xref: /freebsd/sys/kern/kern_descrip.c (revision a6fb86917362e3f6d24e95e940e80845c2cfde8a)
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	fget_unlocked_seq(struct thread *td, int fd,
119 		    cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp);
120 static int	getmaxfd(struct thread *td);
121 static u_long	*filecaps_copy_prep(const struct filecaps *src);
122 static void	filecaps_copy_finish(const struct filecaps *src,
123 		    struct filecaps *dst, u_long *ioctls);
124 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
125 static void	filecaps_free_finish(u_long *ioctls);
126 
127 static struct pwd *pwd_alloc(void);
128 
129 /*
130  * Each process has:
131  *
132  * - An array of open file descriptors (fd_ofiles)
133  * - An array of file flags (fd_ofileflags)
134  * - A bitmap recording which descriptors are in use (fd_map)
135  *
136  * A process starts out with NDFILE descriptors.  The value of NDFILE has
137  * been selected based the historical limit of 20 open files, and an
138  * assumption that the majority of processes, especially short-lived
139  * processes like shells, will never need more.
140  *
141  * If this initial allocation is exhausted, a larger descriptor table and
142  * map are allocated dynamically, and the pointers in the process's struct
143  * filedesc are updated to point to those.  This is repeated every time
144  * the process runs out of file descriptors (provided it hasn't hit its
145  * resource limit).
146  *
147  * Since threads may hold references to individual descriptor table
148  * entries, the tables are never freed.  Instead, they are placed on a
149  * linked list and freed only when the struct filedesc is released.
150  */
151 #define NDFILE		20
152 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
153 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
154 #define NDSLOT(x)	((x) / NDENTRIES)
155 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
156 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
157 
158 /*
159  * SLIST entry used to keep track of ofiles which must be reclaimed when
160  * the process exits.
161  */
162 struct freetable {
163 	struct fdescenttbl *ft_table;
164 	SLIST_ENTRY(freetable) ft_next;
165 };
166 
167 /*
168  * Initial allocation: a filedesc structure + the head of SLIST used to
169  * keep track of old ofiles + enough space for NDFILE descriptors.
170  */
171 
172 struct fdescenttbl0 {
173 	int	fdt_nfiles;
174 	struct	filedescent fdt_ofiles[NDFILE];
175 };
176 
177 struct filedesc0 {
178 	struct filedesc fd_fd;
179 	SLIST_HEAD(, freetable) fd_free;
180 	struct	fdescenttbl0 fd_dfiles;
181 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
182 };
183 
184 /*
185  * Descriptor management.
186  */
187 static int __exclusive_cache_line openfiles; /* actual number of open files */
188 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
189 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
190 
191 /*
192  * If low >= size, just return low. Otherwise find the first zero bit in the
193  * given bitmap, starting at low and not exceeding size - 1. Return size if
194  * not found.
195  */
196 static int
197 fd_first_free(struct filedesc *fdp, int low, int size)
198 {
199 	NDSLOTTYPE *map = fdp->fd_map;
200 	NDSLOTTYPE mask;
201 	int off, maxoff;
202 
203 	if (low >= size)
204 		return (low);
205 
206 	off = NDSLOT(low);
207 	if (low % NDENTRIES) {
208 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
209 		if ((mask &= ~map[off]) != 0UL)
210 			return (off * NDENTRIES + ffsl(mask) - 1);
211 		++off;
212 	}
213 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
214 		if (map[off] != ~0UL)
215 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
216 	return (size);
217 }
218 
219 /*
220  * Find the last used fd.
221  *
222  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
223  * Otherwise use fdlastfile.
224  */
225 int
226 fdlastfile_single(struct filedesc *fdp)
227 {
228 	NDSLOTTYPE *map = fdp->fd_map;
229 	int off, minoff;
230 
231 	off = NDSLOT(fdp->fd_nfiles - 1);
232 	for (minoff = NDSLOT(0); off >= minoff; --off)
233 		if (map[off] != 0)
234 			return (off * NDENTRIES + flsl(map[off]) - 1);
235 	return (-1);
236 }
237 
238 int
239 fdlastfile(struct filedesc *fdp)
240 {
241 
242 	FILEDESC_LOCK_ASSERT(fdp);
243 	return (fdlastfile_single(fdp));
244 }
245 
246 static int
247 fdisused(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
251 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
252 
253 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
254 }
255 
256 /*
257  * Mark a file descriptor as used.
258  */
259 static void
260 fdused_init(struct filedesc *fdp, int fd)
261 {
262 
263 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
264 
265 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
266 }
267 
268 static void
269 fdused(struct filedesc *fdp, int fd)
270 {
271 
272 	FILEDESC_XLOCK_ASSERT(fdp);
273 
274 	fdused_init(fdp, fd);
275 	if (fd == fdp->fd_freefile)
276 		fdp->fd_freefile++;
277 }
278 
279 /*
280  * Mark a file descriptor as unused.
281  */
282 static void
283 fdunused(struct filedesc *fdp, int fd)
284 {
285 
286 	FILEDESC_XLOCK_ASSERT(fdp);
287 
288 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
289 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
290 	    ("fd=%d is still in use", fd));
291 
292 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
293 	if (fd < fdp->fd_freefile)
294 		fdp->fd_freefile = fd;
295 }
296 
297 /*
298  * Free a file descriptor.
299  *
300  * Avoid some work if fdp is about to be destroyed.
301  */
302 static inline void
303 fdefree_last(struct filedescent *fde)
304 {
305 
306 	filecaps_free(&fde->fde_caps);
307 }
308 
309 static inline void
310 fdfree(struct filedesc *fdp, int fd)
311 {
312 	struct filedescent *fde;
313 
314 	FILEDESC_XLOCK_ASSERT(fdp);
315 	fde = &fdp->fd_ofiles[fd];
316 #ifdef CAPABILITIES
317 	seqc_write_begin(&fde->fde_seqc);
318 #endif
319 	fde->fde_file = NULL;
320 #ifdef CAPABILITIES
321 	seqc_write_end(&fde->fde_seqc);
322 #endif
323 	fdefree_last(fde);
324 	fdunused(fdp, fd);
325 }
326 
327 /*
328  * System calls on descriptors.
329  */
330 #ifndef _SYS_SYSPROTO_H_
331 struct getdtablesize_args {
332 	int	dummy;
333 };
334 #endif
335 /* ARGSUSED */
336 int
337 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 {
339 #ifdef	RACCT
340 	uint64_t lim;
341 #endif
342 
343 	td->td_retval[0] = getmaxfd(td);
344 #ifdef	RACCT
345 	PROC_LOCK(td->td_proc);
346 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
347 	PROC_UNLOCK(td->td_proc);
348 	if (lim < td->td_retval[0])
349 		td->td_retval[0] = lim;
350 #endif
351 	return (0);
352 }
353 
354 /*
355  * Duplicate a file descriptor to a particular value.
356  *
357  * Note: keep in mind that a potential race condition exists when closing
358  * descriptors from a shared descriptor table (via rfork).
359  */
360 #ifndef _SYS_SYSPROTO_H_
361 struct dup2_args {
362 	u_int	from;
363 	u_int	to;
364 };
365 #endif
366 /* ARGSUSED */
367 int
368 sys_dup2(struct thread *td, struct dup2_args *uap)
369 {
370 
371 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
372 }
373 
374 /*
375  * Duplicate a file descriptor.
376  */
377 #ifndef _SYS_SYSPROTO_H_
378 struct dup_args {
379 	u_int	fd;
380 };
381 #endif
382 /* ARGSUSED */
383 int
384 sys_dup(struct thread *td, struct dup_args *uap)
385 {
386 
387 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
388 }
389 
390 /*
391  * The file control system call.
392  */
393 #ifndef _SYS_SYSPROTO_H_
394 struct fcntl_args {
395 	int	fd;
396 	int	cmd;
397 	long	arg;
398 };
399 #endif
400 /* ARGSUSED */
401 int
402 sys_fcntl(struct thread *td, struct fcntl_args *uap)
403 {
404 
405 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
406 }
407 
408 int
409 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
410 {
411 	struct flock fl;
412 	struct __oflock ofl;
413 	intptr_t arg1;
414 	int error, newcmd;
415 
416 	error = 0;
417 	newcmd = cmd;
418 	switch (cmd) {
419 	case F_OGETLK:
420 	case F_OSETLK:
421 	case F_OSETLKW:
422 		/*
423 		 * Convert old flock structure to new.
424 		 */
425 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
426 		fl.l_start = ofl.l_start;
427 		fl.l_len = ofl.l_len;
428 		fl.l_pid = ofl.l_pid;
429 		fl.l_type = ofl.l_type;
430 		fl.l_whence = ofl.l_whence;
431 		fl.l_sysid = 0;
432 
433 		switch (cmd) {
434 		case F_OGETLK:
435 			newcmd = F_GETLK;
436 			break;
437 		case F_OSETLK:
438 			newcmd = F_SETLK;
439 			break;
440 		case F_OSETLKW:
441 			newcmd = F_SETLKW;
442 			break;
443 		}
444 		arg1 = (intptr_t)&fl;
445 		break;
446 	case F_GETLK:
447 	case F_SETLK:
448 	case F_SETLKW:
449 	case F_SETLK_REMOTE:
450 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
451 		arg1 = (intptr_t)&fl;
452 		break;
453 	default:
454 		arg1 = arg;
455 		break;
456 	}
457 	if (error)
458 		return (error);
459 	error = kern_fcntl(td, fd, newcmd, arg1);
460 	if (error)
461 		return (error);
462 	if (cmd == F_OGETLK) {
463 		ofl.l_start = fl.l_start;
464 		ofl.l_len = fl.l_len;
465 		ofl.l_pid = fl.l_pid;
466 		ofl.l_type = fl.l_type;
467 		ofl.l_whence = fl.l_whence;
468 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
469 	} else if (cmd == F_GETLK) {
470 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 	}
472 	return (error);
473 }
474 
475 int
476 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
477 {
478 	struct filedesc *fdp;
479 	struct flock *flp;
480 	struct file *fp, *fp2;
481 	struct filedescent *fde;
482 	struct proc *p;
483 	struct vnode *vp;
484 	struct mount *mp;
485 	struct kinfo_file *kif;
486 	int error, flg, kif_sz, seals, tmp;
487 	uint64_t bsize;
488 	off_t foffset;
489 
490 	error = 0;
491 	flg = F_POSIX;
492 	p = td->td_proc;
493 	fdp = p->p_fd;
494 
495 	AUDIT_ARG_FD(cmd);
496 	AUDIT_ARG_CMD(cmd);
497 	switch (cmd) {
498 	case F_DUPFD:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
501 		break;
502 
503 	case F_DUPFD_CLOEXEC:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
511 		break;
512 
513 	case F_DUP2FD_CLOEXEC:
514 		tmp = arg;
515 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
516 		break;
517 
518 	case F_GETFD:
519 		error = EBADF;
520 		FILEDESC_SLOCK(fdp);
521 		fde = fdeget_noref(fdp, fd);
522 		if (fde != NULL) {
523 			td->td_retval[0] =
524 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
525 			error = 0;
526 		}
527 		FILEDESC_SUNLOCK(fdp);
528 		break;
529 
530 	case F_SETFD:
531 		error = EBADF;
532 		FILEDESC_XLOCK(fdp);
533 		fde = fdeget_noref(fdp, fd);
534 		if (fde != NULL) {
535 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
536 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
537 			error = 0;
538 		}
539 		FILEDESC_XUNLOCK(fdp);
540 		break;
541 
542 	case F_GETFL:
543 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
544 		if (error != 0)
545 			break;
546 		td->td_retval[0] = OFLAGS(fp->f_flag);
547 		fdrop(fp, td);
548 		break;
549 
550 	case F_SETFL:
551 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
552 		if (error != 0)
553 			break;
554 		if (fp->f_ops == &path_fileops) {
555 			fdrop(fp, td);
556 			error = EBADF;
557 			break;
558 		}
559 		do {
560 			tmp = flg = fp->f_flag;
561 			tmp &= ~FCNTLFLAGS;
562 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
563 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
564 		tmp = fp->f_flag & FNONBLOCK;
565 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
566 		if (error != 0) {
567 			fdrop(fp, td);
568 			break;
569 		}
570 		tmp = fp->f_flag & FASYNC;
571 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
572 		if (error == 0) {
573 			fdrop(fp, td);
574 			break;
575 		}
576 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
577 		tmp = 0;
578 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_GETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
587 		if (error == 0)
588 			td->td_retval[0] = tmp;
589 		fdrop(fp, td);
590 		break;
591 
592 	case F_SETOWN:
593 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
594 		if (error != 0)
595 			break;
596 		tmp = arg;
597 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
598 		fdrop(fp, td);
599 		break;
600 
601 	case F_SETLK_REMOTE:
602 		error = priv_check(td, PRIV_NFS_LOCKD);
603 		if (error != 0)
604 			return (error);
605 		flg = F_REMOTE;
606 		goto do_setlk;
607 
608 	case F_SETLKW:
609 		flg |= F_WAIT;
610 		/* FALLTHROUGH F_SETLK */
611 
612 	case F_SETLK:
613 	do_setlk:
614 		flp = (struct flock *)arg;
615 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
616 			error = EINVAL;
617 			break;
618 		}
619 
620 		error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
621 		if (error != 0)
622 			break;
623 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
624 			error = EBADF;
625 			fdrop(fp, td);
626 			break;
627 		}
628 
629 		if (flp->l_whence == SEEK_CUR) {
630 			foffset = foffset_get(fp);
631 			if (foffset < 0 ||
632 			    (flp->l_start > 0 &&
633 			     foffset > OFF_MAX - flp->l_start)) {
634 				error = EOVERFLOW;
635 				fdrop(fp, td);
636 				break;
637 			}
638 			flp->l_start += foffset;
639 		}
640 
641 		vp = fp->f_vnode;
642 		switch (flp->l_type) {
643 		case F_RDLCK:
644 			if ((fp->f_flag & FREAD) == 0) {
645 				error = EBADF;
646 				break;
647 			}
648 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
649 				PROC_LOCK(p->p_leader);
650 				p->p_leader->p_flag |= P_ADVLOCK;
651 				PROC_UNLOCK(p->p_leader);
652 			}
653 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 			    flp, flg);
655 			break;
656 		case F_WRLCK:
657 			if ((fp->f_flag & FWRITE) == 0) {
658 				error = EBADF;
659 				break;
660 			}
661 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
662 				PROC_LOCK(p->p_leader);
663 				p->p_leader->p_flag |= P_ADVLOCK;
664 				PROC_UNLOCK(p->p_leader);
665 			}
666 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
667 			    flp, flg);
668 			break;
669 		case F_UNLCK:
670 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
671 			    flp, flg);
672 			break;
673 		case F_UNLCKSYS:
674 			if (flg != F_REMOTE) {
675 				error = EINVAL;
676 				break;
677 			}
678 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
679 			    F_UNLCKSYS, flp, flg);
680 			break;
681 		default:
682 			error = EINVAL;
683 			break;
684 		}
685 		if (error != 0 || flp->l_type == F_UNLCK ||
686 		    flp->l_type == F_UNLCKSYS) {
687 			fdrop(fp, td);
688 			break;
689 		}
690 
691 		/*
692 		 * Check for a race with close.
