474 	struct flock *flp;
475 	struct file *fp, *fp2;
476 	struct filedescent *fde;
477 	struct proc *p;
478 	struct vnode *vp;
479 	struct mount *mp;
480 	int error, flg, seals, tmp;
481 	uint64_t bsize;
482 	off_t foffset;
483 
484 	error = 0;
485 	flg = F_POSIX;
486 	p = td->td_proc;
487 	fdp = p->p_fd;
488 
489 	AUDIT_ARG_FD(cmd);
490 	AUDIT_ARG_CMD(cmd);
491 	switch (cmd) {
492 	case F_DUPFD:
493 		tmp = arg;
494 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
495 		break;
496 
497 	case F_DUPFD_CLOEXEC:
498 		tmp = arg;
499 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD:
503 		tmp = arg;
504 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
505 		break;
506 
507 	case F_DUP2FD_CLOEXEC:
508 		tmp = arg;
509 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 		break;
511 
512 	case F_GETFD:
513 		error = EBADF;
514 		FILEDESC_SLOCK(fdp);
515 		fde = fdeget_locked(fdp, fd);
516 		if (fde != NULL) {
517 			td->td_retval[0] =
518 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
519 			error = 0;
520 		}
521 		FILEDESC_SUNLOCK(fdp);
522 		break;
523 
524 	case F_SETFD:
525 		error = EBADF;
526 		FILEDESC_XLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
531 			error = 0;
532 		}
533 		FILEDESC_XUNLOCK(fdp);
534 		break;
535 
536 	case F_GETFL:
537 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
538 		if (error != 0)
539 			break;
540 		td->td_retval[0] = OFLAGS(fp->f_flag);
541 		fdrop(fp, td);
542 		break;
543 
544 	case F_SETFL:
545 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
546 		if (error != 0)
547 			break;
548 		do {
549 			tmp = flg = fp->f_flag;
550 			tmp &= ~FCNTLFLAGS;
551 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
552 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
553 		tmp = fp->f_flag & FNONBLOCK;
554 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
555 		if (error != 0) {
556 			fdrop(fp, td);
557 			break;
558 		}
559 		tmp = fp->f_flag & FASYNC;
560 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
561 		if (error == 0) {
562 			fdrop(fp, td);
563 			break;
564 		}
565 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
566 		tmp = 0;
567 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 		fdrop(fp, td);
569 		break;
570 
571 	case F_GETOWN:
572 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
573 		if (error != 0)
574 			break;
575 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
576 		if (error == 0)
577 			td->td_retval[0] = tmp;
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_SETOWN:
582 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
583 		if (error != 0)
584 			break;
585 		tmp = arg;
586 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
587 		fdrop(fp, td);
588 		break;
589 
590 	case F_SETLK_REMOTE:
591 		error = priv_check(td, PRIV_NFS_LOCKD);
592 		if (error != 0)
593 			return (error);
594 		flg = F_REMOTE;
595 		goto do_setlk;
596 
597 	case F_SETLKW:
598 		flg |= F_WAIT;
599 		/* FALLTHROUGH F_SETLK */
600 
601 	case F_SETLK:
602 	do_setlk:
603 		flp = (struct flock *)arg;
604 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
605 			error = EINVAL;
606 			break;
607 		}
608 
609 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
610 		if (error != 0)
611 			break;
612 		if (fp->f_type != DTYPE_VNODE) {
613 			error = EBADF;
614 			fdrop(fp, td);
615 			break;
616 		}
617 
618 		if (flp->l_whence == SEEK_CUR) {
619 			foffset = foffset_get(fp);
620 			if (foffset < 0 ||
621 			    (flp->l_start > 0 &&
622 			     foffset > OFF_MAX - flp->l_start)) {
623 				error = EOVERFLOW;
624 				fdrop(fp, td);
625 				break;
626 			}
627 			flp->l_start += foffset;
628 		}
629 
630 		vp = fp->f_vnode;
631 		switch (flp->l_type) {
632 		case F_RDLCK:
633 			if ((fp->f_flag & FREAD) == 0) {
634 				error = EBADF;
635 				break;
636 			}
637 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
638 				PROC_LOCK(p->p_leader);
639 				p->p_leader->p_flag |= P_ADVLOCK;
640 				PROC_UNLOCK(p->p_leader);
641 			}
642 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
643 			    flp, flg);
644 			break;
645 		case F_WRLCK:
646 			if ((fp->f_flag & FWRITE) == 0) {
647 				error = EBADF;
648 				break;
649 			}
650 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 				PROC_LOCK(p->p_leader);
652 				p->p_leader->p_flag |= P_ADVLOCK;
653 				PROC_UNLOCK(p->p_leader);
654 			}
655 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
656 			    flp, flg);
657 			break;
658 		case F_UNLCK:
659 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
660 			    flp, flg);
661 			break;
662 		case F_UNLCKSYS:
663 			if (flg != F_REMOTE) {
664 				error = EINVAL;
665 				break;
666 			}
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
668 			    F_UNLCKSYS, flp, flg);
669 			break;
670 		default:
671 			error = EINVAL;
672 			break;
673 		}
674 		if (error != 0 || flp->l_type == F_UNLCK ||
675 		    flp->l_type == F_UNLCKSYS) {
676 			fdrop(fp, td);
677 			break;
678 		}
679 
680 		/*
681 		 * Check for a race with close.
682 		 *
683 		 * The vnode is now advisory locked (or unlocked, but this case
684 		 * is not really important) as the caller requested.
685 		 * We had to drop the filedesc lock, so we need to recheck if
686 		 * the descriptor is still valid, because if it was closed
687 		 * in the meantime we need to remove advisory lock from the
688 		 * vnode - close on any descriptor leading to an advisory
689 		 * locked vnode, removes that lock.
690 		 * We will return 0 on purpose in that case, as the result of
691 		 * successful advisory lock might have been externally visible
692 		 * already. This is fine - effectively we pretend to the caller
693 		 * that the closing thread was a bit slower and that the
694 		 * advisory lock succeeded before the close.
695 		 */
696 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
697 		if (error != 0) {
698 			fdrop(fp, td);
699 			break;
700 		}
701 		if (fp != fp2) {
702 			flp->l_whence = SEEK_SET;
703 			flp->l_start = 0;
704 			flp->l_len = 0;
705 			flp->l_type = F_UNLCK;
706 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
707 			    F_UNLCK, flp, F_POSIX);
708 		}
709 		fdrop(fp, td);
710 		fdrop(fp2, td);
711 		break;
712 
713 	case F_GETLK:
714 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
715 		if (error != 0)
716 			break;
717 		if (fp->f_type != DTYPE_VNODE) {
718 			error = EBADF;
719 			fdrop(fp, td);
720 			break;
721 		}
722 		flp = (struct flock *)arg;
723 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
724 		    flp->l_type != F_UNLCK) {
725 			error = EINVAL;
726 			fdrop(fp, td);
727 			break;
728 		}
729 		if (flp->l_whence == SEEK_CUR) {
730 			foffset = foffset_get(fp);
731 			if ((flp->l_start > 0 &&
732 			    foffset > OFF_MAX - flp->l_start) ||
733 			    (flp->l_start < 0 &&
734 			    foffset < OFF_MIN - flp->l_start)) {
735 				error = EOVERFLOW;
736 				fdrop(fp, td);
737 				break;
738 			}
739 			flp->l_start += foffset;
740 		}
741 		vp = fp->f_vnode;
742 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
743 		    F_POSIX);
744 		fdrop(fp, td);
745 		break;
746 
747 	case F_ADD_SEALS:
748 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
749 		if (error != 0)
750 			break;
751 		error = fo_add_seals(fp, arg);
752 		fdrop(fp, td);
753 		break;
754 
755 	case F_GET_SEALS:
756 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
757 		if (error != 0)
758 			break;
759 		if (fo_get_seals(fp, &seals) == 0)
760 			td->td_retval[0] = seals;
761 		else
762 			error = EINVAL;
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_RDAHEAD:
767 		arg = arg ? 128 * 1024: 0;
768 		/* FALLTHROUGH */
769 	case F_READAHEAD:
770 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
771 		if (error != 0)
772 			break;
773 		if (fp->f_type != DTYPE_VNODE) {
774 			fdrop(fp, td);
775 			error = EBADF;
776 			break;
777 		}
778 		vp = fp->f_vnode;
779 		if (vp->v_type != VREG) {
780 			fdrop(fp, td);
781 			error = ENOTTY;
782 			break;
783 		}
784 
785 		/*
786 		 * Exclusive lock synchronizes against f_seqcount reads and
787 		 * writes in sequential_heuristic().
788 		 */
789 		error = vn_lock(vp, LK_EXCLUSIVE);
790 		if (error != 0) {
791 			fdrop(fp, td);
792 			break;
793 		}
794 		if (arg >= 0) {
795 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
796 			arg = MIN(arg, INT_MAX - bsize + 1);
797 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
798 			    (arg + bsize - 1) / bsize);
799 			atomic_set_int(&fp->f_flag, FRDAHEAD);
800 		} else {
801 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
802 		}
803 		VOP_UNLOCK(vp);
804 		fdrop(fp, td);
805 		break;
806 
807 	case F_ISUNIONSTACK:
808 		/*
809 		 * Check if the vnode is part of a union stack (either the
810 		 * "union" flag from mount(2) or unionfs).
811 		 *
812 		 * Prior to introduction of this op libc's readdir would call
813 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
814 		 * data just to check fs name and a mount flag.
815 		 *
816 		 * Fixing the code to handle everything in the kernel instead
817 		 * is a non-trivial endeavor and has low priority, thus this
818 		 * horrible kludge facilitates the current behavior in a much
819 		 * cheaper manner until someone(tm) sorts this out.
820 		 */
821 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
822 		if (error != 0)
823 			break;
824 		if (fp->f_type != DTYPE_VNODE) {
825 			fdrop(fp, td);
826 			error = EBADF;
827 			break;
828 		}
829 		vp = fp->f_vnode;
830 		/*
831 		 * Since we don't prevent dooming the vnode even non-null mp
832 		 * found can become immediately stale. This is tolerable since
833 		 * mount points are type-stable (providing safe memory access)
834 		 * and any vfs op on this vnode going forward will return an
835 		 * error (meaning return value in this case is meaningless).
836 		 */
837 		mp = atomic_load_ptr(&vp->v_mount);
838 		if (__predict_false(mp == NULL)) {
839 			fdrop(fp, td);
840 			error = EBADF;
841 			break;
842 		}
843 		td->td_retval[0] = 0;
844 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
845 		    mp->mnt_flag & MNT_UNION)
846 			td->td_retval[0] = 1;
847 		fdrop(fp, td);
848 		break;
849 
850 	default:
851 		error = EINVAL;
852 		break;
853 	}
854 	return (error);
855 }
856 
857 static int
858 getmaxfd(struct thread *td)
859 {
860 
861 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
862 }
863 
864 /*
865  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
866  */
867 int
868 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
869 {
870 	struct filedesc *fdp;
871 	struct filedescent *oldfde, *newfde;
872 	struct proc *p;
873 	struct file *delfp, *oldfp;
874 	u_long *oioctls, *nioctls;
875 	int error, maxfd;
876 
877 	p = td->td_proc;
878 	fdp = p->p_fd;
879 	oioctls = NULL;
880 
881 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
882 	MPASS(mode < FDDUP_LASTMODE);
883 
884 	AUDIT_ARG_FD(old);
885 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
886 
887 	/*
888 	 * Verify we have a valid descriptor to dup from and possibly to
889 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
890 	 * return EINVAL when the new descriptor is out of bounds.
891 	 */
892 	if (old < 0)
893 		return (EBADF);
894 	if (new < 0)
895 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
896 	maxfd = getmaxfd(td);
897 	if (new >= maxfd)
898 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
899 
900 	error = EBADF;
901 	FILEDESC_XLOCK(fdp);
902 	if (fget_locked(fdp, old) == NULL)
903 		goto unlock;
904 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
905 		td->td_retval[0] = new;
906 		if (flags & FDDUP_FLAG_CLOEXEC)
907 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
908 		error = 0;
909 		goto unlock;
910 	}
911 
912 	oldfde = &fdp->fd_ofiles[old];
913 	oldfp = oldfde->fde_file;
914 	if (!fhold(oldfp))
915 		goto unlock;
916 
917 	/*
918 	 * If the caller specified a file descriptor, make sure the file
919 	 * table is large enough to hold it, and grab it.  Otherwise, just
920 	 * allocate a new descriptor the usual way.
921 	 */
922 	switch (mode) {
923 	case FDDUP_NORMAL:
924 	case FDDUP_FCNTL:
925 		if ((error = fdalloc(td, new, &new)) != 0) {
926 			fdrop(oldfp, td);
927 			goto unlock;
928 		}
929 		break;
930 	case FDDUP_MUSTREPLACE:
931 		/* Target file descriptor must exist. */
932 		if (fget_locked(fdp, new) == NULL) {
933 			fdrop(oldfp, td);
934 			goto unlock;
935 		}
936 		break;
937 	case FDDUP_FIXED:
938 		if (new >= fdp->fd_nfiles) {
939 			/*
940 			 * The resource limits are here instead of e.g.
941 			 * fdalloc(), because the file descriptor table may be
942 			 * shared between processes, so we can't really use
943 			 * racct_add()/racct_sub().  Instead of counting the
944 			 * number of actually allocated descriptors, just put
945 			 * the limit on the size of the file descriptor table.
946 			 */
947 #ifdef RACCT
948 			if (RACCT_ENABLED()) {
949 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
950 				if (error != 0) {
951 					error = EMFILE;
952 					fdrop(oldfp, td);
953 					goto unlock;
954 				}
955 			}
956 #endif
957 			fdgrowtable_exp(fdp, new + 1);
958 		}
959 		if (!fdisused(fdp, new))
960 			fdused(fdp, new);
961 		break;
962 	default:
963 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
964 	}
965 
966 	KASSERT(old != new, ("new fd is same as old"));
967 
968 	/* Refetch oldfde because the table may have grown and old one freed. */
969 	oldfde = &fdp->fd_ofiles[old];
970 	KASSERT(oldfp == oldfde->fde_file,
971 	    ("fdt_ofiles shift from growth observed at fd %d",
972 	    old));
973 
974 	newfde = &fdp->fd_ofiles[new];
975 	delfp = newfde->fde_file;
976 
977 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
978 
979 	/*
980 	 * Duplicate the source descriptor.
981 	 */
982 #ifdef CAPABILITIES
983 	seqc_write_begin(&newfde->fde_seqc);
984 #endif
985 	oioctls = filecaps_free_prep(&newfde->fde_caps);
986 	memcpy(newfde, oldfde, fde_change_size);
987 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
988 	    nioctls);
989 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
990 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
991 	else
992 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
993 #ifdef CAPABILITIES
994 	seqc_write_end(&newfde->fde_seqc);
995 #endif
996 	td->td_retval[0] = new;
997 
998 	error = 0;
999 
1000 	if (delfp != NULL) {
1001 		(void) closefp(fdp, new, delfp, td, 1);
1002 		FILEDESC_UNLOCK_ASSERT(fdp);
1003 	} else {
1004 unlock:
1005 		FILEDESC_XUNLOCK(fdp);
1006 	}
1007 
1008 	filecaps_free_finish(oioctls);
1009 	return (error);
1010 }
1011 
1012 static void
1013 sigiofree(struct sigio *sigio)
1014 {
1015 	crfree(sigio->sio_ucred);
1016 	free(sigio, M_SIGIO);
1017 }
1018 
1019 static struct sigio *
1020 funsetown_locked(struct sigio *sigio)
1021 {
1022 	struct proc *p;
1023 	struct pgrp *pg;
1024 
1025 	SIGIO_ASSERT_LOCKED();
1026 
1027 	if (sigio == NULL)
1028 		return (NULL);
1029 	*(sigio->sio_myref) = NULL;
1030 	if (sigio->sio_pgid < 0) {
1031 		pg = sigio->sio_pgrp;
1032 		PGRP_LOCK(pg);
1033 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1034 		    sigio, sio_pgsigio);
1035 		PGRP_UNLOCK(pg);
1036 	} else {
1037 		p = sigio->sio_proc;
1038 		PROC_LOCK(p);
1039 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1040 		    sigio, sio_pgsigio);
1041 		PROC_UNLOCK(p);
1042 	}
1043 	return (sigio);
1044 }
1045 
1046 /*
1047  * If sigio is on the list associated with a process or process group,
1048  * disable signalling from the device, remove sigio from the list and
1049  * free sigio.
