xref: /freebsd/sys/kern/kern_descrip.c (revision c983ec6428ecfa362903cbdba1dc9f4303e2f28c)
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/poll.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sbuf.h>
71 #include <sys/signalvar.h>
72 #include <sys/kdb.h>
73 #include <sys/smr.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
106 VFS_SMR_DECLARE;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, bool holdleaders, bool audit);
110 static void	export_file_to_kinfo(struct file *fp, int fd,
111 		    cap_rights_t *rightsp, struct kinfo_file *kif,
112 		    struct filedesc *fdp, int flags);
113 static int	fd_first_free(struct filedesc *fdp, int low, int size);
114 static void	fdgrowtable(struct filedesc *fdp, int nfd);
115 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
116 static void	fdunused(struct filedesc *fdp, int fd);
117 static void	fdused(struct filedesc *fdp, int fd);
118 static int	getmaxfd(struct thread *td);
119 static u_long	*filecaps_copy_prep(const struct filecaps *src);
120 static void	filecaps_copy_finish(const struct filecaps *src,
121 		    struct filecaps *dst, u_long *ioctls);
122 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
123 static void	filecaps_free_finish(u_long *ioctls);
124 
125 static struct pwd *pwd_alloc(void);
126 
127 /*
128  * Each process has:
129  *
130  * - An array of open file descriptors (fd_ofiles)
131  * - An array of file flags (fd_ofileflags)
132  * - A bitmap recording which descriptors are in use (fd_map)
133  *
134  * A process starts out with NDFILE descriptors.  The value of NDFILE has
135  * been selected based the historical limit of 20 open files, and an
136  * assumption that the majority of processes, especially short-lived
137  * processes like shells, will never need more.
138  *
139  * If this initial allocation is exhausted, a larger descriptor table and
140  * map are allocated dynamically, and the pointers in the process's struct
141  * filedesc are updated to point to those.  This is repeated every time
142  * the process runs out of file descriptors (provided it hasn't hit its
143  * resource limit).
144  *
145  * Since threads may hold references to individual descriptor table
146  * entries, the tables are never freed.  Instead, they are placed on a
147  * linked list and freed only when the struct filedesc is released.
148  */
149 #define NDFILE		20
150 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
151 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
152 #define NDSLOT(x)	((x) / NDENTRIES)
153 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
154 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
155 
156 /*
157  * SLIST entry used to keep track of ofiles which must be reclaimed when
158  * the process exits.
159  */
160 struct freetable {
161 	struct fdescenttbl *ft_table;
162 	SLIST_ENTRY(freetable) ft_next;
163 };
164 
165 /*
166  * Initial allocation: a filedesc structure + the head of SLIST used to
167  * keep track of old ofiles + enough space for NDFILE descriptors.
168  */
169 
170 struct fdescenttbl0 {
171 	int	fdt_nfiles;
172 	struct	filedescent fdt_ofiles[NDFILE];
173 };
174 
175 struct filedesc0 {
176 	struct filedesc fd_fd;
177 	SLIST_HEAD(, freetable) fd_free;
178 	struct	fdescenttbl0 fd_dfiles;
179 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
180 };
181 
182 /*
183  * Descriptor management.
184  */
185 static int __exclusive_cache_line openfiles; /* actual number of open files */
186 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
187 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
188 
189 /*
190  * If low >= size, just return low. Otherwise find the first zero bit in the
191  * given bitmap, starting at low and not exceeding size - 1. Return size if
192  * not found.
193  */
194 static int
195 fd_first_free(struct filedesc *fdp, int low, int size)
196 {
197 	NDSLOTTYPE *map = fdp->fd_map;
198 	NDSLOTTYPE mask;
199 	int off, maxoff;
200 
201 	if (low >= size)
202 		return (low);
203 
204 	off = NDSLOT(low);
205 	if (low % NDENTRIES) {
206 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
207 		if ((mask &= ~map[off]) != 0UL)
208 			return (off * NDENTRIES + ffsl(mask) - 1);
209 		++off;
210 	}
211 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
212 		if (map[off] != ~0UL)
213 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
214 	return (size);
215 }
216 
217 /*
218  * Find the last used fd.
219  *
220  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
221  * Otherwise use fdlastfile.
222  */
223 int
224 fdlastfile_single(struct filedesc *fdp)
225 {
226 	NDSLOTTYPE *map = fdp->fd_map;
227 	int off, minoff;
228 
229 	off = NDSLOT(fdp->fd_nfiles - 1);
230 	for (minoff = NDSLOT(0); off >= minoff; --off)
231 		if (map[off] != 0)
232 			return (off * NDENTRIES + flsl(map[off]) - 1);
233 	return (-1);
234 }
235 
236 int
237 fdlastfile(struct filedesc *fdp)
238 {
239 
240 	FILEDESC_LOCK_ASSERT(fdp);
241 	return (fdlastfile_single(fdp));
242 }
243 
244 static int
245 fdisused(struct filedesc *fdp, int fd)
246 {
247 
248 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
249 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
250 
251 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
252 }
253 
254 /*
255  * Mark a file descriptor as used.
256  */
257 static void
258 fdused_init(struct filedesc *fdp, int fd)
259 {
260 
261 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
262 
263 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
264 }
265 
266 static void
267 fdused(struct filedesc *fdp, int fd)
268 {
269 
270 	FILEDESC_XLOCK_ASSERT(fdp);
271 
272 	fdused_init(fdp, fd);
273 	if (fd == fdp->fd_freefile)
274 		fdp->fd_freefile++;
275 }
276 
277 /*
278  * Mark a file descriptor as unused.
279  */
280 static void
281 fdunused(struct filedesc *fdp, int fd)
282 {
283 
284 	FILEDESC_XLOCK_ASSERT(fdp);
285 
286 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
287 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
288 	    ("fd=%d is still in use", fd));
289 
290 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
291 	if (fd < fdp->fd_freefile)
292 		fdp->fd_freefile = fd;
293 }
294 
295 /*
296  * Free a file descriptor.
297  *
298  * Avoid some work if fdp is about to be destroyed.
299  */
300 static inline void
301 fdefree_last(struct filedescent *fde)
302 {
303 
304 	filecaps_free(&fde->fde_caps);
305 }
306 
307 static inline void
308 fdfree(struct filedesc *fdp, int fd)
309 {
310 	struct filedescent *fde;
311 
312 	FILEDESC_XLOCK_ASSERT(fdp);
313 	fde = &fdp->fd_ofiles[fd];
314 #ifdef CAPABILITIES
315 	seqc_write_begin(&fde->fde_seqc);
316 #endif
317 	fde->fde_file = NULL;
318 #ifdef CAPABILITIES
319 	seqc_write_end(&fde->fde_seqc);
320 #endif
321 	fdefree_last(fde);
322 	fdunused(fdp, fd);
323 }
324 
325 /*
326  * System calls on descriptors.
327  */
328 #ifndef _SYS_SYSPROTO_H_
329 struct getdtablesize_args {
330 	int	dummy;
331 };
332 #endif
333 /* ARGSUSED */
334 int
335 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
336 {
337 #ifdef	RACCT
338 	uint64_t lim;
339 #endif
340 
341 	td->td_retval[0] = getmaxfd(td);
342 #ifdef	RACCT
343 	PROC_LOCK(td->td_proc);
344 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
345 	PROC_UNLOCK(td->td_proc);
346 	if (lim < td->td_retval[0])
347 		td->td_retval[0] = lim;
348 #endif
349 	return (0);
350 }
351 
352 /*
353  * Duplicate a file descriptor to a particular value.
354  *
355  * Note: keep in mind that a potential race condition exists when closing
356  * descriptors from a shared descriptor table (via rfork).
357  */
358 #ifndef _SYS_SYSPROTO_H_
359 struct dup2_args {
360 	u_int	from;
361 	u_int	to;
362 };
363 #endif
364 /* ARGSUSED */
365 int
366 sys_dup2(struct thread *td, struct dup2_args *uap)
367 {
368 
369 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
370 }
371 
372 /*
373  * Duplicate a file descriptor.
374  */
375 #ifndef _SYS_SYSPROTO_H_
376 struct dup_args {
377 	u_int	fd;
378 };
379 #endif
380 /* ARGSUSED */
381 int
382 sys_dup(struct thread *td, struct dup_args *uap)
383 {
384 
385 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
386 }
387 
388 /*
389  * The file control system call.
390  */
391 #ifndef _SYS_SYSPROTO_H_
392 struct fcntl_args {
393 	int	fd;
394 	int	cmd;
395 	long	arg;
396 };
397 #endif
398 /* ARGSUSED */
399 int
400 sys_fcntl(struct thread *td, struct fcntl_args *uap)
401 {
402 
403 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
404 }
405 
406 int
407 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
408 {
409 	struct flock fl;
410 	struct __oflock ofl;
411 	intptr_t arg1;
412 	int error, newcmd;
413 
414 	error = 0;
415 	newcmd = cmd;
416 	switch (cmd) {
417 	case F_OGETLK:
418 	case F_OSETLK:
419 	case F_OSETLKW:
420 		/*
421 		 * Convert old flock structure to new.
422 		 */
423 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
424 		fl.l_start = ofl.l_start;
425 		fl.l_len = ofl.l_len;
426 		fl.l_pid = ofl.l_pid;
427 		fl.l_type = ofl.l_type;
428 		fl.l_whence = ofl.l_whence;
429 		fl.l_sysid = 0;
430 
431 		switch (cmd) {
432 		case F_OGETLK:
433 			newcmd = F_GETLK;
434 			break;
435 		case F_OSETLK:
436 			newcmd = F_SETLK;
437 			break;
438 		case F_OSETLKW:
439 			newcmd = F_SETLKW;
440 			break;
441 		}
442 		arg1 = (intptr_t)&fl;
443 		break;
444 	case F_GETLK:
445 	case F_SETLK:
446 	case F_SETLKW:
447 	case F_SETLK_REMOTE:
448 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
449 		arg1 = (intptr_t)&fl;
450 		break;
451 	default:
452 		arg1 = arg;
453 		break;
454 	}
455 	if (error)
456 		return (error);
457 	error = kern_fcntl(td, fd, newcmd, arg1);
458 	if (error)
459 		return (error);
460 	if (cmd == F_OGETLK) {
461 		ofl.l_start = fl.l_start;
462 		ofl.l_len = fl.l_len;
463 		ofl.l_pid = fl.l_pid;
464 		ofl.l_type = fl.l_type;
465 		ofl.l_whence = fl.l_whence;
466 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
467 	} else if (cmd == F_GETLK) {
468 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
469 	}
470 	return (error);
471 }
472 
473 int
474 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
475 {
476 	struct filedesc *fdp;
477 	struct flock *flp;
478 	struct file *fp, *fp2;
479 	struct filedescent *fde;
480 	struct proc *p;
481 	struct vnode *vp;
482 	struct mount *mp;
483 	struct kinfo_file *kif;
484 	int error, flg, kif_sz, seals, tmp;
485 	uint64_t bsize;
486 	off_t foffset;
487 
488 	error = 0;
489 	flg = F_POSIX;
490 	p = td->td_proc;
491 	fdp = p->p_fd;
492 
493 	AUDIT_ARG_FD(cmd);
494 	AUDIT_ARG_CMD(cmd);
495 	switch (cmd) {
496 	case F_DUPFD:
497 		tmp = arg;
498 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
499 		break;
500 
501 	case F_DUPFD_CLOEXEC:
502 		tmp = arg;
503 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
504 		break;
505 
506 	case F_DUP2FD:
507 		tmp = arg;
508 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
509 		break;
510 
511 	case F_DUP2FD_CLOEXEC:
512 		tmp = arg;
513 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
514 		break;
515 
516 	case F_GETFD:
517 		error = EBADF;
518 		FILEDESC_SLOCK(fdp);
519 		fde = fdeget_locked(fdp, fd);
520 		if (fde != NULL) {
521 			td->td_retval[0] =
522 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
523 			error = 0;
524 		}
525 		FILEDESC_SUNLOCK(fdp);
526 		break;
527 
528 	case F_SETFD:
529 		error = EBADF;
530 		FILEDESC_XLOCK(fdp);
531 		fde = fdeget_locked(fdp, fd);
532 		if (fde != NULL) {
533 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
534 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
535 			error = 0;
536 		}
537 		FILEDESC_XUNLOCK(fdp);
538 		break;
539 
540 	case F_GETFL:
541 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
542 		if (error != 0)
543 			break;
544 		td->td_retval[0] = OFLAGS(fp->f_flag);
545 		fdrop(fp, td);
546 		break;
547 
548 	case F_SETFL:
549 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
550 		if (error != 0)
551 			break;
552 		if (fp->f_ops == &path_fileops) {
553 			fdrop(fp, td);
554 			error = EBADF;
555 			break;
556 		}
557 		do {
558 			tmp = flg = fp->f_flag;
559 			tmp &= ~FCNTLFLAGS;
560 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
561 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
562 		tmp = fp->f_flag & FNONBLOCK;
563 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
564 		if (error != 0) {
565 			fdrop(fp, td);
566 			break;
567 		}
568 		tmp = fp->f_flag & FASYNC;
569 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
570 		if (error == 0) {
571 			fdrop(fp, td);
572 			break;
573 		}
574 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
575 		tmp = 0;
576 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
577 		fdrop(fp, td);
578 		break;
579 
580 	case F_GETOWN:
581 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
582 		if (error != 0)
583 			break;
584 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
585 		if (error == 0)
586 			td->td_retval[0] = tmp;
587 		fdrop(fp, td);
588 		break;
589 
590 	case F_SETOWN:
591 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
592 		if (error != 0)
593 			break;
594 		tmp = arg;
595 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
596 		fdrop(fp, td);
597 		break;
598 
599 	case F_SETLK_REMOTE:
600 		error = priv_check(td, PRIV_NFS_LOCKD);
601 		if (error != 0)
602 			return (error);
603 		flg = F_REMOTE;
604 		goto do_setlk;
605 
606 	case F_SETLKW:
607 		flg |= F_WAIT;
608 		/* FALLTHROUGH F_SETLK */
609 
610 	case F_SETLK:
611 	do_setlk:
612 		flp = (struct flock *)arg;
613 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
614 			error = EINVAL;
615 			break;
616 		}
617 
618 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
619 		if (error != 0)
620 			break;
621 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
622 			error = EBADF;
623 			fdrop(fp, td);
624 			break;
625 		}
626 
627 		if (flp->l_whence == SEEK_CUR) {
628 			foffset = foffset_get(fp);
629 			if (foffset < 0 ||
630 			    (flp->l_start > 0 &&
631 			     foffset > OFF_MAX - flp->l_start)) {
632 				error = EOVERFLOW;
633 				fdrop(fp, td);
634 				break;
635 			}
636 			flp->l_start += foffset;
637 		}
638 
639 		vp = fp->f_vnode;
640 		switch (flp->l_type) {
641 		case F_RDLCK:
642 			if ((fp->f_flag & FREAD) == 0) {
643 				error = EBADF;
644 				break;
645 			}
646 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
647 				PROC_LOCK(p->p_leader);
648 				p->p_leader->p_flag |= P_ADVLOCK;
649 				PROC_UNLOCK(p->p_leader);
650 			}
651 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
652 			    flp, flg);
653 			break;
654 		case F_WRLCK:
655 			if ((fp->f_flag & FWRITE) == 0) {
656 				error = EBADF;
657 				break;
658 			}
659 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
660 				PROC_LOCK(p->p_leader);
661 				p->p_leader->p_flag |= P_ADVLOCK;
662 				PROC_UNLOCK(p->p_leader);
663 			}
664 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
665 			    flp, flg);
666 			break;
667 		case F_UNLCK:
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
669 			    flp, flg);
670 			break;
671 		case F_UNLCKSYS:
672 			if (flg != F_REMOTE) {
673 				error = EINVAL;
674 				break;
675 			}
676 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
677 			    F_UNLCKSYS, flp, flg);
678 			break;
679 		default:
680 			error = EINVAL;
681 			break;
682 		}
683 		if (error != 0 || flp->l_type == F_UNLCK ||
684 		    flp->l_type == F_UNLCKSYS) {
685 			fdrop(fp, td);
686 			break;
687 		}
688 
689 		/*
690 		 * Check for a race with close.
691 		 *
692 		 * The vnode is now advisory locked (or unlocked, but this case
693 		 * is not really important) as the caller requested.
694 		 * We had to drop the filedesc lock, so we need to recheck if
695 		 * the descriptor is still valid, because if it was closed
696 		 * in the meantime we need to remove advisory lock from the
697 		 * vnode - close on any descriptor leading to an advisory
698 		 * locked vnode, removes that lock.
699 		 * We will return 0 on purpose in that case, as the result of
700 		 * successful advisory lock might have been externally visible
701 		 * already. This is fine - effectively we pretend to the caller
702 		 * that the closing thread was a bit slower and that the
703 		 * advisory lock succeeded before the close.
704 		 */
705 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
706 		if (error != 0) {
707 			fdrop(fp, td);
708 			break;
709 		}
710 		if (fp != fp2) {
711 			flp->l_whence = SEEK_SET;
712 			flp->l_start = 0;
713 			flp->l_len = 0;
714 			flp->l_type = F_UNLCK;
715 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
716 			    F_UNLCK, flp, F_POSIX);
717 		}
718 		fdrop(fp, td);
719 		fdrop(fp2, td);
720 		break;
721 
722 	case F_GETLK:
723 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
724 		if (error != 0)
725 			break;
726 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
727 			error = EBADF;
728 			fdrop(fp, td);
729 			break;
730 		}
731 		flp = (struct flock *)arg;
732 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
733 		    flp->l_type != F_UNLCK) {
734 			error = EINVAL;
735 			fdrop(fp, td);
736 			break;
737 		}
738 		if (flp->l_whence == SEEK_CUR) {
739 			foffset = foffset_get(fp);
740 			if ((flp->l_start > 0 &&
741 			    foffset > OFF_MAX - flp->l_start) ||
742 			    (flp->l_start < 0 &&
743 			    foffset < OFF_MIN - flp->l_start)) {
744 				error = EOVERFLOW;
745 				fdrop(fp, td);
746 				break;
747 			}
748 			flp->l_start += foffset;
749 		}
750 		vp = fp->f_vnode;
751 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
752 		    F_POSIX);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_ADD_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		error = fo_add_seals(fp, arg);
761 		fdrop(fp, td);
762 		break;
763 
764 	case F_GET_SEALS:
765 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
766 		if (error != 0)
767 			break;
768 		if (fo_get_seals(fp, &seals) == 0)
769 			td->td_retval[0] = seals;
770 		else
771 			error = EINVAL;
772 		fdrop(fp, td);
773 		break;
774 
775 	case F_RDAHEAD:
776 		arg = arg ? 128 * 1024: 0;
777 		/* FALLTHROUGH */
778 	case F_READAHEAD:
779 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
780 		if (error != 0)
781 			break;
782 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
783 			fdrop(fp, td);
784 			error = EBADF;
785 			break;
786 		}
787 		vp = fp->f_vnode;
788 		if (vp->v_type != VREG) {
789 			fdrop(fp, td);
790 			error = ENOTTY;
791 			break;
792 		}
793 
794 		/*
795 		 * Exclusive lock synchronizes against f_seqcount reads and
796 		 * writes in sequential_heuristic().