693 		 *
694 		 * The vnode is now advisory locked (or unlocked, but this case
695 		 * is not really important) as the caller requested.
696 		 * We had to drop the filedesc lock, so we need to recheck if
697 		 * the descriptor is still valid, because if it was closed
698 		 * in the meantime we need to remove advisory lock from the
699 		 * vnode - close on any descriptor leading to an advisory
700 		 * locked vnode, removes that lock.
701 		 * We will return 0 on purpose in that case, as the result of
702 		 * successful advisory lock might have been externally visible
703 		 * already. This is fine - effectively we pretend to the caller
704 		 * that the closing thread was a bit slower and that the
705 		 * advisory lock succeeded before the close.
706 		 */
707 		error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
708 		if (error != 0) {
709 			fdrop(fp, td);
710 			break;
711 		}
712 		if (fp != fp2) {
713 			flp->l_whence = SEEK_SET;
714 			flp->l_start = 0;
715 			flp->l_len = 0;
716 			flp->l_type = F_UNLCK;
717 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
718 			    F_UNLCK, flp, F_POSIX);
719 		}
720 		fdrop(fp, td);
721 		fdrop(fp2, td);
722 		break;
723 
724 	case F_GETLK:
725 		error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
726 		if (error != 0)
727 			break;
728 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
729 			error = EBADF;
730 			fdrop(fp, td);
731 			break;
732 		}
733 		flp = (struct flock *)arg;
734 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
735 		    flp->l_type != F_UNLCK) {
736 			error = EINVAL;
737 			fdrop(fp, td);
738 			break;
739 		}
740 		if (flp->l_whence == SEEK_CUR) {
741 			foffset = foffset_get(fp);
742 			if ((flp->l_start > 0 &&
743 			    foffset > OFF_MAX - flp->l_start) ||
744 			    (flp->l_start < 0 &&
745 			    foffset < OFF_MIN - flp->l_start)) {
746 				error = EOVERFLOW;
747 				fdrop(fp, td);
748 				break;
749 			}
750 			flp->l_start += foffset;
751 		}
752 		vp = fp->f_vnode;
753 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
754 		    F_POSIX);
755 		fdrop(fp, td);
756 		break;
757 
758 	case F_ADD_SEALS:
759 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
760 		if (error != 0)
761 			break;
762 		error = fo_add_seals(fp, arg);
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_GET_SEALS:
767 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
768 		if (error != 0)
769 			break;
770 		if (fo_get_seals(fp, &seals) == 0)
771 			td->td_retval[0] = seals;
772 		else
773 			error = EINVAL;
774 		fdrop(fp, td);
775 		break;
776 
777 	case F_RDAHEAD:
778 		arg = arg ? 128 * 1024: 0;
779 		/* FALLTHROUGH */
780 	case F_READAHEAD:
781 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
782 		if (error != 0)
783 			break;
784 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
785 			fdrop(fp, td);
786 			error = EBADF;
787 			break;
788 		}
789 		vp = fp->f_vnode;
790 		if (vp->v_type != VREG) {
791 			fdrop(fp, td);
792 			error = ENOTTY;
793 			break;
794 		}
795 
796 		/*
797 		 * Exclusive lock synchronizes against f_seqcount reads and
798 		 * writes in sequential_heuristic().
799 		 */
800 		error = vn_lock(vp, LK_EXCLUSIVE);
801 		if (error != 0) {
802 			fdrop(fp, td);
803 			break;
804 		}
805 		if (arg >= 0) {
806 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
807 			arg = MIN(arg, INT_MAX - bsize + 1);
808 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
809 			    (arg + bsize - 1) / bsize);
810 			atomic_set_int(&fp->f_flag, FRDAHEAD);
811 		} else {
812 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
813 		}
814 		VOP_UNLOCK(vp);
815 		fdrop(fp, td);
816 		break;
817 
818 	case F_ISUNIONSTACK:
819 		/*
820 		 * Check if the vnode is part of a union stack (either the
821 		 * "union" flag from mount(2) or unionfs).
822 		 *
823 		 * Prior to introduction of this op libc's readdir would call
824 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
825 		 * data just to check fs name and a mount flag.
826 		 *
827 		 * Fixing the code to handle everything in the kernel instead
828 		 * is a non-trivial endeavor and has low priority, thus this
829 		 * horrible kludge facilitates the current behavior in a much
830 		 * cheaper manner until someone(tm) sorts this out.
831 		 */
832 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
833 		if (error != 0)
834 			break;
835 		if (fp->f_type != DTYPE_VNODE) {
836 			fdrop(fp, td);
837 			error = EBADF;
838 			break;
839 		}
840 		vp = fp->f_vnode;
841 		/*
842 		 * Since we don't prevent dooming the vnode even non-null mp
843 		 * found can become immediately stale. This is tolerable since
844 		 * mount points are type-stable (providing safe memory access)
845 		 * and any vfs op on this vnode going forward will return an
846 		 * error (meaning return value in this case is meaningless).
847 		 */
848 		mp = atomic_load_ptr(&vp->v_mount);
849 		if (__predict_false(mp == NULL)) {
850 			fdrop(fp, td);
851 			error = EBADF;
852 			break;
853 		}
854 		td->td_retval[0] = 0;
855 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
856 		    mp->mnt_flag & MNT_UNION)
857 			td->td_retval[0] = 1;
858 		fdrop(fp, td);
859 		break;
860 
861 	case F_KINFO:
862 #ifdef CAPABILITY_MODE
863 		if (IN_CAPABILITY_MODE(td)) {
864 			error = ECAPMODE;
865 			break;
866 		}
867 #endif
868 		error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
869 		if (error != 0)
870 			break;
871 		if (kif_sz != sizeof(*kif)) {
872 			error = EINVAL;
873 			break;
874 		}
875 		kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
876 		FILEDESC_SLOCK(fdp);
877 		error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
878 		if (error == 0 && fhold(fp)) {
879 			export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
880 			FILEDESC_SUNLOCK(fdp);
881 			fdrop(fp, td);
882 			if ((kif->kf_status & KF_ATTR_VALID) != 0) {
883 				kif->kf_structsize = sizeof(*kif);
884 				error = copyout(kif, (void *)arg, sizeof(*kif));
885 			} else {
886 				error = EBADF;
887 			}
888 		} else {
889 			FILEDESC_SUNLOCK(fdp);
890 			if (error == 0)
891 				error = EBADF;
892 		}
893 		free(kif, M_TEMP);
894 		break;
895 
896 	default:
897 		error = EINVAL;
898 		break;
899 	}
900 	return (error);
901 }
902 
903 static int
904 getmaxfd(struct thread *td)
905 {
906 
907 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
908 }
909 
910 /*
911  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
912  */
913 int
914 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
915 {
916 	struct filedesc *fdp;
917 	struct filedescent *oldfde, *newfde;
918 	struct proc *p;
919 	struct file *delfp, *oldfp;
920 	u_long *oioctls, *nioctls;
921 	int error, maxfd;
922 
923 	p = td->td_proc;
924 	fdp = p->p_fd;
925 	oioctls = NULL;
926 
927 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
928 	MPASS(mode < FDDUP_LASTMODE);
929 
930 	AUDIT_ARG_FD(old);
931 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
932 
933 	/*
934 	 * Verify we have a valid descriptor to dup from and possibly to
935 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
936 	 * return EINVAL when the new descriptor is out of bounds.
937 	 */
938 	if (old < 0)
939 		return (EBADF);
940 	if (new < 0)
941 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
942 	maxfd = getmaxfd(td);
943 	if (new >= maxfd)
944 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
945 
946 	error = EBADF;
947 	FILEDESC_XLOCK(fdp);
948 	if (fget_noref(fdp, old) == NULL)
949 		goto unlock;
950 	if (mode == FDDUP_FIXED && old == new) {
951 		td->td_retval[0] = new;
952 		if (flags & FDDUP_FLAG_CLOEXEC)
953 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
954 		error = 0;
955 		goto unlock;
956 	}
957 
958 	oldfde = &fdp->fd_ofiles[old];
959 	oldfp = oldfde->fde_file;
960 	if (!fhold(oldfp))
961 		goto unlock;
962 
963 	/*
964 	 * If the caller specified a file descriptor, make sure the file
965 	 * table is large enough to hold it, and grab it.  Otherwise, just
966 	 * allocate a new descriptor the usual way.
967 	 */
968 	switch (mode) {
969 	case FDDUP_NORMAL:
970 	case FDDUP_FCNTL:
971 		if ((error = fdalloc(td, new, &new)) != 0) {
972 			fdrop(oldfp, td);
973 			goto unlock;
974 		}
975 		break;
976 	case FDDUP_FIXED:
977 		if (new >= fdp->fd_nfiles) {
978 			/*
979 			 * The resource limits are here instead of e.g.
980 			 * fdalloc(), because the file descriptor table may be
981 			 * shared between processes, so we can't really use
982 			 * racct_add()/racct_sub().  Instead of counting the
983 			 * number of actually allocated descriptors, just put
984 			 * the limit on the size of the file descriptor table.
985 			 */
986 #ifdef RACCT
987 			if (RACCT_ENABLED()) {
988 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
989 				if (error != 0) {
990 					error = EMFILE;
991 					fdrop(oldfp, td);
992 					goto unlock;
993 				}
994 			}
995 #endif
996 			fdgrowtable_exp(fdp, new + 1);
997 		}
998 		if (!fdisused(fdp, new))
999 			fdused(fdp, new);
1000 		break;
1001 	default:
1002 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1003 	}
1004 
1005 	KASSERT(old != new, ("new fd is same as old"));
1006 
1007 	/* Refetch oldfde because the table may have grown and old one freed. */
1008 	oldfde = &fdp->fd_ofiles[old];
1009 	KASSERT(oldfp == oldfde->fde_file,
1010 	    ("fdt_ofiles shift from growth observed at fd %d",
1011 	    old));
1012 
1013 	newfde = &fdp->fd_ofiles[new];
1014 	delfp = newfde->fde_file;
1015 
1016 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1017 
1018 	/*
1019 	 * Duplicate the source descriptor.
1020 	 */
1021 #ifdef CAPABILITIES
1022 	seqc_write_begin(&newfde->fde_seqc);
1023 #endif
1024 	oioctls = filecaps_free_prep(&newfde->fde_caps);
1025 	fde_copy(oldfde, newfde);
1026 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1027 	    nioctls);
1028 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1029 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1030 	else
1031 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1032 #ifdef CAPABILITIES
1033 	seqc_write_end(&newfde->fde_seqc);
1034 #endif
1035 	td->td_retval[0] = new;
1036 
1037 	error = 0;
1038 
1039 	if (delfp != NULL) {
1040 		(void) closefp(fdp, new, delfp, td, true, false);
1041 		FILEDESC_UNLOCK_ASSERT(fdp);
1042 	} else {
1043 unlock:
1044 		FILEDESC_XUNLOCK(fdp);
1045 	}
1046 
1047 	filecaps_free_finish(oioctls);
1048 	return (error);
1049 }
1050 
1051 static void
1052 sigiofree(struct sigio *sigio)
1053 {
1054 	crfree(sigio->sio_ucred);
1055 	free(sigio, M_SIGIO);
1056 }
1057 
1058 static struct sigio *
1059 funsetown_locked(struct sigio *sigio)
1060 {
1061 	struct proc *p;
1062 	struct pgrp *pg;
1063 
1064 	SIGIO_ASSERT_LOCKED();
1065 
1066 	if (sigio == NULL)
1067 		return (NULL);
1068 	*sigio->sio_myref = NULL;
1069 	if (sigio->sio_pgid < 0) {
1070 		pg = sigio->sio_pgrp;
1071 		PGRP_LOCK(pg);
1072 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1073 		PGRP_UNLOCK(pg);
1074 	} else {
1075 		p = sigio->sio_proc;
1076 		PROC_LOCK(p);
1077 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1078 		PROC_UNLOCK(p);
1079 	}
1080 	return (sigio);
1081 }
1082 
1083 /*
1084  * If sigio is on the list associated with a process or process group,
1085  * disable signalling from the device, remove sigio from the list and
1086  * free sigio.