1050  */
1051 void
1052 funsetown(struct sigio **sigiop)
1053 {
1054 	struct sigio *sigio;
1055 
1056 	/* Racy check, consumers must provide synchronization. */
1057 	if (*sigiop == NULL)
1058 		return;
1059 
1060 	SIGIO_LOCK();
1061 	sigio = funsetown_locked(*sigiop);
1062 	SIGIO_UNLOCK();
1063 	if (sigio != NULL)
1064 		sigiofree(sigio);
1065 }
1066 
1067 /*
1068  * Free a list of sigio structures.  The caller must ensure that new sigio
1069  * structures cannot be added after this point.  For process groups this is
1070  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1071  * as an interlock.
1072  */
1073 void
1074 funsetownlst(struct sigiolst *sigiolst)
1075 {
1076 	struct proc *p;
1077 	struct pgrp *pg;
1078 	struct sigio *sigio, *tmp;
1079 
1080 	/* Racy check. */
1081 	sigio = SLIST_FIRST(sigiolst);
1082 	if (sigio == NULL)
1083 		return;
1084 
1085 	p = NULL;
1086 	pg = NULL;
1087 
1088 	SIGIO_LOCK();
1089 	sigio = SLIST_FIRST(sigiolst);
1090 	if (sigio == NULL) {
1091 		SIGIO_UNLOCK();
1092 		return;
1093 	}
1094 
1095 	/*
1096 	 * Every entry of the list should belong to a single proc or pgrp.
1097 	 */
1098 	if (sigio->sio_pgid < 0) {
1099 		pg = sigio->sio_pgrp;
1100 		sx_assert(&proctree_lock, SX_XLOCKED);
1101 		PGRP_LOCK(pg);
1102 	} else /* if (sigio->sio_pgid > 0) */ {
1103 		p = sigio->sio_proc;
1104 		PROC_LOCK(p);
1105 		KASSERT((p->p_flag & P_WEXIT) != 0,
1106 		    ("%s: process %p is not exiting", __func__, p));
1107 	}
1108 
1109 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1110 		*sigio->sio_myref = NULL;
1111 		if (pg != NULL) {
1112 			KASSERT(sigio->sio_pgid < 0,
1113 			    ("Proc sigio in pgrp sigio list"));
1114 			KASSERT(sigio->sio_pgrp == pg,
1115 			    ("Bogus pgrp in sigio list"));
1116 		} else /* if (p != NULL) */ {
1117 			KASSERT(sigio->sio_pgid > 0,
1118 			    ("Pgrp sigio in proc sigio list"));
1119 			KASSERT(sigio->sio_proc == p,
1120 			    ("Bogus proc in sigio list"));
1121 		}
1122 	}
1123 
1124 	if (pg != NULL)
1125 		PGRP_UNLOCK(pg);
1126 	else
1127 		PROC_UNLOCK(p);
1128 	SIGIO_UNLOCK();
1129 
1130 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1131 		sigiofree(sigio);
1132 }
1133 
1134 /*
1135  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1136  *
1137  * After permission checking, add a sigio structure to the sigio list for
1138  * the process or process group.
1139  */
1140 int
1141 fsetown(pid_t pgid, struct sigio **sigiop)
1142 {
1143 	struct proc *proc;
1144 	struct pgrp *pgrp;
1145 	struct sigio *osigio, *sigio;
1146 	int ret;
1147 
1148 	if (pgid == 0) {
1149 		funsetown(sigiop);
1150 		return (0);
1151 	}
1152 
1153 	ret = 0;
1154 
1155 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1156 	sigio->sio_pgid = pgid;
1157 	sigio->sio_ucred = crhold(curthread->td_ucred);
1158 	sigio->sio_myref = sigiop;
1159 
1160 	sx_slock(&proctree_lock);
1161 	SIGIO_LOCK();
1162 	osigio = funsetown_locked(*sigiop);
1163 	if (pgid > 0) {
1164 		proc = pfind(pgid);
1165 		if (proc == NULL) {
1166 			ret = ESRCH;
1167 			goto fail;
1168 		}
1169 
1170 		/*
1171 		 * Policy - Don't allow a process to FSETOWN a process
1172 		 * in another session.
1173 		 *
1174 		 * Remove this test to allow maximum flexibility or
1175 		 * restrict FSETOWN to the current process or process
1176 		 * group for maximum safety.
1177 		 */
1178 		if (proc->p_session != curthread->td_proc->p_session) {
1179 			PROC_UNLOCK(proc);
1180 			ret = EPERM;
1181 			goto fail;
1182 		}
1183 
1184 		sigio->sio_proc = proc;
1185 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1186 		PROC_UNLOCK(proc);
1187 	} else /* if (pgid < 0) */ {
1188 		pgrp = pgfind(-pgid);
1189 		if (pgrp == NULL) {
1190 			ret = ESRCH;
1191 			goto fail;
1192 		}
1193 
1194 		/*
1195 		 * Policy - Don't allow a process to FSETOWN a process
1196 		 * in another session.
1197 		 *
1198 		 * Remove this test to allow maximum flexibility or
1199 		 * restrict FSETOWN to the current process or process
1200 		 * group for maximum safety.
1201 		 */
1202 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1203 			PGRP_UNLOCK(pgrp);
1204 			ret = EPERM;
1205 			goto fail;
1206 		}
1207 
1208 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1209 		sigio->sio_pgrp = pgrp;
1210 		PGRP_UNLOCK(pgrp);
1211 	}
1212 	sx_sunlock(&proctree_lock);
1213 	*sigiop = sigio;
1214 	SIGIO_UNLOCK();
1215 	if (osigio != NULL)
1216 		sigiofree(osigio);
1217 	return (0);
1218 
1219 fail:
1220 	SIGIO_UNLOCK();
1221 	sx_sunlock(&proctree_lock);
1222 	sigiofree(sigio);
1223 	if (osigio != NULL)
1224 		sigiofree(osigio);
1225 	return (ret);
1226 }
1227 
1228 /*
1229  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1230  */
1231 pid_t
1232 fgetown(struct sigio **sigiop)
1233 {
1234 	pid_t pgid;
1235 
1236 	SIGIO_LOCK();
1237 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1238 	SIGIO_UNLOCK();
1239 	return (pgid);
1240 }
1241 
1242 /*
1243  * Function drops the filedesc lock on return.
1244  */
1245 static int
1246 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1247     int holdleaders)
1248 {
1249 	int error;
1250 
1251 	FILEDESC_XLOCK_ASSERT(fdp);
1252 
1253 	if (holdleaders) {
1254 		if (td->td_proc->p_fdtol != NULL) {
1255 			/*
1256 			 * Ask fdfree() to sleep to ensure that all relevant
1257 			 * process leaders can be traversed in closef().
1258 			 */
1259 			fdp->fd_holdleaderscount++;
1260 		} else {
1261 			holdleaders = 0;
1262 		}
1263 	}
1264 
1265 	/*
1266 	 * We now hold the fp reference that used to be owned by the
1267 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1268 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1269 	 * added, and deleteing a knote for the new fd.
1270 	 */
1271 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1272 		knote_fdclose(td, fd);
1273 
1274 	/*
1275 	 * We need to notify mqueue if the object is of type mqueue.
1276 	 */
1277 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1278 		mq_fdclose(td, fd, fp);
1279 	FILEDESC_XUNLOCK(fdp);
1280 
1281 	error = closef(fp, td);
1282 
1283 	/*
1284 	 * All paths leading up to closefp() will have already removed or
1285 	 * replaced the fd in the filedesc table, so a restart would not
1286 	 * operate on the same file.
1287 	 */
1288 	if (error == ERESTART)
1289 		error = EINTR;
1290 
1291 	if (holdleaders) {
1292 		FILEDESC_XLOCK(fdp);
1293 		fdp->fd_holdleaderscount--;
1294 		if (fdp->fd_holdleaderscount == 0 &&
1295 		    fdp->fd_holdleaderswakeup != 0) {
1296 			fdp->fd_holdleaderswakeup = 0;
1297 			wakeup(&fdp->fd_holdleaderscount);
1298 		}
1299 		FILEDESC_XUNLOCK(fdp);
1300 	}
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Close a file descriptor.
1306  */
1307 #ifndef _SYS_SYSPROTO_H_
1308 struct close_args {
1309 	int     fd;
1310 };
1311 #endif
1312 /* ARGSUSED */
1313 int
1314 sys_close(struct thread *td, struct close_args *uap)
1315 {
1316 
1317 	return (kern_close(td, uap->fd));
1318 }
1319 
1320 int
1321 kern_close(struct thread *td, int fd)
1322 {
1323 	struct filedesc *fdp;
1324 	struct file *fp;
1325 
1326 	fdp = td->td_proc->p_fd;
1327 
1328 	AUDIT_SYSCLOSE(td, fd);
1329 
1330 	FILEDESC_XLOCK(fdp);
1331 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1332 		FILEDESC_XUNLOCK(fdp);
1333 		return (EBADF);
1334 	}
1335 	fdfree(fdp, fd);
1336 
1337 	/* closefp() drops the FILEDESC lock for us. */
1338 	return (closefp(fdp, fd, fp, td, 1));
1339 }
1340 
1341 int
1342 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1343 {
1344 	struct filedesc *fdp;
1345 	int fd, ret, lastfile;
1346 
1347 	ret = 0;
1348 	fdp = td->td_proc->p_fd;
1349 	FILEDESC_SLOCK(fdp);
1350 
1351 	/*
1352 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1353 	 * open should not be a usage error.  From a close_range() perspective,
1354 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1355 	 * be a usage error as all fd above 3 are in-fact already closed.
1356 	 */
1357 	if (highfd < lowfd) {
1358 		ret = EINVAL;
1359 		goto out;
1360 	}
1361 
1362 	/*
1363 	 * If lastfile == -1, we're dealing with either a fresh file
1364 	 * table or one in which every fd has been closed.  Just return
1365 	 * successful; there's nothing left to do.
1366 	 */
1367 	lastfile = fdlastfile(fdp);
1368 	if (lastfile == -1)
1369 		goto out;
1370 	/* Clamped to [lowfd, lastfile] */
1371 	highfd = MIN(highfd, lastfile);
1372 	for (fd = lowfd; fd <= highfd; fd++) {
1373 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1374 			FILEDESC_SUNLOCK(fdp);
1375 			(void)kern_close(td, fd);
1376 			FILEDESC_SLOCK(fdp);
1377 		}
1378 	}
1379 out:
1380 	FILEDESC_SUNLOCK(fdp);
1381 	return (ret);
1382 }
1383 
1384 #ifndef _SYS_SYSPROTO_H_
1385 struct close_range_args {
1386 	u_int	lowfd;
1387 	u_int	highfd;
1388 	int	flags;
1389 };
1390 #endif
1391 int
1392 sys_close_range(struct thread *td, struct close_range_args *uap)
1393 {
1394 
1395 	/* No flags currently defined */
1396 	if (uap->flags != 0)
1397 		return (EINVAL);
1398 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1399 }
1400 
1401 #ifdef COMPAT_FREEBSD12
1402 /*
1403  * Close open file descriptors.
1404  */
1405 #ifndef _SYS_SYSPROTO_H_
1406 struct freebsd12_closefrom_args {
1407 	int	lowfd;
1408 };
1409 #endif
1410 /* ARGSUSED */
1411 int
1412 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1413 {
1414 	u_int lowfd;
1415 
1416 	AUDIT_ARG_FD(uap->lowfd);
1417 
1418 	/*
1419 	 * Treat negative starting file descriptor values identical to
1420 	 * closefrom(0) which closes all files.
1421 	 */
1422 	lowfd = MAX(0, uap->lowfd);
1423 	return (kern_close_range(td, lowfd, ~0U));
1424 }
1425 #endif	/* COMPAT_FREEBSD12 */
1426 
1427 #if defined(COMPAT_43)
1428 /*
1429  * Return status information about a file descriptor.
1430  */
1431 #ifndef _SYS_SYSPROTO_H_
1432 struct ofstat_args {
1433 	int	fd;
1434 	struct	ostat *sb;
1435 };
1436 #endif
1437 /* ARGSUSED */
1438 int
1439 ofstat(struct thread *td, struct ofstat_args *uap)
1440 {
1441 	struct ostat oub;
1442 	struct stat ub;
1443 	int error;
1444 
1445 	error = kern_fstat(td, uap->fd, &ub);
1446 	if (error == 0) {
1447 		cvtstat(&ub, &oub);
1448 		error = copyout(&oub, uap->sb, sizeof(oub));
1449 	}
1450 	return (error);
1451 }
1452 #endif /* COMPAT_43 */
1453 
1454 #if defined(COMPAT_FREEBSD11)
1455 int
1456 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1457 {
1458 	struct stat sb;
1459 	struct freebsd11_stat osb;
1460 	int error;
1461 
1462 	error = kern_fstat(td, uap->fd, &sb);
1463 	if (error != 0)
1464 		return (error);
1465 	error = freebsd11_cvtstat(&sb, &osb);
1466 	if (error == 0)
1467 		error = copyout(&osb, uap->sb, sizeof(osb));
1468 	return (error);
1469 }
1470 #endif	/* COMPAT_FREEBSD11 */
1471 
1472 /*
1473  * Return status information about a file descriptor.
1474  */
1475 #ifndef _SYS_SYSPROTO_H_
1476 struct fstat_args {
1477 	int	fd;
1478 	struct	stat *sb;
1479 };
1480 #endif
1481 /* ARGSUSED */
1482 int
1483 sys_fstat(struct thread *td, struct fstat_args *uap)
1484 {
1485 	struct stat ub;
1486 	int error;
1487 
1488 	error = kern_fstat(td, uap->fd, &ub);
1489 	if (error == 0)
1490 		error = copyout(&ub, uap->sb, sizeof(ub));
1491 	return (error);
1492 }
1493 
1494 int
1495 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1496 {
1497 	struct file *fp;
1498 	int error;
1499 
1500 	AUDIT_ARG_FD(fd);
1501 
1502 	error = fget(td, fd, &cap_fstat_rights, &fp);
1503 	if (__predict_false(error != 0))
1504 		return (error);
1505 
1506 	AUDIT_ARG_FILE(td->td_proc, fp);
1507 
1508 	error = fo_stat(fp, sbp, td->td_ucred, td);
1509 	fdrop(fp, td);
1510 #ifdef __STAT_TIME_T_EXT
1511 	sbp->st_atim_ext = 0;
1512 	sbp->st_mtim_ext = 0;
1513 	sbp->st_ctim_ext = 0;
1514 	sbp->st_btim_ext = 0;
1515 #endif
1516 #ifdef KTRACE
1517 	if (KTRPOINT(td, KTR_STRUCT))
1518 		ktrstat_error(sbp, error);
1519 #endif
1520 	return (error);
1521 }
1522 
1523 #if defined(COMPAT_FREEBSD11)
1524 /*
1525  * Return status information about a file descriptor.
1526  */
1527 #ifndef _SYS_SYSPROTO_H_
1528 struct freebsd11_nfstat_args {
1529 	int	fd;
1530 	struct	nstat *sb;
1531 };
1532 #endif
1533 /* ARGSUSED */
1534 int
1535 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1536 {
1537 	struct nstat nub;
1538 	struct stat ub;
1539 	int error;
1540 
1541 	error = kern_fstat(td, uap->fd, &ub);
1542 	if (error == 0) {
1543 		freebsd11_cvtnstat(&ub, &nub);
1544 		error = copyout(&nub, uap->sb, sizeof(nub));
1545 	}
1546 	return (error);
1547 }
1548 #endif /* COMPAT_FREEBSD11 */
1549 
1550 /*
1551  * Return pathconf information about a file descriptor.
1552  */
1553 #ifndef _SYS_SYSPROTO_H_
1554 struct fpathconf_args {
1555 	int	fd;
1556 	int	name;
1557 };
1558 #endif
1559 /* ARGSUSED */
1560 int
1561 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1562 {
1563 	long value;
1564 	int error;
1565 
1566 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1567 	if (error == 0)
1568 		td->td_retval[0] = value;
1569 	return (error);
1570 }
1571 
1572 int
1573 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1574 {
1575 	struct file *fp;
1576 	struct vnode *vp;
1577 	int error;
1578 
1579 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1580 	if (error != 0)
1581 		return (error);
1582 
1583 	if (name == _PC_ASYNC_IO) {
1584 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1585 		goto out;
1586 	}
1587 	vp = fp->f_vnode;
1588 	if (vp != NULL) {
1589 		vn_lock(vp, LK_SHARED | LK_RETRY);
1590 		error = VOP_PATHCONF(vp, name, valuep);
1591 		VOP_UNLOCK(vp);
1592 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1593 		if (name != _PC_PIPE_BUF) {
1594 			error = EINVAL;
1595 		} else {
1596 			*valuep = PIPE_BUF;
1597 			error = 0;
1598 		}
1599 	} else {
1600 		error = EOPNOTSUPP;
1601 	}
1602 out:
1603 	fdrop(fp, td);
1604 	return (error);
1605 }
1606 
1607 /*
1608  * Copy filecaps structure allocating memory for ioctls array if needed.