797 		 */
798 		error = vn_lock(vp, LK_EXCLUSIVE);
799 		if (error != 0) {
800 			fdrop(fp, td);
801 			break;
802 		}
803 		if (arg >= 0) {
804 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
805 			arg = MIN(arg, INT_MAX - bsize + 1);
806 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
807 			    (arg + bsize - 1) / bsize);
808 			atomic_set_int(&fp->f_flag, FRDAHEAD);
809 		} else {
810 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
811 		}
812 		VOP_UNLOCK(vp);
813 		fdrop(fp, td);
814 		break;
815 
816 	case F_ISUNIONSTACK:
817 		/*
818 		 * Check if the vnode is part of a union stack (either the
819 		 * "union" flag from mount(2) or unionfs).
820 		 *
821 		 * Prior to introduction of this op libc's readdir would call
822 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
823 		 * data just to check fs name and a mount flag.
824 		 *
825 		 * Fixing the code to handle everything in the kernel instead
826 		 * is a non-trivial endeavor and has low priority, thus this
827 		 * horrible kludge facilitates the current behavior in a much
828 		 * cheaper manner until someone(tm) sorts this out.
829 		 */
830 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
831 		if (error != 0)
832 			break;
833 		if (fp->f_type != DTYPE_VNODE) {
834 			fdrop(fp, td);
835 			error = EBADF;
836 			break;
837 		}
838 		vp = fp->f_vnode;
839 		/*
840 		 * Since we don't prevent dooming the vnode even non-null mp
841 		 * found can become immediately stale. This is tolerable since
842 		 * mount points are type-stable (providing safe memory access)
843 		 * and any vfs op on this vnode going forward will return an
844 		 * error (meaning return value in this case is meaningless).
845 		 */
846 		mp = atomic_load_ptr(&vp->v_mount);
847 		if (__predict_false(mp == NULL)) {
848 			fdrop(fp, td);
849 			error = EBADF;
850 			break;
851 		}
852 		td->td_retval[0] = 0;
853 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
854 		    mp->mnt_flag & MNT_UNION)
855 			td->td_retval[0] = 1;
856 		fdrop(fp, td);
857 		break;
858 
859 	case F_KINFO:
860 #ifdef CAPABILITY_MODE
861 		if (IN_CAPABILITY_MODE(td)) {
862 			error = ECAPMODE;
863 			break;
864 		}
865 #endif
866 		error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
867 		if (error != 0)
868 			break;
869 		if (kif_sz != sizeof(*kif)) {
870 			error = EINVAL;
871 			break;
872 		}
873 		kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
874 		FILEDESC_SLOCK(fdp);
875 		error = fget_cap_locked(fdp, fd, &cap_fcntl_rights, &fp, NULL);
876 		if (error == 0 && fhold(fp)) {
877 			export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
878 			FILEDESC_SUNLOCK(fdp);
879 			fdrop(fp, td);
880 			if ((kif->kf_status & KF_ATTR_VALID) != 0) {
881 				kif->kf_structsize = sizeof(*kif);
882 				error = copyout(kif, (void *)arg, sizeof(*kif));
883 			} else {
884 				error = EBADF;
885 			}
886 		} else {
887 			FILEDESC_SUNLOCK(fdp);
888 			if (error == 0)
889 				error = EBADF;
890 		}
891 		free(kif, M_TEMP);
892 		break;
893 
894 	default:
895 		error = EINVAL;
896 		break;
897 	}
898 	return (error);
899 }
900 
901 static int
902 getmaxfd(struct thread *td)
903 {
904 
905 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
906 }
907 
908 /*
909  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
910  */
911 int
912 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
913 {
914 	struct filedesc *fdp;
915 	struct filedescent *oldfde, *newfde;
916 	struct proc *p;
917 	struct file *delfp, *oldfp;
918 	u_long *oioctls, *nioctls;
919 	int error, maxfd;
920 
921 	p = td->td_proc;
922 	fdp = p->p_fd;
923 	oioctls = NULL;
924 
925 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
926 	MPASS(mode < FDDUP_LASTMODE);
927 
928 	AUDIT_ARG_FD(old);
929 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
930 
931 	/*
932 	 * Verify we have a valid descriptor to dup from and possibly to
933 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
934 	 * return EINVAL when the new descriptor is out of bounds.
935 	 */
936 	if (old < 0)
937 		return (EBADF);
938 	if (new < 0)
939 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
940 	maxfd = getmaxfd(td);
941 	if (new >= maxfd)
942 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
943 
944 	error = EBADF;
945 	FILEDESC_XLOCK(fdp);
946 	if (fget_locked(fdp, old) == NULL)
947 		goto unlock;
948 	if (mode == FDDUP_FIXED && old == new) {
949 		td->td_retval[0] = new;
950 		if (flags & FDDUP_FLAG_CLOEXEC)
951 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
952 		error = 0;
953 		goto unlock;
954 	}
955 
956 	oldfde = &fdp->fd_ofiles[old];
957 	oldfp = oldfde->fde_file;
958 	if (!fhold(oldfp))
959 		goto unlock;
960 
961 	/*
962 	 * If the caller specified a file descriptor, make sure the file
963 	 * table is large enough to hold it, and grab it.  Otherwise, just
964 	 * allocate a new descriptor the usual way.
965 	 */
966 	switch (mode) {
967 	case FDDUP_NORMAL:
968 	case FDDUP_FCNTL:
969 		if ((error = fdalloc(td, new, &new)) != 0) {
970 			fdrop(oldfp, td);
971 			goto unlock;
972 		}
973 		break;
974 	case FDDUP_FIXED:
975 		if (new >= fdp->fd_nfiles) {
976 			/*
977 			 * The resource limits are here instead of e.g.
978 			 * fdalloc(), because the file descriptor table may be
979 			 * shared between processes, so we can't really use
980 			 * racct_add()/racct_sub().  Instead of counting the
981 			 * number of actually allocated descriptors, just put
982 			 * the limit on the size of the file descriptor table.
983 			 */
984 #ifdef RACCT
985 			if (RACCT_ENABLED()) {
986 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
987 				if (error != 0) {
988 					error = EMFILE;
989 					fdrop(oldfp, td);
990 					goto unlock;
991 				}
992 			}
993 #endif
994 			fdgrowtable_exp(fdp, new + 1);
995 		}
996 		if (!fdisused(fdp, new))
997 			fdused(fdp, new);
998 		break;
999 	default:
1000 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1001 	}
1002 
1003 	KASSERT(old != new, ("new fd is same as old"));
1004 
1005 	/* Refetch oldfde because the table may have grown and old one freed. */
1006 	oldfde = &fdp->fd_ofiles[old];
1007 	KASSERT(oldfp == oldfde->fde_file,
1008 	    ("fdt_ofiles shift from growth observed at fd %d",
1009 	    old));
1010 
1011 	newfde = &fdp->fd_ofiles[new];
1012 	delfp = newfde->fde_file;
1013 
1014 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1015 
1016 	/*
1017 	 * Duplicate the source descriptor.
1018 	 */
1019 #ifdef CAPABILITIES
1020 	seqc_write_begin(&newfde->fde_seqc);
1021 #endif
1022 	oioctls = filecaps_free_prep(&newfde->fde_caps);
1023 	memcpy(newfde, oldfde, fde_change_size);
1024 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1025 	    nioctls);
1026 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1027 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1028 	else
1029 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1030 #ifdef CAPABILITIES
1031 	seqc_write_end(&newfde->fde_seqc);
1032 #endif
1033 	td->td_retval[0] = new;
1034 
1035 	error = 0;
1036 
1037 	if (delfp != NULL) {
1038 		(void) closefp(fdp, new, delfp, td, true, false);
1039 		FILEDESC_UNLOCK_ASSERT(fdp);
1040 	} else {
1041 unlock:
1042 		FILEDESC_XUNLOCK(fdp);
1043 	}
1044 
1045 	filecaps_free_finish(oioctls);
1046 	return (error);
1047 }
1048 
1049 static void
1050 sigiofree(struct sigio *sigio)
1051 {
1052 	crfree(sigio->sio_ucred);
1053 	free(sigio, M_SIGIO);
1054 }
1055 
1056 static struct sigio *
1057 funsetown_locked(struct sigio *sigio)
1058 {
1059 	struct proc *p;
1060 	struct pgrp *pg;
1061 
1062 	SIGIO_ASSERT_LOCKED();
1063 
1064 	if (sigio == NULL)
1065 		return (NULL);
1066 	*sigio->sio_myref = NULL;
1067 	if (sigio->sio_pgid < 0) {
1068 		pg = sigio->sio_pgrp;
1069 		PGRP_LOCK(pg);
1070 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1071 		PGRP_UNLOCK(pg);
1072 	} else {
1073 		p = sigio->sio_proc;
1074 		PROC_LOCK(p);
1075 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1076 		PROC_UNLOCK(p);
1077 	}
1078 	return (sigio);
1079 }
1080 
1081 /*
1082  * If sigio is on the list associated with a process or process group,
1083  * disable signalling from the device, remove sigio from the list and
1084  * free sigio.
1085  */
1086 void
1087 funsetown(struct sigio **sigiop)
1088 {
1089 	struct sigio *sigio;
1090 
1091 	/* Racy check, consumers must provide synchronization. */
1092 	if (*sigiop == NULL)
1093 		return;
1094 
1095 	SIGIO_LOCK();
1096 	sigio = funsetown_locked(*sigiop);
1097 	SIGIO_UNLOCK();
1098 	if (sigio != NULL)
1099 		sigiofree(sigio);
1100 }
1101 
1102 /*
1103  * Free a list of sigio structures.  The caller must ensure that new sigio
1104  * structures cannot be added after this point.  For process groups this is
1105  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1106  * as an interlock.
1107  */
1108 void
1109 funsetownlst(struct sigiolst *sigiolst)
1110 {
1111 	struct proc *p;
1112 	struct pgrp *pg;
1113 	struct sigio *sigio, *tmp;
1114 
1115 	/* Racy check. */
1116 	sigio = SLIST_FIRST(sigiolst);
1117 	if (sigio == NULL)
1118 		return;
1119 
1120 	p = NULL;
1121 	pg = NULL;
1122 
1123 	SIGIO_LOCK();
1124 	sigio = SLIST_FIRST(sigiolst);
1125 	if (sigio == NULL) {
1126 		SIGIO_UNLOCK();
1127 		return;
1128 	}
1129 
1130 	/*
1131 	 * Every entry of the list should belong to a single proc or pgrp.
1132 	 */
1133 	if (sigio->sio_pgid < 0) {
1134 		pg = sigio->sio_pgrp;
1135 		sx_assert(&proctree_lock, SX_XLOCKED);
1136 		PGRP_LOCK(pg);
1137 	} else /* if (sigio->sio_pgid > 0) */ {
1138 		p = sigio->sio_proc;
1139 		PROC_LOCK(p);
1140 		KASSERT((p->p_flag & P_WEXIT) != 0,
1141 		    ("%s: process %p is not exiting", __func__, p));
1142 	}
1143 
1144 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1145 		*sigio->sio_myref = NULL;
1146 		if (pg != NULL) {
1147 			KASSERT(sigio->sio_pgid < 0,
1148 			    ("Proc sigio in pgrp sigio list"));
1149 			KASSERT(sigio->sio_pgrp == pg,
1150 			    ("Bogus pgrp in sigio list"));
1151 		} else /* if (p != NULL) */ {
1152 			KASSERT(sigio->sio_pgid > 0,
1153 			    ("Pgrp sigio in proc sigio list"));
1154 			KASSERT(sigio->sio_proc == p,
1155 			    ("Bogus proc in sigio list"));
1156 		}
1157 	}
1158 
1159 	if (pg != NULL)
1160 		PGRP_UNLOCK(pg);
1161 	else
1162 		PROC_UNLOCK(p);
1163 	SIGIO_UNLOCK();
1164 
1165 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1166 		sigiofree(sigio);
1167 }
1168 
1169 /*
1170  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1171  *
1172  * After permission checking, add a sigio structure to the sigio list for
1173  * the process or process group.
1174  */
1175 int
1176 fsetown(pid_t pgid, struct sigio **sigiop)
1177 {
1178 	struct proc *proc;
1179 	struct pgrp *pgrp;
1180 	struct sigio *osigio, *sigio;
1181 	int ret;
1182 
1183 	if (pgid == 0) {
1184 		funsetown(sigiop);
1185 		return (0);
1186 	}
1187 
1188 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1189 	sigio->sio_pgid = pgid;
1190 	sigio->sio_ucred = crhold(curthread->td_ucred);
1191 	sigio->sio_myref = sigiop;
1192 
1193 	ret = 0;
1194 	if (pgid > 0) {
1195 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1196 		SIGIO_LOCK();
1197 		osigio = funsetown_locked(*sigiop);
1198 		if (ret == 0) {
1199 			PROC_LOCK(proc);
1200 			_PRELE(proc);
1201 			if ((proc->p_flag & P_WEXIT) != 0) {
1202 				ret = ESRCH;
1203 			} else if (proc->p_session !=
1204 			    curthread->td_proc->p_session) {
1205 				/*
1206 				 * Policy - Don't allow a process to FSETOWN a
1207 				 * process in another session.
1208 				 *
1209 				 * Remove this test to allow maximum flexibility
1210 				 * or restrict FSETOWN to the current process or
1211 				 * process group for maximum safety.
1212 				 */
1213 				ret = EPERM;
1214 			} else {
1215 				sigio->sio_proc = proc;
1216 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1217 				    sio_pgsigio);
1218 			}
1219 			PROC_UNLOCK(proc);
1220 		}
1221 	} else /* if (pgid < 0) */ {
1222 		sx_slock(&proctree_lock);
1223 		SIGIO_LOCK();
1224 		osigio = funsetown_locked(*sigiop);
1225 		pgrp = pgfind(-pgid);
1226 		if (pgrp == NULL) {
1227 			ret = ESRCH;
1228 		} else {
1229 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1230 				/*
1231 				 * Policy - Don't allow a process to FSETOWN a
1232 				 * process in another session.
1233 				 *
1234 				 * Remove this test to allow maximum flexibility
1235 				 * or restrict FSETOWN to the current process or
1236 				 * process group for maximum safety.
1237 				 */
1238 				ret = EPERM;
1239 			} else {
1240 				sigio->sio_pgrp = pgrp;
1241 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1242 				    sio_pgsigio);
1243 			}
1244 			PGRP_UNLOCK(pgrp);
1245 		}
1246 		sx_sunlock(&proctree_lock);
1247 	}
1248 	if (ret == 0)
1249 		*sigiop = sigio;
1250 	SIGIO_UNLOCK();
1251 	if (osigio != NULL)
1252 		sigiofree(osigio);
1253 	return (ret);
1254 }
1255 
1256 /*
1257  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1258  */
1259 pid_t
1260 fgetown(struct sigio **sigiop)
1261 {
1262 	pid_t pgid;
1263 
1264 	SIGIO_LOCK();
1265 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1266 	SIGIO_UNLOCK();
1267 	return (pgid);
1268 }
1269 
1270 static int
1271 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1272     bool audit)
1273 {
1274 	int error;
1275 
1276 	FILEDESC_XLOCK_ASSERT(fdp);
1277 
1278 	/*
1279 	 * We now hold the fp reference that used to be owned by the
1280 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1281 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1282 	 * added, and deleteing a knote for the new fd.
1283 	 */
1284 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1285 		knote_fdclose(td, fd);
1286 
1287 	/*
1288 	 * We need to notify mqueue if the object is of type mqueue.
1289 	 */
1290 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1291 		mq_fdclose(td, fd, fp);
1292 	FILEDESC_XUNLOCK(fdp);
1293 
1294 #ifdef AUDIT
1295 	if (AUDITING_TD(td) && audit)
1296 		audit_sysclose(td, fd, fp);
1297 #endif
1298 	error = closef(fp, td);
1299 
1300 	/*
1301 	 * All paths leading up to closefp() will have already removed or
1302 	 * replaced the fd in the filedesc table, so a restart would not
1303 	 * operate on the same file.
1304 	 */
1305 	if (error == ERESTART)
1306 		error = EINTR;
1307 
1308 	return (error);
1309 }
1310 
1311 static int
1312 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1313     bool holdleaders, bool audit)
1314 {
1315 	int error;
1316 
1317 	FILEDESC_XLOCK_ASSERT(fdp);
1318 
1319 	if (holdleaders) {
1320 		if (td->td_proc->p_fdtol != NULL) {
1321 			/*
1322 			 * Ask fdfree() to sleep to ensure that all relevant
1323 			 * process leaders can be traversed in closef().
1324 			 */
1325 			fdp->fd_holdleaderscount++;
1326 		} else {
1327 			holdleaders = false;
1328 		}
1329 	}
1330 
1331 	error = closefp_impl(fdp, fd, fp, td, audit);
1332 	if (holdleaders) {
1333 		FILEDESC_XLOCK(fdp);
1334 		fdp->fd_holdleaderscount--;
1335 		if (fdp->fd_holdleaderscount == 0 &&
1336 		    fdp->fd_holdleaderswakeup != 0) {
1337 			fdp->fd_holdleaderswakeup = 0;
1338 			wakeup(&fdp->fd_holdleaderscount);
1339 		}
1340 		FILEDESC_XUNLOCK(fdp);
1341 	}
1342 	return (error);
1343 }
1344 
1345 static int
1346 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1347     bool holdleaders, bool audit)
1348 {
1349 
1350 	FILEDESC_XLOCK_ASSERT(fdp);
1351 
1352 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1353 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1354 	} else {
1355 		return (closefp_impl(fdp, fd, fp, td, audit));
1356 	}
1357 }
1358 
1359 /*
1360  * Close a file descriptor.
1361  */
1362 #ifndef _SYS_SYSPROTO_H_
1363 struct close_args {
1364 	int     fd;
1365 };
1366 #endif
1367 /* ARGSUSED */
1368 int
1369 sys_close(struct thread *td, struct close_args *uap)
1370 {
1371 
1372 	return (kern_close(td, uap->fd));
1373 }
1374 
1375 int
1376 kern_close(struct thread *td, int fd)
1377 {
1378 	struct filedesc *fdp;
1379 	struct file *fp;
1380 
1381 	fdp = td->td_proc->p_fd;
1382 
1383 	FILEDESC_XLOCK(fdp);
1384 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1385 		FILEDESC_XUNLOCK(fdp);
1386 		return (EBADF);
1387 	}
1388 	fdfree(fdp, fd);
1389 
1390 	/* closefp() drops the FILEDESC lock for us. */
1391 	return (closefp(fdp, fd, fp, td, true, true));
1392 }
1393 
1394 int
1395 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1396 {
1397 	struct filedesc *fdp;
1398 	const struct fdescenttbl *fdt;
1399 	struct file *fp;
1400 	int fd;
1401 
1402 	/*
1403 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1404 	 * open should not be a usage error.  From a close_range() perspective,
1405 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1406 	 * be a usage error as all fd above 3 are in-fact already closed.