1087  */
1088 void
1089 funsetown(struct sigio **sigiop)
1090 {
1091 	struct sigio *sigio;
1092 
1093 	/* Racy check, consumers must provide synchronization. */
1094 	if (*sigiop == NULL)
1095 		return;
1096 
1097 	SIGIO_LOCK();
1098 	sigio = funsetown_locked(*sigiop);
1099 	SIGIO_UNLOCK();
1100 	if (sigio != NULL)
1101 		sigiofree(sigio);
1102 }
1103 
1104 /*
1105  * Free a list of sigio structures.  The caller must ensure that new sigio
1106  * structures cannot be added after this point.  For process groups this is
1107  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1108  * as an interlock.
1109  */
1110 void
1111 funsetownlst(struct sigiolst *sigiolst)
1112 {
1113 	struct proc *p;
1114 	struct pgrp *pg;
1115 	struct sigio *sigio, *tmp;
1116 
1117 	/* Racy check. */
1118 	sigio = SLIST_FIRST(sigiolst);
1119 	if (sigio == NULL)
1120 		return;
1121 
1122 	p = NULL;
1123 	pg = NULL;
1124 
1125 	SIGIO_LOCK();
1126 	sigio = SLIST_FIRST(sigiolst);
1127 	if (sigio == NULL) {
1128 		SIGIO_UNLOCK();
1129 		return;
1130 	}
1131 
1132 	/*
1133 	 * Every entry of the list should belong to a single proc or pgrp.
1134 	 */
1135 	if (sigio->sio_pgid < 0) {
1136 		pg = sigio->sio_pgrp;
1137 		sx_assert(&proctree_lock, SX_XLOCKED);
1138 		PGRP_LOCK(pg);
1139 	} else /* if (sigio->sio_pgid > 0) */ {
1140 		p = sigio->sio_proc;
1141 		PROC_LOCK(p);
1142 		KASSERT((p->p_flag & P_WEXIT) != 0,
1143 		    ("%s: process %p is not exiting", __func__, p));
1144 	}
1145 
1146 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1147 		*sigio->sio_myref = NULL;
1148 		if (pg != NULL) {
1149 			KASSERT(sigio->sio_pgid < 0,
1150 			    ("Proc sigio in pgrp sigio list"));
1151 			KASSERT(sigio->sio_pgrp == pg,
1152 			    ("Bogus pgrp in sigio list"));
1153 		} else /* if (p != NULL) */ {
1154 			KASSERT(sigio->sio_pgid > 0,
1155 			    ("Pgrp sigio in proc sigio list"));
1156 			KASSERT(sigio->sio_proc == p,
1157 			    ("Bogus proc in sigio list"));
1158 		}
1159 	}
1160 
1161 	if (pg != NULL)
1162 		PGRP_UNLOCK(pg);
1163 	else
1164 		PROC_UNLOCK(p);
1165 	SIGIO_UNLOCK();
1166 
1167 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1168 		sigiofree(sigio);
1169 }
1170 
1171 /*
1172  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1173  *
1174  * After permission checking, add a sigio structure to the sigio list for
1175  * the process or process group.
1176  */
1177 int
1178 fsetown(pid_t pgid, struct sigio **sigiop)
1179 {
1180 	struct proc *proc;
1181 	struct pgrp *pgrp;
1182 	struct sigio *osigio, *sigio;
1183 	int ret;
1184 
1185 	if (pgid == 0) {
1186 		funsetown(sigiop);
1187 		return (0);
1188 	}
1189 
1190 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1191 	sigio->sio_pgid = pgid;
1192 	sigio->sio_ucred = crhold(curthread->td_ucred);
1193 	sigio->sio_myref = sigiop;
1194 
1195 	ret = 0;
1196 	if (pgid > 0) {
1197 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1198 		SIGIO_LOCK();
1199 		osigio = funsetown_locked(*sigiop);
1200 		if (ret == 0) {
1201 			PROC_LOCK(proc);
1202 			_PRELE(proc);
1203 			if ((proc->p_flag & P_WEXIT) != 0) {
1204 				ret = ESRCH;
1205 			} else if (proc->p_session !=
1206 			    curthread->td_proc->p_session) {
1207 				/*
1208 				 * Policy - Don't allow a process to FSETOWN a
1209 				 * process in another session.
1210 				 *
1211 				 * Remove this test to allow maximum flexibility
1212 				 * or restrict FSETOWN to the current process or
1213 				 * process group for maximum safety.
1214 				 */
1215 				ret = EPERM;
1216 			} else {
1217 				sigio->sio_proc = proc;
1218 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1219 				    sio_pgsigio);
1220 			}
1221 			PROC_UNLOCK(proc);
1222 		}
1223 	} else /* if (pgid < 0) */ {
1224 		sx_slock(&proctree_lock);
1225 		SIGIO_LOCK();
1226 		osigio = funsetown_locked(*sigiop);
1227 		pgrp = pgfind(-pgid);
1228 		if (pgrp == NULL) {
1229 			ret = ESRCH;
1230 		} else {
1231 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1232 				/*
1233 				 * Policy - Don't allow a process to FSETOWN a
1234 				 * process in another session.
1235 				 *
1236 				 * Remove this test to allow maximum flexibility
1237 				 * or restrict FSETOWN to the current process or
1238 				 * process group for maximum safety.
1239 				 */
1240 				ret = EPERM;
1241 			} else {
1242 				sigio->sio_pgrp = pgrp;
1243 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1244 				    sio_pgsigio);
1245 			}
1246 			PGRP_UNLOCK(pgrp);
1247 		}
1248 		sx_sunlock(&proctree_lock);
1249 	}
1250 	if (ret == 0)
1251 		*sigiop = sigio;
1252 	SIGIO_UNLOCK();
1253 	if (osigio != NULL)
1254 		sigiofree(osigio);
1255 	return (ret);
1256 }
1257 
1258 /*
1259  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1260  */
1261 pid_t
1262 fgetown(struct sigio **sigiop)
1263 {
1264 	pid_t pgid;
1265 
1266 	SIGIO_LOCK();
1267 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1268 	SIGIO_UNLOCK();
1269 	return (pgid);
1270 }
1271 
1272 static int
1273 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1274     bool audit)
1275 {
1276 	int error;
1277 
1278 	FILEDESC_XLOCK_ASSERT(fdp);
1279 
1280 	/*
1281 	 * We now hold the fp reference that used to be owned by the
1282 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1283 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1284 	 * added, and deleteing a knote for the new fd.
1285 	 */
1286 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1287 		knote_fdclose(td, fd);
1288 
1289 	/*
1290 	 * We need to notify mqueue if the object is of type mqueue.
1291 	 */
1292 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1293 		mq_fdclose(td, fd, fp);
1294 	FILEDESC_XUNLOCK(fdp);
1295 
1296 #ifdef AUDIT
1297 	if (AUDITING_TD(td) && audit)
1298 		audit_sysclose(td, fd, fp);
1299 #endif
1300 	error = closef(fp, td);
1301 
1302 	/*
1303 	 * All paths leading up to closefp() will have already removed or
1304 	 * replaced the fd in the filedesc table, so a restart would not
1305 	 * operate on the same file.
1306 	 */
1307 	if (error == ERESTART)
1308 		error = EINTR;
1309 
1310 	return (error);
1311 }
1312 
1313 static int
1314 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1315     bool holdleaders, bool audit)
1316 {
1317 	int error;
1318 
1319 	FILEDESC_XLOCK_ASSERT(fdp);
1320 
1321 	if (holdleaders) {
1322 		if (td->td_proc->p_fdtol != NULL) {
1323 			/*
1324 			 * Ask fdfree() to sleep to ensure that all relevant
1325 			 * process leaders can be traversed in closef().
1326 			 */
1327 			fdp->fd_holdleaderscount++;
1328 		} else {
1329 			holdleaders = false;
1330 		}
1331 	}
1332 
1333 	error = closefp_impl(fdp, fd, fp, td, audit);
1334 	if (holdleaders) {
1335 		FILEDESC_XLOCK(fdp);
1336 		fdp->fd_holdleaderscount--;
1337 		if (fdp->fd_holdleaderscount == 0 &&
1338 		    fdp->fd_holdleaderswakeup != 0) {
1339 			fdp->fd_holdleaderswakeup = 0;
1340 			wakeup(&fdp->fd_holdleaderscount);
1341 		}
1342 		FILEDESC_XUNLOCK(fdp);
1343 	}
1344 	return (error);
1345 }
1346 
1347 static int
1348 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1349     bool holdleaders, bool audit)
1350 {
1351 
1352 	FILEDESC_XLOCK_ASSERT(fdp);
1353 
1354 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1355 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1356 	} else {
1357 		return (closefp_impl(fdp, fd, fp, td, audit));
1358 	}
1359 }
1360 
1361 /*
1362  * Close a file descriptor.
1363  */
1364 #ifndef _SYS_SYSPROTO_H_
1365 struct close_args {
1366 	int     fd;
1367 };
1368 #endif
1369 /* ARGSUSED */
1370 int
1371 sys_close(struct thread *td, struct close_args *uap)
1372 {
1373 
1374 	return (kern_close(td, uap->fd));
1375 }
1376 
1377 int
1378 kern_close(struct thread *td, int fd)
1379 {
1380 	struct filedesc *fdp;
1381 	struct file *fp;
1382 
1383 	fdp = td->td_proc->p_fd;
1384 
1385 	FILEDESC_XLOCK(fdp);
1386 	if ((fp = fget_noref(fdp, fd)) == NULL) {
1387 		FILEDESC_XUNLOCK(fdp);
1388 		return (EBADF);
1389 	}
1390 	fdfree(fdp, fd);
1391 
1392 	/* closefp() drops the FILEDESC lock for us. */
1393 	return (closefp(fdp, fd, fp, td, true, true));
1394 }
1395 
1396 static int
1397 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd)
1398 {
1399 	struct filedesc *fdp;
1400 	struct fdescenttbl *fdt;
1401 	struct filedescent *fde;
1402 	int fd;
1403 
1404 	fdp = td->td_proc->p_fd;
1405 	FILEDESC_XLOCK(fdp);
1406 	fdt = atomic_load_ptr(&fdp->fd_files);
1407 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1408 	fd = lowfd;
1409 	if (__predict_false(fd > highfd)) {
1410 		goto out_locked;
1411 	}
1412 	for (; fd <= highfd; fd++) {
1413 		fde = &fdt->fdt_ofiles[fd];
1414 		if (fde->fde_file != NULL)
1415 			fde->fde_flags |= UF_EXCLOSE;
1416 	}
1417 out_locked:
1418 	FILEDESC_XUNLOCK(fdp);
1419 	return (0);
1420 }
1421 
1422 static int
1423 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1424 {
1425 	struct filedesc *fdp;
1426 	const struct fdescenttbl *fdt;
1427 	struct file *fp;
1428 	int fd;
1429 
1430 	fdp = td->td_proc->p_fd;
1431 	FILEDESC_XLOCK(fdp);
1432 	fdt = atomic_load_ptr(&fdp->fd_files);
1433 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1434 	fd = lowfd;
1435 	if (__predict_false(fd > highfd)) {
1436 		goto out_locked;
1437 	}
1438 	for (;;) {
1439 		fp = fdt->fdt_ofiles[fd].fde_file;
1440 		if (fp == NULL) {
1441 			if (fd == highfd)
1442 				goto out_locked;
1443 		} else {
1444 			fdfree(fdp, fd);
1445 			(void) closefp(fdp, fd, fp, td, true, true);
1446 			if (fd == highfd)
1447 				goto out_unlocked;
1448 			FILEDESC_XLOCK(fdp);
1449 			fdt = atomic_load_ptr(&fdp->fd_files);
1450 		}
1451 		fd++;
1452 	}
1453 out_locked:
1454 	FILEDESC_XUNLOCK(fdp);
1455 out_unlocked:
1456 	return (0);
1457 }
1458 
1459 int
1460 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1461 {
1462 
1463 	/*
1464 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1465 	 * open should not be a usage error.  From a close_range() perspective,
1466 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1467 	 * be a usage error as all fd above 3 are in-fact already closed.
1468 	 */
1469 	if (highfd < lowfd) {
1470 		return (EINVAL);
1471 	}
1472 
1473 	if ((flags & CLOSE_RANGE_CLOEXEC) != 0)
1474 		return (close_range_cloexec(td, lowfd, highfd));
1475 
1476 	return (close_range_impl(td, lowfd, highfd));
1477 }
1478 
1479 #ifndef _SYS_SYSPROTO_H_
1480 struct close_range_args {
1481 	u_int	lowfd;
1482 	u_int	highfd;
1483 	int	flags;
1484 };
1485 #endif
1486 int
1487 sys_close_range(struct thread *td, struct close_range_args *uap)
1488 {
1489 
1490 	AUDIT_ARG_FD(uap->lowfd);
1491 	AUDIT_ARG_CMD(uap->highfd);
1492 	AUDIT_ARG_FFLAGS(uap->flags);
1493 
1494 	if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0)
1495 		return (EINVAL);
1496 	return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1497 }
1498 
1499 #ifdef COMPAT_FREEBSD12
1500 /*
1501  * Close open file descriptors.
1502  */
1503 #ifndef _SYS_SYSPROTO_H_
1504 struct freebsd12_closefrom_args {
1505 	int	lowfd;
1506 };
1507 #endif
1508 /* ARGSUSED */
1509 int
1510 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1511 {
1512 	u_int lowfd;
1513 
1514 	AUDIT_ARG_FD(uap->lowfd);
1515 
1516 	/*
1517 	 * Treat negative starting file descriptor values identical to
1518 	 * closefrom(0) which closes all files.
1519 	 */
1520 	lowfd = MAX(0, uap->lowfd);
1521 	return (kern_close_range(td, 0, lowfd, ~0U));
1522 }
1523 #endif	/* COMPAT_FREEBSD12 */
1524 
1525 #if defined(COMPAT_43)
1526 /*
1527  * Return status information about a file descriptor.