1609  *
1610  * The last parameter indicates whether the fdtable is locked. If it is not and
1611  * ioctls are encountered, copying fails and the caller must lock the table.
1612  *
1613  * Note that if the table was not locked, the caller has to check the relevant
1614  * sequence counter to determine whether the operation was successful.
1615  */
1616 bool
1617 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1618 {
1619 	size_t size;
1620 
1621 	if (src->fc_ioctls != NULL && !locked)
1622 		return (false);
1623 	memcpy(dst, src, sizeof(*src));
1624 	if (src->fc_ioctls == NULL)
1625 		return (true);
1626 
1627 	KASSERT(src->fc_nioctls > 0,
1628 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1629 
1630 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1631 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1632 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1633 	return (true);
1634 }
1635 
1636 static u_long *
1637 filecaps_copy_prep(const struct filecaps *src)
1638 {
1639 	u_long *ioctls;
1640 	size_t size;
1641 
1642 	if (__predict_true(src->fc_ioctls == NULL))
1643 		return (NULL);
1644 
1645 	KASSERT(src->fc_nioctls > 0,
1646 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1647 
1648 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1649 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1650 	return (ioctls);
1651 }
1652 
1653 static void
1654 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1655     u_long *ioctls)
1656 {
1657 	size_t size;
1658 
1659 	*dst = *src;
1660 	if (__predict_true(src->fc_ioctls == NULL)) {
1661 		MPASS(ioctls == NULL);
1662 		return;
1663 	}
1664 
1665 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1666 	dst->fc_ioctls = ioctls;
1667 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1668 }
1669 
1670 /*
1671  * Move filecaps structure to the new place and clear the old place.
1672  */
1673 void
1674 filecaps_move(struct filecaps *src, struct filecaps *dst)
1675 {
1676 
1677 	*dst = *src;
1678 	bzero(src, sizeof(*src));
1679 }
1680 
1681 /*
1682  * Fill the given filecaps structure with full rights.
1683  */
1684 static void
1685 filecaps_fill(struct filecaps *fcaps)
1686 {
1687 
1688 	CAP_ALL(&fcaps->fc_rights);
1689 	fcaps->fc_ioctls = NULL;
1690 	fcaps->fc_nioctls = -1;
1691 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1692 }
1693 
1694 /*
1695  * Free memory allocated within filecaps structure.
1696  */
1697 void
1698 filecaps_free(struct filecaps *fcaps)
1699 {
1700 
1701 	free(fcaps->fc_ioctls, M_FILECAPS);
1702 	bzero(fcaps, sizeof(*fcaps));
1703 }
1704 
1705 static u_long *
1706 filecaps_free_prep(struct filecaps *fcaps)
1707 {
1708 	u_long *ioctls;
1709 
1710 	ioctls = fcaps->fc_ioctls;
1711 	bzero(fcaps, sizeof(*fcaps));
1712 	return (ioctls);
1713 }
1714 
1715 static void
1716 filecaps_free_finish(u_long *ioctls)
1717 {
1718 
1719 	free(ioctls, M_FILECAPS);
1720 }
1721 
1722 /*
1723  * Validate the given filecaps structure.
1724  */
1725 static void
1726 filecaps_validate(const struct filecaps *fcaps, const char *func)
1727 {
1728 
1729 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1730 	    ("%s: invalid rights", func));
1731 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1732 	    ("%s: invalid fcntls", func));
1733 	KASSERT(fcaps->fc_fcntls == 0 ||
1734 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1735 	    ("%s: fcntls without CAP_FCNTL", func));
1736 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1737 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1738 	    ("%s: invalid ioctls", func));
1739 	KASSERT(fcaps->fc_nioctls == 0 ||
1740 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1741 	    ("%s: ioctls without CAP_IOCTL", func));
1742 }
1743 
1744 static void
1745 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1746 {
1747 	int nfd1;
1748 
1749 	FILEDESC_XLOCK_ASSERT(fdp);
1750 
1751 	nfd1 = fdp->fd_nfiles * 2;
1752 	if (nfd1 < nfd)
1753 		nfd1 = nfd;
1754 	fdgrowtable(fdp, nfd1);
1755 }
1756 
1757 /*
1758  * Grow the file table to accommodate (at least) nfd descriptors.
1759  */
1760 static void
1761 fdgrowtable(struct filedesc *fdp, int nfd)
1762 {
1763 	struct filedesc0 *fdp0;
1764 	struct freetable *ft;
1765 	struct fdescenttbl *ntable;
1766 	struct fdescenttbl *otable;
1767 	int nnfiles, onfiles;
1768 	NDSLOTTYPE *nmap, *omap;
1769 
1770 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1771 
1772 	/* save old values */
1773 	onfiles = fdp->fd_nfiles;
1774 	otable = fdp->fd_files;
1775 	omap = fdp->fd_map;
1776 
1777 	/* compute the size of the new table */
1778 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1779 	if (nnfiles <= onfiles)
1780 		/* the table is already large enough */
1781 		return;
1782 
1783 	/*
1784 	 * Allocate a new table.  We need enough space for the number of
1785 	 * entries, file entries themselves and the struct freetable we will use
1786 	 * when we decommission the table and place it on the freelist.
1787 	 * We place the struct freetable in the middle so we don't have
1788 	 * to worry about padding.
1789 	 */
1790 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1791 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1792 	    sizeof(struct freetable),
1793 	    M_FILEDESC, M_ZERO | M_WAITOK);
1794 	/* copy the old data */
1795 	ntable->fdt_nfiles = nnfiles;
1796 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1797 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1798 
1799 	/*
1800 	 * Allocate a new map only if the old is not large enough.  It will
1801 	 * grow at a slower rate than the table as it can map more
1802 	 * entries than the table can hold.
1803 	 */
1804 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1805 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1806 		    M_ZERO | M_WAITOK);
1807 		/* copy over the old data and update the pointer */
1808 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1809 		fdp->fd_map = nmap;
1810 	}
1811 
1812 	/*
1813 	 * Make sure that ntable is correctly initialized before we replace
1814 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1815 	 * data.
1816 	 */
1817 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1818 
1819 	/*
1820 	 * Free the old file table when not shared by other threads or processes.
1821 	 * The old file table is considered to be shared when either are true:
1822 	 * - The process has more than one thread.
1823 	 * - The file descriptor table has been shared via fdshare().
1824 	 *
1825 	 * When shared, the old file table will be placed on a freelist
1826 	 * which will be processed when the struct filedesc is released.
1827 	 *
1828 	 * Note that if onfiles == NDFILE, we're dealing with the original
1829 	 * static allocation contained within (struct filedesc0 *)fdp,
1830 	 * which must not be freed.
1831 	 */
1832 	if (onfiles > NDFILE) {
1833 		if (curproc->p_numthreads == 1 &&
1834 		    refcount_load(&fdp->fd_refcnt) == 1)
1835 			free(otable, M_FILEDESC);
1836 		else {
1837 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1838 			fdp0 = (struct filedesc0 *)fdp;
1839 			ft->ft_table = otable;
1840 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1841 		}
1842 	}
1843 	/*
1844 	 * The map does not have the same possibility of threads still
1845 	 * holding references to it.  So always free it as long as it
1846 	 * does not reference the original static allocation.
1847 	 */
1848 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1849 		free(omap, M_FILEDESC);
1850 }
1851 
1852 /*
1853  * Allocate a file descriptor for the process.
1854  */
1855 int
1856 fdalloc(struct thread *td, int minfd, int *result)
1857 {
1858 	struct proc *p = td->td_proc;
1859 	struct filedesc *fdp = p->p_fd;
1860 	int fd, maxfd, allocfd;
1861 #ifdef RACCT
1862 	int error;
1863 #endif
1864 
1865 	FILEDESC_XLOCK_ASSERT(fdp);
1866 
1867 	if (fdp->fd_freefile > minfd)
1868 		minfd = fdp->fd_freefile;
1869 
1870 	maxfd = getmaxfd(td);
1871 
1872 	/*
1873 	 * Search the bitmap for a free descriptor starting at minfd.
1874 	 * If none is found, grow the file table.
1875 	 */
1876 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1877 	if (__predict_false(fd >= maxfd))
1878 		return (EMFILE);
1879 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1880 		allocfd = min(fd * 2, maxfd);
1881 #ifdef RACCT
1882 		if (RACCT_ENABLED()) {
1883 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1884 			if (error != 0)
1885 				return (EMFILE);
1886 		}
1887 #endif
1888 		/*
1889 		 * fd is already equal to first free descriptor >= minfd, so
1890 		 * we only need to grow the table and we are done.
1891 		 */
1892 		fdgrowtable_exp(fdp, allocfd);
1893 	}
1894 
1895 	/*
1896 	 * Perform some sanity checks, then mark the file descriptor as
1897 	 * used and return it to the caller.
1898 	 */
1899 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1900 	    ("invalid descriptor %d", fd));
1901 	KASSERT(!fdisused(fdp, fd),
1902 	    ("fd_first_free() returned non-free descriptor"));
1903 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1904 	    ("file descriptor isn't free"));
1905 	fdused(fdp, fd);
1906 	*result = fd;
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Allocate n file descriptors for the process.
1912  */
1913 int
1914 fdallocn(struct thread *td, int minfd, int *fds, int n)
1915 {
1916 	struct proc *p = td->td_proc;
1917 	struct filedesc *fdp = p->p_fd;
1918 	int i;
1919 
1920 	FILEDESC_XLOCK_ASSERT(fdp);
1921 
1922 	for (i = 0; i < n; i++)
1923 		if (fdalloc(td, 0, &fds[i]) != 0)
1924 			break;
1925 
1926 	if (i < n) {
1927 		for (i--; i >= 0; i--)
1928 			fdunused(fdp, fds[i]);
1929 		return (EMFILE);
1930 	}
1931 
1932 	return (0);
1933 }
1934 
1935 /*
1936  * Create a new open file structure and allocate a file descriptor for the
1937  * process that refers to it.  We add one reference to the file for the
1938  * descriptor table and one reference for resultfp. This is to prevent us
1939  * being preempted and the entry in the descriptor table closed after we
1940  * release the FILEDESC lock.
1941  */
1942 int
1943 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1944     struct filecaps *fcaps)
1945 {
1946 	struct file *fp;
1947 	int error, fd;
1948 
1949 	error = falloc_noinstall(td, &fp);
1950 	if (error)
1951 		return (error);		/* no reference held on error */
1952 
1953 	error = finstall(td, fp, &fd, flags, fcaps);
1954 	if (error) {
1955 		fdrop(fp, td);		/* one reference (fp only) */
1956 		return (error);
1957 	}
1958 
1959 	if (resultfp != NULL)
1960 		*resultfp = fp;		/* copy out result */
1961 	else
1962 		fdrop(fp, td);		/* release local reference */
1963 
1964 	if (resultfd != NULL)
1965 		*resultfd = fd;
1966 
1967 	return (0);
1968 }
1969 
1970 /*
1971  * Create a new open file structure without allocating a file descriptor.
1972  */
1973 int
1974 falloc_noinstall(struct thread *td, struct file **resultfp)
1975 {
1976 	struct file *fp;
1977 	int maxuserfiles = maxfiles - (maxfiles / 20);
1978 	int openfiles_new;
1979 	static struct timeval lastfail;
1980 	static int curfail;
1981 
1982 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1983 
1984 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1985 	if ((openfiles_new >= maxuserfiles &&
1986 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1987 	    openfiles_new >= maxfiles) {
1988 		atomic_subtract_int(&openfiles, 1);
1989 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1990 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1991 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1992 		}
1993 		return (ENFILE);
1994 	}
1995 	fp = uma_zalloc(file_zone, M_WAITOK);
1996 	bzero(fp, sizeof(*fp));
1997 	refcount_init(&fp->f_count, 1);
1998 	fp->f_cred = crhold(td->td_ucred);
1999 	fp->f_ops = &badfileops;
2000 	*resultfp = fp;
2001 	return (0);
2002 }
2003 
2004 /*
2005  * Install a file in a file descriptor table.
2006  */
2007 void
2008 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2009     struct filecaps *fcaps)
2010 {
2011 	struct filedescent *fde;
2012 
2013 	MPASS(fp != NULL);
2014 	if (fcaps != NULL)
2015 		filecaps_validate(fcaps, __func__);
2016 	FILEDESC_XLOCK_ASSERT(fdp);
2017 
2018 	fde = &fdp->fd_ofiles[fd];
2019 #ifdef CAPABILITIES
2020 	seqc_write_begin(&fde->fde_seqc);
2021 #endif
2022 	fde->fde_file = fp;
2023 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2024 	if (fcaps != NULL)
2025 		filecaps_move(fcaps, &fde->fde_caps);
2026 	else
2027 		filecaps_fill(&fde->fde_caps);
2028 #ifdef CAPABILITIES
2029 	seqc_write_end(&fde->fde_seqc);
2030 #endif
2031 }
2032 
2033 int
2034 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2035     struct filecaps *fcaps)
2036 {
2037 	struct filedesc *fdp = td->td_proc->p_fd;
2038 	int error;
2039 
2040 	MPASS(fd != NULL);
2041 
2042 	if (!fhold(fp))
2043 		return (EBADF);
2044 	FILEDESC_XLOCK(fdp);
2045 	error = fdalloc(td, 0, fd);
2046 	if (__predict_false(error != 0)) {
2047 		FILEDESC_XUNLOCK(fdp);
2048 		fdrop(fp, td);
2049 		return (error);
2050 	}
2051 	_finstall(fdp, fp, *fd, flags, fcaps);
2052 	FILEDESC_XUNLOCK(fdp);
2053 	return (0);
2054 }
2055 
2056 /*
2057  * Build a new filedesc structure from another.
2058  *
2059  * If fdp is not NULL, return with it shared locked.
2060  */
2061 struct filedesc *
2062 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2063 {
2064 	struct filedesc0 *newfdp0;
2065 	struct filedesc *newfdp;
2066 
2067 	if (prepfiles)
2068 		MPASS(lastfile != NULL);
2069 	else
2070 		MPASS(lastfile == NULL);
2071 
2072 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2073 	newfdp = &newfdp0->fd_fd;
2074 
2075 	/* Create the file descriptor table. */
2076 	FILEDESC_LOCK_INIT(newfdp);
2077 	refcount_init(&newfdp->fd_refcnt, 1);
2078 	refcount_init(&newfdp->fd_holdcnt, 1);
2079 	newfdp->fd_map = newfdp0->fd_dmap;
2080 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2081 	newfdp->fd_files->fdt_nfiles = NDFILE;
2082 
2083 	if (fdp == NULL)
2084 		return (newfdp);
2085 
2086 	FILEDESC_SLOCK(fdp);
2087 	if (!prepfiles) {
2088 		FILEDESC_SUNLOCK(fdp);
2089 		return (newfdp);
2090 	}
2091 
2092 	for (;;) {
2093 		*lastfile = fdlastfile(fdp);
2094 		if (*lastfile < newfdp->fd_nfiles)
2095 			break;
2096 		FILEDESC_SUNLOCK(fdp);
2097 		fdgrowtable(newfdp, *lastfile + 1);
2098 		FILEDESC_SLOCK(fdp);
2099 	}
2100 
2101 	return (newfdp);
2102 }
2103 
2104 /*
2105  * Build a pwddesc structure from another.
2106  * Copy the current, root, and jail root vnode references.
2107  *
2108  * If pdp is not NULL, return with it shared locked.
2109  */
2110 struct pwddesc *
2111 pdinit(struct pwddesc *pdp, bool keeplock)
2112 {
2113 	struct pwddesc *newpdp;
2114 	struct pwd *newpwd;
2115 
2116 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2117 
2118 	PWDDESC_LOCK_INIT(newpdp);
2119 	refcount_init(&newpdp->pd_refcount, 1);
2120 	newpdp->pd_cmask = CMASK;
2121 
2122 	if (pdp == NULL) {
2123 		newpwd = pwd_alloc();
2124 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2125 		return (newpdp);
2126 	}
2127 
2128 	PWDDESC_XLOCK(pdp);
2129 	newpwd = pwd_hold_pwddesc(pdp);
2130 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2131 	if (!keeplock)
2132 		PWDDESC_XUNLOCK(pdp);
2133 	return (newpdp);
2134 }
2135 
2136 static struct filedesc *
2137 fdhold(struct proc *p)
2138 {
2139 	struct filedesc *fdp;
2140 
2141 	PROC_LOCK_ASSERT(p, MA_OWNED);
2142 	fdp = p->p_fd;
2143 	if (fdp != NULL)
2144 		refcount_acquire(&fdp->fd_holdcnt);
2145 	return (fdp);
2146 }
2147 
2148 static struct pwddesc *
2149 pdhold(struct proc *p)
2150 {
2151 	struct pwddesc *pdp;
2152 
2153 	PROC_LOCK_ASSERT(p, MA_OWNED);
2154 	pdp = p->p_pd;
2155 	if (pdp != NULL)
2156 		refcount_acquire(&pdp->pd_refcount);
2157 	return (pdp);
2158 }
2159 
2160 static void
2161 fddrop(struct filedesc *fdp)
2162 {
2163 
2164 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2165 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2166 			return;
2167 	}
2168 
2169 	FILEDESC_LOCK_DESTROY(fdp);
2170 	uma_zfree(filedesc0_zone, fdp);
2171 }
2172 
2173 static void
2174 pddrop(struct pwddesc *pdp)
2175 {
2176 	struct pwd *pwd;
2177 
2178 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2179 		return;
2180 
2181 	PWDDESC_XLOCK(pdp);
2182 	if (refcount_release(&pdp->pd_refcount) == 0) {
2183 		PWDDESC_XUNLOCK(pdp);
2184 		return;
2185 	}
2186 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2187 	pwd_set(pdp, NULL);
2188 	PWDDESC_XUNLOCK(pdp);
2189 	pwd_drop(pwd);
2190 
2191 	PWDDESC_LOCK_DESTROY(pdp);
2192 	free(pdp, M_PWDDESC);
2193 }
2194 
2195 /*
2196  * Share a filedesc structure.