1407 	 */
1408 	if (highfd < lowfd) {
1409 		return (EINVAL);
1410 	}
1411 
1412 	fdp = td->td_proc->p_fd;
1413 	FILEDESC_XLOCK(fdp);
1414 	fdt = atomic_load_ptr(&fdp->fd_files);
1415 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1416 	fd = lowfd;
1417 	if (__predict_false(fd > highfd)) {
1418 		goto out_locked;
1419 	}
1420 	for (;;) {
1421 		fp = fdt->fdt_ofiles[fd].fde_file;
1422 		if (fp == NULL) {
1423 			if (fd == highfd)
1424 				goto out_locked;
1425 		} else {
1426 			fdfree(fdp, fd);
1427 			(void) closefp(fdp, fd, fp, td, true, true);
1428 			if (fd == highfd)
1429 				goto out_unlocked;
1430 			FILEDESC_XLOCK(fdp);
1431 			fdt = atomic_load_ptr(&fdp->fd_files);
1432 		}
1433 		fd++;
1434 	}
1435 out_locked:
1436 	FILEDESC_XUNLOCK(fdp);
1437 out_unlocked:
1438 	return (0);
1439 }
1440 
1441 #ifndef _SYS_SYSPROTO_H_
1442 struct close_range_args {
1443 	u_int	lowfd;
1444 	u_int	highfd;
1445 	int	flags;
1446 };
1447 #endif
1448 int
1449 sys_close_range(struct thread *td, struct close_range_args *uap)
1450 {
1451 
1452 	AUDIT_ARG_FD(uap->lowfd);
1453 	AUDIT_ARG_CMD(uap->highfd);
1454 	AUDIT_ARG_FFLAGS(uap->flags);
1455 
1456 	/* No flags currently defined */
1457 	if (uap->flags != 0)
1458 		return (EINVAL);
1459 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1460 }
1461 
1462 #ifdef COMPAT_FREEBSD12
1463 /*
1464  * Close open file descriptors.
1465  */
1466 #ifndef _SYS_SYSPROTO_H_
1467 struct freebsd12_closefrom_args {
1468 	int	lowfd;
1469 };
1470 #endif
1471 /* ARGSUSED */
1472 int
1473 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1474 {
1475 	u_int lowfd;
1476 
1477 	AUDIT_ARG_FD(uap->lowfd);
1478 
1479 	/*
1480 	 * Treat negative starting file descriptor values identical to
1481 	 * closefrom(0) which closes all files.
1482 	 */
1483 	lowfd = MAX(0, uap->lowfd);
1484 	return (kern_close_range(td, lowfd, ~0U));
1485 }
1486 #endif	/* COMPAT_FREEBSD12 */
1487 
1488 #if defined(COMPAT_43)
1489 /*
1490  * Return status information about a file descriptor.
1491  */
1492 #ifndef _SYS_SYSPROTO_H_
1493 struct ofstat_args {
1494 	int	fd;
1495 	struct	ostat *sb;
1496 };
1497 #endif
1498 /* ARGSUSED */
1499 int
1500 ofstat(struct thread *td, struct ofstat_args *uap)
1501 {
1502 	struct ostat oub;
1503 	struct stat ub;
1504 	int error;
1505 
1506 	error = kern_fstat(td, uap->fd, &ub);
1507 	if (error == 0) {
1508 		cvtstat(&ub, &oub);
1509 		error = copyout(&oub, uap->sb, sizeof(oub));
1510 	}
1511 	return (error);
1512 }
1513 #endif /* COMPAT_43 */
1514 
1515 #if defined(COMPAT_FREEBSD11)
1516 int
1517 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1518 {
1519 	struct stat sb;
1520 	struct freebsd11_stat osb;
1521 	int error;
1522 
1523 	error = kern_fstat(td, uap->fd, &sb);
1524 	if (error != 0)
1525 		return (error);
1526 	error = freebsd11_cvtstat(&sb, &osb);
1527 	if (error == 0)
1528 		error = copyout(&osb, uap->sb, sizeof(osb));
1529 	return (error);
1530 }
1531 #endif	/* COMPAT_FREEBSD11 */
1532 
1533 /*
1534  * Return status information about a file descriptor.
1535  */
1536 #ifndef _SYS_SYSPROTO_H_
1537 struct fstat_args {
1538 	int	fd;
1539 	struct	stat *sb;
1540 };
1541 #endif
1542 /* ARGSUSED */
1543 int
1544 sys_fstat(struct thread *td, struct fstat_args *uap)
1545 {
1546 	struct stat ub;
1547 	int error;
1548 
1549 	error = kern_fstat(td, uap->fd, &ub);
1550 	if (error == 0)
1551 		error = copyout(&ub, uap->sb, sizeof(ub));
1552 	return (error);
1553 }
1554 
1555 int
1556 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1557 {
1558 	struct file *fp;
1559 	int error;
1560 
1561 	AUDIT_ARG_FD(fd);
1562 
1563 	error = fget(td, fd, &cap_fstat_rights, &fp);
1564 	if (__predict_false(error != 0))
1565 		return (error);
1566 
1567 	AUDIT_ARG_FILE(td->td_proc, fp);
1568 
1569 	error = fo_stat(fp, sbp, td->td_ucred);
1570 	fdrop(fp, td);
1571 #ifdef __STAT_TIME_T_EXT
1572 	sbp->st_atim_ext = 0;
1573 	sbp->st_mtim_ext = 0;
1574 	sbp->st_ctim_ext = 0;
1575 	sbp->st_btim_ext = 0;
1576 #endif
1577 #ifdef KTRACE
1578 	if (KTRPOINT(td, KTR_STRUCT))
1579 		ktrstat_error(sbp, error);
1580 #endif
1581 	return (error);
1582 }
1583 
1584 #if defined(COMPAT_FREEBSD11)
1585 /*
1586  * Return status information about a file descriptor.
1587  */
1588 #ifndef _SYS_SYSPROTO_H_
1589 struct freebsd11_nfstat_args {
1590 	int	fd;
1591 	struct	nstat *sb;
1592 };
1593 #endif
1594 /* ARGSUSED */
1595 int
1596 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1597 {
1598 	struct nstat nub;
1599 	struct stat ub;
1600 	int error;
1601 
1602 	error = kern_fstat(td, uap->fd, &ub);
1603 	if (error != 0)
1604 		return (error);
1605 	error = freebsd11_cvtnstat(&ub, &nub);
1606 	if (error != 0)
1607 		error = copyout(&nub, uap->sb, sizeof(nub));
1608 	return (error);
1609 }
1610 #endif /* COMPAT_FREEBSD11 */
1611 
1612 /*
1613  * Return pathconf information about a file descriptor.
1614  */
1615 #ifndef _SYS_SYSPROTO_H_
1616 struct fpathconf_args {
1617 	int	fd;
1618 	int	name;
1619 };
1620 #endif
1621 /* ARGSUSED */
1622 int
1623 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1624 {
1625 	long value;
1626 	int error;
1627 
1628 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1629 	if (error == 0)
1630 		td->td_retval[0] = value;
1631 	return (error);
1632 }
1633 
1634 int
1635 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1636 {
1637 	struct file *fp;
1638 	struct vnode *vp;
1639 	int error;
1640 
1641 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1642 	if (error != 0)
1643 		return (error);
1644 
1645 	if (name == _PC_ASYNC_IO) {
1646 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1647 		goto out;
1648 	}
1649 	vp = fp->f_vnode;
1650 	if (vp != NULL) {
1651 		vn_lock(vp, LK_SHARED | LK_RETRY);
1652 		error = VOP_PATHCONF(vp, name, valuep);
1653 		VOP_UNLOCK(vp);
1654 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1655 		if (name != _PC_PIPE_BUF) {
1656 			error = EINVAL;
1657 		} else {
1658 			*valuep = PIPE_BUF;
1659 			error = 0;
1660 		}
1661 	} else {
1662 		error = EOPNOTSUPP;
1663 	}
1664 out:
1665 	fdrop(fp, td);
1666 	return (error);
1667 }
1668 
1669 /*
1670  * Copy filecaps structure allocating memory for ioctls array if needed.
1671  *
1672  * The last parameter indicates whether the fdtable is locked. If it is not and
1673  * ioctls are encountered, copying fails and the caller must lock the table.
1674  *
1675  * Note that if the table was not locked, the caller has to check the relevant
1676  * sequence counter to determine whether the operation was successful.
1677  */
1678 bool
1679 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1680 {
1681 	size_t size;
1682 
1683 	if (src->fc_ioctls != NULL && !locked)
1684 		return (false);
1685 	memcpy(dst, src, sizeof(*src));
1686 	if (src->fc_ioctls == NULL)
1687 		return (true);
1688 
1689 	KASSERT(src->fc_nioctls > 0,
1690 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1691 
1692 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1693 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1694 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1695 	return (true);
1696 }
1697 
1698 static u_long *
1699 filecaps_copy_prep(const struct filecaps *src)
1700 {
1701 	u_long *ioctls;
1702 	size_t size;
1703 
1704 	if (__predict_true(src->fc_ioctls == NULL))
1705 		return (NULL);
1706 
1707 	KASSERT(src->fc_nioctls > 0,
1708 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1709 
1710 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1711 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1712 	return (ioctls);
1713 }
1714 
1715 static void
1716 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1717     u_long *ioctls)
1718 {
1719 	size_t size;
1720 
1721 	*dst = *src;
1722 	if (__predict_true(src->fc_ioctls == NULL)) {
1723 		MPASS(ioctls == NULL);
1724 		return;
1725 	}
1726 
1727 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1728 	dst->fc_ioctls = ioctls;
1729 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1730 }
1731 
1732 /*
1733  * Move filecaps structure to the new place and clear the old place.
1734  */
1735 void
1736 filecaps_move(struct filecaps *src, struct filecaps *dst)
1737 {
1738 
1739 	*dst = *src;
1740 	bzero(src, sizeof(*src));
1741 }
1742 
1743 /*
1744  * Fill the given filecaps structure with full rights.
1745  */
1746 static void
1747 filecaps_fill(struct filecaps *fcaps)
1748 {
1749 
1750 	CAP_ALL(&fcaps->fc_rights);
1751 	fcaps->fc_ioctls = NULL;
1752 	fcaps->fc_nioctls = -1;
1753 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1754 }
1755 
1756 /*
1757  * Free memory allocated within filecaps structure.
1758  */
1759 void
1760 filecaps_free(struct filecaps *fcaps)
1761 {
1762 
1763 	free(fcaps->fc_ioctls, M_FILECAPS);
1764 	bzero(fcaps, sizeof(*fcaps));
1765 }
1766 
1767 static u_long *
1768 filecaps_free_prep(struct filecaps *fcaps)
1769 {
1770 	u_long *ioctls;
1771 
1772 	ioctls = fcaps->fc_ioctls;
1773 	bzero(fcaps, sizeof(*fcaps));
1774 	return (ioctls);
1775 }
1776 
1777 static void
1778 filecaps_free_finish(u_long *ioctls)
1779 {
1780 
1781 	free(ioctls, M_FILECAPS);
1782 }
1783 
1784 /*
1785  * Validate the given filecaps structure.
1786  */
1787 static void
1788 filecaps_validate(const struct filecaps *fcaps, const char *func)
1789 {
1790 
1791 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1792 	    ("%s: invalid rights", func));
1793 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1794 	    ("%s: invalid fcntls", func));
1795 	KASSERT(fcaps->fc_fcntls == 0 ||
1796 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1797 	    ("%s: fcntls without CAP_FCNTL", func));
1798 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1799 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1800 	    ("%s: invalid ioctls", func));
1801 	KASSERT(fcaps->fc_nioctls == 0 ||
1802 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1803 	    ("%s: ioctls without CAP_IOCTL", func));
1804 }
1805 
1806 static void
1807 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1808 {
1809 	int nfd1;
1810 
1811 	FILEDESC_XLOCK_ASSERT(fdp);
1812 
1813 	nfd1 = fdp->fd_nfiles * 2;
1814 	if (nfd1 < nfd)
1815 		nfd1 = nfd;
1816 	fdgrowtable(fdp, nfd1);
1817 }
1818 
1819 /*
1820  * Grow the file table to accommodate (at least) nfd descriptors.
1821  */
1822 static void
1823 fdgrowtable(struct filedesc *fdp, int nfd)
1824 {
1825 	struct filedesc0 *fdp0;
1826 	struct freetable *ft;
1827 	struct fdescenttbl *ntable;
1828 	struct fdescenttbl *otable;
1829 	int nnfiles, onfiles;
1830 	NDSLOTTYPE *nmap, *omap;
1831 
1832 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1833 
1834 	/* save old values */
1835 	onfiles = fdp->fd_nfiles;
1836 	otable = fdp->fd_files;
1837 	omap = fdp->fd_map;
1838 
1839 	/* compute the size of the new table */
1840 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1841 	if (nnfiles <= onfiles)
1842 		/* the table is already large enough */
1843 		return;
1844 
1845 	/*
1846 	 * Allocate a new table.  We need enough space for the number of
1847 	 * entries, file entries themselves and the struct freetable we will use
1848 	 * when we decommission the table and place it on the freelist.
1849 	 * We place the struct freetable in the middle so we don't have
1850 	 * to worry about padding.
1851 	 */
1852 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1853 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1854 	    sizeof(struct freetable),
1855 	    M_FILEDESC, M_ZERO | M_WAITOK);
1856 	/* copy the old data */
1857 	ntable->fdt_nfiles = nnfiles;
1858 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1859 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1860 
1861 	/*
1862 	 * Allocate a new map only if the old is not large enough.  It will
1863 	 * grow at a slower rate than the table as it can map more
1864 	 * entries than the table can hold.
1865 	 */
1866 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1867 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1868 		    M_ZERO | M_WAITOK);
1869 		/* copy over the old data and update the pointer */
1870 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1871 		fdp->fd_map = nmap;
1872 	}
1873 
1874 	/*
1875 	 * Make sure that ntable is correctly initialized before we replace
1876 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1877 	 * data.
1878 	 */
1879 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1880 
1881 	/*
1882 	 * Free the old file table when not shared by other threads or processes.
1883 	 * The old file table is considered to be shared when either are true:
1884 	 * - The process has more than one thread.
1885 	 * - The file descriptor table has been shared via fdshare().
1886 	 *
1887 	 * When shared, the old file table will be placed on a freelist
1888 	 * which will be processed when the struct filedesc is released.
1889 	 *
1890 	 * Note that if onfiles == NDFILE, we're dealing with the original
1891 	 * static allocation contained within (struct filedesc0 *)fdp,
1892 	 * which must not be freed.
1893 	 */
1894 	if (onfiles > NDFILE) {
1895 		/*
1896 		 * Note we may be called here from fdinit while allocating a
1897 		 * table for a new process in which case ->p_fd points
1898 		 * elsewhere.
1899 		 */
1900 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1901 			free(otable, M_FILEDESC);
1902 		} else {
1903 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1904 			fdp0 = (struct filedesc0 *)fdp;
1905 			ft->ft_table = otable;
1906 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1907 		}
1908 	}
1909 	/*
1910 	 * The map does not have the same possibility of threads still
1911 	 * holding references to it.  So always free it as long as it
1912 	 * does not reference the original static allocation.
1913 	 */
1914 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1915 		free(omap, M_FILEDESC);
1916 }
1917 
1918 /*
1919  * Allocate a file descriptor for the process.
1920  */
1921 int
1922 fdalloc(struct thread *td, int minfd, int *result)
1923 {
1924 	struct proc *p = td->td_proc;
1925 	struct filedesc *fdp = p->p_fd;
1926 	int fd, maxfd, allocfd;
1927 #ifdef RACCT
1928 	int error;
1929 #endif
1930 
1931 	FILEDESC_XLOCK_ASSERT(fdp);
1932 
1933 	if (fdp->fd_freefile > minfd)
1934 		minfd = fdp->fd_freefile;
1935 
1936 	maxfd = getmaxfd(td);
1937 
1938 	/*
1939 	 * Search the bitmap for a free descriptor starting at minfd.
1940 	 * If none is found, grow the file table.
1941 	 */
1942 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1943 	if (__predict_false(fd >= maxfd))
1944 		return (EMFILE);
1945 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1946 		allocfd = min(fd * 2, maxfd);
1947 #ifdef RACCT
1948 		if (RACCT_ENABLED()) {
1949 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1950 			if (error != 0)
1951 				return (EMFILE);
1952 		}
1953 #endif
1954 		/*
1955 		 * fd is already equal to first free descriptor >= minfd, so
1956 		 * we only need to grow the table and we are done.
1957 		 */
1958 		fdgrowtable_exp(fdp, allocfd);
1959 	}
1960 
1961 	/*
1962 	 * Perform some sanity checks, then mark the file descriptor as
1963 	 * used and return it to the caller.
1964 	 */
1965 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1966 	    ("invalid descriptor %d", fd));
1967 	KASSERT(!fdisused(fdp, fd),
1968 	    ("fd_first_free() returned non-free descriptor"));
1969 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1970 	    ("file descriptor isn't free"));
1971 	fdused(fdp, fd);
1972 	*result = fd;
1973 	return (0);
1974 }
1975 
1976 /*
1977  * Allocate n file descriptors for the process.
1978  */
1979 int
1980 fdallocn(struct thread *td, int minfd, int *fds, int n)
1981 {
1982 	struct proc *p = td->td_proc;
1983 	struct filedesc *fdp = p->p_fd;
1984 	int i;
1985 
1986 	FILEDESC_XLOCK_ASSERT(fdp);
1987 
1988 	for (i = 0; i < n; i++)
1989 		if (fdalloc(td, 0, &fds[i]) != 0)
1990 			break;
1991 
1992 	if (i < n) {
1993 		for (i--; i >= 0; i--)
1994 			fdunused(fdp, fds[i]);
1995 		return (EMFILE);
1996 	}
1997 
1998 	return (0);
1999 }
2000 
2001 /*
2002  * Create a new open file structure and allocate a file descriptor for the
2003  * process that refers to it.  We add one reference to the file for the
2004  * descriptor table and one reference for resultfp. This is to prevent us
2005  * being preempted and the entry in the descriptor table closed after we
2006  * release the FILEDESC lock.