1528  */
1529 #ifndef _SYS_SYSPROTO_H_
1530 struct ofstat_args {
1531 	int	fd;
1532 	struct	ostat *sb;
1533 };
1534 #endif
1535 /* ARGSUSED */
1536 int
1537 ofstat(struct thread *td, struct ofstat_args *uap)
1538 {
1539 	struct ostat oub;
1540 	struct stat ub;
1541 	int error;
1542 
1543 	error = kern_fstat(td, uap->fd, &ub);
1544 	if (error == 0) {
1545 		cvtstat(&ub, &oub);
1546 		error = copyout(&oub, uap->sb, sizeof(oub));
1547 	}
1548 	return (error);
1549 }
1550 #endif /* COMPAT_43 */
1551 
1552 #if defined(COMPAT_FREEBSD11)
1553 int
1554 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1555 {
1556 	struct stat sb;
1557 	struct freebsd11_stat osb;
1558 	int error;
1559 
1560 	error = kern_fstat(td, uap->fd, &sb);
1561 	if (error != 0)
1562 		return (error);
1563 	error = freebsd11_cvtstat(&sb, &osb);
1564 	if (error == 0)
1565 		error = copyout(&osb, uap->sb, sizeof(osb));
1566 	return (error);
1567 }
1568 #endif	/* COMPAT_FREEBSD11 */
1569 
1570 /*
1571  * Return status information about a file descriptor.
1572  */
1573 #ifndef _SYS_SYSPROTO_H_
1574 struct fstat_args {
1575 	int	fd;
1576 	struct	stat *sb;
1577 };
1578 #endif
1579 /* ARGSUSED */
1580 int
1581 sys_fstat(struct thread *td, struct fstat_args *uap)
1582 {
1583 	struct stat ub;
1584 	int error;
1585 
1586 	error = kern_fstat(td, uap->fd, &ub);
1587 	if (error == 0)
1588 		error = copyout(&ub, uap->sb, sizeof(ub));
1589 	return (error);
1590 }
1591 
1592 int
1593 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1594 {
1595 	struct file *fp;
1596 	int error;
1597 
1598 	AUDIT_ARG_FD(fd);
1599 
1600 	error = fget(td, fd, &cap_fstat_rights, &fp);
1601 	if (__predict_false(error != 0))
1602 		return (error);
1603 
1604 	AUDIT_ARG_FILE(td->td_proc, fp);
1605 
1606 	error = fo_stat(fp, sbp, td->td_ucred);
1607 	fdrop(fp, td);
1608 #ifdef __STAT_TIME_T_EXT
1609 	sbp->st_atim_ext = 0;
1610 	sbp->st_mtim_ext = 0;
1611 	sbp->st_ctim_ext = 0;
1612 	sbp->st_btim_ext = 0;
1613 #endif
1614 #ifdef KTRACE
1615 	if (KTRPOINT(td, KTR_STRUCT))
1616 		ktrstat_error(sbp, error);
1617 #endif
1618 	return (error);
1619 }
1620 
1621 #if defined(COMPAT_FREEBSD11)
1622 /*
1623  * Return status information about a file descriptor.
1624  */
1625 #ifndef _SYS_SYSPROTO_H_
1626 struct freebsd11_nfstat_args {
1627 	int	fd;
1628 	struct	nstat *sb;
1629 };
1630 #endif
1631 /* ARGSUSED */
1632 int
1633 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1634 {
1635 	struct nstat nub;
1636 	struct stat ub;
1637 	int error;
1638 
1639 	error = kern_fstat(td, uap->fd, &ub);
1640 	if (error != 0)
1641 		return (error);
1642 	error = freebsd11_cvtnstat(&ub, &nub);
1643 	if (error != 0)
1644 		error = copyout(&nub, uap->sb, sizeof(nub));
1645 	return (error);
1646 }
1647 #endif /* COMPAT_FREEBSD11 */
1648 
1649 /*
1650  * Return pathconf information about a file descriptor.
1651  */
1652 #ifndef _SYS_SYSPROTO_H_
1653 struct fpathconf_args {
1654 	int	fd;
1655 	int	name;
1656 };
1657 #endif
1658 /* ARGSUSED */
1659 int
1660 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1661 {
1662 	long value;
1663 	int error;
1664 
1665 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1666 	if (error == 0)
1667 		td->td_retval[0] = value;
1668 	return (error);
1669 }
1670 
1671 int
1672 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1673 {
1674 	struct file *fp;
1675 	struct vnode *vp;
1676 	int error;
1677 
1678 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1679 	if (error != 0)
1680 		return (error);
1681 
1682 	if (name == _PC_ASYNC_IO) {
1683 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1684 		goto out;
1685 	}
1686 	vp = fp->f_vnode;
1687 	if (vp != NULL) {
1688 		vn_lock(vp, LK_SHARED | LK_RETRY);
1689 		error = VOP_PATHCONF(vp, name, valuep);
1690 		VOP_UNLOCK(vp);
1691 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1692 		if (name != _PC_PIPE_BUF) {
1693 			error = EINVAL;
1694 		} else {
1695 			*valuep = PIPE_BUF;
1696 			error = 0;
1697 		}
1698 	} else {
1699 		error = EOPNOTSUPP;
1700 	}
1701 out:
1702 	fdrop(fp, td);
1703 	return (error);
1704 }
1705 
1706 /*
1707  * Copy filecaps structure allocating memory for ioctls array if needed.
1708  *
1709  * The last parameter indicates whether the fdtable is locked. If it is not and
1710  * ioctls are encountered, copying fails and the caller must lock the table.
1711  *
1712  * Note that if the table was not locked, the caller has to check the relevant
1713  * sequence counter to determine whether the operation was successful.
1714  */
1715 bool
1716 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1717 {
1718 	size_t size;
1719 
1720 	if (src->fc_ioctls != NULL && !locked)
1721 		return (false);
1722 	memcpy(dst, src, sizeof(*src));
1723 	if (src->fc_ioctls == NULL)
1724 		return (true);
1725 
1726 	KASSERT(src->fc_nioctls > 0,
1727 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1728 
1729 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1730 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1731 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1732 	return (true);
1733 }
1734 
1735 static u_long *
1736 filecaps_copy_prep(const struct filecaps *src)
1737 {
1738 	u_long *ioctls;
1739 	size_t size;
1740 
1741 	if (__predict_true(src->fc_ioctls == NULL))
1742 		return (NULL);
1743 
1744 	KASSERT(src->fc_nioctls > 0,
1745 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1746 
1747 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1748 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1749 	return (ioctls);
1750 }
1751 
1752 static void
1753 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1754     u_long *ioctls)
1755 {
1756 	size_t size;
1757 
1758 	*dst = *src;
1759 	if (__predict_true(src->fc_ioctls == NULL)) {
1760 		MPASS(ioctls == NULL);
1761 		return;
1762 	}
1763 
1764 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1765 	dst->fc_ioctls = ioctls;
1766 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1767 }
1768 
1769 /*
1770  * Move filecaps structure to the new place and clear the old place.
1771  */
1772 void
1773 filecaps_move(struct filecaps *src, struct filecaps *dst)
1774 {
1775 
1776 	*dst = *src;
1777 	bzero(src, sizeof(*src));
1778 }
1779 
1780 /*
1781  * Fill the given filecaps structure with full rights.
1782  */
1783 static void
1784 filecaps_fill(struct filecaps *fcaps)
1785 {
1786 
1787 	CAP_ALL(&fcaps->fc_rights);
1788 	fcaps->fc_ioctls = NULL;
1789 	fcaps->fc_nioctls = -1;
1790 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1791 }
1792 
1793 /*
1794  * Free memory allocated within filecaps structure.
1795  */
1796 void
1797 filecaps_free(struct filecaps *fcaps)
1798 {
1799 
1800 	free(fcaps->fc_ioctls, M_FILECAPS);
1801 	bzero(fcaps, sizeof(*fcaps));
1802 }
1803 
1804 static u_long *
1805 filecaps_free_prep(struct filecaps *fcaps)
1806 {
1807 	u_long *ioctls;
1808 
1809 	ioctls = fcaps->fc_ioctls;
1810 	bzero(fcaps, sizeof(*fcaps));
1811 	return (ioctls);
1812 }
1813 
1814 static void
1815 filecaps_free_finish(u_long *ioctls)
1816 {
1817 
1818 	free(ioctls, M_FILECAPS);
1819 }
1820 
1821 /*
1822  * Validate the given filecaps structure.
1823  */
1824 static void
1825 filecaps_validate(const struct filecaps *fcaps, const char *func)
1826 {
1827 
1828 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1829 	    ("%s: invalid rights", func));
1830 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1831 	    ("%s: invalid fcntls", func));
1832 	KASSERT(fcaps->fc_fcntls == 0 ||
1833 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1834 	    ("%s: fcntls without CAP_FCNTL", func));
1835 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1836 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1837 	    ("%s: invalid ioctls", func));
1838 	KASSERT(fcaps->fc_nioctls == 0 ||
1839 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1840 	    ("%s: ioctls without CAP_IOCTL", func));
1841 }
1842 
1843 static void
1844 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1845 {
1846 	int nfd1;
1847 
1848 	FILEDESC_XLOCK_ASSERT(fdp);
1849 
1850 	nfd1 = fdp->fd_nfiles * 2;
1851 	if (nfd1 < nfd)
1852 		nfd1 = nfd;
1853 	fdgrowtable(fdp, nfd1);
1854 }
1855 
1856 /*
1857  * Grow the file table to accommodate (at least) nfd descriptors.
1858  */
1859 static void
1860 fdgrowtable(struct filedesc *fdp, int nfd)
1861 {
1862 	struct filedesc0 *fdp0;
1863 	struct freetable *ft;
1864 	struct fdescenttbl *ntable;
1865 	struct fdescenttbl *otable;
1866 	int nnfiles, onfiles;
1867 	NDSLOTTYPE *nmap, *omap;
1868 
1869 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1870 
1871 	/* save old values */
1872 	onfiles = fdp->fd_nfiles;
1873 	otable = fdp->fd_files;
1874 	omap = fdp->fd_map;
1875 
1876 	/* compute the size of the new table */
1877 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1878 	if (nnfiles <= onfiles)
1879 		/* the table is already large enough */
1880 		return;
1881 
1882 	/*
1883 	 * Allocate a new table.  We need enough space for the number of
1884 	 * entries, file entries themselves and the struct freetable we will use
1885 	 * when we decommission the table and place it on the freelist.
1886 	 * We place the struct freetable in the middle so we don't have
1887 	 * to worry about padding.
1888 	 */
1889 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1890 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1891 	    sizeof(struct freetable),
1892 	    M_FILEDESC, M_ZERO | M_WAITOK);
1893 	/* copy the old data */
1894 	ntable->fdt_nfiles = nnfiles;
1895 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1896 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1897 
1898 	/*
1899 	 * Allocate a new map only if the old is not large enough.  It will
1900 	 * grow at a slower rate than the table as it can map more
1901 	 * entries than the table can hold.
1902 	 */
1903 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1904 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1905 		    M_ZERO | M_WAITOK);
1906 		/* copy over the old data and update the pointer */
1907 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1908 		fdp->fd_map = nmap;
1909 	}
1910 
1911 	/*
1912 	 * Make sure that ntable is correctly initialized before we replace
1913 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1914 	 * data.
1915 	 */
1916 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1917 
1918 	/*
1919 	 * Free the old file table when not shared by other threads or processes.
1920 	 * The old file table is considered to be shared when either are true:
1921 	 * - The process has more than one thread.
1922 	 * - The file descriptor table has been shared via fdshare().
1923 	 *
1924 	 * When shared, the old file table will be placed on a freelist
1925 	 * which will be processed when the struct filedesc is released.
1926 	 *
1927 	 * Note that if onfiles == NDFILE, we're dealing with the original
1928 	 * static allocation contained within (struct filedesc0 *)fdp,
1929 	 * which must not be freed.
1930 	 */
1931 	if (onfiles > NDFILE) {
1932 		/*
1933 		 * Note we may be called here from fdinit while allocating a
1934 		 * table for a new process in which case ->p_fd points
1935 		 * elsewhere.
1936 		 */
1937 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1938 			free(otable, M_FILEDESC);
1939 		} else {
1940 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1941 			fdp0 = (struct filedesc0 *)fdp;
1942 			ft->ft_table = otable;
1943 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1944 		}
1945 	}
1946 	/*
1947 	 * The map does not have the same possibility of threads still
1948 	 * holding references to it.  So always free it as long as it
1949 	 * does not reference the original static allocation.
1950 	 */
1951 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1952 		free(omap, M_FILEDESC);
1953 }
1954 
1955 /*
1956  * Allocate a file descriptor for the process.
1957  */
1958 int
1959 fdalloc(struct thread *td, int minfd, int *result)
1960 {
1961 	struct proc *p = td->td_proc;
1962 	struct filedesc *fdp = p->p_fd;
1963 	int fd, maxfd, allocfd;
1964 #ifdef RACCT
1965 	int error;
1966 #endif
1967 
1968 	FILEDESC_XLOCK_ASSERT(fdp);
1969 
1970 	if (fdp->fd_freefile > minfd)
1971 		minfd = fdp->fd_freefile;
1972 
1973 	maxfd = getmaxfd(td);
1974 
1975 	/*
1976 	 * Search the bitmap for a free descriptor starting at minfd.
1977 	 * If none is found, grow the file table.
1978 	 */
1979 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1980 	if (__predict_false(fd >= maxfd))
1981 		return (EMFILE);
1982 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1983 		allocfd = min(fd * 2, maxfd);
1984 #ifdef RACCT
1985 		if (RACCT_ENABLED()) {
1986 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1987 			if (error != 0)
1988 				return (EMFILE);
1989 		}
1990 #endif
1991 		/*
1992 		 * fd is already equal to first free descriptor >= minfd, so
1993 		 * we only need to grow the table and we are done.