2197  */
2198 struct filedesc *
2199 fdshare(struct filedesc *fdp)
2200 {
2201 
2202 	refcount_acquire(&fdp->fd_refcnt);
2203 	return (fdp);
2204 }
2205 
2206 /*
2207  * Share a pwddesc structure.
2208  */
2209 struct pwddesc *
2210 pdshare(struct pwddesc *pdp)
2211 {
2212 	refcount_acquire(&pdp->pd_refcount);
2213 	return (pdp);
2214 }
2215 
2216 /*
2217  * Unshare a filedesc structure, if necessary by making a copy
2218  */
2219 void
2220 fdunshare(struct thread *td)
2221 {
2222 	struct filedesc *tmp;
2223 	struct proc *p = td->td_proc;
2224 
2225 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2226 		return;
2227 
2228 	tmp = fdcopy(p->p_fd);
2229 	fdescfree(td);
2230 	p->p_fd = tmp;
2231 }
2232 
2233 /*
2234  * Unshare a pwddesc structure.
2235  */
2236 void
2237 pdunshare(struct thread *td)
2238 {
2239 	struct pwddesc *pdp;
2240 	struct proc *p;
2241 
2242 	p = td->td_proc;
2243 	/* Not shared. */
2244 	if (p->p_pd->pd_refcount == 1)
2245 		return;
2246 
2247 	pdp = pdcopy(p->p_pd);
2248 	pdescfree(td);
2249 	p->p_pd = pdp;
2250 }
2251 
2252 void
2253 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2254 {
2255 
2256 	fdescfree(td);
2257 	td->td_proc->p_fd = fdp;
2258 }
2259 
2260 /*
2261  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2262  * this is to ease callers, not catch errors.
2263  */
2264 struct filedesc *
2265 fdcopy(struct filedesc *fdp)
2266 {
2267 	struct filedesc *newfdp;
2268 	struct filedescent *nfde, *ofde;
2269 	int i, lastfile;
2270 
2271 	MPASS(fdp != NULL);
2272 
2273 	newfdp = fdinit(fdp, true, &lastfile);
2274 	/* copy all passable descriptors (i.e. not kqueue) */
2275 	newfdp->fd_freefile = -1;
2276 	for (i = 0; i <= lastfile; ++i) {
2277 		ofde = &fdp->fd_ofiles[i];
2278 		if (ofde->fde_file == NULL ||
2279 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2280 		    !fhold(ofde->fde_file)) {
2281 			if (newfdp->fd_freefile == -1)
2282 				newfdp->fd_freefile = i;
2283 			continue;
2284 		}
2285 		nfde = &newfdp->fd_ofiles[i];
2286 		*nfde = *ofde;
2287 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2288 		fdused_init(newfdp, i);
2289 	}
2290 	if (newfdp->fd_freefile == -1)
2291 		newfdp->fd_freefile = i;
2292 	FILEDESC_SUNLOCK(fdp);
2293 	return (newfdp);
2294 }
2295 
2296 /*
2297  * Copy a pwddesc structure.
2298  */
2299 struct pwddesc *
2300 pdcopy(struct pwddesc *pdp)
2301 {
2302 	struct pwddesc *newpdp;
2303 
2304 	MPASS(pdp != NULL);
2305 
2306 	newpdp = pdinit(pdp, true);
2307 	newpdp->pd_cmask = pdp->pd_cmask;
2308 	PWDDESC_XUNLOCK(pdp);
2309 	return (newpdp);
2310 }
2311 
2312 /*
2313  * Copies a filedesc structure, while remapping all file descriptors
2314  * stored inside using a translation table.
2315  *
2316  * File descriptors are copied over to the new file descriptor table,
2317  * regardless of whether the close-on-exec flag is set.
2318  */
2319 int
2320 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2321     struct filedesc **ret)
2322 {
2323 	struct filedesc *newfdp;
2324 	struct filedescent *nfde, *ofde;
2325 	int error, i, lastfile;
2326 
2327 	MPASS(fdp != NULL);
2328 
2329 	newfdp = fdinit(fdp, true, &lastfile);
2330 	if (nfds > lastfile + 1) {
2331 		/* New table cannot be larger than the old one. */
2332 		error = E2BIG;
2333 		goto bad;
2334 	}
2335 	/* Copy all passable descriptors (i.e. not kqueue). */
2336 	newfdp->fd_freefile = nfds;
2337 	for (i = 0; i < nfds; ++i) {
2338 		if (fds[i] < 0 || fds[i] > lastfile) {
2339 			/* File descriptor out of bounds. */
2340 			error = EBADF;
2341 			goto bad;
2342 		}
2343 		ofde = &fdp->fd_ofiles[fds[i]];
2344 		if (ofde->fde_file == NULL) {
2345 			/* Unused file descriptor. */
2346 			error = EBADF;
2347 			goto bad;
2348 		}
2349 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2350 			/* File descriptor cannot be passed. */
2351 			error = EINVAL;
2352 			goto bad;
2353 		}
2354 		if (!fhold(ofde->fde_file)) {
2355 			error = EBADF;
2356 			goto bad;
2357 		}
2358 		nfde = &newfdp->fd_ofiles[i];
2359 		*nfde = *ofde;
2360 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2361 		fdused_init(newfdp, i);
2362 	}
2363 	FILEDESC_SUNLOCK(fdp);
2364 	*ret = newfdp;
2365 	return (0);
2366 bad:
2367 	FILEDESC_SUNLOCK(fdp);
2368 	fdescfree_remapped(newfdp);
2369 	return (error);
2370 }
2371 
2372 /*
2373  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2374  * one of processes using it exits) and the table used to be shared.
2375  */
2376 static void
2377 fdclearlocks(struct thread *td)
2378 {
2379 	struct filedesc *fdp;
2380 	struct filedesc_to_leader *fdtol;
2381 	struct flock lf;
2382 	struct file *fp;
2383 	struct proc *p;
2384 	struct vnode *vp;
2385 	int i, lastfile;
2386 
2387 	p = td->td_proc;
2388 	fdp = p->p_fd;
2389 	fdtol = p->p_fdtol;
2390 	MPASS(fdtol != NULL);
2391 
2392 	FILEDESC_XLOCK(fdp);
2393 	KASSERT(fdtol->fdl_refcount > 0,
2394 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2395 	    fdtol->fdl_refcount));
2396 	if (fdtol->fdl_refcount == 1 &&
2397 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2398 		lastfile = fdlastfile(fdp);
2399 		for (i = 0; i <= lastfile; i++) {
2400 			fp = fdp->fd_ofiles[i].fde_file;
2401 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2402 			    !fhold(fp))
2403 				continue;
2404 			FILEDESC_XUNLOCK(fdp);
2405 			lf.l_whence = SEEK_SET;
2406 			lf.l_start = 0;
2407 			lf.l_len = 0;
2408 			lf.l_type = F_UNLCK;
2409 			vp = fp->f_vnode;
2410 			(void) VOP_ADVLOCK(vp,
2411 			    (caddr_t)p->p_leader, F_UNLCK,
2412 			    &lf, F_POSIX);
2413 			FILEDESC_XLOCK(fdp);
2414 			fdrop(fp, td);
2415 		}
2416 	}
2417 retry:
2418 	if (fdtol->fdl_refcount == 1) {
2419 		if (fdp->fd_holdleaderscount > 0 &&
2420 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2421 			/*
2422 			 * close() or kern_dup() has cleared a reference
2423 			 * in a shared file descriptor table.
2424 			 */
2425 			fdp->fd_holdleaderswakeup = 1;
2426 			sx_sleep(&fdp->fd_holdleaderscount,
2427 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2428 			goto retry;
2429 		}
2430 		if (fdtol->fdl_holdcount > 0) {
2431 			/*
2432 			 * Ensure that fdtol->fdl_leader remains
2433 			 * valid in closef().
2434 			 */
2435 			fdtol->fdl_wakeup = 1;
2436 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2437 			    "fdlhold", 0);
2438 			goto retry;
2439 		}
2440 	}
2441 	fdtol->fdl_refcount--;
2442 	if (fdtol->fdl_refcount == 0 &&
2443 	    fdtol->fdl_holdcount == 0) {
2444 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2445 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2446 	} else
2447 		fdtol = NULL;
2448 	p->p_fdtol = NULL;
2449 	FILEDESC_XUNLOCK(fdp);
2450 	if (fdtol != NULL)
2451 		free(fdtol, M_FILEDESC_TO_LEADER);
2452 }
2453 
2454 /*
2455  * Release a filedesc structure.
2456  */
2457 static void
2458 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2459 {
2460 	struct filedesc0 *fdp0;
2461 	struct freetable *ft, *tft;
2462 	struct filedescent *fde;
2463 	struct file *fp;
2464 	int i, lastfile;
2465 
2466 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2467 	    ("%s: fd table %p carries references", __func__, fdp));
2468 
2469 	/*
2470 	 * Serialize with threads iterating over the table, if any.
2471 	 */
2472 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2473 		FILEDESC_XLOCK(fdp);
2474 		FILEDESC_XUNLOCK(fdp);
2475 	}
2476 
2477 	lastfile = fdlastfile_single(fdp);
2478 	for (i = 0; i <= lastfile; i++) {
2479 		fde = &fdp->fd_ofiles[i];
2480 		fp = fde->fde_file;
2481 		if (fp != NULL) {
2482 			fdefree_last(fde);
2483 			if (needclose)
2484 				(void) closef(fp, td);
2485 			else
2486 				fdrop(fp, td);
2487 		}
2488 	}
2489 
2490 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2491 		free(fdp->fd_map, M_FILEDESC);
2492 	if (fdp->fd_nfiles > NDFILE)
2493 		free(fdp->fd_files, M_FILEDESC);
2494 
2495 	fdp0 = (struct filedesc0 *)fdp;
2496 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2497 		free(ft->ft_table, M_FILEDESC);
2498 
2499 	fddrop(fdp);
2500 }
2501 
2502 void
2503 fdescfree(struct thread *td)
2504 {
2505 	struct proc *p;
2506 	struct filedesc *fdp;
2507 
2508 	p = td->td_proc;
2509 	fdp = p->p_fd;
2510 	MPASS(fdp != NULL);
2511 
2512 #ifdef RACCT
2513 	if (RACCT_ENABLED())
2514 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2515 #endif
2516 
2517 	if (p->p_fdtol != NULL)
2518 		fdclearlocks(td);
2519 
2520 	PROC_LOCK(p);
2521 	p->p_fd = NULL;
2522 	PROC_UNLOCK(p);
2523 
2524 	if (refcount_release(&fdp->fd_refcnt) == 0)
2525 		return;
2526 
2527 	fdescfree_fds(td, fdp, 1);
2528 }
2529 
2530 void
2531 pdescfree(struct thread *td)
2532 {
2533 	struct proc *p;
2534 	struct pwddesc *pdp;
2535 
2536 	p = td->td_proc;
2537 	pdp = p->p_pd;
2538 	MPASS(pdp != NULL);
2539 
2540 	PROC_LOCK(p);
2541 	p->p_pd = NULL;
2542 	PROC_UNLOCK(p);
2543 
2544 	pddrop(pdp);
2545 }
2546 
2547 void
2548 fdescfree_remapped(struct filedesc *fdp)
2549 {
2550 #ifdef INVARIANTS
2551 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2552 	if (!refcount_release(&fdp->fd_refcnt))
2553 		panic("%s: fd table %p has extra references", __func__, fdp);
2554 #endif
2555 	fdescfree_fds(curthread, fdp, 0);
2556 }
2557 
2558 /*
2559  * For setugid programs, we don't want to people to use that setugidness
2560  * to generate error messages which write to a file which otherwise would
2561  * otherwise be off-limits to the process.  We check for filesystems where
2562  * the vnode can change out from under us after execve (like [lin]procfs).
2563  *
2564  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2565  * sufficient.  We also don't check for setugidness since we know we are.
2566  */
2567 static bool
2568 is_unsafe(struct file *fp)
2569 {
2570 	struct vnode *vp;
2571 
2572 	if (fp->f_type != DTYPE_VNODE)
2573 		return (false);
2574 
2575 	vp = fp->f_vnode;
2576 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2577 }
2578 
2579 /*
2580  * Make this setguid thing safe, if at all possible.
2581  */
2582 void
2583 fdsetugidsafety(struct thread *td)
2584 {
2585 	struct filedesc *fdp;
2586 	struct file *fp;
2587 	int i;
2588 
2589 	fdp = td->td_proc->p_fd;
2590 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2591 	    ("the fdtable should not be shared"));
2592 	MPASS(fdp->fd_nfiles >= 3);
2593 	for (i = 0; i <= 2; i++) {
2594 		fp = fdp->fd_ofiles[i].fde_file;
2595 		if (fp != NULL && is_unsafe(fp)) {
2596 			FILEDESC_XLOCK(fdp);
2597 			knote_fdclose(td, i);
2598 			/*
2599 			 * NULL-out descriptor prior to close to avoid
2600 			 * a race while close blocks.
2601 			 */
2602 			fdfree(fdp, i);
2603 			FILEDESC_XUNLOCK(fdp);
2604 			(void) closef(fp, td);
2605 		}
2606 	}
2607 }
2608 
2609 /*
2610  * If a specific file object occupies a specific file descriptor, close the
2611  * file descriptor entry and drop a reference on the file object.  This is a
2612  * convenience function to handle a subsequent error in a function that calls
2613  * falloc() that handles the race that another thread might have closed the
2614  * file descriptor out from under the thread creating the file object.
2615  */
2616 void
2617 fdclose(struct thread *td, struct file *fp, int idx)
2618 {
2619 	struct filedesc *fdp = td->td_proc->p_fd;
2620 
2621 	FILEDESC_XLOCK(fdp);
2622 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2623 		fdfree(fdp, idx);
2624 		FILEDESC_XUNLOCK(fdp);
2625 		fdrop(fp, td);
2626 	} else
2627 		FILEDESC_XUNLOCK(fdp);
2628 }
2629 
2630 /*
2631  * Close any files on exec?
2632  */
2633 void
2634 fdcloseexec(struct thread *td)
2635 {
2636 	struct filedesc *fdp;
2637 	struct filedescent *fde;
2638 	struct file *fp;
2639 	int i, lastfile;
2640 
2641 	fdp = td->td_proc->p_fd;
2642 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2643 	    ("the fdtable should not be shared"));
2644 	lastfile = fdlastfile_single(fdp);
2645 	for (i = 0; i <= lastfile; i++) {
2646 		fde = &fdp->fd_ofiles[i];
2647 		fp = fde->fde_file;
2648 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2649 		    (fde->fde_flags & UF_EXCLOSE))) {
2650 			FILEDESC_XLOCK(fdp);
2651 			fdfree(fdp, i);
2652 			(void) closefp(fdp, i, fp, td, 0);
2653 			FILEDESC_UNLOCK_ASSERT(fdp);
2654 		}
2655 	}
2656 }
2657 
2658 /*
2659  * It is unsafe for set[ug]id processes to be started with file
2660  * descriptors 0..2 closed, as these descriptors are given implicit
2661  * significance in the Standard C library.  fdcheckstd() will create a
2662  * descriptor referencing /dev/null for each of stdin, stdout, and
2663  * stderr that is not already open.