2007  */
2008 int
2009 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2010     struct filecaps *fcaps)
2011 {
2012 	struct file *fp;
2013 	int error, fd;
2014 
2015 	MPASS(resultfp != NULL);
2016 	MPASS(resultfd != NULL);
2017 
2018 	error = _falloc_noinstall(td, &fp, 2);
2019 	if (__predict_false(error != 0)) {
2020 		return (error);
2021 	}
2022 
2023 	error = finstall_refed(td, fp, &fd, flags, fcaps);
2024 	if (__predict_false(error != 0)) {
2025 		falloc_abort(td, fp);
2026 		return (error);
2027 	}
2028 
2029 	*resultfp = fp;
2030 	*resultfd = fd;
2031 
2032 	return (0);
2033 }
2034 
2035 /*
2036  * Create a new open file structure without allocating a file descriptor.
2037  */
2038 int
2039 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2040 {
2041 	struct file *fp;
2042 	int maxuserfiles = maxfiles - (maxfiles / 20);
2043 	int openfiles_new;
2044 	static struct timeval lastfail;
2045 	static int curfail;
2046 
2047 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2048 	MPASS(n > 0);
2049 
2050 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2051 	if ((openfiles_new >= maxuserfiles &&
2052 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2053 	    openfiles_new >= maxfiles) {
2054 		atomic_subtract_int(&openfiles, 1);
2055 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2056 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2057 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2058 		}
2059 		return (ENFILE);
2060 	}
2061 	fp = uma_zalloc(file_zone, M_WAITOK);
2062 	bzero(fp, sizeof(*fp));
2063 	refcount_init(&fp->f_count, n);
2064 	fp->f_cred = crhold(td->td_ucred);
2065 	fp->f_ops = &badfileops;
2066 	*resultfp = fp;
2067 	return (0);
2068 }
2069 
2070 void
2071 falloc_abort(struct thread *td, struct file *fp)
2072 {
2073 
2074 	/*
2075 	 * For assertion purposes.
2076 	 */
2077 	refcount_init(&fp->f_count, 0);
2078 	_fdrop(fp, td);
2079 }
2080 
2081 /*
2082  * Install a file in a file descriptor table.
2083  */
2084 void
2085 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2086     struct filecaps *fcaps)
2087 {
2088 	struct filedescent *fde;
2089 
2090 	MPASS(fp != NULL);
2091 	if (fcaps != NULL)
2092 		filecaps_validate(fcaps, __func__);
2093 	FILEDESC_XLOCK_ASSERT(fdp);
2094 
2095 	fde = &fdp->fd_ofiles[fd];
2096 #ifdef CAPABILITIES
2097 	seqc_write_begin(&fde->fde_seqc);
2098 #endif
2099 	fde->fde_file = fp;
2100 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2101 	if (fcaps != NULL)
2102 		filecaps_move(fcaps, &fde->fde_caps);
2103 	else
2104 		filecaps_fill(&fde->fde_caps);
2105 #ifdef CAPABILITIES
2106 	seqc_write_end(&fde->fde_seqc);
2107 #endif
2108 }
2109 
2110 int
2111 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2112     struct filecaps *fcaps)
2113 {
2114 	struct filedesc *fdp = td->td_proc->p_fd;
2115 	int error;
2116 
2117 	MPASS(fd != NULL);
2118 
2119 	FILEDESC_XLOCK(fdp);
2120 	error = fdalloc(td, 0, fd);
2121 	if (__predict_true(error == 0)) {
2122 		_finstall(fdp, fp, *fd, flags, fcaps);
2123 	}
2124 	FILEDESC_XUNLOCK(fdp);
2125 	return (error);
2126 }
2127 
2128 int
2129 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2130     struct filecaps *fcaps)
2131 {
2132 	int error;
2133 
2134 	MPASS(fd != NULL);
2135 
2136 	if (!fhold(fp))
2137 		return (EBADF);
2138 	error = finstall_refed(td, fp, fd, flags, fcaps);
2139 	if (__predict_false(error != 0)) {
2140 		fdrop(fp, td);
2141 	}
2142 	return (error);
2143 }
2144 
2145 /*
2146  * Build a new filedesc structure from another.
2147  *
2148  * If fdp is not NULL, return with it shared locked.
2149  */
2150 struct filedesc *
2151 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2152 {
2153 	struct filedesc0 *newfdp0;
2154 	struct filedesc *newfdp;
2155 
2156 	if (prepfiles)
2157 		MPASS(lastfile != NULL);
2158 	else
2159 		MPASS(lastfile == NULL);
2160 
2161 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2162 	newfdp = &newfdp0->fd_fd;
2163 
2164 	/* Create the file descriptor table. */
2165 	FILEDESC_LOCK_INIT(newfdp);
2166 	refcount_init(&newfdp->fd_refcnt, 1);
2167 	refcount_init(&newfdp->fd_holdcnt, 1);
2168 	newfdp->fd_map = newfdp0->fd_dmap;
2169 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2170 	newfdp->fd_files->fdt_nfiles = NDFILE;
2171 
2172 	if (fdp == NULL)
2173 		return (newfdp);
2174 
2175 	FILEDESC_SLOCK(fdp);
2176 	if (!prepfiles) {
2177 		FILEDESC_SUNLOCK(fdp);
2178 		return (newfdp);
2179 	}
2180 
2181 	for (;;) {
2182 		*lastfile = fdlastfile(fdp);
2183 		if (*lastfile < newfdp->fd_nfiles)
2184 			break;
2185 		FILEDESC_SUNLOCK(fdp);
2186 		fdgrowtable(newfdp, *lastfile + 1);
2187 		FILEDESC_SLOCK(fdp);
2188 	}
2189 
2190 	return (newfdp);
2191 }
2192 
2193 /*
2194  * Build a pwddesc structure from another.
2195  * Copy the current, root, and jail root vnode references.
2196  *
2197  * If pdp is not NULL, return with it shared locked.
2198  */
2199 struct pwddesc *
2200 pdinit(struct pwddesc *pdp, bool keeplock)
2201 {
2202 	struct pwddesc *newpdp;
2203 	struct pwd *newpwd;
2204 
2205 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2206 
2207 	PWDDESC_LOCK_INIT(newpdp);
2208 	refcount_init(&newpdp->pd_refcount, 1);
2209 	newpdp->pd_cmask = CMASK;
2210 
2211 	if (pdp == NULL) {
2212 		newpwd = pwd_alloc();
2213 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2214 		return (newpdp);
2215 	}
2216 
2217 	PWDDESC_XLOCK(pdp);
2218 	newpwd = pwd_hold_pwddesc(pdp);
2219 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2220 	if (!keeplock)
2221 		PWDDESC_XUNLOCK(pdp);
2222 	return (newpdp);
2223 }
2224 
2225 /*
2226  * Hold either filedesc or pwddesc of the passed process.
2227  *
2228  * The process lock is used to synchronize against the target exiting and
2229  * freeing the data.
2230  *
2231  * Clearing can be ilustrated in 3 steps:
2232  * 1. set the pointer to NULL. Either routine can race against it, hence
2233  *   atomic_load_ptr.
2234  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2235  *   race to either still see the pointer or find NULL. It is still safe to
2236  *   grab a reference as clearing is stalled.
2237  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2238  *   guaranteed to see NULL, making it safe to finish clearing
2239  */
2240 static struct filedesc *
2241 fdhold(struct proc *p)
2242 {
2243 	struct filedesc *fdp;
2244 
2245 	PROC_LOCK_ASSERT(p, MA_OWNED);
2246 	fdp = atomic_load_ptr(&p->p_fd);
2247 	if (fdp != NULL)
2248 		refcount_acquire(&fdp->fd_holdcnt);
2249 	return (fdp);
2250 }
2251 
2252 static struct pwddesc *
2253 pdhold(struct proc *p)
2254 {
2255 	struct pwddesc *pdp;
2256 
2257 	PROC_LOCK_ASSERT(p, MA_OWNED);
2258 	pdp = atomic_load_ptr(&p->p_pd);
2259 	if (pdp != NULL)
2260 		refcount_acquire(&pdp->pd_refcount);
2261 	return (pdp);
2262 }
2263 
2264 static void
2265 fddrop(struct filedesc *fdp)
2266 {
2267 
2268 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2269 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2270 			return;
2271 	}
2272 
2273 	FILEDESC_LOCK_DESTROY(fdp);
2274 	uma_zfree(filedesc0_zone, fdp);
2275 }
2276 
2277 static void
2278 pddrop(struct pwddesc *pdp)
2279 {
2280 	struct pwd *pwd;
2281 
2282 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2283 		return;
2284 
2285 	PWDDESC_XLOCK(pdp);
2286 	if (refcount_release(&pdp->pd_refcount) == 0) {
2287 		PWDDESC_XUNLOCK(pdp);
2288 		return;
2289 	}
2290 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2291 	pwd_set(pdp, NULL);
2292 	PWDDESC_XUNLOCK(pdp);
2293 	pwd_drop(pwd);
2294 
2295 	PWDDESC_LOCK_DESTROY(pdp);
2296 	free(pdp, M_PWDDESC);
2297 }
2298 
2299 /*
2300  * Share a filedesc structure.
2301  */
2302 struct filedesc *
2303 fdshare(struct filedesc *fdp)
2304 {
2305 
2306 	refcount_acquire(&fdp->fd_refcnt);
2307 	return (fdp);
2308 }
2309 
2310 /*
2311  * Share a pwddesc structure.
2312  */
2313 struct pwddesc *
2314 pdshare(struct pwddesc *pdp)
2315 {
2316 	refcount_acquire(&pdp->pd_refcount);
2317 	return (pdp);
2318 }
2319 
2320 /*
2321  * Unshare a filedesc structure, if necessary by making a copy
2322  */
2323 void
2324 fdunshare(struct thread *td)
2325 {
2326 	struct filedesc *tmp;
2327 	struct proc *p = td->td_proc;
2328 
2329 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2330 		return;
2331 
2332 	tmp = fdcopy(p->p_fd);
2333 	fdescfree(td);
2334 	p->p_fd = tmp;
2335 }
2336 
2337 /*
2338  * Unshare a pwddesc structure.
2339  */
2340 void
2341 pdunshare(struct thread *td)
2342 {
2343 	struct pwddesc *pdp;
2344 	struct proc *p;
2345 
2346 	p = td->td_proc;
2347 	/* Not shared. */
2348 	if (p->p_pd->pd_refcount == 1)
2349 		return;
2350 
2351 	pdp = pdcopy(p->p_pd);
2352 	pdescfree(td);
2353 	p->p_pd = pdp;
2354 }
2355 
2356 /*
2357  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2358  * this is to ease callers, not catch errors.
2359  */
2360 struct filedesc *
2361 fdcopy(struct filedesc *fdp)
2362 {
2363 	struct filedesc *newfdp;
2364 	struct filedescent *nfde, *ofde;
2365 	int i, lastfile;
2366 
2367 	MPASS(fdp != NULL);
2368 
2369 	newfdp = fdinit(fdp, true, &lastfile);
2370 	/* copy all passable descriptors (i.e. not kqueue) */
2371 	newfdp->fd_freefile = -1;
2372 	for (i = 0; i <= lastfile; ++i) {
2373 		ofde = &fdp->fd_ofiles[i];
2374 		if (ofde->fde_file == NULL ||
2375 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2376 		    !fhold(ofde->fde_file)) {
2377 			if (newfdp->fd_freefile == -1)
2378 				newfdp->fd_freefile = i;
2379 			continue;
2380 		}
2381 		nfde = &newfdp->fd_ofiles[i];
2382 		*nfde = *ofde;
2383 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2384 		fdused_init(newfdp, i);
2385 	}
2386 	if (newfdp->fd_freefile == -1)
2387 		newfdp->fd_freefile = i;
2388 	FILEDESC_SUNLOCK(fdp);
2389 	return (newfdp);
2390 }
2391 
2392 /*
2393  * Copy a pwddesc structure.
2394  */
2395 struct pwddesc *
2396 pdcopy(struct pwddesc *pdp)
2397 {
2398 	struct pwddesc *newpdp;
2399 
2400 	MPASS(pdp != NULL);
2401 
2402 	newpdp = pdinit(pdp, true);
2403 	newpdp->pd_cmask = pdp->pd_cmask;
2404 	PWDDESC_XUNLOCK(pdp);
2405 	return (newpdp);
2406 }
2407 
2408 /*
2409  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2410  * one of processes using it exits) and the table used to be shared.
2411  */
2412 static void
2413 fdclearlocks(struct thread *td)
2414 {
2415 	struct filedesc *fdp;
2416 	struct filedesc_to_leader *fdtol;
2417 	struct flock lf;
2418 	struct file *fp;
2419 	struct proc *p;
2420 	struct vnode *vp;
2421 	int i, lastfile;
2422 
2423 	p = td->td_proc;
2424 	fdp = p->p_fd;
2425 	fdtol = p->p_fdtol;
2426 	MPASS(fdtol != NULL);
2427 
2428 	FILEDESC_XLOCK(fdp);
2429 	KASSERT(fdtol->fdl_refcount > 0,
2430 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2431 	    fdtol->fdl_refcount));
2432 	if (fdtol->fdl_refcount == 1 &&
2433 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2434 		lastfile = fdlastfile(fdp);
2435 		for (i = 0; i <= lastfile; i++) {
2436 			fp = fdp->fd_ofiles[i].fde_file;
2437 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2438 			    !fhold(fp))
2439 				continue;
2440 			FILEDESC_XUNLOCK(fdp);
2441 			lf.l_whence = SEEK_SET;
2442 			lf.l_start = 0;
2443 			lf.l_len = 0;
2444 			lf.l_type = F_UNLCK;
2445 			vp = fp->f_vnode;
2446 			(void) VOP_ADVLOCK(vp,
2447 			    (caddr_t)p->p_leader, F_UNLCK,
2448 			    &lf, F_POSIX);
2449 			FILEDESC_XLOCK(fdp);
2450 			fdrop(fp, td);
2451 		}
2452 	}
2453 retry:
2454 	if (fdtol->fdl_refcount == 1) {
2455 		if (fdp->fd_holdleaderscount > 0 &&
2456 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2457 			/*
2458 			 * close() or kern_dup() has cleared a reference
2459 			 * in a shared file descriptor table.
2460 			 */
2461 			fdp->fd_holdleaderswakeup = 1;
2462 			sx_sleep(&fdp->fd_holdleaderscount,
2463 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2464 			goto retry;
2465 		}
2466 		if (fdtol->fdl_holdcount > 0) {
2467 			/*
2468 			 * Ensure that fdtol->fdl_leader remains
2469 			 * valid in closef().
2470 			 */
2471 			fdtol->fdl_wakeup = 1;
2472 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2473 			    "fdlhold", 0);
2474 			goto retry;
2475 		}
2476 	}
2477 	fdtol->fdl_refcount--;
2478 	if (fdtol->fdl_refcount == 0 &&
2479 	    fdtol->fdl_holdcount == 0) {
2480 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2481 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2482 	} else
2483 		fdtol = NULL;
2484 	p->p_fdtol = NULL;
2485 	FILEDESC_XUNLOCK(fdp);
2486 	if (fdtol != NULL)
2487 		free(fdtol, M_FILEDESC_TO_LEADER);
2488 }
2489 
2490 /*
2491  * Release a filedesc structure.
2492  */
2493 static void
2494 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2495 {
2496 	struct filedesc0 *fdp0;
2497 	struct freetable *ft, *tft;
2498 	struct filedescent *fde;
2499 	struct file *fp;
2500 	int i, lastfile;
2501 
2502 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2503 	    ("%s: fd table %p carries references", __func__, fdp));
2504 
2505 	/*
2506 	 * Serialize with threads iterating over the table, if any.
2507 	 */
2508 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2509 		FILEDESC_XLOCK(fdp);
2510 		FILEDESC_XUNLOCK(fdp);
2511 	}
2512 
2513 	lastfile = fdlastfile_single(fdp);
2514 	for (i = 0; i <= lastfile; i++) {
2515 		fde = &fdp->fd_ofiles[i];
2516 		fp = fde->fde_file;
2517 		if (fp != NULL) {
2518 			fdefree_last(fde);
2519 			(void) closef(fp, td);
2520 		}
2521 	}
2522 
2523 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2524 		free(fdp->fd_map, M_FILEDESC);
2525 	if (fdp->fd_nfiles > NDFILE)
2526 		free(fdp->fd_files, M_FILEDESC);
2527 
2528 	fdp0 = (struct filedesc0 *)fdp;
2529 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2530 		free(ft->ft_table, M_FILEDESC);
2531 
2532 	fddrop(fdp);
2533 }
2534 
2535 void
2536 fdescfree(struct thread *td)
2537 {
2538 	struct proc *p;
2539 	struct filedesc *fdp;
2540 
2541 	p = td->td_proc;
2542 	fdp = p->p_fd;
2543 	MPASS(fdp != NULL);
2544 
2545 #ifdef RACCT
2546 	if (RACCT_ENABLED())
2547 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2548 #endif
2549 
2550 	if (p->p_fdtol != NULL)
2551 		fdclearlocks(td);
2552 
2553 	/*
2554 	 * Check fdhold for an explanation.
2555 	 */
2556 	atomic_store_ptr(&p->p_fd, NULL);
2557 	atomic_thread_fence_seq_cst();
2558 	PROC_WAIT_UNLOCKED(p);
2559 
2560 	if (refcount_release(&fdp->fd_refcnt) == 0)
2561 		return;
2562 
2563 	fdescfree_fds(td, fdp);
2564 }
2565 
2566 void
2567 pdescfree(struct thread *td)
2568 {
2569 	struct proc *p;
2570 	struct pwddesc *pdp;
2571 
2572 	p = td->td_proc;
2573 	pdp = p->p_pd;
2574 	MPASS(pdp != NULL);
2575 
2576 	/*
2577 	 * Check pdhold for an explanation.
2578 	 */
2579 	atomic_store_ptr(&p->p_pd, NULL);
2580 	atomic_thread_fence_seq_cst();
2581 	PROC_WAIT_UNLOCKED(p);
2582 
2583 	pddrop(pdp);
2584 }
2585 
2586 /*
2587  * For setugid programs, we don't want to people to use that setugidness
2588  * to generate error messages which write to a file which otherwise would
2589  * otherwise be off-limits to the process.  We check for filesystems where
2590  * the vnode can change out from under us after execve (like [lin]procfs).
2591  *
2592  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2593  * sufficient.  We also don't check for setugidness since we know we are.
2594  */
2595 static bool
2596 is_unsafe(struct file *fp)
2597 {
2598 	struct vnode *vp;
2599 
2600 	if (fp->f_type != DTYPE_VNODE)
2601 		return (false);
2602 
2603 	vp = fp->f_vnode;
2604 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2605 }
2606 
2607 /*
2608  * Make this setguid thing safe, if at all possible.
2609  */
2610 void
2611 fdsetugidsafety(struct thread *td)
2612 {
2613 	struct filedesc *fdp;
2614 	struct file *fp;
2615 	int i;
2616 
2617 	fdp = td->td_proc->p_fd;
2618 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2619 	    ("the fdtable should not be shared"));
2620 	MPASS(fdp->fd_nfiles >= 3);
2621 	for (i = 0; i <= 2; i++) {
2622 		fp = fdp->fd_ofiles[i].fde_file;
2623 		if (fp != NULL && is_unsafe(fp)) {
2624 			FILEDESC_XLOCK(fdp);
2625 			knote_fdclose(td, i);
2626 			/*
2627 			 * NULL-out descriptor prior to close to avoid
2628 			 * a race while close blocks.