1994 		 */
1995 		fdgrowtable_exp(fdp, allocfd);
1996 	}
1997 
1998 	/*
1999 	 * Perform some sanity checks, then mark the file descriptor as
2000 	 * used and return it to the caller.
2001 	 */
2002 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2003 	    ("invalid descriptor %d", fd));
2004 	KASSERT(!fdisused(fdp, fd),
2005 	    ("fd_first_free() returned non-free descriptor"));
2006 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2007 	    ("file descriptor isn't free"));
2008 	fdused(fdp, fd);
2009 	*result = fd;
2010 	return (0);
2011 }
2012 
2013 /*
2014  * Allocate n file descriptors for the process.
2015  */
2016 int
2017 fdallocn(struct thread *td, int minfd, int *fds, int n)
2018 {
2019 	struct proc *p = td->td_proc;
2020 	struct filedesc *fdp = p->p_fd;
2021 	int i;
2022 
2023 	FILEDESC_XLOCK_ASSERT(fdp);
2024 
2025 	for (i = 0; i < n; i++)
2026 		if (fdalloc(td, 0, &fds[i]) != 0)
2027 			break;
2028 
2029 	if (i < n) {
2030 		for (i--; i >= 0; i--)
2031 			fdunused(fdp, fds[i]);
2032 		return (EMFILE);
2033 	}
2034 
2035 	return (0);
2036 }
2037 
2038 /*
2039  * Create a new open file structure and allocate a file descriptor for the
2040  * process that refers to it.  We add one reference to the file for the
2041  * descriptor table and one reference for resultfp. This is to prevent us
2042  * being preempted and the entry in the descriptor table closed after we
2043  * release the FILEDESC lock.
2044  */
2045 int
2046 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2047     struct filecaps *fcaps)
2048 {
2049 	struct file *fp;
2050 	int error, fd;
2051 
2052 	MPASS(resultfp != NULL);
2053 	MPASS(resultfd != NULL);
2054 
2055 	error = _falloc_noinstall(td, &fp, 2);
2056 	if (__predict_false(error != 0)) {
2057 		return (error);
2058 	}
2059 
2060 	error = finstall_refed(td, fp, &fd, flags, fcaps);
2061 	if (__predict_false(error != 0)) {
2062 		falloc_abort(td, fp);
2063 		return (error);
2064 	}
2065 
2066 	*resultfp = fp;
2067 	*resultfd = fd;
2068 
2069 	return (0);
2070 }
2071 
2072 /*
2073  * Create a new open file structure without allocating a file descriptor.
2074  */
2075 int
2076 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2077 {
2078 	struct file *fp;
2079 	int maxuserfiles = maxfiles - (maxfiles / 20);
2080 	int openfiles_new;
2081 	static struct timeval lastfail;
2082 	static int curfail;
2083 
2084 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2085 	MPASS(n > 0);
2086 
2087 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2088 	if ((openfiles_new >= maxuserfiles &&
2089 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2090 	    openfiles_new >= maxfiles) {
2091 		atomic_subtract_int(&openfiles, 1);
2092 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2093 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2094 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2095 		}
2096 		return (ENFILE);
2097 	}
2098 	fp = uma_zalloc(file_zone, M_WAITOK);
2099 	bzero(fp, sizeof(*fp));
2100 	refcount_init(&fp->f_count, n);
2101 	fp->f_cred = crhold(td->td_ucred);
2102 	fp->f_ops = &badfileops;
2103 	*resultfp = fp;
2104 	return (0);
2105 }
2106 
2107 void
2108 falloc_abort(struct thread *td, struct file *fp)
2109 {
2110 
2111 	/*
2112 	 * For assertion purposes.
2113 	 */
2114 	refcount_init(&fp->f_count, 0);
2115 	_fdrop(fp, td);
2116 }
2117 
2118 /*
2119  * Install a file in a file descriptor table.
2120  */
2121 void
2122 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2123     struct filecaps *fcaps)
2124 {
2125 	struct filedescent *fde;
2126 
2127 	MPASS(fp != NULL);
2128 	if (fcaps != NULL)
2129 		filecaps_validate(fcaps, __func__);
2130 	FILEDESC_XLOCK_ASSERT(fdp);
2131 
2132 	fde = &fdp->fd_ofiles[fd];
2133 #ifdef CAPABILITIES
2134 	seqc_write_begin(&fde->fde_seqc);
2135 #endif
2136 	fde->fde_file = fp;
2137 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2138 	if (fcaps != NULL)
2139 		filecaps_move(fcaps, &fde->fde_caps);
2140 	else
2141 		filecaps_fill(&fde->fde_caps);
2142 #ifdef CAPABILITIES
2143 	seqc_write_end(&fde->fde_seqc);
2144 #endif
2145 }
2146 
2147 int
2148 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2149     struct filecaps *fcaps)
2150 {
2151 	struct filedesc *fdp = td->td_proc->p_fd;
2152 	int error;
2153 
2154 	MPASS(fd != NULL);
2155 
2156 	FILEDESC_XLOCK(fdp);
2157 	error = fdalloc(td, 0, fd);
2158 	if (__predict_true(error == 0)) {
2159 		_finstall(fdp, fp, *fd, flags, fcaps);
2160 	}
2161 	FILEDESC_XUNLOCK(fdp);
2162 	return (error);
2163 }
2164 
2165 int
2166 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2167     struct filecaps *fcaps)
2168 {
2169 	int error;
2170 
2171 	MPASS(fd != NULL);
2172 
2173 	if (!fhold(fp))
2174 		return (EBADF);
2175 	error = finstall_refed(td, fp, fd, flags, fcaps);
2176 	if (__predict_false(error != 0)) {
2177 		fdrop(fp, td);
2178 	}
2179 	return (error);
2180 }
2181 
2182 /*
2183  * Build a new filedesc structure from another.
2184  *
2185  * If fdp is not NULL, return with it shared locked.
2186  */
2187 struct filedesc *
2188 fdinit(void)
2189 {
2190 	struct filedesc0 *newfdp0;
2191 	struct filedesc *newfdp;
2192 
2193 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2194 	newfdp = &newfdp0->fd_fd;
2195 
2196 	/* Create the file descriptor table. */
2197 	FILEDESC_LOCK_INIT(newfdp);
2198 	refcount_init(&newfdp->fd_refcnt, 1);
2199 	refcount_init(&newfdp->fd_holdcnt, 1);
2200 	newfdp->fd_map = newfdp0->fd_dmap;
2201 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2202 	newfdp->fd_files->fdt_nfiles = NDFILE;
2203 
2204 	return (newfdp);
2205 }
2206 
2207 /*
2208  * Build a pwddesc structure from another.
2209  * Copy the current, root, and jail root vnode references.
2210  *
2211  * If pdp is not NULL, return with it shared locked.
2212  */
2213 struct pwddesc *
2214 pdinit(struct pwddesc *pdp, bool keeplock)
2215 {
2216 	struct pwddesc *newpdp;
2217 	struct pwd *newpwd;
2218 
2219 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2220 
2221 	PWDDESC_LOCK_INIT(newpdp);
2222 	refcount_init(&newpdp->pd_refcount, 1);
2223 	newpdp->pd_cmask = CMASK;
2224 
2225 	if (pdp == NULL) {
2226 		newpwd = pwd_alloc();
2227 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2228 		return (newpdp);
2229 	}
2230 
2231 	PWDDESC_XLOCK(pdp);
2232 	newpwd = pwd_hold_pwddesc(pdp);
2233 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2234 	if (!keeplock)
2235 		PWDDESC_XUNLOCK(pdp);
2236 	return (newpdp);
2237 }
2238 
2239 /*
2240  * Hold either filedesc or pwddesc of the passed process.
2241  *
2242  * The process lock is used to synchronize against the target exiting and
2243  * freeing the data.
2244  *
2245  * Clearing can be ilustrated in 3 steps:
2246  * 1. set the pointer to NULL. Either routine can race against it, hence
2247  *   atomic_load_ptr.
2248  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2249  *   race to either still see the pointer or find NULL. It is still safe to
2250  *   grab a reference as clearing is stalled.
2251  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2252  *   guaranteed to see NULL, making it safe to finish clearing
2253  */
2254 static struct filedesc *
2255 fdhold(struct proc *p)
2256 {
2257 	struct filedesc *fdp;
2258 
2259 	PROC_LOCK_ASSERT(p, MA_OWNED);
2260 	fdp = atomic_load_ptr(&p->p_fd);
2261 	if (fdp != NULL)
2262 		refcount_acquire(&fdp->fd_holdcnt);
2263 	return (fdp);
2264 }
2265 
2266 static struct pwddesc *
2267 pdhold(struct proc *p)
2268 {
2269 	struct pwddesc *pdp;
2270 
2271 	PROC_LOCK_ASSERT(p, MA_OWNED);
2272 	pdp = atomic_load_ptr(&p->p_pd);
2273 	if (pdp != NULL)
2274 		refcount_acquire(&pdp->pd_refcount);
2275 	return (pdp);
2276 }
2277 
2278 static void
2279 fddrop(struct filedesc *fdp)
2280 {
2281 
2282 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2283 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2284 			return;
2285 	}
2286 
2287 	FILEDESC_LOCK_DESTROY(fdp);
2288 	uma_zfree(filedesc0_zone, fdp);
2289 }
2290 
2291 static void
2292 pddrop(struct pwddesc *pdp)
2293 {
2294 	struct pwd *pwd;
2295 
2296 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2297 		return;
2298 
2299 	PWDDESC_XLOCK(pdp);
2300 	if (refcount_release(&pdp->pd_refcount) == 0) {
2301 		PWDDESC_XUNLOCK(pdp);
2302 		return;
2303 	}
2304 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2305 	pwd_set(pdp, NULL);
2306 	PWDDESC_XUNLOCK(pdp);
2307 	pwd_drop(pwd);
2308 
2309 	PWDDESC_LOCK_DESTROY(pdp);
2310 	free(pdp, M_PWDDESC);
2311 }
2312 
2313 /*
2314  * Share a filedesc structure.
2315  */
2316 struct filedesc *
2317 fdshare(struct filedesc *fdp)
2318 {
2319 
2320 	refcount_acquire(&fdp->fd_refcnt);
2321 	return (fdp);
2322 }
2323 
2324 /*
2325  * Share a pwddesc structure.
2326  */
2327 struct pwddesc *
2328 pdshare(struct pwddesc *pdp)
2329 {
2330 	refcount_acquire(&pdp->pd_refcount);
2331 	return (pdp);
2332 }
2333 
2334 /*
2335  * Unshare a filedesc structure, if necessary by making a copy
2336  */
2337 void
2338 fdunshare(struct thread *td)
2339 {
2340 	struct filedesc *tmp;
2341 	struct proc *p = td->td_proc;
2342 
2343 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2344 		return;
2345 
2346 	tmp = fdcopy(p->p_fd);
2347 	fdescfree(td);
2348 	p->p_fd = tmp;
2349 }
2350 
2351 /*
2352  * Unshare a pwddesc structure.
2353  */
2354 void
2355 pdunshare(struct thread *td)
2356 {
2357 	struct pwddesc *pdp;
2358 	struct proc *p;
2359 
2360 	p = td->td_proc;
2361 	/* Not shared. */
2362 	if (refcount_load(&p->p_pd->pd_refcount) == 1)
2363 		return;
2364 
2365 	pdp = pdcopy(p->p_pd);
2366 	pdescfree(td);
2367 	p->p_pd = pdp;
2368 }
2369 
2370 /*
2371  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2372  * this is to ease callers, not catch errors.
2373  */
2374 struct filedesc *
2375 fdcopy(struct filedesc *fdp)
2376 {
2377 	struct filedesc *newfdp;
2378 	struct filedescent *nfde, *ofde;
2379 	int i, lastfile;
2380 
2381 	MPASS(fdp != NULL);
2382 
2383 	newfdp = fdinit();
2384 	FILEDESC_SLOCK(fdp);
2385 	for (;;) {
2386 		lastfile = fdlastfile(fdp);
2387 		if (lastfile < newfdp->fd_nfiles)
2388 			break;
2389 		FILEDESC_SUNLOCK(fdp);
2390 		fdgrowtable(newfdp, lastfile + 1);
2391 		FILEDESC_SLOCK(fdp);
2392 	}
2393 	/* copy all passable descriptors (i.e. not kqueue) */
2394 	newfdp->fd_freefile = fdp->fd_freefile;
2395 	FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2396 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2397 		    !fhold(ofde->fde_file)) {
2398 			if (newfdp->fd_freefile == fdp->fd_freefile)
2399 				newfdp->fd_freefile = i;
2400 			continue;
2401 		}
2402 		nfde = &newfdp->fd_ofiles[i];
2403 		*nfde = *ofde;
2404 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2405 		fdused_init(newfdp, i);
2406 	}
2407 	MPASS(newfdp->fd_freefile != -1);
2408 	FILEDESC_SUNLOCK(fdp);
2409 	return (newfdp);
2410 }
2411 
2412 /*
2413  * Copy a pwddesc structure.
2414  */
2415 struct pwddesc *
2416 pdcopy(struct pwddesc *pdp)
2417 {
2418 	struct pwddesc *newpdp;
2419 
2420 	MPASS(pdp != NULL);
2421 
2422 	newpdp = pdinit(pdp, true);
2423 	newpdp->pd_cmask = pdp->pd_cmask;
2424 	PWDDESC_XUNLOCK(pdp);
2425 	return (newpdp);
2426 }
2427 
2428 /*
2429  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2430  * one of processes using it exits) and the table used to be shared.