2664  */
2665 int
2666 fdcheckstd(struct thread *td)
2667 {
2668 	struct filedesc *fdp;
2669 	register_t save;
2670 	int i, error, devnull;
2671 
2672 	fdp = td->td_proc->p_fd;
2673 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2674 	    ("the fdtable should not be shared"));
2675 	MPASS(fdp->fd_nfiles >= 3);
2676 	devnull = -1;
2677 	for (i = 0; i <= 2; i++) {
2678 		if (fdp->fd_ofiles[i].fde_file != NULL)
2679 			continue;
2680 
2681 		save = td->td_retval[0];
2682 		if (devnull != -1) {
2683 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2684 		} else {
2685 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2686 			    UIO_SYSSPACE, O_RDWR, 0);
2687 			if (error == 0) {
2688 				devnull = td->td_retval[0];
2689 				KASSERT(devnull == i, ("we didn't get our fd"));
2690 			}
2691 		}
2692 		td->td_retval[0] = save;
2693 		if (error != 0)
2694 			return (error);
2695 	}
2696 	return (0);
2697 }
2698 
2699 /*
2700  * Internal form of close.  Decrement reference count on file structure.
2701  * Note: td may be NULL when closing a file that was being passed in a
2702  * message.
2703  */
2704 int
2705 closef(struct file *fp, struct thread *td)
2706 {
2707 	struct vnode *vp;
2708 	struct flock lf;
2709 	struct filedesc_to_leader *fdtol;
2710 	struct filedesc *fdp;
2711 
2712 	/*
2713 	 * POSIX record locking dictates that any close releases ALL
2714 	 * locks owned by this process.  This is handled by setting
2715 	 * a flag in the unlock to free ONLY locks obeying POSIX
2716 	 * semantics, and not to free BSD-style file locks.
2717 	 * If the descriptor was in a message, POSIX-style locks
2718 	 * aren't passed with the descriptor, and the thread pointer
2719 	 * will be NULL.  Callers should be careful only to pass a
2720 	 * NULL thread pointer when there really is no owning
2721 	 * context that might have locks, or the locks will be
2722 	 * leaked.
2723 	 */
2724 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2725 		vp = fp->f_vnode;
2726 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2727 			lf.l_whence = SEEK_SET;
2728 			lf.l_start = 0;
2729 			lf.l_len = 0;
2730 			lf.l_type = F_UNLCK;
2731 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2732 			    F_UNLCK, &lf, F_POSIX);
2733 		}
2734 		fdtol = td->td_proc->p_fdtol;
2735 		if (fdtol != NULL) {
2736 			/*
2737 			 * Handle special case where file descriptor table is
2738 			 * shared between multiple process leaders.
2739 			 */
2740 			fdp = td->td_proc->p_fd;
2741 			FILEDESC_XLOCK(fdp);
2742 			for (fdtol = fdtol->fdl_next;
2743 			    fdtol != td->td_proc->p_fdtol;
2744 			    fdtol = fdtol->fdl_next) {
2745 				if ((fdtol->fdl_leader->p_flag &
2746 				    P_ADVLOCK) == 0)
2747 					continue;
2748 				fdtol->fdl_holdcount++;
2749 				FILEDESC_XUNLOCK(fdp);
2750 				lf.l_whence = SEEK_SET;
2751 				lf.l_start = 0;
2752 				lf.l_len = 0;
2753 				lf.l_type = F_UNLCK;
2754 				vp = fp->f_vnode;
2755 				(void) VOP_ADVLOCK(vp,
2756 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2757 				    F_POSIX);
2758 				FILEDESC_XLOCK(fdp);
2759 				fdtol->fdl_holdcount--;
2760 				if (fdtol->fdl_holdcount == 0 &&
2761 				    fdtol->fdl_wakeup != 0) {
2762 					fdtol->fdl_wakeup = 0;
2763 					wakeup(fdtol);
2764 				}
2765 			}
2766 			FILEDESC_XUNLOCK(fdp);
2767 		}
2768 	}
2769 	return (fdrop_close(fp, td));
2770 }
2771 
2772 /*
2773  * Initialize the file pointer with the specified properties.
2774  *
2775  * The ops are set with release semantics to be certain that the flags, type,
2776  * and data are visible when ops is.  This is to prevent ops methods from being
2777  * called with bad data.
2778  */
2779 void
2780 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2781 {
2782 	fp->f_data = data;
2783 	fp->f_flag = flag;
2784 	fp->f_type = type;
2785 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2786 }
2787 
2788 void
2789 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2790 {
2791 	fp->f_seqcount[UIO_READ] = 1;
2792 	fp->f_seqcount[UIO_WRITE] = 1;
2793 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2794 	    data, ops);
2795 }
2796 
2797 int
2798 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2799     struct file **fpp, struct filecaps *havecapsp)
2800 {
2801 	struct filedescent *fde;
2802 	int error;
2803 
2804 	FILEDESC_LOCK_ASSERT(fdp);
2805 
2806 	fde = fdeget_locked(fdp, fd);
2807 	if (fde == NULL) {
2808 		error = EBADF;
2809 		goto out;
2810 	}
2811 
2812 #ifdef CAPABILITIES
2813 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2814 	if (error != 0)
2815 		goto out;
2816 #endif
2817 
2818 	if (havecapsp != NULL)
2819 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2820 
2821 	*fpp = fde->fde_file;
2822 
2823 	error = 0;
2824 out:
2825 	return (error);
2826 }
2827 
2828 int
2829 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2830     struct file **fpp, struct filecaps *havecapsp)
2831 {
2832 	struct filedesc *fdp = td->td_proc->p_fd;
2833 	int error;
2834 #ifndef CAPABILITIES
2835 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2836 	if (havecapsp != NULL && error == 0)
2837 		filecaps_fill(havecapsp);
2838 #else
2839 	struct file *fp;
2840 	seqc_t seq;
2841 
2842 	*fpp = NULL;
2843 	for (;;) {
2844 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2845 		if (error != 0)
2846 			return (error);
2847 
2848 		if (havecapsp != NULL) {
2849 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2850 			    havecapsp, false)) {
2851 				fdrop(fp, td);
2852 				goto get_locked;
2853 			}
2854 		}
2855 
2856 		if (!fd_modified(fdp, fd, seq))
2857 			break;
2858 		fdrop(fp, td);
2859 	}
2860 
2861 	*fpp = fp;
2862 	return (0);
2863 
2864 get_locked:
2865 	FILEDESC_SLOCK(fdp);
2866 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2867 	if (error == 0 && !fhold(*fpp))
2868 		error = EBADF;
2869 	FILEDESC_SUNLOCK(fdp);
2870 #endif
2871 	return (error);
2872 }
2873 
2874 #ifdef CAPABILITIES
2875 int
2876 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2877 {
2878 	const struct filedescent *fde;
2879 	const struct fdescenttbl *fdt;
2880 	struct filedesc *fdp;
2881 	struct file *fp;
2882 	struct vnode *vp;
2883 	const cap_rights_t *haverights;
2884 	cap_rights_t rights;
2885 	seqc_t seq;
2886 
2887 	VFS_SMR_ASSERT_ENTERED();
2888 
2889 	rights = *ndp->ni_rightsneeded;
2890 	cap_rights_set_one(&rights, CAP_LOOKUP);
2891 
2892 	fdp = curproc->p_fd;
2893 	fdt = fdp->fd_files;
2894 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2895 		return (EBADF);
2896 	seq = seqc_read_any(fd_seqc(fdt, fd));
2897 	if (__predict_false(seqc_in_modify(seq)))
2898 		return (EAGAIN);
2899 	fde = &fdt->fdt_ofiles[fd];
2900 	haverights = cap_rights_fde_inline(fde);
2901 	fp = fde->fde_file;
2902 	if (__predict_false(fp == NULL))
2903 		return (EAGAIN);
2904 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2905 		return (EAGAIN);
2906 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2907 	vp = fp->f_vnode;
2908 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2909 		return (EAGAIN);
2910 	}
2911 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2912 		return (EAGAIN);
2913 	}
2914 	/*
2915 	 * Use an acquire barrier to force re-reading of fdt so it is
2916 	 * refreshed for verification.
2917 	 */
2918 	atomic_thread_fence_acq();
2919 	fdt = fdp->fd_files;
2920 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2921 		return (EAGAIN);
2922 	/*
2923 	 * If file descriptor doesn't have all rights,
2924 	 * all lookups relative to it must also be
2925 	 * strictly relative.
2926 	 *
2927 	 * Not yet supported by fast path.
2928 	 */
2929 	CAP_ALL(&rights);
2930 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2931 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2932 	    ndp->ni_filecaps.fc_nioctls != -1) {
2933 #ifdef notyet
2934 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2935 #else
2936 		return (EAGAIN);
2937 #endif
2938 	}
2939 	*vpp = vp;
2940 	return (0);
2941 }
2942 #else
2943 int
2944 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2945 {
2946 	const struct fdescenttbl *fdt;
2947 	struct filedesc *fdp;
2948 	struct file *fp;
2949 	struct vnode *vp;
2950 
2951 	VFS_SMR_ASSERT_ENTERED();
2952 
2953 	fdp = curproc->p_fd;
2954 	fdt = fdp->fd_files;
2955 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2956 		return (EBADF);
2957 	fp = fdt->fdt_ofiles[fd].fde_file;
2958 	if (__predict_false(fp == NULL))
2959 		return (EAGAIN);
2960 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2961 	vp = fp->f_vnode;
2962 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2963 		return (EAGAIN);
2964 	}
2965 	/*
2966 	 * Use an acquire barrier to force re-reading of fdt so it is
2967 	 * refreshed for verification.
2968 	 */
2969 	atomic_thread_fence_acq();
2970 	fdt = fdp->fd_files;
2971 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2972 		return (EAGAIN);
2973 	filecaps_fill(&ndp->ni_filecaps);
2974 	*vpp = vp;
2975 	return (0);
2976 }
2977 #endif
2978 
2979 int
2980 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2981     struct file **fpp, seqc_t *seqp)
2982 {
2983 #ifdef CAPABILITIES
2984 	const struct filedescent *fde;
2985 #endif
2986 	const struct fdescenttbl *fdt;
2987 	struct file *fp;
2988 #ifdef CAPABILITIES
2989 	seqc_t seq;
2990 	cap_rights_t haverights;
2991 	int error;
2992 #endif
2993 
2994 	fdt = fdp->fd_files;
2995 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2996 		return (EBADF);
2997 	/*
2998 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2999 	 * never raising a refcount above 0.  To accomplish this we have
3000 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3001 	 * must be re-verified once we acquire a reference to be certain
3002 	 * that the identity is still correct and we did not lose a race
3003 	 * due to preemption.
3004 	 */
3005 	for (;;) {
3006 #ifdef CAPABILITIES
3007 		seq = seqc_read(fd_seqc(fdt, fd));
3008 		fde = &fdt->fdt_ofiles[fd];
3009 		haverights = *cap_rights_fde_inline(fde);
3010 		fp = fde->fde_file;
3011 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3012 			continue;
3013 #else
3014 		fp = fdt->fdt_ofiles[fd].fde_file;
3015 #endif
3016 		if (fp == NULL)
3017 			return (EBADF);
3018 #ifdef CAPABILITIES
3019 		error = cap_check_inline(&haverights, needrightsp);
3020 		if (error != 0)
3021 			return (error);
3022 #endif
3023 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3024 			/*
3025 			 * Force a reload. Other thread could reallocate the
3026 			 * table before this fd was closed, so it is possible
3027 			 * that there is a stale fp pointer in cached version.
3028 			 */
3029 			fdt = atomic_load_ptr(&fdp->fd_files);
3030 			continue;
3031 		}
3032 		/*
3033 		 * Use an acquire barrier to force re-reading of fdt so it is
3034 		 * refreshed for verification.
3035 		 */
3036 		atomic_thread_fence_acq();
3037 		fdt = fdp->fd_files;
3038 #ifdef	CAPABILITIES
3039 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3040 #else
3041 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3042 #endif
3043 			break;
3044 		fdrop(fp, curthread);
3045 	}
3046 	*fpp = fp;
3047 	if (seqp != NULL) {
3048 #ifdef CAPABILITIES
3049 		*seqp = seq;
3050 #endif
3051 	}
3052 	return (0);
3053 }
3054 
3055 /*
3056  * See the comments in fget_unlocked_seq for an explanation of how this works.
3057  *
3058  * This is a simplified variant which bails out to the aforementioned routine
3059  * if anything goes wrong. In practice this only happens when userspace is
3060  * racing with itself.
3061  */
3062 int
3063 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3064     struct file **fpp)
3065 {
3066 #ifdef CAPABILITIES
3067 	const struct filedescent *fde;
3068 #endif
3069 	const struct fdescenttbl *fdt;
3070 	struct file *fp;
3071 #ifdef CAPABILITIES
3072 	seqc_t seq;
3073 	const cap_rights_t *haverights;
3074 #endif
3075 
3076 	fdt = fdp->fd_files;
3077 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3078 		return (EBADF);
3079 #ifdef CAPABILITIES
3080 	seq = seqc_read_any(fd_seqc(fdt, fd));
3081 	if (__predict_false(seqc_in_modify(seq)))
3082 		goto out_fallback;
3083 	fde = &fdt->fdt_ofiles[fd];
3084 	haverights = cap_rights_fde_inline(fde);
3085 	fp = fde->fde_file;
3086 #else
3087 	fp = fdt->fdt_ofiles[fd].fde_file;
3088 #endif
3089 	if (__predict_false(fp == NULL))
3090 		goto out_fallback;
3091 #ifdef CAPABILITIES
3092 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3093 		goto out_fallback;
3094 #endif
3095 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3096 		goto out_fallback;
3097 
3098 	/*
3099 	 * Use an acquire barrier to force re-reading of fdt so it is
3100 	 * refreshed for verification.
3101 	 */
3102 	atomic_thread_fence_acq();
3103 	fdt = fdp->fd_files;
3104 #ifdef	CAPABILITIES
3105 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3106 #else
3107 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3108 #endif
3109 		goto out_fdrop;
3110 	*fpp = fp;
3111 	return (0);
3112 out_fdrop:
3113 	fdrop(fp, curthread);
3114 out_fallback:
3115 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3116 }
3117 
3118 /*
3119  * Extract the file pointer associated with the specified descriptor for the
3120  * current user process.
3121  *
3122  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3123  * returned.
3124  *
3125  * File's rights will be checked against the capability rights mask.
3126  *
3127  * If an error occurred the non-zero error is returned and *fpp is set to
3128  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3129  * responsible for fdrop().
3130  */
3131 static __inline int
3132 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3133     cap_rights_t *needrightsp)
3134 {
3135 	struct filedesc *fdp;
3136 	struct file *fp;
3137 	int error;
3138 
3139 	*fpp = NULL;
3140 	fdp = td->td_proc->p_fd;
3141 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3142 	if (__predict_false(error != 0))
3143 		return (error);
3144 	if (__predict_false(fp->f_ops == &badfileops)) {
3145 		fdrop(fp, td);
3146 		return (EBADF);
3147 	}
3148 
3149 	/*
3150 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3151 	 */
3152 	error = 0;
3153 	switch (flags) {
3154 	case FREAD:
3155 	case FWRITE:
3156 		if ((fp->f_flag & flags) == 0)
3157 			error = EBADF;
3158 		break;
3159 	case FEXEC:
3160 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3161 		    ((fp->f_flag & FWRITE) != 0))
3162 			error = EBADF;
3163 		break;
3164 	case 0:
3165 		break;
3166 	default:
3167 		KASSERT(0, ("wrong flags"));
3168 	}
3169 
3170 	if (error != 0) {
3171 		fdrop(fp, td);
3172 		return (error);
3173 	}
3174 
3175 	*fpp = fp;
3176 	return (0);
3177 }
3178 
3179 int
3180 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3181 {
3182 
3183 	return (_fget(td, fd, fpp, 0, rightsp));
3184 }
3185 
3186 int
3187 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3188     struct file **fpp)
3189 {
3190 	int error;
3191 #ifndef CAPABILITIES
3192 	error = _fget(td, fd, fpp, 0, rightsp);
3193 	if (maxprotp != NULL)
3194 		*maxprotp = VM_PROT_ALL;
3195 	return (error);
3196 #else
3197 	cap_rights_t fdrights;
3198 	struct filedesc *fdp;
3199 	struct file *fp;
3200 	seqc_t seq;
3201 
3202 	*fpp = NULL;
3203 	fdp = td->td_proc->p_fd;
3204 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3205 	for (;;) {
3206 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3207 		if (__predict_false(error != 0))
3208 			return (error);
3209 		if (__predict_false(fp->f_ops == &badfileops)) {
3210 			fdrop(fp, td);
3211 			return (EBADF);
3212 		}
3213 		if (maxprotp != NULL)
3214 			fdrights = *cap_rights(fdp, fd);
3215 		if (!fd_modified(fdp, fd, seq))
3216 			break;
3217 		fdrop(fp, td);
3218 	}
3219 
3220 	/*
3221 	 * If requested, convert capability rights to access flags.