2629 			 */
2630 			fdfree(fdp, i);
2631 			FILEDESC_XUNLOCK(fdp);
2632 			(void) closef(fp, td);
2633 		}
2634 	}
2635 }
2636 
2637 /*
2638  * If a specific file object occupies a specific file descriptor, close the
2639  * file descriptor entry and drop a reference on the file object.  This is a
2640  * convenience function to handle a subsequent error in a function that calls
2641  * falloc() that handles the race that another thread might have closed the
2642  * file descriptor out from under the thread creating the file object.
2643  */
2644 void
2645 fdclose(struct thread *td, struct file *fp, int idx)
2646 {
2647 	struct filedesc *fdp = td->td_proc->p_fd;
2648 
2649 	FILEDESC_XLOCK(fdp);
2650 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2651 		fdfree(fdp, idx);
2652 		FILEDESC_XUNLOCK(fdp);
2653 		fdrop(fp, td);
2654 	} else
2655 		FILEDESC_XUNLOCK(fdp);
2656 }
2657 
2658 /*
2659  * Close any files on exec?
2660  */
2661 void
2662 fdcloseexec(struct thread *td)
2663 {
2664 	struct filedesc *fdp;
2665 	struct filedescent *fde;
2666 	struct file *fp;
2667 	int i, lastfile;
2668 
2669 	fdp = td->td_proc->p_fd;
2670 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2671 	    ("the fdtable should not be shared"));
2672 	lastfile = fdlastfile_single(fdp);
2673 	for (i = 0; i <= lastfile; i++) {
2674 		fde = &fdp->fd_ofiles[i];
2675 		fp = fde->fde_file;
2676 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2677 		    (fde->fde_flags & UF_EXCLOSE))) {
2678 			FILEDESC_XLOCK(fdp);
2679 			fdfree(fdp, i);
2680 			(void) closefp(fdp, i, fp, td, false, false);
2681 			FILEDESC_UNLOCK_ASSERT(fdp);
2682 		}
2683 	}
2684 }
2685 
2686 /*
2687  * It is unsafe for set[ug]id processes to be started with file
2688  * descriptors 0..2 closed, as these descriptors are given implicit
2689  * significance in the Standard C library.  fdcheckstd() will create a
2690  * descriptor referencing /dev/null for each of stdin, stdout, and
2691  * stderr that is not already open.
2692  */
2693 int
2694 fdcheckstd(struct thread *td)
2695 {
2696 	struct filedesc *fdp;
2697 	register_t save;
2698 	int i, error, devnull;
2699 
2700 	fdp = td->td_proc->p_fd;
2701 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2702 	    ("the fdtable should not be shared"));
2703 	MPASS(fdp->fd_nfiles >= 3);
2704 	devnull = -1;
2705 	for (i = 0; i <= 2; i++) {
2706 		if (fdp->fd_ofiles[i].fde_file != NULL)
2707 			continue;
2708 
2709 		save = td->td_retval[0];
2710 		if (devnull != -1) {
2711 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2712 		} else {
2713 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2714 			    UIO_SYSSPACE, O_RDWR, 0);
2715 			if (error == 0) {
2716 				devnull = td->td_retval[0];
2717 				KASSERT(devnull == i, ("we didn't get our fd"));
2718 			}
2719 		}
2720 		td->td_retval[0] = save;
2721 		if (error != 0)
2722 			return (error);
2723 	}
2724 	return (0);
2725 }
2726 
2727 /*
2728  * Internal form of close.  Decrement reference count on file structure.
2729  * Note: td may be NULL when closing a file that was being passed in a
2730  * message.
2731  */
2732 int
2733 closef(struct file *fp, struct thread *td)
2734 {
2735 	struct vnode *vp;
2736 	struct flock lf;
2737 	struct filedesc_to_leader *fdtol;
2738 	struct filedesc *fdp;
2739 
2740 	MPASS(td != NULL);
2741 
2742 	/*
2743 	 * POSIX record locking dictates that any close releases ALL
2744 	 * locks owned by this process.  This is handled by setting
2745 	 * a flag in the unlock to free ONLY locks obeying POSIX
2746 	 * semantics, and not to free BSD-style file locks.
2747 	 * If the descriptor was in a message, POSIX-style locks
2748 	 * aren't passed with the descriptor, and the thread pointer
2749 	 * will be NULL.  Callers should be careful only to pass a
2750 	 * NULL thread pointer when there really is no owning
2751 	 * context that might have locks, or the locks will be
2752 	 * leaked.
2753 	 */
2754 	if (fp->f_type == DTYPE_VNODE) {
2755 		vp = fp->f_vnode;
2756 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2757 			lf.l_whence = SEEK_SET;
2758 			lf.l_start = 0;
2759 			lf.l_len = 0;
2760 			lf.l_type = F_UNLCK;
2761 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2762 			    F_UNLCK, &lf, F_POSIX);
2763 		}
2764 		fdtol = td->td_proc->p_fdtol;
2765 		if (fdtol != NULL) {
2766 			/*
2767 			 * Handle special case where file descriptor table is
2768 			 * shared between multiple process leaders.
2769 			 */
2770 			fdp = td->td_proc->p_fd;
2771 			FILEDESC_XLOCK(fdp);
2772 			for (fdtol = fdtol->fdl_next;
2773 			    fdtol != td->td_proc->p_fdtol;
2774 			    fdtol = fdtol->fdl_next) {
2775 				if ((fdtol->fdl_leader->p_flag &
2776 				    P_ADVLOCK) == 0)
2777 					continue;
2778 				fdtol->fdl_holdcount++;
2779 				FILEDESC_XUNLOCK(fdp);
2780 				lf.l_whence = SEEK_SET;
2781 				lf.l_start = 0;
2782 				lf.l_len = 0;
2783 				lf.l_type = F_UNLCK;
2784 				vp = fp->f_vnode;
2785 				(void) VOP_ADVLOCK(vp,
2786 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2787 				    F_POSIX);
2788 				FILEDESC_XLOCK(fdp);
2789 				fdtol->fdl_holdcount--;
2790 				if (fdtol->fdl_holdcount == 0 &&
2791 				    fdtol->fdl_wakeup != 0) {
2792 					fdtol->fdl_wakeup = 0;
2793 					wakeup(fdtol);
2794 				}
2795 			}
2796 			FILEDESC_XUNLOCK(fdp);
2797 		}
2798 	}
2799 	return (fdrop_close(fp, td));
2800 }
2801 
2802 /*
2803  * Hack for file descriptor passing code.
2804  */
2805 void
2806 closef_nothread(struct file *fp)
2807 {
2808 
2809 	fdrop(fp, NULL);
2810 }
2811 
2812 /*
2813  * Initialize the file pointer with the specified properties.
2814  *
2815  * The ops are set with release semantics to be certain that the flags, type,
2816  * and data are visible when ops is.  This is to prevent ops methods from being
2817  * called with bad data.
2818  */
2819 void
2820 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2821 {
2822 	fp->f_data = data;
2823 	fp->f_flag = flag;
2824 	fp->f_type = type;
2825 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2826 }
2827 
2828 void
2829 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2830 {
2831 	fp->f_seqcount[UIO_READ] = 1;
2832 	fp->f_seqcount[UIO_WRITE] = 1;
2833 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2834 	    data, ops);
2835 }
2836 
2837 int
2838 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2839     struct file **fpp, struct filecaps *havecapsp)
2840 {
2841 	struct filedescent *fde;
2842 	int error;
2843 
2844 	FILEDESC_LOCK_ASSERT(fdp);
2845 
2846 	fde = fdeget_locked(fdp, fd);
2847 	if (fde == NULL) {
2848 		error = EBADF;
2849 		goto out;
2850 	}
2851 
2852 #ifdef CAPABILITIES
2853 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2854 	if (error != 0)
2855 		goto out;
2856 #endif
2857 
2858 	if (havecapsp != NULL)
2859 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2860 
2861 	*fpp = fde->fde_file;
2862 
2863 	error = 0;
2864 out:
2865 	return (error);
2866 }
2867 
2868 int
2869 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2870     struct file **fpp, struct filecaps *havecapsp)
2871 {
2872 	struct filedesc *fdp = td->td_proc->p_fd;
2873 	int error;
2874 #ifndef CAPABILITIES
2875 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2876 	if (havecapsp != NULL && error == 0)
2877 		filecaps_fill(havecapsp);
2878 #else
2879 	struct file *fp;
2880 	seqc_t seq;
2881 
2882 	*fpp = NULL;
2883 	for (;;) {
2884 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2885 		if (error != 0)
2886 			return (error);
2887 
2888 		if (havecapsp != NULL) {
2889 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2890 			    havecapsp, false)) {
2891 				fdrop(fp, td);
2892 				goto get_locked;
2893 			}
2894 		}
2895 
2896 		if (!fd_modified(fdp, fd, seq))
2897 			break;
2898 		fdrop(fp, td);
2899 	}
2900 
2901 	*fpp = fp;
2902 	return (0);
2903 
2904 get_locked:
2905 	FILEDESC_SLOCK(fdp);
2906 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2907 	if (error == 0 && !fhold(*fpp))
2908 		error = EBADF;
2909 	FILEDESC_SUNLOCK(fdp);
2910 #endif
2911 	return (error);
2912 }
2913 
2914 #ifdef CAPABILITIES
2915 int
2916 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2917 {
2918 	const struct filedescent *fde;
2919 	const struct fdescenttbl *fdt;
2920 	struct filedesc *fdp;
2921 	struct file *fp;
2922 	struct vnode *vp;
2923 	const cap_rights_t *haverights;
2924 	cap_rights_t rights;
2925 	seqc_t seq;
2926 
2927 	VFS_SMR_ASSERT_ENTERED();
2928 
2929 	rights = *ndp->ni_rightsneeded;
2930 	cap_rights_set_one(&rights, CAP_LOOKUP);
2931 
2932 	fdp = curproc->p_fd;
2933 	fdt = fdp->fd_files;
2934 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2935 		return (EBADF);
2936 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2937 	fde = &fdt->fdt_ofiles[fd];
2938 	haverights = cap_rights_fde_inline(fde);
2939 	fp = fde->fde_file;
2940 	if (__predict_false(fp == NULL))
2941 		return (EAGAIN);
2942 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2943 		return (EAGAIN);
2944 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2945 	vp = fp->f_vnode;
2946 	if (__predict_false(vp == NULL)) {
2947 		return (EAGAIN);
2948 	}
2949 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2950 		return (EAGAIN);
2951 	}
2952 	/*
2953 	 * Use an acquire barrier to force re-reading of fdt so it is
2954 	 * refreshed for verification.
2955 	 */
2956 	atomic_thread_fence_acq();
2957 	fdt = fdp->fd_files;
2958 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2959 		return (EAGAIN);
2960 	/*
2961 	 * If file descriptor doesn't have all rights,
2962 	 * all lookups relative to it must also be
2963 	 * strictly relative.
2964 	 *
2965 	 * Not yet supported by fast path.
2966 	 */
2967 	CAP_ALL(&rights);
2968 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2969 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2970 	    ndp->ni_filecaps.fc_nioctls != -1) {
2971 #ifdef notyet
2972 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2973 #else
2974 		return (EAGAIN);
2975 #endif
2976 	}
2977 	*vpp = vp;
2978 	return (0);
2979 }
2980 #else
2981 int
2982 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2983 {
2984 	const struct fdescenttbl *fdt;
2985 	struct filedesc *fdp;
2986 	struct file *fp;
2987 	struct vnode *vp;
2988 
2989 	VFS_SMR_ASSERT_ENTERED();
2990 
2991 	fdp = curproc->p_fd;
2992 	fdt = fdp->fd_files;
2993 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2994 		return (EBADF);
2995 	fp = fdt->fdt_ofiles[fd].fde_file;
2996 	if (__predict_false(fp == NULL))
2997 		return (EAGAIN);
2998 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2999 	vp = fp->f_vnode;
3000 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3001 		return (EAGAIN);
3002 	}
3003 	/*
3004 	 * Use an acquire barrier to force re-reading of fdt so it is
3005 	 * refreshed for verification.
3006 	 */
3007 	atomic_thread_fence_acq();
3008 	fdt = fdp->fd_files;
3009 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3010 		return (EAGAIN);
3011 	filecaps_fill(&ndp->ni_filecaps);
3012 	*vpp = vp;
3013 	return (0);
3014 }
3015 #endif
3016 
3017 int
3018 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3019     struct file **fpp, seqc_t *seqp)
3020 {
3021 #ifdef CAPABILITIES
3022 	const struct filedescent *fde;
3023 #endif
3024 	const struct fdescenttbl *fdt;
3025 	struct file *fp;
3026 #ifdef CAPABILITIES
3027 	seqc_t seq;
3028 	cap_rights_t haverights;
3029 	int error;
3030 #endif
3031 
3032 	fdt = fdp->fd_files;
3033 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3034 		return (EBADF);
3035 	/*
3036 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3037 	 * never raising a refcount above 0.  To accomplish this we have
3038 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3039 	 * must be re-verified once we acquire a reference to be certain
3040 	 * that the identity is still correct and we did not lose a race
3041 	 * due to preemption.
3042 	 */
3043 	for (;;) {
3044 #ifdef CAPABILITIES
3045 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3046 		fde = &fdt->fdt_ofiles[fd];
3047 		haverights = *cap_rights_fde_inline(fde);
3048 		fp = fde->fde_file;
3049 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3050 			continue;
3051 #else
3052 		fp = fdt->fdt_ofiles[fd].fde_file;
3053 #endif
3054 		if (fp == NULL)
3055 			return (EBADF);
3056 #ifdef CAPABILITIES
3057 		error = cap_check_inline(&haverights, needrightsp);
3058 		if (error != 0)
3059 			return (error);
3060 #endif
3061 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3062 			/*
3063 			 * Force a reload. Other thread could reallocate the
3064 			 * table before this fd was closed, so it is possible
3065 			 * that there is a stale fp pointer in cached version.
3066 			 */
3067 			fdt = atomic_load_ptr(&fdp->fd_files);
3068 			continue;
3069 		}
3070 		/*
3071 		 * Use an acquire barrier to force re-reading of fdt so it is
3072 		 * refreshed for verification.
3073 		 */
3074 		atomic_thread_fence_acq();
3075 		fdt = fdp->fd_files;
3076 #ifdef	CAPABILITIES
3077 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3078 #else
3079 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3080 #endif
3081 			break;
3082 		fdrop(fp, curthread);
3083 	}
3084 	*fpp = fp;
3085 	if (seqp != NULL) {
3086 #ifdef CAPABILITIES
3087 		*seqp = seq;
3088 #endif
3089 	}
3090 	return (0);
3091 }
3092 
3093 /*
3094  * See the comments in fget_unlocked_seq for an explanation of how this works.
3095  *
3096  * This is a simplified variant which bails out to the aforementioned routine
3097  * if anything goes wrong. In practice this only happens when userspace is
3098  * racing with itself.
3099  */
3100 int
3101 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3102     struct file **fpp)
3103 {
3104 #ifdef CAPABILITIES
3105 	const struct filedescent *fde;
3106 #endif
3107 	const struct fdescenttbl *fdt;
3108 	struct file *fp;
3109 #ifdef CAPABILITIES
3110 	seqc_t seq;
3111 	const cap_rights_t *haverights;
3112 #endif
3113 
3114 	fdt = fdp->fd_files;
3115 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3116 		return (EBADF);
3117 #ifdef CAPABILITIES
3118 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3119 	fde = &fdt->fdt_ofiles[fd];
3120 	haverights = cap_rights_fde_inline(fde);
3121 	fp = fde->fde_file;
3122 #else
3123 	fp = fdt->fdt_ofiles[fd].fde_file;
3124 #endif
3125 	if (__predict_false(fp == NULL))
3126 		goto out_fallback;
3127 #ifdef CAPABILITIES
3128 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3129 		goto out_fallback;
3130 #endif
3131 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3132 		goto out_fallback;
3133 
3134 	/*
3135 	 * Use an acquire barrier to force re-reading of fdt so it is
3136 	 * refreshed for verification.
3137 	 */
3138 	atomic_thread_fence_acq();
3139 	fdt = fdp->fd_files;
3140 #ifdef	CAPABILITIES
3141 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3142 #else
3143 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3144 #endif
3145 		goto out_fdrop;
3146 	*fpp = fp;
3147 	return (0);
3148 out_fdrop:
3149 	fdrop(fp, curthread);
3150 out_fallback:
3151 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3152 }
3153 
3154 /*
3155  * Translate fd -> file when the caller guarantees the file descriptor table
3156  * can't be changed by others.
3157  *
3158  * Note this does not mean the file object itself is only visible to the caller,
3159  * merely that it wont disappear without having to be referenced.
3160  *
3161  * Must be paired with fput_only_user.
3162  */
3163 #ifdef	CAPABILITIES
3164 int
3165 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3166     struct file **fpp)
3167 {
3168 	const struct filedescent *fde;
3169 	const struct fdescenttbl *fdt;
3170 	const cap_rights_t *haverights;
3171 	struct file *fp;
3172 	int error;
3173 
3174 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3175 
3176 	if (__predict_false(fd >= fdp->fd_nfiles))
3177 		return (EBADF);
3178 
3179 	fdt = fdp->fd_files;
3180 	fde = &fdt->fdt_ofiles[fd];
3181 	fp = fde->fde_file;
3182 	if (__predict_false(fp == NULL))
3183 		return (EBADF);
3184 	MPASS(refcount_load(&fp->f_count) > 0);
3185 	haverights = cap_rights_fde_inline(fde);
3186 	error = cap_check_inline(haverights, needrightsp);
3187 	if (__predict_false(error != 0))
3188 		return (error);
3189 	*fpp = fp;
3190 	return (0);
3191 }
3192 #else
3193 int
3194 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3195     struct file **fpp)
3196 {
3197 	struct file *fp;
3198 
3199 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3200 
3201 	if (__predict_false(fd >= fdp->fd_nfiles))
3202 		return (EBADF);
3203 
3204 	fp = fdp->fd_ofiles[fd].fde_file;
3205 	if (__predict_false(fp == NULL))
3206 		return (EBADF);
3207 
3208 	MPASS(refcount_load(&fp->f_count) > 0);
3209 	*fpp = fp;
3210 	return (0);
3211 }
3212 #endif
3213 
3214 /*
3215  * Extract the file pointer associated with the specified descriptor for the
3216  * current user process.
3217  *
3218  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3219  * returned.
3220  *
3221  * File's rights will be checked against the capability rights mask.