2431  */
2432 static void
2433 fdclearlocks(struct thread *td)
2434 {
2435 	struct filedesc *fdp;
2436 	struct filedesc_to_leader *fdtol;
2437 	struct flock lf;
2438 	struct file *fp;
2439 	struct proc *p;
2440 	struct vnode *vp;
2441 	int i;
2442 
2443 	p = td->td_proc;
2444 	fdp = p->p_fd;
2445 	fdtol = p->p_fdtol;
2446 	MPASS(fdtol != NULL);
2447 
2448 	FILEDESC_XLOCK(fdp);
2449 	KASSERT(fdtol->fdl_refcount > 0,
2450 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2451 	    fdtol->fdl_refcount));
2452 	if (fdtol->fdl_refcount == 1 &&
2453 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2454 		FILEDESC_FOREACH_FP(fdp, i, fp) {
2455 			if (fp->f_type != DTYPE_VNODE ||
2456 			    !fhold(fp))
2457 				continue;
2458 			FILEDESC_XUNLOCK(fdp);
2459 			lf.l_whence = SEEK_SET;
2460 			lf.l_start = 0;
2461 			lf.l_len = 0;
2462 			lf.l_type = F_UNLCK;
2463 			vp = fp->f_vnode;
2464 			(void) VOP_ADVLOCK(vp,
2465 			    (caddr_t)p->p_leader, F_UNLCK,
2466 			    &lf, F_POSIX);
2467 			FILEDESC_XLOCK(fdp);
2468 			fdrop(fp, td);
2469 		}
2470 	}
2471 retry:
2472 	if (fdtol->fdl_refcount == 1) {
2473 		if (fdp->fd_holdleaderscount > 0 &&
2474 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2475 			/*
2476 			 * close() or kern_dup() has cleared a reference
2477 			 * in a shared file descriptor table.
2478 			 */
2479 			fdp->fd_holdleaderswakeup = 1;
2480 			sx_sleep(&fdp->fd_holdleaderscount,
2481 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2482 			goto retry;
2483 		}
2484 		if (fdtol->fdl_holdcount > 0) {
2485 			/*
2486 			 * Ensure that fdtol->fdl_leader remains
2487 			 * valid in closef().
2488 			 */
2489 			fdtol->fdl_wakeup = 1;
2490 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2491 			    "fdlhold", 0);
2492 			goto retry;
2493 		}
2494 	}
2495 	fdtol->fdl_refcount--;
2496 	if (fdtol->fdl_refcount == 0 &&
2497 	    fdtol->fdl_holdcount == 0) {
2498 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2499 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2500 	} else
2501 		fdtol = NULL;
2502 	p->p_fdtol = NULL;
2503 	FILEDESC_XUNLOCK(fdp);
2504 	if (fdtol != NULL)
2505 		free(fdtol, M_FILEDESC_TO_LEADER);
2506 }
2507 
2508 /*
2509  * Release a filedesc structure.
2510  */
2511 static void
2512 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2513 {
2514 	struct filedesc0 *fdp0;
2515 	struct freetable *ft, *tft;
2516 	struct filedescent *fde;
2517 	struct file *fp;
2518 	int i;
2519 
2520 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2521 	    ("%s: fd table %p carries references", __func__, fdp));
2522 
2523 	/*
2524 	 * Serialize with threads iterating over the table, if any.
2525 	 */
2526 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2527 		FILEDESC_XLOCK(fdp);
2528 		FILEDESC_XUNLOCK(fdp);
2529 	}
2530 
2531 	FILEDESC_FOREACH_FDE(fdp, i, fde) {
2532 		fp = fde->fde_file;
2533 		fdefree_last(fde);
2534 		(void) closef(fp, td);
2535 	}
2536 
2537 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2538 		free(fdp->fd_map, M_FILEDESC);
2539 	if (fdp->fd_nfiles > NDFILE)
2540 		free(fdp->fd_files, M_FILEDESC);
2541 
2542 	fdp0 = (struct filedesc0 *)fdp;
2543 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2544 		free(ft->ft_table, M_FILEDESC);
2545 
2546 	fddrop(fdp);
2547 }
2548 
2549 void
2550 fdescfree(struct thread *td)
2551 {
2552 	struct proc *p;
2553 	struct filedesc *fdp;
2554 
2555 	p = td->td_proc;
2556 	fdp = p->p_fd;
2557 	MPASS(fdp != NULL);
2558 
2559 #ifdef RACCT
2560 	if (RACCT_ENABLED())
2561 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2562 #endif
2563 
2564 	if (p->p_fdtol != NULL)
2565 		fdclearlocks(td);
2566 
2567 	/*
2568 	 * Check fdhold for an explanation.
2569 	 */
2570 	atomic_store_ptr(&p->p_fd, NULL);
2571 	atomic_thread_fence_seq_cst();
2572 	PROC_WAIT_UNLOCKED(p);
2573 
2574 	if (refcount_release(&fdp->fd_refcnt) == 0)
2575 		return;
2576 
2577 	fdescfree_fds(td, fdp);
2578 }
2579 
2580 void
2581 pdescfree(struct thread *td)
2582 {
2583 	struct proc *p;
2584 	struct pwddesc *pdp;
2585 
2586 	p = td->td_proc;
2587 	pdp = p->p_pd;
2588 	MPASS(pdp != NULL);
2589 
2590 	/*
2591 	 * Check pdhold for an explanation.
2592 	 */
2593 	atomic_store_ptr(&p->p_pd, NULL);
2594 	atomic_thread_fence_seq_cst();
2595 	PROC_WAIT_UNLOCKED(p);
2596 
2597 	pddrop(pdp);
2598 }
2599 
2600 /*
2601  * For setugid programs, we don't want to people to use that setugidness
2602  * to generate error messages which write to a file which otherwise would
2603  * otherwise be off-limits to the process.  We check for filesystems where
2604  * the vnode can change out from under us after execve (like [lin]procfs).
2605  *
2606  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2607  * sufficient.  We also don't check for setugidness since we know we are.
2608  */
2609 static bool
2610 is_unsafe(struct file *fp)
2611 {
2612 	struct vnode *vp;
2613 
2614 	if (fp->f_type != DTYPE_VNODE)
2615 		return (false);
2616 
2617 	vp = fp->f_vnode;
2618 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2619 }
2620 
2621 /*
2622  * Make this setguid thing safe, if at all possible.
2623  */
2624 void
2625 fdsetugidsafety(struct thread *td)
2626 {
2627 	struct filedesc *fdp;
2628 	struct file *fp;
2629 	int i;
2630 
2631 	fdp = td->td_proc->p_fd;
2632 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2633 	    ("the fdtable should not be shared"));
2634 	MPASS(fdp->fd_nfiles >= 3);
2635 	for (i = 0; i <= 2; i++) {
2636 		fp = fdp->fd_ofiles[i].fde_file;
2637 		if (fp != NULL && is_unsafe(fp)) {
2638 			FILEDESC_XLOCK(fdp);
2639 			knote_fdclose(td, i);
2640 			/*
2641 			 * NULL-out descriptor prior to close to avoid
2642 			 * a race while close blocks.
2643 			 */
2644 			fdfree(fdp, i);
2645 			FILEDESC_XUNLOCK(fdp);
2646 			(void) closef(fp, td);
2647 		}
2648 	}
2649 }
2650 
2651 /*
2652  * If a specific file object occupies a specific file descriptor, close the
2653  * file descriptor entry and drop a reference on the file object.  This is a
2654  * convenience function to handle a subsequent error in a function that calls
2655  * falloc() that handles the race that another thread might have closed the
2656  * file descriptor out from under the thread creating the file object.
2657  */
2658 void
2659 fdclose(struct thread *td, struct file *fp, int idx)
2660 {
2661 	struct filedesc *fdp = td->td_proc->p_fd;
2662 
2663 	FILEDESC_XLOCK(fdp);
2664 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2665 		fdfree(fdp, idx);
2666 		FILEDESC_XUNLOCK(fdp);
2667 		fdrop(fp, td);
2668 	} else
2669 		FILEDESC_XUNLOCK(fdp);
2670 }
2671 
2672 /*
2673  * Close any files on exec?
2674  */
2675 void
2676 fdcloseexec(struct thread *td)
2677 {
2678 	struct filedesc *fdp;
2679 	struct filedescent *fde;
2680 	struct file *fp;
2681 	int i;
2682 
2683 	fdp = td->td_proc->p_fd;
2684 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2685 	    ("the fdtable should not be shared"));
2686 	FILEDESC_FOREACH_FDE(fdp, i, fde) {
2687 		fp = fde->fde_file;
2688 		if (fp->f_type == DTYPE_MQUEUE ||
2689 		    (fde->fde_flags & UF_EXCLOSE)) {
2690 			FILEDESC_XLOCK(fdp);
2691 			fdfree(fdp, i);
2692 			(void) closefp(fdp, i, fp, td, false, false);
2693 			FILEDESC_UNLOCK_ASSERT(fdp);
2694 		}
2695 	}
2696 }
2697 
2698 /*
2699  * It is unsafe for set[ug]id processes to be started with file
2700  * descriptors 0..2 closed, as these descriptors are given implicit
2701  * significance in the Standard C library.  fdcheckstd() will create a
2702  * descriptor referencing /dev/null for each of stdin, stdout, and
2703  * stderr that is not already open.
2704  */
2705 int
2706 fdcheckstd(struct thread *td)
2707 {
2708 	struct filedesc *fdp;
2709 	register_t save;
2710 	int i, error, devnull;
2711 
2712 	fdp = td->td_proc->p_fd;
2713 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2714 	    ("the fdtable should not be shared"));
2715 	MPASS(fdp->fd_nfiles >= 3);
2716 	devnull = -1;
2717 	for (i = 0; i <= 2; i++) {
2718 		if (fdp->fd_ofiles[i].fde_file != NULL)
2719 			continue;
2720 
2721 		save = td->td_retval[0];
2722 		if (devnull != -1) {
2723 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2724 		} else {
2725 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2726 			    UIO_SYSSPACE, O_RDWR, 0);
2727 			if (error == 0) {
2728 				devnull = td->td_retval[0];
2729 				KASSERT(devnull == i, ("we didn't get our fd"));
2730 			}
2731 		}
2732 		td->td_retval[0] = save;
2733 		if (error != 0)
2734 			return (error);
2735 	}
2736 	return (0);
2737 }
2738 
2739 /*
2740  * Internal form of close.  Decrement reference count on file structure.
2741  * Note: td may be NULL when closing a file that was being passed in a
2742  * message.
2743  */
2744 int
2745 closef(struct file *fp, struct thread *td)
2746 {
2747 	struct vnode *vp;
2748 	struct flock lf;
2749 	struct filedesc_to_leader *fdtol;
2750 	struct filedesc *fdp;
2751 
2752 	MPASS(td != NULL);
2753 
2754 	/*
2755 	 * POSIX record locking dictates that any close releases ALL
2756 	 * locks owned by this process.  This is handled by setting
2757 	 * a flag in the unlock to free ONLY locks obeying POSIX
2758 	 * semantics, and not to free BSD-style file locks.
2759 	 * If the descriptor was in a message, POSIX-style locks
2760 	 * aren't passed with the descriptor, and the thread pointer
2761 	 * will be NULL.  Callers should be careful only to pass a
2762 	 * NULL thread pointer when there really is no owning
2763 	 * context that might have locks, or the locks will be
2764 	 * leaked.
2765 	 */
2766 	if (fp->f_type == DTYPE_VNODE) {
2767 		vp = fp->f_vnode;
2768 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2769 			lf.l_whence = SEEK_SET;
2770 			lf.l_start = 0;
2771 			lf.l_len = 0;
2772 			lf.l_type = F_UNLCK;
2773 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2774 			    F_UNLCK, &lf, F_POSIX);
2775 		}
2776 		fdtol = td->td_proc->p_fdtol;
2777 		if (fdtol != NULL) {
2778 			/*
2779 			 * Handle special case where file descriptor table is
2780 			 * shared between multiple process leaders.
2781 			 */
2782 			fdp = td->td_proc->p_fd;
2783 			FILEDESC_XLOCK(fdp);
2784 			for (fdtol = fdtol->fdl_next;
2785 			    fdtol != td->td_proc->p_fdtol;
2786 			    fdtol = fdtol->fdl_next) {
2787 				if ((fdtol->fdl_leader->p_flag &
2788 				    P_ADVLOCK) == 0)
2789 					continue;
2790 				fdtol->fdl_holdcount++;
2791 				FILEDESC_XUNLOCK(fdp);
2792 				lf.l_whence = SEEK_SET;
2793 				lf.l_start = 0;
2794 				lf.l_len = 0;
2795 				lf.l_type = F_UNLCK;
2796 				vp = fp->f_vnode;
2797 				(void) VOP_ADVLOCK(vp,
2798 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2799 				    F_POSIX);
2800 				FILEDESC_XLOCK(fdp);
2801 				fdtol->fdl_holdcount--;
2802 				if (fdtol->fdl_holdcount == 0 &&
2803 				    fdtol->fdl_wakeup != 0) {
2804 					fdtol->fdl_wakeup = 0;
2805 					wakeup(fdtol);
2806 				}
2807 			}
2808 			FILEDESC_XUNLOCK(fdp);
2809 		}
2810 	}
2811 	return (fdrop_close(fp, td));
2812 }
2813 
2814 /*
2815  * Hack for file descriptor passing code.
2816  */
2817 void
2818 closef_nothread(struct file *fp)
2819 {
2820 
2821 	fdrop(fp, NULL);
2822 }
2823 
2824 /*
2825  * Initialize the file pointer with the specified properties.
2826  *
2827  * The ops are set with release semantics to be certain that the flags, type,
2828  * and data are visible when ops is.  This is to prevent ops methods from being
2829  * called with bad data.