3222 	 */
3223 	if (maxprotp != NULL)
3224 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3225 	*fpp = fp;
3226 	return (0);
3227 #endif
3228 }
3229 
3230 int
3231 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3232 {
3233 
3234 	return (_fget(td, fd, fpp, FREAD, rightsp));
3235 }
3236 
3237 int
3238 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3239 {
3240 
3241 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3242 }
3243 
3244 int
3245 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3246     struct file **fpp)
3247 {
3248 	struct filedesc *fdp = td->td_proc->p_fd;
3249 #ifndef CAPABILITIES
3250 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3251 #else
3252 	struct file *fp;
3253 	int error;
3254 	seqc_t seq;
3255 
3256 	*fpp = NULL;
3257 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3258 	for (;;) {
3259 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3260 		if (error != 0)
3261 			return (error);
3262 		error = cap_fcntl_check(fdp, fd, needfcntl);
3263 		if (!fd_modified(fdp, fd, seq))
3264 			break;
3265 		fdrop(fp, td);
3266 	}
3267 	if (error != 0) {
3268 		fdrop(fp, td);
3269 		return (error);
3270 	}
3271 	*fpp = fp;
3272 	return (0);
3273 #endif
3274 }
3275 
3276 /*
3277  * Like fget() but loads the underlying vnode, or returns an error if the
3278  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3279  * never have VM objects.  The returned vnode will be vref()'d.
3280  *
3281  * XXX: what about the unused flags ?
3282  */
3283 static __inline int
3284 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3285     struct vnode **vpp)
3286 {
3287 	struct file *fp;
3288 	int error;
3289 
3290 	*vpp = NULL;
3291 	error = _fget(td, fd, &fp, flags, needrightsp);
3292 	if (error != 0)
3293 		return (error);
3294 	if (fp->f_vnode == NULL) {
3295 		error = EINVAL;
3296 	} else {
3297 		*vpp = fp->f_vnode;
3298 		vrefact(*vpp);
3299 	}
3300 	fdrop(fp, td);
3301 
3302 	return (error);
3303 }
3304 
3305 int
3306 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3307 {
3308 
3309 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3310 }
3311 
3312 int
3313 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3314     struct filecaps *havecaps, struct vnode **vpp)
3315 {
3316 	struct filecaps caps;
3317 	struct file *fp;
3318 	int error;
3319 
3320 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3321 	if (error != 0)
3322 		return (error);
3323 	if (fp->f_ops == &badfileops) {
3324 		error = EBADF;
3325 		goto out;
3326 	}
3327 	if (fp->f_vnode == NULL) {
3328 		error = EINVAL;
3329 		goto out;
3330 	}
3331 
3332 	*havecaps = caps;
3333 	*vpp = fp->f_vnode;
3334 	vrefact(*vpp);
3335 	fdrop(fp, td);
3336 
3337 	return (0);
3338 out:
3339 	filecaps_free(&caps);
3340 	fdrop(fp, td);
3341 	return (error);
3342 }
3343 
3344 int
3345 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3346 {
3347 
3348 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3349 }
3350 
3351 int
3352 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3353 {
3354 
3355 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3356 }
3357 
3358 #ifdef notyet
3359 int
3360 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3361     struct vnode **vpp)
3362 {
3363 
3364 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3365 }
3366 #endif
3367 
3368 /*
3369  * Handle the last reference to a file being closed.
3370  *
3371  * Without the noinline attribute clang keeps inlining the func thorough this
3372  * file when fdrop is used.
3373  */
3374 int __noinline
3375 _fdrop(struct file *fp, struct thread *td)
3376 {
3377 	int error;
3378 #ifdef INVARIANTS
3379 	int count;
3380 
3381 	count = refcount_load(&fp->f_count);
3382 	if (count != 0)
3383 		panic("fdrop: fp %p count %d", fp, count);
3384 #endif
3385 	error = fo_close(fp, td);
3386 	atomic_subtract_int(&openfiles, 1);
3387 	crfree(fp->f_cred);
3388 	free(fp->f_advice, M_FADVISE);
3389 	uma_zfree(file_zone, fp);
3390 
3391 	return (error);
3392 }
3393 
3394 /*
3395  * Apply an advisory lock on a file descriptor.
3396  *
3397  * Just attempt to get a record lock of the requested type on the entire file
3398  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3399  */
3400 #ifndef _SYS_SYSPROTO_H_
3401 struct flock_args {
3402 	int	fd;
3403 	int	how;
3404 };
3405 #endif
3406 /* ARGSUSED */
3407 int
3408 sys_flock(struct thread *td, struct flock_args *uap)
3409 {
3410 	struct file *fp;
3411 	struct vnode *vp;
3412 	struct flock lf;
3413 	int error;
3414 
3415 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3416 	if (error != 0)
3417 		return (error);
3418 	if (fp->f_type != DTYPE_VNODE) {
3419 		fdrop(fp, td);
3420 		return (EOPNOTSUPP);
3421 	}
3422 
3423 	vp = fp->f_vnode;
3424 	lf.l_whence = SEEK_SET;
3425 	lf.l_start = 0;
3426 	lf.l_len = 0;
3427 	if (uap->how & LOCK_UN) {
3428 		lf.l_type = F_UNLCK;
3429 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3430 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3431 		goto done2;
3432 	}
3433 	if (uap->how & LOCK_EX)
3434 		lf.l_type = F_WRLCK;
3435 	else if (uap->how & LOCK_SH)
3436 		lf.l_type = F_RDLCK;
3437 	else {
3438 		error = EBADF;
3439 		goto done2;
3440 	}
3441 	atomic_set_int(&fp->f_flag, FHASLOCK);
3442 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3443 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3444 done2:
3445 	fdrop(fp, td);
3446 	return (error);
3447 }
3448 /*
3449  * Duplicate the specified descriptor to a free descriptor.
3450  */
3451 int
3452 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3453     int openerror, int *indxp)
3454 {
3455 	struct filedescent *newfde, *oldfde;
3456 	struct file *fp;
3457 	u_long *ioctls;
3458 	int error, indx;
3459 
3460 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3461 	    ("unexpected error %d in %s", openerror, __func__));
3462 
3463 	/*
3464 	 * If the to-be-dup'd fd number is greater than the allowed number
3465 	 * of file descriptors, or the fd to be dup'd has already been
3466 	 * closed, then reject.
3467 	 */
3468 	FILEDESC_XLOCK(fdp);
3469 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3470 		FILEDESC_XUNLOCK(fdp);
3471 		return (EBADF);
3472 	}
3473 
3474 	error = fdalloc(td, 0, &indx);
3475 	if (error != 0) {
3476 		FILEDESC_XUNLOCK(fdp);
3477 		return (error);
3478 	}
3479 
3480 	/*
3481 	 * There are two cases of interest here.
3482 	 *
3483 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3484 	 *
3485 	 * For ENXIO steal away the file structure from (dfd) and store it in
3486 	 * (indx).  (dfd) is effectively closed by this operation.
3487 	 */
3488 	switch (openerror) {
3489 	case ENODEV:
3490 		/*
3491 		 * Check that the mode the file is being opened for is a
3492 		 * subset of the mode of the existing descriptor.
3493 		 */
3494 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3495 			fdunused(fdp, indx);
3496 			FILEDESC_XUNLOCK(fdp);
3497 			return (EACCES);
3498 		}
3499 		if (!fhold(fp)) {
3500 			fdunused(fdp, indx);
3501 			FILEDESC_XUNLOCK(fdp);
3502 			return (EBADF);
3503 		}
3504 		newfde = &fdp->fd_ofiles[indx];
3505 		oldfde = &fdp->fd_ofiles[dfd];
3506 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3507 #ifdef CAPABILITIES
3508 		seqc_write_begin(&newfde->fde_seqc);
3509 #endif
3510 		memcpy(newfde, oldfde, fde_change_size);
3511 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3512 		    ioctls);
3513 #ifdef CAPABILITIES
3514 		seqc_write_end(&newfde->fde_seqc);
3515 #endif
3516 		break;
3517 	case ENXIO:
3518 		/*
3519 		 * Steal away the file pointer from dfd and stuff it into indx.
3520 		 */
3521 		newfde = &fdp->fd_ofiles[indx];
3522 		oldfde = &fdp->fd_ofiles[dfd];
3523 #ifdef CAPABILITIES
3524 		seqc_write_begin(&newfde->fde_seqc);
3525 #endif
3526 		memcpy(newfde, oldfde, fde_change_size);
3527 		oldfde->fde_file = NULL;
3528 		fdunused(fdp, dfd);
3529 #ifdef CAPABILITIES
3530 		seqc_write_end(&newfde->fde_seqc);
3531 #endif
3532 		break;
3533 	}
3534 	FILEDESC_XUNLOCK(fdp);
3535 	*indxp = indx;
3536 	return (0);
3537 }
3538 
3539 /*
3540  * This sysctl determines if we will allow a process to chroot(2) if it
3541  * has a directory open:
3542  *	0: disallowed for all processes.
3543  *	1: allowed for processes that were not already chroot(2)'ed.
3544  *	2: allowed for all processes.
3545  */
3546 
3547 static int chroot_allow_open_directories = 1;
3548 
3549 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3550     &chroot_allow_open_directories, 0,
3551     "Allow a process to chroot(2) if it has a directory open");
3552 
3553 /*
3554  * Helper function for raised chroot(2) security function:  Refuse if
3555  * any filedescriptors are open directories.
3556  */
3557 static int
3558 chroot_refuse_vdir_fds(struct filedesc *fdp)
3559 {
3560 	struct vnode *vp;
3561 	struct file *fp;
3562 	int fd, lastfile;
3563 
3564 	FILEDESC_LOCK_ASSERT(fdp);
3565 
3566 	lastfile = fdlastfile(fdp);
3567 	for (fd = 0; fd <= lastfile; fd++) {
3568 		fp = fget_locked(fdp, fd);
3569 		if (fp == NULL)
3570 			continue;
3571 		if (fp->f_type == DTYPE_VNODE) {
3572 			vp = fp->f_vnode;
3573 			if (vp->v_type == VDIR)
3574 				return (EPERM);
3575 		}
3576 	}
3577 	return (0);
3578 }
3579 
3580 static void
3581 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3582 {
3583 
3584 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3585 		vrefact(oldpwd->pwd_cdir);
3586 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3587 	}
3588 
3589 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3590 		vrefact(oldpwd->pwd_rdir);
3591 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3592 	}
3593 
3594 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3595 		vrefact(oldpwd->pwd_jdir);
3596 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3597 	}
3598 }
3599 
3600 struct pwd *
3601 pwd_hold_pwddesc(struct pwddesc *pdp)
3602 {
3603 	struct pwd *pwd;
3604 
3605 	PWDDESC_ASSERT_XLOCKED(pdp);
3606 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3607 	if (pwd != NULL)
3608 		refcount_acquire(&pwd->pwd_refcount);
3609 	return (pwd);
3610 }
3611 
3612 bool
3613 pwd_hold_smr(struct pwd *pwd)
3614 {
3615 
3616 	MPASS(pwd != NULL);
3617 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3618 		return (true);
3619 	}
3620 	return (false);
3621 }
3622 
3623 struct pwd *
3624 pwd_hold(struct thread *td)
3625 {
3626 	struct pwddesc *pdp;
3627 	struct pwd *pwd;
3628 
3629 	pdp = td->td_proc->p_pd;
3630 
3631 	vfs_smr_enter();
3632 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3633 	if (pwd_hold_smr(pwd)) {
3634 		vfs_smr_exit();
3635 		return (pwd);
3636 	}
3637 	vfs_smr_exit();
3638 	PWDDESC_XLOCK(pdp);
3639 	pwd = pwd_hold_pwddesc(pdp);
3640 	MPASS(pwd != NULL);
3641 	PWDDESC_XUNLOCK(pdp);
3642 	return (pwd);
3643 }
3644 
3645 struct pwd *
3646 pwd_get_smr(void)
3647 {
3648 	struct pwd *pwd;
3649 
3650 	pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3651 	MPASS(pwd != NULL);
3652 	return (pwd);
3653 }
3654 
3655 static struct pwd *
3656 pwd_alloc(void)
3657 {
3658 	struct pwd *pwd;
3659 
3660 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3661 	bzero(pwd, sizeof(*pwd));
3662 	refcount_init(&pwd->pwd_refcount, 1);
3663 	return (pwd);
3664 }
3665 
3666 void
3667 pwd_drop(struct pwd *pwd)
3668 {
3669 
3670 	if (!refcount_release(&pwd->pwd_refcount))
3671 		return;
3672 
3673 	if (pwd->pwd_cdir != NULL)
3674 		vrele(pwd->pwd_cdir);
3675 	if (pwd->pwd_rdir != NULL)
3676 		vrele(pwd->pwd_rdir);
3677 	if (pwd->pwd_jdir != NULL)
3678 		vrele(pwd->pwd_jdir);
3679 	uma_zfree_smr(pwd_zone, pwd);
3680 }
3681 
3682 /*
3683 * Common routine for kern_chroot() and jail_attach().  The caller is
3684 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3685 * authorize this operation.
3686 */
3687 int
3688 pwd_chroot(struct thread *td, struct vnode *vp)
3689 {
3690 	struct pwddesc *pdp;
3691 	struct filedesc *fdp;
3692 	struct pwd *newpwd, *oldpwd;
3693 	int error;
3694 
3695 	fdp = td->td_proc->p_fd;
3696 	pdp = td->td_proc->p_pd;
3697 	newpwd = pwd_alloc();
3698 	FILEDESC_SLOCK(fdp);
3699 	PWDDESC_XLOCK(pdp);
3700 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3701 	if (chroot_allow_open_directories == 0 ||
3702 	    (chroot_allow_open_directories == 1 &&
3703 	    oldpwd->pwd_rdir != rootvnode)) {
3704 		error = chroot_refuse_vdir_fds(fdp);
3705 		FILEDESC_SUNLOCK(fdp);
3706 		if (error != 0) {
3707 			PWDDESC_XUNLOCK(pdp);
3708 			pwd_drop(newpwd);
3709 			return (error);
3710 		}
3711 	} else {
3712 		FILEDESC_SUNLOCK(fdp);
3713 	}
3714 
3715 	vrefact(vp);
3716 	newpwd->pwd_rdir = vp;
3717 	if (oldpwd->pwd_jdir == NULL) {
3718 		vrefact(vp);
3719 		newpwd->pwd_jdir = vp;
3720 	}
3721 	pwd_fill(oldpwd, newpwd);
3722 	pwd_set(pdp, newpwd);
3723 	PWDDESC_XUNLOCK(pdp);
3724 	pwd_drop(oldpwd);
3725 	return (0);
3726 }
3727 
3728 void
3729 pwd_chdir(struct thread *td, struct vnode *vp)
3730 {
3731 	struct pwddesc *pdp;
3732 	struct pwd *newpwd, *oldpwd;
3733 
3734 	VNPASS(vp->v_usecount > 0, vp);
3735 
3736 	newpwd = pwd_alloc();
3737 	pdp = td->td_proc->p_pd;
3738 	PWDDESC_XLOCK(pdp);
3739 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3740 	newpwd->pwd_cdir = vp;
3741 	pwd_fill(oldpwd, newpwd);
3742 	pwd_set(pdp, newpwd);
3743 	PWDDESC_XUNLOCK(pdp);
3744 	pwd_drop(oldpwd);
3745 }
3746 
3747 void
3748 pwd_ensure_dirs(void)
3749 {
3750 	struct pwddesc *pdp;
3751 	struct pwd *oldpwd, *newpwd;
3752 
3753 	pdp = curproc->p_pd;
3754 	PWDDESC_XLOCK(pdp);
3755 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3756 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3757 		PWDDESC_XUNLOCK(pdp);
3758 		return;
3759 	}
3760 	PWDDESC_XUNLOCK(pdp);
3761 
3762 	newpwd = pwd_alloc();
3763 	PWDDESC_XLOCK(pdp);
3764 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3765 	pwd_fill(oldpwd, newpwd);
3766 	if (newpwd->pwd_cdir == NULL) {
3767 		vrefact(rootvnode);
3768 		newpwd->pwd_cdir = rootvnode;
3769 	}
3770 	if (newpwd->pwd_rdir == NULL) {
3771 		vrefact(rootvnode);
3772 		newpwd->pwd_rdir = rootvnode;
3773 	}
3774 	pwd_set(pdp, newpwd);
3775 	PWDDESC_XUNLOCK(pdp);
3776 	pwd_drop(oldpwd);
3777 }
3778 
3779 void
3780 pwd_set_rootvnode(void)
3781 {
3782 	struct pwddesc *pdp;
3783 	struct pwd *oldpwd, *newpwd;
3784 
3785 	pdp = curproc->p_pd;
3786 
3787 	newpwd = pwd_alloc();
3788 	PWDDESC_XLOCK(pdp);
3789 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3790 	vrefact(rootvnode);
3791 	newpwd->pwd_cdir = rootvnode;
3792 	vrefact(rootvnode);
3793 	newpwd->pwd_rdir = rootvnode;
3794 	pwd_fill(oldpwd, newpwd);
3795 	pwd_set(pdp, newpwd);
3796 	PWDDESC_XUNLOCK(pdp);
3797 	pwd_drop(oldpwd);
3798 }
3799 
3800 /*
3801  * Scan all active processes and prisons to see if any of them have a current
3802  * or root directory of `olddp'. If so, replace them with the new mount point.