3222  *
3223  * If an error occurred the non-zero error is returned and *fpp is set to
3224  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3225  * responsible for fdrop().
3226  */
3227 static __inline int
3228 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3229     cap_rights_t *needrightsp)
3230 {
3231 	struct filedesc *fdp;
3232 	struct file *fp;
3233 	int error;
3234 
3235 	*fpp = NULL;
3236 	fdp = td->td_proc->p_fd;
3237 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3238 	if (__predict_false(error != 0))
3239 		return (error);
3240 	if (__predict_false(fp->f_ops == &badfileops)) {
3241 		fdrop(fp, td);
3242 		return (EBADF);
3243 	}
3244 
3245 	/*
3246 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3247 	 */
3248 	error = 0;
3249 	switch (flags) {
3250 	case FREAD:
3251 	case FWRITE:
3252 		if ((fp->f_flag & flags) == 0)
3253 			error = EBADF;
3254 		break;
3255 	case FEXEC:
3256 		if (fp->f_ops != &path_fileops &&
3257 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3258 		    (fp->f_flag & FWRITE) != 0))
3259 			error = EBADF;
3260 		break;
3261 	case 0:
3262 		break;
3263 	default:
3264 		KASSERT(0, ("wrong flags"));
3265 	}
3266 
3267 	if (error != 0) {
3268 		fdrop(fp, td);
3269 		return (error);
3270 	}
3271 
3272 	*fpp = fp;
3273 	return (0);
3274 }
3275 
3276 int
3277 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3278 {
3279 
3280 	return (_fget(td, fd, fpp, 0, rightsp));
3281 }
3282 
3283 int
3284 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3285     struct file **fpp)
3286 {
3287 	int error;
3288 #ifndef CAPABILITIES
3289 	error = _fget(td, fd, fpp, 0, rightsp);
3290 	if (maxprotp != NULL)
3291 		*maxprotp = VM_PROT_ALL;
3292 	return (error);
3293 #else
3294 	cap_rights_t fdrights;
3295 	struct filedesc *fdp;
3296 	struct file *fp;
3297 	seqc_t seq;
3298 
3299 	*fpp = NULL;
3300 	fdp = td->td_proc->p_fd;
3301 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3302 	for (;;) {
3303 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3304 		if (__predict_false(error != 0))
3305 			return (error);
3306 		if (__predict_false(fp->f_ops == &badfileops)) {
3307 			fdrop(fp, td);
3308 			return (EBADF);
3309 		}
3310 		if (maxprotp != NULL)
3311 			fdrights = *cap_rights(fdp, fd);
3312 		if (!fd_modified(fdp, fd, seq))
3313 			break;
3314 		fdrop(fp, td);
3315 	}
3316 
3317 	/*
3318 	 * If requested, convert capability rights to access flags.
3319 	 */
3320 	if (maxprotp != NULL)
3321 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3322 	*fpp = fp;
3323 	return (0);
3324 #endif
3325 }
3326 
3327 int
3328 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3329 {
3330 
3331 	return (_fget(td, fd, fpp, FREAD, rightsp));
3332 }
3333 
3334 int
3335 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3336 {
3337 
3338 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3339 }
3340 
3341 int
3342 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3343     struct file **fpp)
3344 {
3345 	struct filedesc *fdp = td->td_proc->p_fd;
3346 #ifndef CAPABILITIES
3347 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3348 #else
3349 	struct file *fp;
3350 	int error;
3351 	seqc_t seq;
3352 
3353 	*fpp = NULL;
3354 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3355 	for (;;) {
3356 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3357 		if (error != 0)
3358 			return (error);
3359 		error = cap_fcntl_check(fdp, fd, needfcntl);
3360 		if (!fd_modified(fdp, fd, seq))
3361 			break;
3362 		fdrop(fp, td);
3363 	}
3364 	if (error != 0) {
3365 		fdrop(fp, td);
3366 		return (error);
3367 	}
3368 	*fpp = fp;
3369 	return (0);
3370 #endif
3371 }
3372 
3373 /*
3374  * Like fget() but loads the underlying vnode, or returns an error if the
3375  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3376  * never have VM objects.  The returned vnode will be vref()'d.
3377  *
3378  * XXX: what about the unused flags ?
3379  */
3380 static __inline int
3381 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3382     struct vnode **vpp)
3383 {
3384 	struct file *fp;
3385 	int error;
3386 
3387 	*vpp = NULL;
3388 	error = _fget(td, fd, &fp, flags, needrightsp);
3389 	if (error != 0)
3390 		return (error);
3391 	if (fp->f_vnode == NULL) {
3392 		error = EINVAL;
3393 	} else {
3394 		*vpp = fp->f_vnode;
3395 		vref(*vpp);
3396 	}
3397 	fdrop(fp, td);
3398 
3399 	return (error);
3400 }
3401 
3402 int
3403 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3404 {
3405 
3406 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3407 }
3408 
3409 int
3410 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3411     struct filecaps *havecaps, struct vnode **vpp)
3412 {
3413 	struct filecaps caps;
3414 	struct file *fp;
3415 	int error;
3416 
3417 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3418 	if (error != 0)
3419 		return (error);
3420 	if (fp->f_ops == &badfileops) {
3421 		error = EBADF;
3422 		goto out;
3423 	}
3424 	if (fp->f_vnode == NULL) {
3425 		error = EINVAL;
3426 		goto out;
3427 	}
3428 
3429 	*havecaps = caps;
3430 	*vpp = fp->f_vnode;
3431 	vref(*vpp);
3432 	fdrop(fp, td);
3433 
3434 	return (0);
3435 out:
3436 	filecaps_free(&caps);
3437 	fdrop(fp, td);
3438 	return (error);
3439 }
3440 
3441 int
3442 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3443 {
3444 
3445 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3446 }
3447 
3448 int
3449 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3450 {
3451 
3452 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3453 }
3454 
3455 #ifdef notyet
3456 int
3457 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3458     struct vnode **vpp)
3459 {
3460 
3461 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3462 }
3463 #endif
3464 
3465 /*
3466  * Handle the last reference to a file being closed.
3467  *
3468  * Without the noinline attribute clang keeps inlining the func thorough this
3469  * file when fdrop is used.
3470  */
3471 int __noinline
3472 _fdrop(struct file *fp, struct thread *td)
3473 {
3474 	int error;
3475 #ifdef INVARIANTS
3476 	int count;
3477 
3478 	count = refcount_load(&fp->f_count);
3479 	if (count != 0)
3480 		panic("fdrop: fp %p count %d", fp, count);
3481 #endif
3482 	error = fo_close(fp, td);
3483 	atomic_subtract_int(&openfiles, 1);
3484 	crfree(fp->f_cred);
3485 	free(fp->f_advice, M_FADVISE);
3486 	uma_zfree(file_zone, fp);
3487 
3488 	return (error);
3489 }
3490 
3491 /*
3492  * Apply an advisory lock on a file descriptor.
3493  *
3494  * Just attempt to get a record lock of the requested type on the entire file
3495  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3496  */
3497 #ifndef _SYS_SYSPROTO_H_
3498 struct flock_args {
3499 	int	fd;
3500 	int	how;
3501 };
3502 #endif
3503 /* ARGSUSED */
3504 int
3505 sys_flock(struct thread *td, struct flock_args *uap)
3506 {
3507 	struct file *fp;
3508 	struct vnode *vp;
3509 	struct flock lf;
3510 	int error;
3511 
3512 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3513 	if (error != 0)
3514 		return (error);
3515 	error = EOPNOTSUPP;
3516 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3517 		goto done;
3518 	}
3519 	if (fp->f_ops == &path_fileops) {
3520 		goto done;
3521 	}
3522 
3523 	error = 0;
3524 	vp = fp->f_vnode;
3525 	lf.l_whence = SEEK_SET;
3526 	lf.l_start = 0;
3527 	lf.l_len = 0;
3528 	if (uap->how & LOCK_UN) {
3529 		lf.l_type = F_UNLCK;
3530 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3531 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3532 		goto done;
3533 	}
3534 	if (uap->how & LOCK_EX)
3535 		lf.l_type = F_WRLCK;
3536 	else if (uap->how & LOCK_SH)
3537 		lf.l_type = F_RDLCK;
3538 	else {
3539 		error = EBADF;
3540 		goto done;
3541 	}
3542 	atomic_set_int(&fp->f_flag, FHASLOCK);
3543 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3544 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3545 done:
3546 	fdrop(fp, td);
3547 	return (error);
3548 }
3549 /*
3550  * Duplicate the specified descriptor to a free descriptor.
3551  */
3552 int
3553 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3554     int openerror, int *indxp)
3555 {
3556 	struct filedescent *newfde, *oldfde;
3557 	struct file *fp;
3558 	u_long *ioctls;
3559 	int error, indx;
3560 
3561 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3562 	    ("unexpected error %d in %s", openerror, __func__));
3563 
3564 	/*
3565 	 * If the to-be-dup'd fd number is greater than the allowed number
3566 	 * of file descriptors, or the fd to be dup'd has already been
3567 	 * closed, then reject.
3568 	 */
3569 	FILEDESC_XLOCK(fdp);
3570 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3571 		FILEDESC_XUNLOCK(fdp);
3572 		return (EBADF);
3573 	}
3574 
3575 	error = fdalloc(td, 0, &indx);
3576 	if (error != 0) {
3577 		FILEDESC_XUNLOCK(fdp);
3578 		return (error);
3579 	}
3580 
3581 	/*
3582 	 * There are two cases of interest here.
3583 	 *
3584 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3585 	 *
3586 	 * For ENXIO steal away the file structure from (dfd) and store it in
3587 	 * (indx).  (dfd) is effectively closed by this operation.
3588 	 */
3589 	switch (openerror) {
3590 	case ENODEV:
3591 		/*
3592 		 * Check that the mode the file is being opened for is a
3593 		 * subset of the mode of the existing descriptor.
3594 		 */
3595 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3596 			fdunused(fdp, indx);
3597 			FILEDESC_XUNLOCK(fdp);
3598 			return (EACCES);
3599 		}
3600 		if (!fhold(fp)) {
3601 			fdunused(fdp, indx);
3602 			FILEDESC_XUNLOCK(fdp);
3603 			return (EBADF);
3604 		}
3605 		newfde = &fdp->fd_ofiles[indx];
3606 		oldfde = &fdp->fd_ofiles[dfd];
3607 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3608 #ifdef CAPABILITIES
3609 		seqc_write_begin(&newfde->fde_seqc);
3610 #endif
3611 		memcpy(newfde, oldfde, fde_change_size);
3612 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3613 		    ioctls);
3614 #ifdef CAPABILITIES
3615 		seqc_write_end(&newfde->fde_seqc);
3616 #endif
3617 		break;
3618 	case ENXIO:
3619 		/*
3620 		 * Steal away the file pointer from dfd and stuff it into indx.
3621 		 */
3622 		newfde = &fdp->fd_ofiles[indx];
3623 		oldfde = &fdp->fd_ofiles[dfd];
3624 #ifdef CAPABILITIES
3625 		seqc_write_begin(&newfde->fde_seqc);
3626 #endif
3627 		memcpy(newfde, oldfde, fde_change_size);
3628 		oldfde->fde_file = NULL;
3629 		fdunused(fdp, dfd);
3630 #ifdef CAPABILITIES
3631 		seqc_write_end(&newfde->fde_seqc);
3632 #endif
3633 		break;
3634 	}
3635 	FILEDESC_XUNLOCK(fdp);
3636 	*indxp = indx;
3637 	return (0);
3638 }
3639 
3640 /*
3641  * This sysctl determines if we will allow a process to chroot(2) if it
3642  * has a directory open:
3643  *	0: disallowed for all processes.
3644  *	1: allowed for processes that were not already chroot(2)'ed.
3645  *	2: allowed for all processes.
3646  */
3647 
3648 static int chroot_allow_open_directories = 1;
3649 
3650 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3651     &chroot_allow_open_directories, 0,
3652     "Allow a process to chroot(2) if it has a directory open");
3653 
3654 /*
3655  * Helper function for raised chroot(2) security function:  Refuse if
3656  * any filedescriptors are open directories.
3657  */
3658 static int
3659 chroot_refuse_vdir_fds(struct filedesc *fdp)
3660 {
3661 	struct vnode *vp;
3662 	struct file *fp;
3663 	int fd, lastfile;
3664 
3665 	FILEDESC_LOCK_ASSERT(fdp);
3666 
3667 	lastfile = fdlastfile(fdp);
3668 	for (fd = 0; fd <= lastfile; fd++) {
3669 		fp = fget_locked(fdp, fd);
3670 		if (fp == NULL)
3671 			continue;
3672 		if (fp->f_type == DTYPE_VNODE) {
3673 			vp = fp->f_vnode;
3674 			if (vp->v_type == VDIR)
3675 				return (EPERM);
3676 		}
3677 	}
3678 	return (0);
3679 }
3680 
3681 static void
3682 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3683 {
3684 
3685 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3686 		vrefact(oldpwd->pwd_cdir);
3687 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3688 	}
3689 
3690 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3691 		vrefact(oldpwd->pwd_rdir);
3692 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3693 	}
3694 
3695 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3696 		vrefact(oldpwd->pwd_jdir);
3697 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3698 	}
3699 }
3700 
3701 struct pwd *
3702 pwd_hold_pwddesc(struct pwddesc *pdp)
3703 {
3704 	struct pwd *pwd;
3705 
3706 	PWDDESC_ASSERT_XLOCKED(pdp);
3707 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3708 	if (pwd != NULL)
3709 		refcount_acquire(&pwd->pwd_refcount);
3710 	return (pwd);
3711 }
3712 
3713 bool
3714 pwd_hold_smr(struct pwd *pwd)
3715 {
3716 
3717 	MPASS(pwd != NULL);
3718 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3719 		return (true);
3720 	}
3721 	return (false);
3722 }
3723 
3724 struct pwd *
3725 pwd_hold(struct thread *td)
3726 {
3727 	struct pwddesc *pdp;
3728 	struct pwd *pwd;
3729 
3730 	pdp = td->td_proc->p_pd;
3731 
3732 	vfs_smr_enter();
3733 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3734 	if (pwd_hold_smr(pwd)) {
3735 		vfs_smr_exit();
3736 		return (pwd);
3737 	}
3738 	vfs_smr_exit();
3739 	PWDDESC_XLOCK(pdp);
3740 	pwd = pwd_hold_pwddesc(pdp);
3741 	MPASS(pwd != NULL);
3742 	PWDDESC_XUNLOCK(pdp);
3743 	return (pwd);
3744 }
3745 
3746 struct pwd *
3747 pwd_hold_proc(struct proc *p)
3748 {
3749 	struct pwddesc *pdp;
3750 	struct pwd *pwd;
3751 
3752 	PROC_ASSERT_HELD(p);
3753 	PROC_LOCK(p);
3754 	pdp = pdhold(p);
3755 	MPASS(pdp != NULL);
3756 	PROC_UNLOCK(p);
3757 
3758 	PWDDESC_XLOCK(pdp);
3759 	pwd = pwd_hold_pwddesc(pdp);
3760 	MPASS(pwd != NULL);
3761 	PWDDESC_XUNLOCK(pdp);
3762 	pddrop(pdp);
3763 	return (pwd);
3764 }
3765 
3766 static struct pwd *
3767 pwd_alloc(void)
3768 {
3769 	struct pwd *pwd;
3770 
3771 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3772 	bzero(pwd, sizeof(*pwd));
3773 	refcount_init(&pwd->pwd_refcount, 1);
3774 	return (pwd);
3775 }
3776 
3777 void
3778 pwd_drop(struct pwd *pwd)
3779 {
3780 
3781 	if (!refcount_release(&pwd->pwd_refcount))
3782 		return;
3783 
3784 	if (pwd->pwd_cdir != NULL)
3785 		vrele(pwd->pwd_cdir);
3786 	if (pwd->pwd_rdir != NULL)
3787 		vrele(pwd->pwd_rdir);
3788 	if (pwd->pwd_jdir != NULL)
3789 		vrele(pwd->pwd_jdir);
3790 	uma_zfree_smr(pwd_zone, pwd);
3791 }
3792 
3793 /*
3794 * The caller is responsible for invoking priv_check() and
3795 * mac_vnode_check_chroot() to authorize this operation.
3796 */
3797 int
3798 pwd_chroot(struct thread *td, struct vnode *vp)
3799 {
3800 	struct pwddesc *pdp;
3801 	struct filedesc *fdp;
3802 	struct pwd *newpwd, *oldpwd;
3803 	int error;
3804 
3805 	fdp = td->td_proc->p_fd;
3806 	pdp = td->td_proc->p_pd;
3807 	newpwd = pwd_alloc();
3808 	FILEDESC_SLOCK(fdp);
3809 	PWDDESC_XLOCK(pdp);
3810 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3811 	if (chroot_allow_open_directories == 0 ||
3812 	    (chroot_allow_open_directories == 1 &&
3813 	    oldpwd->pwd_rdir != rootvnode)) {
3814 		error = chroot_refuse_vdir_fds(fdp);
3815 		FILEDESC_SUNLOCK(fdp);
3816 		if (error != 0) {
3817 			PWDDESC_XUNLOCK(pdp);
3818 			pwd_drop(newpwd);
3819 			return (error);
3820 		}
3821 	} else {
3822 		FILEDESC_SUNLOCK(fdp);
3823 	}
3824 
3825 	vrefact(vp);
3826 	newpwd->pwd_rdir = vp;
3827 	if (oldpwd->pwd_jdir == NULL) {
3828 		vrefact(vp);
3829 		newpwd->pwd_jdir = vp;
3830 	}
3831 	pwd_fill(oldpwd, newpwd);
3832 	pwd_set(pdp, newpwd);
3833 	PWDDESC_XUNLOCK(pdp);
3834 	pwd_drop(oldpwd);
3835 	return (0);
3836 }
3837 
3838 void
3839 pwd_chdir(struct thread *td, struct vnode *vp)
3840 {
3841 	struct pwddesc *pdp;
3842 	struct pwd *newpwd, *oldpwd;
3843 
3844 	VNPASS(vp->v_usecount > 0, vp);
3845 
3846 	newpwd = pwd_alloc();
3847 	pdp = td->td_proc->p_pd;
3848 	PWDDESC_XLOCK(pdp);
3849 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3850 	newpwd->pwd_cdir = vp;
3851 	pwd_fill(oldpwd, newpwd);
3852 	pwd_set(pdp, newpwd);
3853 	PWDDESC_XUNLOCK(pdp);
3854 	pwd_drop(oldpwd);
3855 }
3856 
3857 /*
3858  * jail_attach(2) changes both root and working directories.