2830  */
2831 void
2832 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2833 {
2834 	fp->f_data = data;
2835 	fp->f_flag = flag;
2836 	fp->f_type = type;
2837 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2838 }
2839 
2840 void
2841 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2842 {
2843 	fp->f_seqcount[UIO_READ] = 1;
2844 	fp->f_seqcount[UIO_WRITE] = 1;
2845 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2846 	    data, ops);
2847 }
2848 
2849 int
2850 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2851     struct file **fpp, struct filecaps *havecapsp)
2852 {
2853 	struct filedescent *fde;
2854 	int error;
2855 
2856 	FILEDESC_LOCK_ASSERT(fdp);
2857 
2858 	*fpp = NULL;
2859 	fde = fdeget_noref(fdp, fd);
2860 	if (fde == NULL) {
2861 		error = EBADF;
2862 		goto out;
2863 	}
2864 
2865 #ifdef CAPABILITIES
2866 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2867 	if (error != 0)
2868 		goto out;
2869 #endif
2870 
2871 	if (havecapsp != NULL)
2872 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2873 
2874 	*fpp = fde->fde_file;
2875 
2876 	error = 0;
2877 out:
2878 	return (error);
2879 }
2880 
2881 #ifdef CAPABILITIES
2882 int
2883 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2884     struct file **fpp, struct filecaps *havecapsp)
2885 {
2886 	struct filedesc *fdp = td->td_proc->p_fd;
2887 	int error;
2888 	struct file *fp;
2889 	seqc_t seq;
2890 
2891 	*fpp = NULL;
2892 	for (;;) {
2893 		error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2894 		if (error != 0)
2895 			return (error);
2896 
2897 		if (havecapsp != NULL) {
2898 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2899 			    havecapsp, false)) {
2900 				fdrop(fp, td);
2901 				goto get_locked;
2902 			}
2903 		}
2904 
2905 		if (!fd_modified(fdp, fd, seq))
2906 			break;
2907 		fdrop(fp, td);
2908 	}
2909 
2910 	*fpp = fp;
2911 	return (0);
2912 
2913 get_locked:
2914 	FILEDESC_SLOCK(fdp);
2915 	error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2916 	if (error == 0 && !fhold(*fpp))
2917 		error = EBADF;
2918 	FILEDESC_SUNLOCK(fdp);
2919 	return (error);
2920 }
2921 #else
2922 int
2923 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2924     struct file **fpp, struct filecaps *havecapsp)
2925 {
2926 	int error;
2927 	error = fget_unlocked(td, fd, needrightsp, fpp);
2928 	if (havecapsp != NULL && error == 0)
2929 		filecaps_fill(havecapsp);
2930 
2931 	return (error);
2932 }
2933 #endif
2934 
2935 #ifdef CAPABILITIES
2936 int
2937 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2938 {
2939 	const struct filedescent *fde;
2940 	const struct fdescenttbl *fdt;
2941 	struct filedesc *fdp;
2942 	struct file *fp;
2943 	struct vnode *vp;
2944 	const cap_rights_t *haverights;
2945 	cap_rights_t rights;
2946 	seqc_t seq;
2947 
2948 	VFS_SMR_ASSERT_ENTERED();
2949 
2950 	rights = *ndp->ni_rightsneeded;
2951 	cap_rights_set_one(&rights, CAP_LOOKUP);
2952 
2953 	fdp = curproc->p_fd;
2954 	fdt = fdp->fd_files;
2955 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2956 		return (EBADF);
2957 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2958 	fde = &fdt->fdt_ofiles[fd];
2959 	haverights = cap_rights_fde_inline(fde);
2960 	fp = fde->fde_file;
2961 	if (__predict_false(fp == NULL))
2962 		return (EAGAIN);
2963 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2964 		return (EAGAIN);
2965 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2966 	vp = fp->f_vnode;
2967 	if (__predict_false(vp == NULL)) {
2968 		return (EAGAIN);
2969 	}
2970 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2971 		return (EAGAIN);
2972 	}
2973 	/*
2974 	 * Use an acquire barrier to force re-reading of fdt so it is
2975 	 * refreshed for verification.
2976 	 */
2977 	atomic_thread_fence_acq();
2978 	fdt = fdp->fd_files;
2979 	if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
2980 		return (EAGAIN);
2981 	/*
2982 	 * If file descriptor doesn't have all rights,
2983 	 * all lookups relative to it must also be
2984 	 * strictly relative.
2985 	 *
2986 	 * Not yet supported by fast path.
2987 	 */
2988 	CAP_ALL(&rights);
2989 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2990 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2991 	    ndp->ni_filecaps.fc_nioctls != -1) {
2992 #ifdef notyet
2993 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2994 #else
2995 		return (EAGAIN);
2996 #endif
2997 	}
2998 	*vpp = vp;
2999 	return (0);
3000 }
3001 #else
3002 int
3003 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3004 {
3005 	const struct fdescenttbl *fdt;
3006 	struct filedesc *fdp;
3007 	struct file *fp;
3008 	struct vnode *vp;
3009 
3010 	VFS_SMR_ASSERT_ENTERED();
3011 
3012 	fdp = curproc->p_fd;
3013 	fdt = fdp->fd_files;
3014 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3015 		return (EBADF);
3016 	fp = fdt->fdt_ofiles[fd].fde_file;
3017 	if (__predict_false(fp == NULL))
3018 		return (EAGAIN);
3019 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3020 	vp = fp->f_vnode;
3021 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3022 		return (EAGAIN);
3023 	}
3024 	/*
3025 	 * Use an acquire barrier to force re-reading of fdt so it is
3026 	 * refreshed for verification.
3027 	 */
3028 	atomic_thread_fence_acq();
3029 	fdt = fdp->fd_files;
3030 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3031 		return (EAGAIN);
3032 	filecaps_fill(&ndp->ni_filecaps);
3033 	*vpp = vp;
3034 	return (0);
3035 }
3036 #endif
3037 
3038 /*
3039  * Fetch the descriptor locklessly.
3040  *
3041  * We avoid fdrop() races by never raising a refcount above 0.  To accomplish
3042  * this we have to use a cmpset loop rather than an atomic_add.  The descriptor
3043  * must be re-verified once we acquire a reference to be certain that the
3044  * identity is still correct and we did not lose a race due to preemption.
3045  *
3046  * Force a reload of fdt when looping. Another thread could reallocate
3047  * the table before this fd was closed, so it is possible that there is
3048  * a stale fp pointer in cached version.
3049  */
3050 #ifdef CAPABILITIES
3051 static int
3052 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3053     struct file **fpp, seqc_t *seqp)
3054 {
3055 	struct filedesc *fdp;
3056 	const struct filedescent *fde;
3057 	const struct fdescenttbl *fdt;
3058 	struct file *fp;
3059 	seqc_t seq;
3060 	cap_rights_t haverights;
3061 	int error;
3062 
3063 	fdp = td->td_proc->p_fd;
3064 	fdt = fdp->fd_files;
3065 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3066 		return (EBADF);
3067 
3068 	for (;;) {
3069 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3070 		fde = &fdt->fdt_ofiles[fd];
3071 		haverights = *cap_rights_fde_inline(fde);
3072 		fp = fde->fde_file;
3073 		if (__predict_false(fp == NULL)) {
3074 			if (seqc_consistent(fd_seqc(fdt, fd), seq))
3075 				return (EBADF);
3076 			fdt = atomic_load_ptr(&fdp->fd_files);
3077 			continue;
3078 		}
3079 		error = cap_check_inline(&haverights, needrightsp);
3080 		if (__predict_false(error != 0)) {
3081 			if (seqc_consistent(fd_seqc(fdt, fd), seq))
3082 				return (error);
3083 			fdt = atomic_load_ptr(&fdp->fd_files);
3084 			continue;
3085 		}
3086 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3087 			fdt = atomic_load_ptr(&fdp->fd_files);
3088 			continue;
3089 		}
3090 		/*
3091 		 * Use an acquire barrier to force re-reading of fdt so it is
3092 		 * refreshed for verification.
3093 		 */
3094 		atomic_thread_fence_acq();
3095 		fdt = fdp->fd_files;
3096 		if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3097 			break;
3098 		fdrop(fp, td);
3099 	}
3100 	*fpp = fp;
3101 	if (seqp != NULL) {
3102 		*seqp = seq;
3103 	}
3104 	return (0);
3105 }
3106 #else
3107 static int
3108 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3109     struct file **fpp, seqc_t *seqp __unused)
3110 {
3111 	struct filedesc *fdp;
3112 	const struct fdescenttbl *fdt;
3113 	struct file *fp;
3114 
3115 	fdp = td->td_proc->p_fd;
3116 	fdt = fdp->fd_files;
3117 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3118 		return (EBADF);
3119 
3120 	for (;;) {
3121 		fp = fdt->fdt_ofiles[fd].fde_file;
3122 		if (__predict_false(fp == NULL))
3123 			return (EBADF);
3124 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3125 			fdt = atomic_load_ptr(&fdp->fd_files);
3126 			continue;
3127 		}
3128 		/*
3129 		 * Use an acquire barrier to force re-reading of fdt so it is
3130 		 * refreshed for verification.
3131 		 */
3132 		atomic_thread_fence_acq();
3133 		fdt = fdp->fd_files;
3134 		if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3135 			break;
3136 		fdrop(fp, td);
3137 	}
3138 	*fpp = fp;
3139 	return (0);
3140 }
3141 #endif
3142 
3143 /*
3144  * See the comments in fget_unlocked_seq for an explanation of how this works.
3145  *
3146  * This is a simplified variant which bails out to the aforementioned routine
3147  * if anything goes wrong. In practice this only happens when userspace is
3148  * racing with itself.
3149  */
3150 int
3151 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3152     struct file **fpp)
3153 {
3154 	struct filedesc *fdp;
3155 #ifdef CAPABILITIES
3156 	const struct filedescent *fde;
3157 #endif
3158 	const struct fdescenttbl *fdt;
3159 	struct file *fp;
3160 #ifdef CAPABILITIES
3161 	seqc_t seq;
3162 	const cap_rights_t *haverights;
3163 #endif
3164 
3165 	fdp = td->td_proc->p_fd;
3166 	fdt = fdp->fd_files;
3167 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3168 		*fpp = NULL;
3169 		return (EBADF);
3170 	}
3171 #ifdef CAPABILITIES
3172 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3173 	fde = &fdt->fdt_ofiles[fd];
3174 	haverights = cap_rights_fde_inline(fde);
3175 	fp = fde->fde_file;
3176 #else
3177 	fp = fdt->fdt_ofiles[fd].fde_file;
3178 #endif
3179 	if (__predict_false(fp == NULL))
3180 		goto out_fallback;
3181 #ifdef CAPABILITIES
3182 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3183 		goto out_fallback;
3184 #endif
3185 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3186 		goto out_fallback;
3187 
3188 	/*
3189 	 * Use an acquire barrier to force re-reading of fdt so it is
3190 	 * refreshed for verification.
3191 	 */
3192 	atomic_thread_fence_acq();
3193 	fdt = fdp->fd_files;
3194 #ifdef	CAPABILITIES
3195 	if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3196 #else
3197 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3198 #endif
3199 		goto out_fdrop;
3200 	*fpp = fp;
3201 	return (0);
3202 out_fdrop:
3203 	fdrop(fp, td);
3204 out_fallback:
3205 	*fpp = NULL;
3206 	return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3207 }
3208 
3209 /*
3210  * Translate fd -> file when the caller guarantees the file descriptor table
3211  * can't be changed by others.
3212  *
3213  * Note this does not mean the file object itself is only visible to the caller,
3214  * merely that it wont disappear without having to be referenced.
3215  *
3216  * Must be paired with fput_only_user.
3217  */
3218 #ifdef	CAPABILITIES
3219 int
3220 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3221     struct file **fpp)
3222 {
3223 	const struct filedescent *fde;
3224 	const struct fdescenttbl *fdt;
3225 	const cap_rights_t *haverights;
3226 	struct file *fp;
3227 	int error;
3228 
3229 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3230 
3231 	*fpp = NULL;
3232 	if (__predict_false(fd >= fdp->fd_nfiles))
3233 		return (EBADF);
3234 
3235 	fdt = fdp->fd_files;
3236 	fde = &fdt->fdt_ofiles[fd];
3237 	fp = fde->fde_file;
3238 	if (__predict_false(fp == NULL))
3239 		return (EBADF);
3240 	MPASS(refcount_load(&fp->f_count) > 0);
3241 	haverights = cap_rights_fde_inline(fde);
3242 	error = cap_check_inline(haverights, needrightsp);
3243 	if (__predict_false(error != 0))
3244 		return (error);
3245 	*fpp = fp;
3246 	return (0);
3247 }
3248 #else
3249 int
3250 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3251     struct file **fpp)
3252 {
3253 	struct file *fp;
3254 
3255 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3256 
3257 	*fpp = NULL;
3258 	if (__predict_false(fd >= fdp->fd_nfiles))
3259 		return (EBADF);
3260 
3261 	fp = fdp->fd_ofiles[fd].fde_file;
3262 	if (__predict_false(fp == NULL))
3263 		return (EBADF);
3264 
3265 	MPASS(refcount_load(&fp->f_count) > 0);
3266 	*fpp = fp;
3267 	return (0);
3268 }
3269 #endif
3270 
3271 /*
3272  * Extract the file pointer associated with the specified descriptor for the
3273  * current user process.
3274  *
3275  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3276  * returned.
3277  *
3278  * File's rights will be checked against the capability rights mask.
3279  *
3280  * If an error occurred the non-zero error is returned and *fpp is set to
3281  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3282  * responsible for fdrop().