3803  */
3804 void
3805 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3806 {
3807 	struct pwddesc *pdp;
3808 	struct pwd *newpwd, *oldpwd;
3809 	struct prison *pr;
3810 	struct proc *p;
3811 	int nrele;
3812 
3813 	if (vrefcnt(olddp) == 1)
3814 		return;
3815 	nrele = 0;
3816 	newpwd = pwd_alloc();
3817 	sx_slock(&allproc_lock);
3818 	FOREACH_PROC_IN_SYSTEM(p) {
3819 		PROC_LOCK(p);
3820 		pdp = pdhold(p);
3821 		PROC_UNLOCK(p);
3822 		if (pdp == NULL)
3823 			continue;
3824 		PWDDESC_XLOCK(pdp);
3825 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3826 		if (oldpwd == NULL ||
3827 		    (oldpwd->pwd_cdir != olddp &&
3828 		    oldpwd->pwd_rdir != olddp &&
3829 		    oldpwd->pwd_jdir != olddp)) {
3830 			PWDDESC_XUNLOCK(pdp);
3831 			pddrop(pdp);
3832 			continue;
3833 		}
3834 		if (oldpwd->pwd_cdir == olddp) {
3835 			vrefact(newdp);
3836 			newpwd->pwd_cdir = newdp;
3837 		}
3838 		if (oldpwd->pwd_rdir == olddp) {
3839 			vrefact(newdp);
3840 			newpwd->pwd_rdir = newdp;
3841 		}
3842 		if (oldpwd->pwd_jdir == olddp) {
3843 			vrefact(newdp);
3844 			newpwd->pwd_jdir = newdp;
3845 		}
3846 		pwd_fill(oldpwd, newpwd);
3847 		pwd_set(pdp, newpwd);
3848 		PWDDESC_XUNLOCK(pdp);
3849 		pwd_drop(oldpwd);
3850 		pddrop(pdp);
3851 		newpwd = pwd_alloc();
3852 	}
3853 	sx_sunlock(&allproc_lock);
3854 	pwd_drop(newpwd);
3855 	if (rootvnode == olddp) {
3856 		vrefact(newdp);
3857 		rootvnode = newdp;
3858 		nrele++;
3859 	}
3860 	mtx_lock(&prison0.pr_mtx);
3861 	if (prison0.pr_root == olddp) {
3862 		vrefact(newdp);
3863 		prison0.pr_root = newdp;
3864 		nrele++;
3865 	}
3866 	mtx_unlock(&prison0.pr_mtx);
3867 	sx_slock(&allprison_lock);
3868 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3869 		mtx_lock(&pr->pr_mtx);
3870 		if (pr->pr_root == olddp) {
3871 			vrefact(newdp);
3872 			pr->pr_root = newdp;
3873 			nrele++;
3874 		}
3875 		mtx_unlock(&pr->pr_mtx);
3876 	}
3877 	sx_sunlock(&allprison_lock);
3878 	while (nrele--)
3879 		vrele(olddp);
3880 }
3881 
3882 struct filedesc_to_leader *
3883 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3884 {
3885 	struct filedesc_to_leader *fdtol;
3886 
3887 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3888 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3889 	fdtol->fdl_refcount = 1;
3890 	fdtol->fdl_holdcount = 0;
3891 	fdtol->fdl_wakeup = 0;
3892 	fdtol->fdl_leader = leader;
3893 	if (old != NULL) {
3894 		FILEDESC_XLOCK(fdp);
3895 		fdtol->fdl_next = old->fdl_next;
3896 		fdtol->fdl_prev = old;
3897 		old->fdl_next = fdtol;
3898 		fdtol->fdl_next->fdl_prev = fdtol;
3899 		FILEDESC_XUNLOCK(fdp);
3900 	} else {
3901 		fdtol->fdl_next = fdtol;
3902 		fdtol->fdl_prev = fdtol;
3903 	}
3904 	return (fdtol);
3905 }
3906 
3907 static int
3908 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3909 {
3910 	NDSLOTTYPE *map;
3911 	struct filedesc *fdp;
3912 	int count, off, minoff;
3913 
3914 	if (*(int *)arg1 != 0)
3915 		return (EINVAL);
3916 
3917 	fdp = curproc->p_fd;
3918 	count = 0;
3919 	FILEDESC_SLOCK(fdp);
3920 	map = fdp->fd_map;
3921 	off = NDSLOT(fdp->fd_nfiles - 1);
3922 	for (minoff = NDSLOT(0); off >= minoff; --off)
3923 		count += bitcountl(map[off]);
3924 	FILEDESC_SUNLOCK(fdp);
3925 
3926 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3927 }
3928 
3929 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3930     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3931     "Number of open file descriptors");
3932 
3933 /*
3934  * Get file structures globally.
3935  */
3936 static int
3937 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3938 {
3939 	struct xfile xf;
3940 	struct filedesc *fdp;
3941 	struct file *fp;
3942 	struct proc *p;
3943 	int error, n, lastfile;
3944 
3945 	error = sysctl_wire_old_buffer(req, 0);
3946 	if (error != 0)
3947 		return (error);
3948 	if (req->oldptr == NULL) {
3949 		n = 0;
3950 		sx_slock(&allproc_lock);
3951 		FOREACH_PROC_IN_SYSTEM(p) {
3952 			PROC_LOCK(p);
3953 			if (p->p_state == PRS_NEW) {
3954 				PROC_UNLOCK(p);
3955 				continue;
3956 			}
3957 			fdp = fdhold(p);
3958 			PROC_UNLOCK(p);
3959 			if (fdp == NULL)
3960 				continue;
3961 			/* overestimates sparse tables. */
3962 			n += fdp->fd_nfiles;
3963 			fddrop(fdp);
3964 		}
3965 		sx_sunlock(&allproc_lock);
3966 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3967 	}
3968 	error = 0;
3969 	bzero(&xf, sizeof(xf));
3970 	xf.xf_size = sizeof(xf);
3971 	sx_slock(&allproc_lock);
3972 	FOREACH_PROC_IN_SYSTEM(p) {
3973 		PROC_LOCK(p);
3974 		if (p->p_state == PRS_NEW) {
3975 			PROC_UNLOCK(p);
3976 			continue;
3977 		}
3978 		if (p_cansee(req->td, p) != 0) {
3979 			PROC_UNLOCK(p);
3980 			continue;
3981 		}
3982 		xf.xf_pid = p->p_pid;
3983 		xf.xf_uid = p->p_ucred->cr_uid;
3984 		fdp = fdhold(p);
3985 		PROC_UNLOCK(p);
3986 		if (fdp == NULL)
3987 			continue;
3988 		FILEDESC_SLOCK(fdp);
3989 		lastfile = fdlastfile(fdp);
3990 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
3991 		    n++) {
3992 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3993 				continue;
3994 			xf.xf_fd = n;
3995 			xf.xf_file = (uintptr_t)fp;
3996 			xf.xf_data = (uintptr_t)fp->f_data;
3997 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3998 			xf.xf_type = (uintptr_t)fp->f_type;
3999 			xf.xf_count = refcount_load(&fp->f_count);
4000 			xf.xf_msgcount = 0;
4001 			xf.xf_offset = foffset_get(fp);
4002 			xf.xf_flag = fp->f_flag;
4003 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4004 			if (error)
4005 				break;
4006 		}
4007 		FILEDESC_SUNLOCK(fdp);
4008 		fddrop(fdp);
4009 		if (error)
4010 			break;
4011 	}
4012 	sx_sunlock(&allproc_lock);
4013 	return (error);
4014 }
4015 
4016 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4017     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4018 
4019 #ifdef KINFO_FILE_SIZE
4020 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4021 #endif
4022 
4023 static int
4024 xlate_fflags(int fflags)
4025 {
4026 	static const struct {
4027 		int	fflag;
4028 		int	kf_fflag;
4029 	} fflags_table[] = {
4030 		{ FAPPEND, KF_FLAG_APPEND },
4031 		{ FASYNC, KF_FLAG_ASYNC },
4032 		{ FFSYNC, KF_FLAG_FSYNC },
4033 		{ FHASLOCK, KF_FLAG_HASLOCK },
4034 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4035 		{ FREAD, KF_FLAG_READ },
4036 		{ FWRITE, KF_FLAG_WRITE },
4037 		{ O_CREAT, KF_FLAG_CREAT },
4038 		{ O_DIRECT, KF_FLAG_DIRECT },
4039 		{ O_EXCL, KF_FLAG_EXCL },
4040 		{ O_EXEC, KF_FLAG_EXEC },
4041 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4042 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4043 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4044 		{ O_TRUNC, KF_FLAG_TRUNC }
4045 	};
4046 	unsigned int i;
4047 	int kflags;
4048 
4049 	kflags = 0;
4050 	for (i = 0; i < nitems(fflags_table); i++)
4051 		if (fflags & fflags_table[i].fflag)
4052 			kflags |=  fflags_table[i].kf_fflag;
4053 	return (kflags);
4054 }
4055 
4056 /* Trim unused data from kf_path by truncating the structure size. */
4057 void
4058 pack_kinfo(struct kinfo_file *kif)
4059 {
4060 
4061 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4062 	    strlen(kif->kf_path) + 1;
4063 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4064 }
4065 
4066 static void
4067 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4068     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4069 {
4070 	int error;
4071 
4072 	bzero(kif, sizeof(*kif));
4073 
4074 	/* Set a default type to allow for empty fill_kinfo() methods. */
4075 	kif->kf_type = KF_TYPE_UNKNOWN;
4076 	kif->kf_flags = xlate_fflags(fp->f_flag);
4077 	if (rightsp != NULL)
4078 		kif->kf_cap_rights = *rightsp;
4079 	else
4080 		cap_rights_init_zero(&kif->kf_cap_rights);
4081 	kif->kf_fd = fd;
4082 	kif->kf_ref_count = refcount_load(&fp->f_count);
4083 	kif->kf_offset = foffset_get(fp);
4084 
4085 	/*
4086 	 * This may drop the filedesc lock, so the 'fp' cannot be
4087 	 * accessed after this call.
4088 	 */
4089 	error = fo_fill_kinfo(fp, kif, fdp);
4090 	if (error == 0)
4091 		kif->kf_status |= KF_ATTR_VALID;
4092 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4093 		pack_kinfo(kif);
4094 	else
4095 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4096 }
4097 
4098 static void
4099 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4100     struct kinfo_file *kif, int flags)
4101 {
4102 	int error;
4103 
4104 	bzero(kif, sizeof(*kif));
4105 
4106 	kif->kf_type = KF_TYPE_VNODE;
4107 	error = vn_fill_kinfo_vnode(vp, kif);
4108 	if (error == 0)
4109 		kif->kf_status |= KF_ATTR_VALID;
4110 	kif->kf_flags = xlate_fflags(fflags);
4111 	cap_rights_init_zero(&kif->kf_cap_rights);
4112 	kif->kf_fd = fd;
4113 	kif->kf_ref_count = -1;
4114 	kif->kf_offset = -1;
4115 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4116 		pack_kinfo(kif);
4117 	else
4118 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4119 	vrele(vp);
4120 }
4121 
4122 struct export_fd_buf {
4123 	struct filedesc		*fdp;
4124 	struct pwddesc	*pdp;
4125 	struct sbuf 		*sb;
4126 	ssize_t			remainder;
4127 	struct kinfo_file	kif;
4128 	int			flags;
4129 };
4130 
4131 static int
4132 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4133 {
4134 	struct kinfo_file *kif;
4135 
4136 	kif = &efbuf->kif;
4137 	if (efbuf->remainder != -1) {
4138 		if (efbuf->remainder < kif->kf_structsize) {
4139 			/* Terminate export. */
4140 			efbuf->remainder = 0;
4141 			return (0);
4142 		}
4143 		efbuf->remainder -= kif->kf_structsize;
4144 	}
4145 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4146 }
4147 
4148 static int
4149 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4150     struct export_fd_buf *efbuf)
4151 {
4152 	int error;
4153 
4154 	if (efbuf->remainder == 0)
4155 		return (0);
4156 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4157 	    efbuf->flags);
4158 	FILEDESC_SUNLOCK(efbuf->fdp);
4159 	error = export_kinfo_to_sb(efbuf);
4160 	FILEDESC_SLOCK(efbuf->fdp);
4161 	return (error);
4162 }
4163 
4164 static int
4165 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4166     struct export_fd_buf *efbuf)
4167 {
4168 	int error;
4169 
4170 	if (efbuf->remainder == 0)
4171 		return (0);
4172 	if (efbuf->pdp != NULL)
4173 		PWDDESC_XUNLOCK(efbuf->pdp);
4174 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4175 	error = export_kinfo_to_sb(efbuf);
4176 	if (efbuf->pdp != NULL)
4177 		PWDDESC_XLOCK(efbuf->pdp);
4178 	return (error);
4179 }
4180 
4181 /*
4182  * Store a process file descriptor information to sbuf.
4183  *
4184  * Takes a locked proc as argument, and returns with the proc unlocked.
4185  */
4186 int
4187 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4188     int flags)
4189 {
4190 	struct file *fp;
4191 	struct filedesc *fdp;
4192 	struct pwddesc *pdp;
4193 	struct export_fd_buf *efbuf;
4194 	struct vnode *cttyvp, *textvp, *tracevp;
4195 	struct pwd *pwd;
4196 	int error, i, lastfile;
4197 	cap_rights_t rights;
4198 
4199 	PROC_LOCK_ASSERT(p, MA_OWNED);
4200 
4201 	/* ktrace vnode */
4202 	tracevp = p->p_tracevp;
4203 	if (tracevp != NULL)
4204 		vrefact(tracevp);
4205 	/* text vnode */
4206 	textvp = p->p_textvp;
4207 	if (textvp != NULL)
4208 		vrefact(textvp);
4209 	/* Controlling tty. */
4210 	cttyvp = NULL;
4211 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4212 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4213 		if (cttyvp != NULL)
4214 			vrefact(cttyvp);
4215 	}
4216 	fdp = fdhold(p);
4217 	pdp = pdhold(p);
4218 	PROC_UNLOCK(p);
4219 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4220 	efbuf->fdp = NULL;
4221 	efbuf->pdp = NULL;
4222 	efbuf->sb = sb;
4223 	efbuf->remainder = maxlen;
4224 	efbuf->flags = flags;
4225 	if (tracevp != NULL)
4226 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4227 		    efbuf);
4228 	if (textvp != NULL)
4229 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4230 	if (cttyvp != NULL)
4231 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4232 		    efbuf);
4233 	error = 0;
4234 	if (pdp == NULL || fdp == NULL)
4235 		goto fail;
4236 	efbuf->fdp = fdp;
4237 	efbuf->pdp = pdp;
4238 	PWDDESC_XLOCK(pdp);
4239 	pwd = pwd_hold_pwddesc(pdp);
4240 	if (pwd != NULL) {
4241 		/* working directory */
4242 		if (pwd->pwd_cdir != NULL) {
4243 			vrefact(pwd->pwd_cdir);
4244 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4245 		}
4246 		/* root directory */
4247 		if (pwd->pwd_rdir != NULL) {
4248 			vrefact(pwd->pwd_rdir);
4249 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4250 		}
4251 		/* jail directory */
4252 		if (pwd->pwd_jdir != NULL) {
4253 			vrefact(pwd->pwd_jdir);
4254 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4255 		}
4256 	}
4257 	PWDDESC_XUNLOCK(pdp);
4258 	if (pwd != NULL)
4259 		pwd_drop(pwd);
4260 	FILEDESC_SLOCK(fdp);
4261 	lastfile = fdlastfile(fdp);
4262 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4263 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4264 			continue;
4265 #ifdef CAPABILITIES
4266 		rights = *cap_rights(fdp, i);
4267 #else /* !CAPABILITIES */
4268 		rights = cap_no_rights;
4269 #endif
4270 		/*
4271 		 * Create sysctl entry.  It is OK to drop the filedesc
4272 		 * lock inside of export_file_to_sb() as we will
4273 		 * re-validate and re-evaluate its properties when the
4274 		 * loop continues.