3859  */
3860 int
3861 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3862 {
3863 	struct pwddesc *pdp;
3864 	struct filedesc *fdp;
3865 	struct pwd *newpwd, *oldpwd;
3866 	int error;
3867 
3868 	fdp = td->td_proc->p_fd;
3869 	pdp = td->td_proc->p_pd;
3870 	newpwd = pwd_alloc();
3871 	FILEDESC_SLOCK(fdp);
3872 	PWDDESC_XLOCK(pdp);
3873 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3874 	error = chroot_refuse_vdir_fds(fdp);
3875 	FILEDESC_SUNLOCK(fdp);
3876 	if (error != 0) {
3877 		PWDDESC_XUNLOCK(pdp);
3878 		pwd_drop(newpwd);
3879 		return (error);
3880 	}
3881 
3882 	vrefact(vp);
3883 	newpwd->pwd_rdir = vp;
3884 	vrefact(vp);
3885 	newpwd->pwd_cdir = vp;
3886 	if (oldpwd->pwd_jdir == NULL) {
3887 		vrefact(vp);
3888 		newpwd->pwd_jdir = vp;
3889 	}
3890 	pwd_fill(oldpwd, newpwd);
3891 	pwd_set(pdp, newpwd);
3892 	PWDDESC_XUNLOCK(pdp);
3893 	pwd_drop(oldpwd);
3894 	return (0);
3895 }
3896 
3897 void
3898 pwd_ensure_dirs(void)
3899 {
3900 	struct pwddesc *pdp;
3901 	struct pwd *oldpwd, *newpwd;
3902 
3903 	pdp = curproc->p_pd;
3904 	PWDDESC_XLOCK(pdp);
3905 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3906 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3907 		PWDDESC_XUNLOCK(pdp);
3908 		return;
3909 	}
3910 	PWDDESC_XUNLOCK(pdp);
3911 
3912 	newpwd = pwd_alloc();
3913 	PWDDESC_XLOCK(pdp);
3914 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3915 	pwd_fill(oldpwd, newpwd);
3916 	if (newpwd->pwd_cdir == NULL) {
3917 		vrefact(rootvnode);
3918 		newpwd->pwd_cdir = rootvnode;
3919 	}
3920 	if (newpwd->pwd_rdir == NULL) {
3921 		vrefact(rootvnode);
3922 		newpwd->pwd_rdir = rootvnode;
3923 	}
3924 	pwd_set(pdp, newpwd);
3925 	PWDDESC_XUNLOCK(pdp);
3926 	pwd_drop(oldpwd);
3927 }
3928 
3929 void
3930 pwd_set_rootvnode(void)
3931 {
3932 	struct pwddesc *pdp;
3933 	struct pwd *oldpwd, *newpwd;
3934 
3935 	pdp = curproc->p_pd;
3936 
3937 	newpwd = pwd_alloc();
3938 	PWDDESC_XLOCK(pdp);
3939 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3940 	vrefact(rootvnode);
3941 	newpwd->pwd_cdir = rootvnode;
3942 	vrefact(rootvnode);
3943 	newpwd->pwd_rdir = rootvnode;
3944 	pwd_fill(oldpwd, newpwd);
3945 	pwd_set(pdp, newpwd);
3946 	PWDDESC_XUNLOCK(pdp);
3947 	pwd_drop(oldpwd);
3948 }
3949 
3950 /*
3951  * Scan all active processes and prisons to see if any of them have a current
3952  * or root directory of `olddp'. If so, replace them with the new mount point.
3953  */
3954 void
3955 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3956 {
3957 	struct pwddesc *pdp;
3958 	struct pwd *newpwd, *oldpwd;
3959 	struct prison *pr;
3960 	struct proc *p;
3961 	int nrele;
3962 
3963 	if (vrefcnt(olddp) == 1)
3964 		return;
3965 	nrele = 0;
3966 	newpwd = pwd_alloc();
3967 	sx_slock(&allproc_lock);
3968 	FOREACH_PROC_IN_SYSTEM(p) {
3969 		PROC_LOCK(p);
3970 		pdp = pdhold(p);
3971 		PROC_UNLOCK(p);
3972 		if (pdp == NULL)
3973 			continue;
3974 		PWDDESC_XLOCK(pdp);
3975 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3976 		if (oldpwd == NULL ||
3977 		    (oldpwd->pwd_cdir != olddp &&
3978 		    oldpwd->pwd_rdir != olddp &&
3979 		    oldpwd->pwd_jdir != olddp)) {
3980 			PWDDESC_XUNLOCK(pdp);
3981 			pddrop(pdp);
3982 			continue;
3983 		}
3984 		if (oldpwd->pwd_cdir == olddp) {
3985 			vrefact(newdp);
3986 			newpwd->pwd_cdir = newdp;
3987 		}
3988 		if (oldpwd->pwd_rdir == olddp) {
3989 			vrefact(newdp);
3990 			newpwd->pwd_rdir = newdp;
3991 		}
3992 		if (oldpwd->pwd_jdir == olddp) {
3993 			vrefact(newdp);
3994 			newpwd->pwd_jdir = newdp;
3995 		}
3996 		pwd_fill(oldpwd, newpwd);
3997 		pwd_set(pdp, newpwd);
3998 		PWDDESC_XUNLOCK(pdp);
3999 		pwd_drop(oldpwd);
4000 		pddrop(pdp);
4001 		newpwd = pwd_alloc();
4002 	}
4003 	sx_sunlock(&allproc_lock);
4004 	pwd_drop(newpwd);
4005 	if (rootvnode == olddp) {
4006 		vrefact(newdp);
4007 		rootvnode = newdp;
4008 		nrele++;
4009 	}
4010 	mtx_lock(&prison0.pr_mtx);
4011 	if (prison0.pr_root == olddp) {
4012 		vrefact(newdp);
4013 		prison0.pr_root = newdp;
4014 		nrele++;
4015 	}
4016 	mtx_unlock(&prison0.pr_mtx);
4017 	sx_slock(&allprison_lock);
4018 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4019 		mtx_lock(&pr->pr_mtx);
4020 		if (pr->pr_root == olddp) {
4021 			vrefact(newdp);
4022 			pr->pr_root = newdp;
4023 			nrele++;
4024 		}
4025 		mtx_unlock(&pr->pr_mtx);
4026 	}
4027 	sx_sunlock(&allprison_lock);
4028 	while (nrele--)
4029 		vrele(olddp);
4030 }
4031 
4032 struct filedesc_to_leader *
4033 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4034 {
4035 	struct filedesc_to_leader *fdtol;
4036 
4037 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4038 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4039 	fdtol->fdl_refcount = 1;
4040 	fdtol->fdl_holdcount = 0;
4041 	fdtol->fdl_wakeup = 0;
4042 	fdtol->fdl_leader = leader;
4043 	if (old != NULL) {
4044 		FILEDESC_XLOCK(fdp);
4045 		fdtol->fdl_next = old->fdl_next;
4046 		fdtol->fdl_prev = old;
4047 		old->fdl_next = fdtol;
4048 		fdtol->fdl_next->fdl_prev = fdtol;
4049 		FILEDESC_XUNLOCK(fdp);
4050 	} else {
4051 		fdtol->fdl_next = fdtol;
4052 		fdtol->fdl_prev = fdtol;
4053 	}
4054 	return (fdtol);
4055 }
4056 
4057 static int
4058 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4059 {
4060 	NDSLOTTYPE *map;
4061 	struct filedesc *fdp;
4062 	u_int namelen;
4063 	int count, off, minoff;
4064 
4065 	namelen = arg2;
4066 	if (namelen != 1)
4067 		return (EINVAL);
4068 
4069 	if (*(int *)arg1 != 0)
4070 		return (EINVAL);
4071 
4072 	fdp = curproc->p_fd;
4073 	count = 0;
4074 	FILEDESC_SLOCK(fdp);
4075 	map = fdp->fd_map;
4076 	off = NDSLOT(fdp->fd_nfiles - 1);
4077 	for (minoff = NDSLOT(0); off >= minoff; --off)
4078 		count += bitcountl(map[off]);
4079 	FILEDESC_SUNLOCK(fdp);
4080 
4081 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4082 }
4083 
4084 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4085     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4086     "Number of open file descriptors");
4087 
4088 /*
4089  * Get file structures globally.
4090  */
4091 static int
4092 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4093 {
4094 	struct xfile xf;
4095 	struct filedesc *fdp;
4096 	struct file *fp;
4097 	struct proc *p;
4098 	int error, n, lastfile;
4099 
4100 	error = sysctl_wire_old_buffer(req, 0);
4101 	if (error != 0)
4102 		return (error);
4103 	if (req->oldptr == NULL) {
4104 		n = 0;
4105 		sx_slock(&allproc_lock);
4106 		FOREACH_PROC_IN_SYSTEM(p) {
4107 			PROC_LOCK(p);
4108 			if (p->p_state == PRS_NEW) {
4109 				PROC_UNLOCK(p);
4110 				continue;
4111 			}
4112 			fdp = fdhold(p);
4113 			PROC_UNLOCK(p);
4114 			if (fdp == NULL)
4115 				continue;
4116 			/* overestimates sparse tables. */
4117 			n += fdp->fd_nfiles;
4118 			fddrop(fdp);
4119 		}
4120 		sx_sunlock(&allproc_lock);
4121 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4122 	}
4123 	error = 0;
4124 	bzero(&xf, sizeof(xf));
4125 	xf.xf_size = sizeof(xf);
4126 	sx_slock(&allproc_lock);
4127 	FOREACH_PROC_IN_SYSTEM(p) {
4128 		PROC_LOCK(p);
4129 		if (p->p_state == PRS_NEW) {
4130 			PROC_UNLOCK(p);
4131 			continue;
4132 		}
4133 		if (p_cansee(req->td, p) != 0) {
4134 			PROC_UNLOCK(p);
4135 			continue;
4136 		}
4137 		xf.xf_pid = p->p_pid;
4138 		xf.xf_uid = p->p_ucred->cr_uid;
4139 		fdp = fdhold(p);
4140 		PROC_UNLOCK(p);
4141 		if (fdp == NULL)
4142 			continue;
4143 		FILEDESC_SLOCK(fdp);
4144 		lastfile = fdlastfile(fdp);
4145 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4146 		    n++) {
4147 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4148 				continue;
4149 			xf.xf_fd = n;
4150 			xf.xf_file = (uintptr_t)fp;
4151 			xf.xf_data = (uintptr_t)fp->f_data;
4152 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4153 			xf.xf_type = (uintptr_t)fp->f_type;
4154 			xf.xf_count = refcount_load(&fp->f_count);
4155 			xf.xf_msgcount = 0;
4156 			xf.xf_offset = foffset_get(fp);
4157 			xf.xf_flag = fp->f_flag;
4158 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4159 			if (error)
4160 				break;
4161 		}
4162 		FILEDESC_SUNLOCK(fdp);
4163 		fddrop(fdp);
4164 		if (error)
4165 			break;
4166 	}
4167 	sx_sunlock(&allproc_lock);
4168 	return (error);
4169 }
4170 
4171 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4172     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4173 
4174 #ifdef KINFO_FILE_SIZE
4175 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4176 #endif
4177 
4178 static int
4179 xlate_fflags(int fflags)
4180 {
4181 	static const struct {
4182 		int	fflag;
4183 		int	kf_fflag;
4184 	} fflags_table[] = {
4185 		{ FAPPEND, KF_FLAG_APPEND },
4186 		{ FASYNC, KF_FLAG_ASYNC },
4187 		{ FFSYNC, KF_FLAG_FSYNC },
4188 		{ FHASLOCK, KF_FLAG_HASLOCK },
4189 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4190 		{ FREAD, KF_FLAG_READ },
4191 		{ FWRITE, KF_FLAG_WRITE },
4192 		{ O_CREAT, KF_FLAG_CREAT },
4193 		{ O_DIRECT, KF_FLAG_DIRECT },
4194 		{ O_EXCL, KF_FLAG_EXCL },
4195 		{ O_EXEC, KF_FLAG_EXEC },
4196 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4197 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4198 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4199 		{ O_TRUNC, KF_FLAG_TRUNC }
4200 	};
4201 	unsigned int i;
4202 	int kflags;
4203 
4204 	kflags = 0;
4205 	for (i = 0; i < nitems(fflags_table); i++)
4206 		if (fflags & fflags_table[i].fflag)
4207 			kflags |=  fflags_table[i].kf_fflag;
4208 	return (kflags);
4209 }
4210 
4211 /* Trim unused data from kf_path by truncating the structure size. */
4212 void
4213 pack_kinfo(struct kinfo_file *kif)
4214 {
4215 
4216 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4217 	    strlen(kif->kf_path) + 1;
4218 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4219 }
4220 
4221 static void
4222 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4223     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4224 {
4225 	int error;
4226 
4227 	bzero(kif, sizeof(*kif));
4228 
4229 	/* Set a default type to allow for empty fill_kinfo() methods. */
4230 	kif->kf_type = KF_TYPE_UNKNOWN;
4231 	kif->kf_flags = xlate_fflags(fp->f_flag);
4232 	if (rightsp != NULL)
4233 		kif->kf_cap_rights = *rightsp;
4234 	else
4235 		cap_rights_init_zero(&kif->kf_cap_rights);
4236 	kif->kf_fd = fd;
4237 	kif->kf_ref_count = refcount_load(&fp->f_count);
4238 	kif->kf_offset = foffset_get(fp);
4239 
4240 	/*
4241 	 * This may drop the filedesc lock, so the 'fp' cannot be
4242 	 * accessed after this call.
4243 	 */
4244 	error = fo_fill_kinfo(fp, kif, fdp);
4245 	if (error == 0)
4246 		kif->kf_status |= KF_ATTR_VALID;
4247 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4248 		pack_kinfo(kif);
4249 	else
4250 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4251 }
4252 
4253 static void
4254 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4255     struct kinfo_file *kif, int flags)
4256 {
4257 	int error;
4258 
4259 	bzero(kif, sizeof(*kif));
4260 
4261 	kif->kf_type = KF_TYPE_VNODE;
4262 	error = vn_fill_kinfo_vnode(vp, kif);
4263 	if (error == 0)
4264 		kif->kf_status |= KF_ATTR_VALID;
4265 	kif->kf_flags = xlate_fflags(fflags);
4266 	cap_rights_init_zero(&kif->kf_cap_rights);
4267 	kif->kf_fd = fd;
4268 	kif->kf_ref_count = -1;
4269 	kif->kf_offset = -1;
4270 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4271 		pack_kinfo(kif);
4272 	else
4273 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4274 	vrele(vp);
4275 }
4276 
4277 struct export_fd_buf {
4278 	struct filedesc		*fdp;
4279 	struct pwddesc	*pdp;
4280 	struct sbuf 		*sb;
4281 	ssize_t			remainder;
4282 	struct kinfo_file	kif;
4283 	int			flags;
4284 };
4285 
4286 static int
4287 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4288 {
4289 	struct kinfo_file *kif;
4290 
4291 	kif = &efbuf->kif;
4292 	if (efbuf->remainder != -1) {
4293 		if (efbuf->remainder < kif->kf_structsize) {
4294 			/* Terminate export. */
4295 			efbuf->remainder = 0;
4296 			return (0);
4297 		}
4298 		efbuf->remainder -= kif->kf_structsize;
4299 	}
4300 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4301 }
4302 
4303 static int
4304 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4305     struct export_fd_buf *efbuf)
4306 {
4307 	int error;
4308 
4309 	if (efbuf->remainder == 0)
4310 		return (0);
4311 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4312 	    efbuf->flags);
4313 	FILEDESC_SUNLOCK(efbuf->fdp);
4314 	error = export_kinfo_to_sb(efbuf);
4315 	FILEDESC_SLOCK(efbuf->fdp);
4316 	return (error);
4317 }
4318 
4319 static int
4320 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4321     struct export_fd_buf *efbuf)
4322 {
4323 	int error;
4324 
4325 	if (efbuf->remainder == 0)
4326 		return (0);
4327 	if (efbuf->pdp != NULL)
4328 		PWDDESC_XUNLOCK(efbuf->pdp);
4329 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4330 	error = export_kinfo_to_sb(efbuf);
4331 	if (efbuf->pdp != NULL)
4332 		PWDDESC_XLOCK(efbuf->pdp);
4333 	return (error);
4334 }
4335 
4336 /*
4337  * Store a process file descriptor information to sbuf.
4338  *
4339  * Takes a locked proc as argument, and returns with the proc unlocked.
4340  */
4341 int
4342 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4343     int flags)
4344 {
4345 	struct file *fp;
4346 	struct filedesc *fdp;
4347 	struct pwddesc *pdp;
4348 	struct export_fd_buf *efbuf;
4349 	struct vnode *cttyvp, *textvp, *tracevp;
4350 	struct pwd *pwd;
4351 	int error, i, lastfile;
4352 	cap_rights_t rights;
4353 
4354 	PROC_LOCK_ASSERT(p, MA_OWNED);
4355 
4356 	/* ktrace vnode */
4357 	tracevp = ktr_get_tracevp(p, true);
4358 	/* text vnode */
4359 	textvp = p->p_textvp;
4360 	if (textvp != NULL)
4361 		vrefact(textvp);
4362 	/* Controlling tty. */
4363 	cttyvp = NULL;
4364 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4365 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4366 		if (cttyvp != NULL)
4367 			vrefact(cttyvp);
4368 	}
4369 	fdp = fdhold(p);
4370 	pdp = pdhold(p);
4371 	PROC_UNLOCK(p);
4372 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4373 	efbuf->fdp = NULL;
4374 	efbuf->pdp = NULL;
4375 	efbuf->sb = sb;
4376 	efbuf->remainder = maxlen;
4377 	efbuf->flags = flags;
4378 	if (tracevp != NULL)
4379 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4380 		    efbuf);
4381 	if (textvp != NULL)
4382 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4383 	if (cttyvp != NULL)
4384 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4385 		    efbuf);
4386 	error = 0;
4387 	if (pdp == NULL || fdp == NULL)
4388 		goto fail;
4389 	efbuf->fdp = fdp;
4390 	efbuf->pdp = pdp;
4391 	PWDDESC_XLOCK(pdp);
4392 	pwd = pwd_hold_pwddesc(pdp);
4393 	if (pwd != NULL) {
4394 		/* working directory */
4395 		if (pwd->pwd_cdir != NULL) {
4396 			vrefact(pwd->pwd_cdir);
4397 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4398 			    FREAD, efbuf);
4399 		}
4400 		/* root directory */
4401 		if (pwd->pwd_rdir != NULL) {
4402 			vrefact(pwd->pwd_rdir);
4403 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4404 			    FREAD, efbuf);
4405 		}
4406 		/* jail directory */
4407 		if (pwd->pwd_jdir != NULL) {
4408 			vrefact(pwd->pwd_jdir);
4409 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4410 			    FREAD, efbuf);
4411 		}
4412 	}
4413 	PWDDESC_XUNLOCK(pdp);
4414 	if (pwd != NULL)
4415 		pwd_drop(pwd);
4416 	FILEDESC_SLOCK(fdp);
4417 	lastfile = fdlastfile(fdp);
4418 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4419 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4420 			continue;
4421 #ifdef CAPABILITIES
4422 		rights = *cap_rights(fdp, i);
4423 #else /* !CAPABILITIES */
4424 		rights = cap_no_rights;
4425 #endif
4426 		/*
4427 		 * Create sysctl entry.  It is OK to drop the filedesc
4428 		 * lock inside of export_file_to_sb() as we will
4429 		 * re-validate and re-evaluate its properties when the
4430 		 * loop continues.
4431 		 */
4432 		error = export_file_to_sb(fp, i, &rights, efbuf);
4433 		if (error != 0 || efbuf->remainder == 0)
4434 			break;
4435 	}
4436 	FILEDESC_SUNLOCK(fdp);
4437 fail:
4438 	if (fdp != NULL)
4439 		fddrop(fdp);
4440 	if (pdp != NULL)
4441 		pddrop(pdp);
4442 	free(efbuf, M_TEMP);
4443 	return (error);
4444 }
4445 
4446 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4447 
4448 /*
4449  * Get per-process file descriptors for use by procstat(1), et al.
4450  */
4451 static int
4452 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4453 {
4454 	struct sbuf sb;
4455 	struct proc *p;
4456 	ssize_t maxlen;
4457 	u_int namelen;
4458 	int error, error2, *name;
4459 
4460 	namelen = arg2;
4461 	if (namelen != 1)
4462 		return (EINVAL);
4463 
4464 	name = (int *)arg1;
4465 
4466 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4467 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4468 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4469 	if (error != 0) {
4470 		sbuf_delete(&sb);
4471 		return (error);
4472 	}
4473 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4474 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4475 	    KERN_FILEDESC_PACK_KINFO);
4476 	error2 = sbuf_finish(&sb);
4477 	sbuf_delete(&sb);
4478 	return (error != 0 ? error : error2);
4479 }
4480 
4481 #ifdef COMPAT_FREEBSD7
4482 #ifdef KINFO_OFILE_SIZE
4483 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4484 #endif
4485 
4486 static void
4487 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4488 {
4489 
4490 	okif->kf_structsize = sizeof(*okif);
4491 	okif->kf_type = kif->kf_type;
4492 	okif->kf_fd = kif->kf_fd;
4493 	okif->kf_ref_count = kif->kf_ref_count;
4494 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4495 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4496 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4497 	okif->kf_offset = kif->kf_offset;
4498 	if (kif->kf_type == KF_TYPE_VNODE)
4499 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4500 	else
4501 		okif->kf_vnode_type = KF_VTYPE_VNON;
4502 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4503 	if (kif->kf_type == KF_TYPE_SOCKET) {
4504 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4505 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4506 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4507 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4508 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4509 	} else {
4510 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4511 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4512 	}
4513 }
4514 
4515 static int
4516 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4517     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4518 {
4519 	int error;
4520 
4521 	vrefact(vp);
4522 	PWDDESC_XUNLOCK(pdp);
4523 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4524 	kinfo_to_okinfo(kif, okif);
4525 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4526 	PWDDESC_XLOCK(pdp);
4527 	return (error);
4528 }
4529 
4530 /*
4531  * Get per-process file descriptors for use by procstat(1), et al.
4532  */
4533 static int
4534 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4535 {
4536 	struct kinfo_ofile *okif;
4537 	struct kinfo_file *kif;
4538 	struct filedesc *fdp;
4539 	struct pwddesc *pdp;
4540 	struct pwd *pwd;
4541 	u_int namelen;
4542 	int error, i, lastfile, *name;
4543 	struct file *fp;
4544 	struct proc *p;
4545 
4546 	namelen = arg2;
4547 	if (namelen != 1)
4548 		return (EINVAL);
4549 
4550 	name = (int *)arg1;
4551 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4552 	if (error != 0)
4553 		return (error);
4554 	fdp = fdhold(p);
4555 	if (fdp != NULL)
4556 		pdp = pdhold(p);
4557 	PROC_UNLOCK(p);
4558 	if (fdp == NULL || pdp == NULL) {
4559 		if (fdp != NULL)
4560 			fddrop(fdp);
4561 		return (ENOENT);
4562 	}
4563 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4564 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4565 	PWDDESC_XLOCK(pdp);
4566 	pwd = pwd_hold_pwddesc(pdp);
4567 	if (pwd != NULL) {
4568 		if (pwd->pwd_cdir != NULL)
4569 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4570 			    okif, pdp, req);
4571 		if (pwd->pwd_rdir != NULL)
4572 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4573 			    okif, pdp, req);
4574 		if (pwd->pwd_jdir != NULL)
4575 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4576 			    okif, pdp, req);
4577 	}
4578 	PWDDESC_XUNLOCK(pdp);
4579 	if (pwd != NULL)
4580 		pwd_drop(pwd);
4581 	FILEDESC_SLOCK(fdp);
4582 	lastfile = fdlastfile(fdp);
4583 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4584 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4585 			continue;
4586 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4587 		    KERN_FILEDESC_PACK_KINFO);
4588 		FILEDESC_SUNLOCK(fdp);
4589 		kinfo_to_okinfo(kif, okif);
4590 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4591 		FILEDESC_SLOCK(fdp);
4592 		if (error)
4593 			break;
4594 	}
4595 	FILEDESC_SUNLOCK(fdp);
4596 	fddrop(fdp);
4597 	pddrop(pdp);
4598 	free(kif, M_TEMP);
4599 	free(okif, M_TEMP);
4600 	return (0);
4601 }
4602 
4603 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4604     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4605     "Process ofiledesc entries");
4606 #endif	/* COMPAT_FREEBSD7 */
4607 
4608 int
4609 vntype_to_kinfo(int vtype)
4610 {
4611 	struct {
4612 		int	vtype;
4613 		int	kf_vtype;
4614 	} vtypes_table[] = {
4615 		{ VBAD, KF_VTYPE_VBAD },
4616 		{ VBLK, KF_VTYPE_VBLK },
4617 		{ VCHR, KF_VTYPE_VCHR },
4618 		{ VDIR, KF_VTYPE_VDIR },
4619 		{ VFIFO, KF_VTYPE_VFIFO },
4620 		{ VLNK, KF_VTYPE_VLNK },
4621 		{ VNON, KF_VTYPE_VNON },
4622 		{ VREG, KF_VTYPE_VREG },
4623 		{ VSOCK, KF_VTYPE_VSOCK }
4624 	};
4625 	unsigned int i;
4626 
4627 	/*
4628 	 * Perform vtype translation.
4629 	 */
4630 	for (i = 0; i < nitems(vtypes_table); i++)
4631 		if (vtypes_table[i].vtype == vtype)
4632 			return (vtypes_table[i].kf_vtype);
4633 
4634 	return (KF_VTYPE_UNKNOWN);
4635 }
4636 
4637 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4638     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4639     "Process filedesc entries");
4640 
4641 /*
4642  * Store a process current working directory information to sbuf.
4643  *
4644  * Takes a locked proc as argument, and returns with the proc unlocked.
4645  */
4646 int
4647 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4648 {
4649 	struct pwddesc *pdp;
4650 	struct pwd *pwd;
4651 	struct export_fd_buf *efbuf;
4652 	struct vnode *cdir;
4653 	int error;
4654 
4655 	PROC_LOCK_ASSERT(p, MA_OWNED);
4656 
4657 	pdp = pdhold(p);
4658 	PROC_UNLOCK(p);
4659 	if (pdp == NULL)
4660 		return (EINVAL);
4661 
4662 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4663 	efbuf->pdp = pdp;
4664 	efbuf->sb = sb;
4665 	efbuf->remainder = maxlen;
4666 
4667 	PWDDESC_XLOCK(pdp);
4668 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4669 	cdir = pwd->pwd_cdir;
4670 	if (cdir == NULL) {
4671 		error = EINVAL;
4672 	} else {
4673 		vrefact(cdir);
4674 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4675 	}
4676 	PWDDESC_XUNLOCK(pdp);
4677 	pddrop(pdp);
4678 	free(efbuf, M_TEMP);
4679 	return (error);
4680 }
4681 
4682 /*
4683  * Get per-process current working directory.
4684  */
4685 static int
4686 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4687 {
4688 	struct sbuf sb;
4689 	struct proc *p;
4690 	ssize_t maxlen;
4691 	u_int namelen;
4692 	int error, error2, *name;
4693 
4694 	namelen = arg2;
4695 	if (namelen != 1)
4696 		return (EINVAL);
4697 
4698 	name = (int *)arg1;
4699 
4700 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4701 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4702 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4703 	if (error != 0) {
4704 		sbuf_delete(&sb);
4705 		return (error);
4706 	}
4707 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4708 	error = kern_proc_cwd_out(p, &sb, maxlen);
4709 	error2 = sbuf_finish(&sb);
4710 	sbuf_delete(&sb);
4711 	return (error != 0 ? error : error2);
4712 }
4713 
4714 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4715     sysctl_kern_proc_cwd, "Process current working directory");
4716 
4717 #ifdef DDB
4718 /*
4719  * For the purposes of debugging, generate a human-readable string for the
4720  * file type.
4721  */
4722 static const char *
4723 file_type_to_name(short type)
4724 {
4725 
4726 	switch (type) {
4727 	case 0:
4728 		return ("zero");
4729 	case DTYPE_VNODE:
4730 		return ("vnode");
4731 	case DTYPE_SOCKET:
4732 		return ("socket");
4733 	case DTYPE_PIPE:
4734 		return ("pipe");
4735 	case DTYPE_FIFO:
4736 		return ("fifo");
4737 	case DTYPE_KQUEUE:
4738 		return ("kqueue");
4739 	case DTYPE_CRYPTO:
4740 		return ("crypto");
4741 	case DTYPE_MQUEUE:
4742 		return ("mqueue");
4743 	case DTYPE_SHM:
4744 		return ("shm");
4745 	case DTYPE_SEM:
4746 		return ("ksem");
4747 	case DTYPE_PTS:
4748 		return ("pts");
4749 	case DTYPE_DEV:
4750 		return ("dev");
4751 	case DTYPE_PROCDESC:
4752 		return ("proc");
4753 	case DTYPE_EVENTFD:
4754 		return ("eventfd");
4755 	case DTYPE_LINUXTFD:
4756 		return ("ltimer");
4757 	default:
4758 		return ("unkn");
4759 	}
4760 }
4761 
4762 /*
4763  * For the purposes of debugging, identify a process (if any, perhaps one of
4764  * many) that references the passed file in its file descriptor array. Return
4765  * NULL if none.
4766  */
4767 static struct proc *
4768 file_to_first_proc(struct file *fp)
4769 {
4770 	struct filedesc *fdp;
4771 	struct proc *p;
4772 	int n;
4773 
4774 	FOREACH_PROC_IN_SYSTEM(p) {
4775 		if (p->p_state == PRS_NEW)
4776 			continue;
4777 		fdp = p->p_fd;
4778 		if (fdp == NULL)
4779 			continue;
4780 		for (n = 0; n < fdp->fd_nfiles; n++) {
4781 			if (fp == fdp->fd_ofiles[n].fde_file)
4782 				return (p);
4783 		}
4784 	}
4785 	return (NULL);
4786 }
4787 
4788 static void
4789 db_print_file(struct file *fp, int header)
4790 {
4791 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4792 	struct proc *p;
4793 
4794 	if (header)
4795 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4796 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4797 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4798 		    "FCmd");
4799 	p = file_to_first_proc(fp);
4800 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4801 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4802 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4803 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4804 
4805 #undef XPTRWIDTH
4806 }
4807 
4808 DB_SHOW_COMMAND(file, db_show_file)
4809 {
4810 	struct file *fp;
4811 
4812 	if (!have_addr) {
4813 		db_printf("usage: show file <addr>\n");
4814 		return;
4815 	}
4816 	fp = (struct file *)addr;
4817 	db_print_file(fp, 1);
4818 }
4819 
4820 DB_SHOW_COMMAND(files, db_show_files)
4821 {
4822 	struct filedesc *fdp;
4823 	struct file *fp;
4824 	struct proc *p;
4825 	int header;
4826 	int n;
4827 
4828 	header = 1;
4829 	FOREACH_PROC_IN_SYSTEM(p) {
4830 		if (p->p_state == PRS_NEW)
4831 			continue;
4832 		if ((fdp = p->p_fd) == NULL)
4833 			continue;
4834 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4835 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4836 				continue;
4837 			db_print_file(fp, header);
4838 			header = 0;
4839 		}
4840 	}
4841 }
4842 #endif
4843 
4844 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4845     &maxfilesperproc, 0, "Maximum files allowed open per process");
4846 
4847 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4848     &maxfiles, 0, "Maximum number of files");
4849 
4850 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4851     &openfiles, 0, "System-wide number of open files");
4852 
4853 /* ARGSUSED*/
4854 static void
4855 filelistinit(void *dummy)
4856 {
4857 
4858 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4859 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4860 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4861 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4862 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4863 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4864 	/*
4865 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4866 	 * the vnode zone.
4867 	 */
4868 	vfs_smr = uma_zone_get_smr(pwd_zone);
4869 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4870 }
4871 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4872 
4873 /*-------------------------------------------------------------------*/
4874 
4875 static int
4876 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4877     int flags, struct thread *td)
4878 {
4879 
4880 	return (EBADF);
4881 }
4882 
4883 static int
4884 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4885     struct thread *td)
4886 {
4887 
4888 	return (EINVAL);
4889 }
4890 
4891 static int
4892 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4893     struct thread *td)
4894 {
4895 
4896 	return (EBADF);
4897 }
4898 
4899 static int
4900 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4901     struct thread *td)
4902 {
4903 
4904 	return (0);
4905 }
4906 
4907 static int
4908 badfo_kqfilter(struct file *fp, struct knote *kn)
4909 {
4910 
4911 	return (EBADF);
4912 }
4913 
4914 static int
4915 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4916 {
4917 
4918 	return (EBADF);
4919 }
4920 
4921 static int
4922 badfo_close(struct file *fp, struct thread *td)
4923 {
4924 
4925 	return (0);
4926 }
4927 
4928 static int
4929 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4930     struct thread *td)
4931 {
4932 
4933 	return (EBADF);
4934 }
4935 
4936 static int
4937 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4938     struct thread *td)
4939 {
4940 
4941 	return (EBADF);
4942 }
4943 
4944 static int
4945 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4946     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4947     struct thread *td)
4948 {
4949 
4950 	return (EBADF);
4951 }
4952 
4953 static int
4954 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4955 {
4956 
4957 	return (0);
4958 }
4959 
4960 struct fileops badfileops = {
4961 	.fo_read = badfo_readwrite,
4962 	.fo_write = badfo_readwrite,
4963 	.fo_truncate = badfo_truncate,
4964 	.fo_ioctl = badfo_ioctl,
4965 	.fo_poll = badfo_poll,
4966 	.fo_kqfilter = badfo_kqfilter,
4967 	.fo_stat = badfo_stat,
4968 	.fo_close = badfo_close,
4969 	.fo_chmod = badfo_chmod,
4970 	.fo_chown = badfo_chown,
4971 	.fo_sendfile = badfo_sendfile,
4972 	.fo_fill_kinfo = badfo_fill_kinfo,
4973 };
4974 
4975 static int
4976 path_poll(struct file *fp, int events, struct ucred *active_cred,
4977     struct thread *td)
4978 {
4979 	return (POLLNVAL);
4980 }
4981 
4982 static int
4983 path_close(struct file *fp, struct thread *td)
4984 {
4985 	MPASS(fp->f_type == DTYPE_VNODE);
4986 	fp->f_ops = &badfileops;
4987 	vdrop(fp->f_vnode);
4988 	return (0);
4989 }
4990 
4991 struct fileops path_fileops = {
4992 	.fo_read = badfo_readwrite,
4993 	.fo_write = badfo_readwrite,
4994 	.fo_truncate = badfo_truncate,
4995 	.fo_ioctl = badfo_ioctl,
4996 	.fo_poll = path_poll,
4997 	.fo_kqfilter = vn_kqfilter_opath,
4998 	.fo_stat = vn_statfile,
4999 	.fo_close = path_close,
5000 	.fo_chmod = badfo_chmod,
5001 	.fo_chown = badfo_chown,
5002 	.fo_sendfile = badfo_sendfile,
5003 	.fo_fill_kinfo = vn_fill_kinfo,
5004 	.fo_flags = DFLAG_PASSABLE,
5005 };
5006 
5007 int
5008 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5009     int flags, struct thread *td)
5010 {
5011 
5012 	return (EOPNOTSUPP);
5013 }
5014 
5015 int
5016 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5017     struct thread *td)
5018 {
5019 
5020 	return (EINVAL);
5021 }
5022 
5023 int
5024 invfo_ioctl(struct file *fp, u_long com, void *data,
5025     struct ucred *active_cred, struct thread *td)
5026 {
5027 
5028 	return (ENOTTY);
5029 }
5030 
5031 int
5032 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5033     struct thread *td)
5034 {
5035 
5036 	return (poll_no_poll(events));
5037 }
5038 
5039 int
5040 invfo_kqfilter(struct file *fp, struct knote *kn)
5041 {
5042 
5043 	return (EINVAL);
5044 }
5045 
5046 int
5047 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5048     struct thread *td)
5049 {
5050 
5051 	return (EINVAL);
5052 }
5053 
5054 int
5055 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5056     struct thread *td)
5057 {
5058 
5059 	return (EINVAL);
5060 }
5061 
5062 int
5063 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5064     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5065     struct thread *td)
5066 {
5067 
5068 	return (EINVAL);
5069 }
5070 
5071 /*-------------------------------------------------------------------*/
5072 
5073 /*
5074  * File Descriptor pseudo-device driver (/dev/fd/).
5075  *
5076  * Opening minor device N dup()s the file (if any) connected to file
5077  * descriptor N belonging to the calling process.  Note that this driver
5078  * consists of only the ``open()'' routine, because all subsequent
5079  * references to this file will be direct to the other driver.
5080  *
5081  * XXX: we could give this one a cloning event handler if necessary.
5082  */
5083 
5084 /* ARGSUSED */
5085 static int
5086 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5087 {
5088 
5089 	/*
5090 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5091 	 * the file descriptor being sought for duplication. The error
5092 	 * return ensures that the vnode for this device will be released
5093 	 * by vn_open. Open will detect this special error and take the
5094 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5095 	 * will simply report the error.
5096 	 */
5097 	td->td_dupfd = dev2unit(dev);
5098 	return (ENODEV);
5099 }
5100 
5101 static struct cdevsw fildesc_cdevsw = {
5102 	.d_version =	D_VERSION,
5103 	.d_open =	fdopen,
5104 	.d_name =	"FD",
5105 };
5106 
5107 static void
5108 fildesc_drvinit(void *unused)
5109 {
5110 	struct cdev *dev;
5111 
5112 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5113 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5114 	make_dev_alias(dev, "stdin");
5115 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5116 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5117 	make_dev_alias(dev, "stdout");
5118 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5119 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5120 	make_dev_alias(dev, "stderr");
5121 }
5122 
5123 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5124