4275 		 */
4276 		error = export_file_to_sb(fp, i, &rights, efbuf);
4277 		if (error != 0 || efbuf->remainder == 0)
4278 			break;
4279 	}
4280 	FILEDESC_SUNLOCK(fdp);
4281 fail:
4282 	if (fdp != NULL)
4283 		fddrop(fdp);
4284 	if (pdp != NULL)
4285 		pddrop(pdp);
4286 	free(efbuf, M_TEMP);
4287 	return (error);
4288 }
4289 
4290 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4291 
4292 /*
4293  * Get per-process file descriptors for use by procstat(1), et al.
4294  */
4295 static int
4296 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4297 {
4298 	struct sbuf sb;
4299 	struct proc *p;
4300 	ssize_t maxlen;
4301 	int error, error2, *name;
4302 
4303 	name = (int *)arg1;
4304 
4305 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4306 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4307 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4308 	if (error != 0) {
4309 		sbuf_delete(&sb);
4310 		return (error);
4311 	}
4312 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4313 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4314 	    KERN_FILEDESC_PACK_KINFO);
4315 	error2 = sbuf_finish(&sb);
4316 	sbuf_delete(&sb);
4317 	return (error != 0 ? error : error2);
4318 }
4319 
4320 #ifdef COMPAT_FREEBSD7
4321 #ifdef KINFO_OFILE_SIZE
4322 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4323 #endif
4324 
4325 static void
4326 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4327 {
4328 
4329 	okif->kf_structsize = sizeof(*okif);
4330 	okif->kf_type = kif->kf_type;
4331 	okif->kf_fd = kif->kf_fd;
4332 	okif->kf_ref_count = kif->kf_ref_count;
4333 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4334 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4335 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4336 	okif->kf_offset = kif->kf_offset;
4337 	if (kif->kf_type == KF_TYPE_VNODE)
4338 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4339 	else
4340 		okif->kf_vnode_type = KF_VTYPE_VNON;
4341 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4342 	if (kif->kf_type == KF_TYPE_SOCKET) {
4343 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4344 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4345 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4346 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4347 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4348 	} else {
4349 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4350 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4351 	}
4352 }
4353 
4354 static int
4355 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4356     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4357 {
4358 	int error;
4359 
4360 	vrefact(vp);
4361 	PWDDESC_XUNLOCK(pdp);
4362 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4363 	kinfo_to_okinfo(kif, okif);
4364 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4365 	PWDDESC_XLOCK(pdp);
4366 	return (error);
4367 }
4368 
4369 /*
4370  * Get per-process file descriptors for use by procstat(1), et al.
4371  */
4372 static int
4373 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4374 {
4375 	struct kinfo_ofile *okif;
4376 	struct kinfo_file *kif;
4377 	struct filedesc *fdp;
4378 	struct pwddesc *pdp;
4379 	struct pwd *pwd;
4380 	int error, i, lastfile, *name;
4381 	struct file *fp;
4382 	struct proc *p;
4383 
4384 	name = (int *)arg1;
4385 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4386 	if (error != 0)
4387 		return (error);
4388 	fdp = fdhold(p);
4389 	if (fdp != NULL)
4390 		pdp = pdhold(p);
4391 	PROC_UNLOCK(p);
4392 	if (fdp == NULL || pdp == NULL) {
4393 		if (fdp != NULL)
4394 			fddrop(fdp);
4395 		return (ENOENT);
4396 	}
4397 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4398 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4399 	PWDDESC_XLOCK(pdp);
4400 	pwd = pwd_hold_pwddesc(pdp);
4401 	if (pwd != NULL) {
4402 		if (pwd->pwd_cdir != NULL)
4403 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4404 			    okif, pdp, req);
4405 		if (pwd->pwd_rdir != NULL)
4406 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4407 			    okif, pdp, req);
4408 		if (pwd->pwd_jdir != NULL)
4409 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4410 			    okif, pdp, req);
4411 	}
4412 	PWDDESC_XUNLOCK(pdp);
4413 	if (pwd != NULL)
4414 		pwd_drop(pwd);
4415 	FILEDESC_SLOCK(fdp);
4416 	lastfile = fdlastfile(fdp);
4417 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4418 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4419 			continue;
4420 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4421 		    KERN_FILEDESC_PACK_KINFO);
4422 		FILEDESC_SUNLOCK(fdp);
4423 		kinfo_to_okinfo(kif, okif);
4424 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4425 		FILEDESC_SLOCK(fdp);
4426 		if (error)
4427 			break;
4428 	}
4429 	FILEDESC_SUNLOCK(fdp);
4430 	fddrop(fdp);
4431 	pddrop(pdp);
4432 	free(kif, M_TEMP);
4433 	free(okif, M_TEMP);
4434 	return (0);
4435 }
4436 
4437 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4438     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4439     "Process ofiledesc entries");
4440 #endif	/* COMPAT_FREEBSD7 */
4441 
4442 int
4443 vntype_to_kinfo(int vtype)
4444 {
4445 	struct {
4446 		int	vtype;
4447 		int	kf_vtype;
4448 	} vtypes_table[] = {
4449 		{ VBAD, KF_VTYPE_VBAD },
4450 		{ VBLK, KF_VTYPE_VBLK },
4451 		{ VCHR, KF_VTYPE_VCHR },
4452 		{ VDIR, KF_VTYPE_VDIR },
4453 		{ VFIFO, KF_VTYPE_VFIFO },
4454 		{ VLNK, KF_VTYPE_VLNK },
4455 		{ VNON, KF_VTYPE_VNON },
4456 		{ VREG, KF_VTYPE_VREG },
4457 		{ VSOCK, KF_VTYPE_VSOCK }
4458 	};
4459 	unsigned int i;
4460 
4461 	/*
4462 	 * Perform vtype translation.
4463 	 */
4464 	for (i = 0; i < nitems(vtypes_table); i++)
4465 		if (vtypes_table[i].vtype == vtype)
4466 			return (vtypes_table[i].kf_vtype);
4467 
4468 	return (KF_VTYPE_UNKNOWN);
4469 }
4470 
4471 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4472     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4473     "Process filedesc entries");
4474 
4475 /*
4476  * Store a process current working directory information to sbuf.
4477  *
4478  * Takes a locked proc as argument, and returns with the proc unlocked.
4479  */
4480 int
4481 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4482 {
4483 	struct pwddesc *pdp;
4484 	struct pwd *pwd;
4485 	struct export_fd_buf *efbuf;
4486 	struct vnode *cdir;
4487 	int error;
4488 
4489 	PROC_LOCK_ASSERT(p, MA_OWNED);
4490 
4491 	pdp = pdhold(p);
4492 	PROC_UNLOCK(p);
4493 	if (pdp == NULL)
4494 		return (EINVAL);
4495 
4496 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4497 	efbuf->pdp = pdp;
4498 	efbuf->sb = sb;
4499 	efbuf->remainder = maxlen;
4500 
4501 	PWDDESC_XLOCK(pdp);
4502 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4503 	cdir = pwd->pwd_cdir;
4504 	if (cdir == NULL) {
4505 		error = EINVAL;
4506 	} else {
4507 		vrefact(cdir);
4508 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4509 	}
4510 	PWDDESC_XUNLOCK(pdp);
4511 	pddrop(pdp);
4512 	free(efbuf, M_TEMP);
4513 	return (error);
4514 }
4515 
4516 /*
4517  * Get per-process current working directory.
4518  */
4519 static int
4520 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4521 {
4522 	struct sbuf sb;
4523 	struct proc *p;
4524 	ssize_t maxlen;
4525 	int error, error2, *name;
4526 
4527 	name = (int *)arg1;
4528 
4529 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4530 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4531 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4532 	if (error != 0) {
4533 		sbuf_delete(&sb);
4534 		return (error);
4535 	}
4536 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4537 	error = kern_proc_cwd_out(p, &sb, maxlen);
4538 	error2 = sbuf_finish(&sb);
4539 	sbuf_delete(&sb);
4540 	return (error != 0 ? error : error2);
4541 }
4542 
4543 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4544     sysctl_kern_proc_cwd, "Process current working directory");
4545 
4546 #ifdef DDB
4547 /*
4548  * For the purposes of debugging, generate a human-readable string for the
4549  * file type.
4550  */
4551 static const char *
4552 file_type_to_name(short type)
4553 {
4554 
4555 	switch (type) {
4556 	case 0:
4557 		return ("zero");
4558 	case DTYPE_VNODE:
4559 		return ("vnode");
4560 	case DTYPE_SOCKET:
4561 		return ("socket");
4562 	case DTYPE_PIPE:
4563 		return ("pipe");
4564 	case DTYPE_FIFO:
4565 		return ("fifo");
4566 	case DTYPE_KQUEUE:
4567 		return ("kqueue");
4568 	case DTYPE_CRYPTO:
4569 		return ("crypto");
4570 	case DTYPE_MQUEUE:
4571 		return ("mqueue");
4572 	case DTYPE_SHM:
4573 		return ("shm");
4574 	case DTYPE_SEM:
4575 		return ("ksem");
4576 	case DTYPE_PTS:
4577 		return ("pts");
4578 	case DTYPE_DEV:
4579 		return ("dev");
4580 	case DTYPE_PROCDESC:
4581 		return ("proc");
4582 	case DTYPE_LINUXEFD:
4583 		return ("levent");
4584 	case DTYPE_LINUXTFD:
4585 		return ("ltimer");
4586 	default:
4587 		return ("unkn");
4588 	}
4589 }
4590 
4591 /*
4592  * For the purposes of debugging, identify a process (if any, perhaps one of
4593  * many) that references the passed file in its file descriptor array. Return
4594  * NULL if none.
4595  */
4596 static struct proc *
4597 file_to_first_proc(struct file *fp)
4598 {
4599 	struct filedesc *fdp;
4600 	struct proc *p;
4601 	int n;
4602 
4603 	FOREACH_PROC_IN_SYSTEM(p) {
4604 		if (p->p_state == PRS_NEW)
4605 			continue;
4606 		fdp = p->p_fd;
4607 		if (fdp == NULL)
4608 			continue;
4609 		for (n = 0; n < fdp->fd_nfiles; n++) {
4610 			if (fp == fdp->fd_ofiles[n].fde_file)
4611 				return (p);
4612 		}
4613 	}
4614 	return (NULL);
4615 }
4616 
4617 static void
4618 db_print_file(struct file *fp, int header)
4619 {
4620 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4621 	struct proc *p;
4622 
4623 	if (header)
4624 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4625 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4626 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4627 		    "FCmd");
4628 	p = file_to_first_proc(fp);
4629 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4630 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4631 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4632 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4633 
4634 #undef XPTRWIDTH
4635 }
4636 
4637 DB_SHOW_COMMAND(file, db_show_file)
4638 {
4639 	struct file *fp;
4640 
4641 	if (!have_addr) {
4642 		db_printf("usage: show file <addr>\n");
4643 		return;
4644 	}
4645 	fp = (struct file *)addr;
4646 	db_print_file(fp, 1);
4647 }
4648 
4649 DB_SHOW_COMMAND(files, db_show_files)
4650 {
4651 	struct filedesc *fdp;
4652 	struct file *fp;
4653 	struct proc *p;
4654 	int header;
4655 	int n;
4656 
4657 	header = 1;
4658 	FOREACH_PROC_IN_SYSTEM(p) {
4659 		if (p->p_state == PRS_NEW)
4660 			continue;
4661 		if ((fdp = p->p_fd) == NULL)
4662 			continue;
4663 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4664 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4665 				continue;
4666 			db_print_file(fp, header);
4667 			header = 0;
4668 		}
4669 	}
4670 }
4671 #endif
4672 
4673 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4674     &maxfilesperproc, 0, "Maximum files allowed open per process");
4675 
4676 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4677     &maxfiles, 0, "Maximum number of files");
4678 
4679 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4680     &openfiles, 0, "System-wide number of open files");
4681 
4682 /* ARGSUSED*/
4683 static void
4684 filelistinit(void *dummy)
4685 {
4686 
4687 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4688 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4689 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4690 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4691 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4692 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4693 	/*
4694 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4695 	 * the vnode zone.
4696 	 */
4697 	vfs_smr = uma_zone_get_smr(pwd_zone);
4698 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4699 }
4700 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4701 
4702 /*-------------------------------------------------------------------*/
4703 
4704 static int
4705 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4706     int flags, struct thread *td)
4707 {
4708 
4709 	return (EBADF);
4710 }
4711 
4712 static int
4713 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4714     struct thread *td)
4715 {
4716 
4717 	return (EINVAL);
4718 }
4719 
4720 static int
4721 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4722     struct thread *td)
4723 {
4724 
4725 	return (EBADF);
4726 }
4727 
4728 static int
4729 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4730     struct thread *td)
4731 {
4732 
4733 	return (0);
4734 }
4735 
4736 static int
4737 badfo_kqfilter(struct file *fp, struct knote *kn)
4738 {
4739 
4740 	return (EBADF);
4741 }
4742 
4743 static int
4744 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4745     struct thread *td)
4746 {
4747 
4748 	return (EBADF);
4749 }
4750 
4751 static int
4752 badfo_close(struct file *fp, struct thread *td)
4753 {
4754 
4755 	return (0);
4756 }
4757 
4758 static int
4759 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4760     struct thread *td)
4761 {
4762 
4763 	return (EBADF);
4764 }
4765 
4766 static int
4767 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4768     struct thread *td)
4769 {
4770 
4771 	return (EBADF);
4772 }
4773 
4774 static int
4775 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4776     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4777     struct thread *td)
4778 {
4779 
4780 	return (EBADF);
4781 }
4782 
4783 static int
4784 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4785 {
4786 
4787 	return (0);
4788 }
4789 
4790 struct fileops badfileops = {
4791 	.fo_read = badfo_readwrite,
4792 	.fo_write = badfo_readwrite,
4793 	.fo_truncate = badfo_truncate,
4794 	.fo_ioctl = badfo_ioctl,
4795 	.fo_poll = badfo_poll,
4796 	.fo_kqfilter = badfo_kqfilter,
4797 	.fo_stat = badfo_stat,
4798 	.fo_close = badfo_close,
4799 	.fo_chmod = badfo_chmod,
4800 	.fo_chown = badfo_chown,
4801 	.fo_sendfile = badfo_sendfile,
4802 	.fo_fill_kinfo = badfo_fill_kinfo,
4803 };
4804 
4805 int
4806 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4807     int flags, struct thread *td)
4808 {
4809 
4810 	return (EOPNOTSUPP);
4811 }
4812 
4813 int
4814 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4815     struct thread *td)
4816 {
4817 
4818 	return (EINVAL);
4819 }
4820 
4821 int
4822 invfo_ioctl(struct file *fp, u_long com, void *data,
4823     struct ucred *active_cred, struct thread *td)
4824 {
4825 
4826 	return (ENOTTY);
4827 }
4828 
4829 int
4830 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4831     struct thread *td)
4832 {
4833 
4834 	return (poll_no_poll(events));
4835 }
4836 
4837 int
4838 invfo_kqfilter(struct file *fp, struct knote *kn)
4839 {
4840 
4841 	return (EINVAL);
4842 }
4843 
4844 int
4845 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4846     struct thread *td)
4847 {
4848 
4849 	return (EINVAL);
4850 }
4851 
4852 int
4853 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4854     struct thread *td)
4855 {
4856 
4857 	return (EINVAL);
4858 }
4859 
4860 int
4861 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4862     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4863     struct thread *td)
4864 {
4865 
4866 	return (EINVAL);
4867 }
4868 
4869 /*-------------------------------------------------------------------*/
4870 
4871 /*
4872  * File Descriptor pseudo-device driver (/dev/fd/).
4873  *
4874  * Opening minor device N dup()s the file (if any) connected to file
4875  * descriptor N belonging to the calling process.  Note that this driver
4876  * consists of only the ``open()'' routine, because all subsequent
4877  * references to this file will be direct to the other driver.
4878  *
4879  * XXX: we could give this one a cloning event handler if necessary.
4880  */
4881 
4882 /* ARGSUSED */
4883 static int
4884 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4885 {
4886 
4887 	/*
4888 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4889 	 * the file descriptor being sought for duplication. The error
4890 	 * return ensures that the vnode for this device will be released
4891 	 * by vn_open. Open will detect this special error and take the
4892 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4893 	 * will simply report the error.
4894 	 */
4895 	td->td_dupfd = dev2unit(dev);
4896 	return (ENODEV);
4897 }
4898 
4899 static struct cdevsw fildesc_cdevsw = {
4900 	.d_version =	D_VERSION,
4901 	.d_open =	fdopen,
4902 	.d_name =	"FD",
4903 };
4904 
4905 static void
4906 fildesc_drvinit(void *unused)
4907 {
4908 	struct cdev *dev;
4909 
4910 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4911 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4912 	make_dev_alias(dev, "stdin");
4913 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4914 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4915 	make_dev_alias(dev, "stdout");
4916 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4917 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4918 	make_dev_alias(dev, "stderr");
4919 }
4920 
4921 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4922