3283  */
3284 static __inline int
3285 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3286     cap_rights_t *needrightsp)
3287 {
3288 	struct file *fp;
3289 	int error;
3290 
3291 	*fpp = NULL;
3292 	error = fget_unlocked(td, fd, needrightsp, &fp);
3293 	if (__predict_false(error != 0))
3294 		return (error);
3295 	if (__predict_false(fp->f_ops == &badfileops)) {
3296 		fdrop(fp, td);
3297 		return (EBADF);
3298 	}
3299 
3300 	/*
3301 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3302 	 */
3303 	error = 0;
3304 	switch (flags) {
3305 	case FREAD:
3306 	case FWRITE:
3307 		if ((fp->f_flag & flags) == 0)
3308 			error = EBADF;
3309 		break;
3310 	case FEXEC:
3311 		if (fp->f_ops != &path_fileops &&
3312 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3313 		    (fp->f_flag & FWRITE) != 0))
3314 			error = EBADF;
3315 		break;
3316 	case 0:
3317 		break;
3318 	default:
3319 		KASSERT(0, ("wrong flags"));
3320 	}
3321 
3322 	if (error != 0) {
3323 		fdrop(fp, td);
3324 		return (error);
3325 	}
3326 
3327 	*fpp = fp;
3328 	return (0);
3329 }
3330 
3331 int
3332 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3333 {
3334 
3335 	return (_fget(td, fd, fpp, 0, rightsp));
3336 }
3337 
3338 int
3339 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3340     struct file **fpp)
3341 {
3342 	int error;
3343 #ifndef CAPABILITIES
3344 	error = _fget(td, fd, fpp, 0, rightsp);
3345 	if (maxprotp != NULL)
3346 		*maxprotp = VM_PROT_ALL;
3347 	return (error);
3348 #else
3349 	cap_rights_t fdrights;
3350 	struct filedesc *fdp;
3351 	struct file *fp;
3352 	seqc_t seq;
3353 
3354 	*fpp = NULL;
3355 	fdp = td->td_proc->p_fd;
3356 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3357 	for (;;) {
3358 		error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3359 		if (__predict_false(error != 0))
3360 			return (error);
3361 		if (__predict_false(fp->f_ops == &badfileops)) {
3362 			fdrop(fp, td);
3363 			return (EBADF);
3364 		}
3365 		if (maxprotp != NULL)
3366 			fdrights = *cap_rights(fdp, fd);
3367 		if (!fd_modified(fdp, fd, seq))
3368 			break;
3369 		fdrop(fp, td);
3370 	}
3371 
3372 	/*
3373 	 * If requested, convert capability rights to access flags.
3374 	 */
3375 	if (maxprotp != NULL)
3376 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3377 	*fpp = fp;
3378 	return (0);
3379 #endif
3380 }
3381 
3382 int
3383 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3384 {
3385 
3386 	return (_fget(td, fd, fpp, FREAD, rightsp));
3387 }
3388 
3389 int
3390 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3391 {
3392 
3393 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3394 }
3395 
3396 int
3397 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3398     struct file **fpp)
3399 {
3400 #ifndef CAPABILITIES
3401 	return (fget_unlocked(td, fd, rightsp, fpp));
3402 #else
3403 	struct filedesc *fdp = td->td_proc->p_fd;
3404 	struct file *fp;
3405 	int error;
3406 	seqc_t seq;
3407 
3408 	*fpp = NULL;
3409 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3410 	for (;;) {
3411 		error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3412 		if (error != 0)
3413 			return (error);
3414 		error = cap_fcntl_check(fdp, fd, needfcntl);
3415 		if (!fd_modified(fdp, fd, seq))
3416 			break;
3417 		fdrop(fp, td);
3418 	}
3419 	if (error != 0) {
3420 		fdrop(fp, td);
3421 		return (error);
3422 	}
3423 	*fpp = fp;
3424 	return (0);
3425 #endif
3426 }
3427 
3428 /*
3429  * Like fget() but loads the underlying vnode, or returns an error if the
3430  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3431  * never have VM objects.  The returned vnode will be vref()'d.
3432  *
3433  * XXX: what about the unused flags ?
3434  */
3435 static __inline int
3436 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3437     struct vnode **vpp)
3438 {
3439 	struct file *fp;
3440 	int error;
3441 
3442 	*vpp = NULL;
3443 	error = _fget(td, fd, &fp, flags, needrightsp);
3444 	if (error != 0)
3445 		return (error);
3446 	if (fp->f_vnode == NULL) {
3447 		error = EINVAL;
3448 	} else {
3449 		*vpp = fp->f_vnode;
3450 		vref(*vpp);
3451 	}
3452 	fdrop(fp, td);
3453 
3454 	return (error);
3455 }
3456 
3457 int
3458 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3459 {
3460 
3461 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3462 }
3463 
3464 int
3465 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3466     struct filecaps *havecaps, struct vnode **vpp)
3467 {
3468 	struct filecaps caps;
3469 	struct file *fp;
3470 	int error;
3471 
3472 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3473 	if (error != 0)
3474 		return (error);
3475 	if (fp->f_ops == &badfileops) {
3476 		error = EBADF;
3477 		goto out;
3478 	}
3479 	if (fp->f_vnode == NULL) {
3480 		error = EINVAL;
3481 		goto out;
3482 	}
3483 
3484 	*havecaps = caps;
3485 	*vpp = fp->f_vnode;
3486 	vref(*vpp);
3487 	fdrop(fp, td);
3488 
3489 	return (0);
3490 out:
3491 	filecaps_free(&caps);
3492 	fdrop(fp, td);
3493 	return (error);
3494 }
3495 
3496 int
3497 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3498 {
3499 
3500 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3501 }
3502 
3503 int
3504 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3505 {
3506 
3507 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3508 }
3509 
3510 #ifdef notyet
3511 int
3512 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3513     struct vnode **vpp)
3514 {
3515 
3516 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3517 }
3518 #endif
3519 
3520 /*
3521  * Handle the last reference to a file being closed.
3522  *
3523  * Without the noinline attribute clang keeps inlining the func thorough this
3524  * file when fdrop is used.
3525  */
3526 int __noinline
3527 _fdrop(struct file *fp, struct thread *td)
3528 {
3529 	int error;
3530 #ifdef INVARIANTS
3531 	int count;
3532 
3533 	count = refcount_load(&fp->f_count);
3534 	if (count != 0)
3535 		panic("fdrop: fp %p count %d", fp, count);
3536 #endif
3537 	error = fo_close(fp, td);
3538 	atomic_subtract_int(&openfiles, 1);
3539 	crfree(fp->f_cred);
3540 	free(fp->f_advice, M_FADVISE);
3541 	uma_zfree(file_zone, fp);
3542 
3543 	return (error);
3544 }
3545 
3546 /*
3547  * Apply an advisory lock on a file descriptor.
3548  *
3549  * Just attempt to get a record lock of the requested type on the entire file
3550  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3551  */
3552 #ifndef _SYS_SYSPROTO_H_
3553 struct flock_args {
3554 	int	fd;
3555 	int	how;
3556 };
3557 #endif
3558 /* ARGSUSED */
3559 int
3560 sys_flock(struct thread *td, struct flock_args *uap)
3561 {
3562 	struct file *fp;
3563 	struct vnode *vp;
3564 	struct flock lf;
3565 	int error;
3566 
3567 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3568 	if (error != 0)
3569 		return (error);
3570 	error = EOPNOTSUPP;
3571 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3572 		goto done;
3573 	}
3574 	if (fp->f_ops == &path_fileops) {
3575 		goto done;
3576 	}
3577 
3578 	error = 0;
3579 	vp = fp->f_vnode;
3580 	lf.l_whence = SEEK_SET;
3581 	lf.l_start = 0;
3582 	lf.l_len = 0;
3583 	if (uap->how & LOCK_UN) {
3584 		lf.l_type = F_UNLCK;
3585 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3586 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3587 		goto done;
3588 	}
3589 	if (uap->how & LOCK_EX)
3590 		lf.l_type = F_WRLCK;
3591 	else if (uap->how & LOCK_SH)
3592 		lf.l_type = F_RDLCK;
3593 	else {
3594 		error = EBADF;
3595 		goto done;
3596 	}
3597 	atomic_set_int(&fp->f_flag, FHASLOCK);
3598 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3599 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3600 done:
3601 	fdrop(fp, td);
3602 	return (error);
3603 }
3604 /*
3605  * Duplicate the specified descriptor to a free descriptor.
3606  */
3607 int
3608 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3609     int openerror, int *indxp)
3610 {
3611 	struct filedescent *newfde, *oldfde;
3612 	struct file *fp;
3613 	u_long *ioctls;
3614 	int error, indx;
3615 
3616 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3617 	    ("unexpected error %d in %s", openerror, __func__));
3618 
3619 	/*
3620 	 * If the to-be-dup'd fd number is greater than the allowed number
3621 	 * of file descriptors, or the fd to be dup'd has already been
3622 	 * closed, then reject.
3623 	 */
3624 	FILEDESC_XLOCK(fdp);
3625 	if ((fp = fget_noref(fdp, dfd)) == NULL) {
3626 		FILEDESC_XUNLOCK(fdp);
3627 		return (EBADF);
3628 	}
3629 
3630 	error = fdalloc(td, 0, &indx);
3631 	if (error != 0) {
3632 		FILEDESC_XUNLOCK(fdp);
3633 		return (error);
3634 	}
3635 
3636 	/*
3637 	 * There are two cases of interest here.
3638 	 *
3639 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3640 	 *
3641 	 * For ENXIO steal away the file structure from (dfd) and store it in
3642 	 * (indx).  (dfd) is effectively closed by this operation.
3643 	 */
3644 	switch (openerror) {
3645 	case ENODEV:
3646 		/*
3647 		 * Check that the mode the file is being opened for is a
3648 		 * subset of the mode of the existing descriptor.
3649 		 */
3650 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3651 			fdunused(fdp, indx);
3652 			FILEDESC_XUNLOCK(fdp);
3653 			return (EACCES);
3654 		}
3655 		if (!fhold(fp)) {
3656 			fdunused(fdp, indx);
3657 			FILEDESC_XUNLOCK(fdp);
3658 			return (EBADF);
3659 		}
3660 		newfde = &fdp->fd_ofiles[indx];
3661 		oldfde = &fdp->fd_ofiles[dfd];
3662 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3663 #ifdef CAPABILITIES
3664 		seqc_write_begin(&newfde->fde_seqc);
3665 #endif
3666 		fde_copy(oldfde, newfde);
3667 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3668 		    ioctls);
3669 #ifdef CAPABILITIES
3670 		seqc_write_end(&newfde->fde_seqc);
3671 #endif
3672 		break;
3673 	case ENXIO:
3674 		/*
3675 		 * Steal away the file pointer from dfd and stuff it into indx.
3676 		 */
3677 		newfde = &fdp->fd_ofiles[indx];
3678 		oldfde = &fdp->fd_ofiles[dfd];
3679 #ifdef CAPABILITIES
3680 		seqc_write_begin(&oldfde->fde_seqc);
3681 		seqc_write_begin(&newfde->fde_seqc);
3682 #endif
3683 		fde_copy(oldfde, newfde);
3684 		oldfde->fde_file = NULL;
3685 		fdunused(fdp, dfd);
3686 #ifdef CAPABILITIES
3687 		seqc_write_end(&newfde->fde_seqc);
3688 		seqc_write_end(&oldfde->fde_seqc);
3689 #endif
3690 		break;
3691 	}
3692 	FILEDESC_XUNLOCK(fdp);
3693 	*indxp = indx;
3694 	return (0);
3695 }
3696 
3697 /*
3698  * This sysctl determines if we will allow a process to chroot(2) if it
3699  * has a directory open:
3700  *	0: disallowed for all processes.
3701  *	1: allowed for processes that were not already chroot(2)'ed.
3702  *	2: allowed for all processes.
3703  */
3704 
3705 static int chroot_allow_open_directories = 1;
3706 
3707 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3708     &chroot_allow_open_directories, 0,
3709     "Allow a process to chroot(2) if it has a directory open");
3710 
3711 /*
3712  * Helper function for raised chroot(2) security function:  Refuse if
3713  * any filedescriptors are open directories.
3714  */
3715 static int
3716 chroot_refuse_vdir_fds(struct filedesc *fdp)
3717 {
3718 	struct vnode *vp;
3719 	struct file *fp;
3720 	int i;
3721 
3722 	FILEDESC_LOCK_ASSERT(fdp);
3723 
3724 	FILEDESC_FOREACH_FP(fdp, i, fp) {
3725 		if (fp->f_type == DTYPE_VNODE) {
3726 			vp = fp->f_vnode;
3727 			if (vp->v_type == VDIR)
3728 				return (EPERM);
3729 		}
3730 	}
3731 	return (0);
3732 }
3733 
3734 static void
3735 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3736 {
3737 
3738 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3739 		vrefact(oldpwd->pwd_cdir);
3740 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3741 	}
3742 
3743 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3744 		vrefact(oldpwd->pwd_rdir);
3745 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3746 	}
3747 
3748 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3749 		vrefact(oldpwd->pwd_jdir);
3750 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3751 	}
3752 }
3753 
3754 struct pwd *
3755 pwd_hold_pwddesc(struct pwddesc *pdp)
3756 {
3757 	struct pwd *pwd;
3758 
3759 	PWDDESC_ASSERT_XLOCKED(pdp);
3760 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3761 	if (pwd != NULL)
3762 		refcount_acquire(&pwd->pwd_refcount);
3763 	return (pwd);
3764 }
3765 
3766 bool
3767 pwd_hold_smr(struct pwd *pwd)
3768 {
3769 
3770 	MPASS(pwd != NULL);
3771 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3772 		return (true);
3773 	}
3774 	return (false);
3775 }
3776 
3777 struct pwd *
3778 pwd_hold(struct thread *td)
3779 {
3780 	struct pwddesc *pdp;
3781 	struct pwd *pwd;
3782 
3783 	pdp = td->td_proc->p_pd;
3784 
3785 	vfs_smr_enter();
3786 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3787 	if (pwd_hold_smr(pwd)) {
3788 		vfs_smr_exit();
3789 		return (pwd);
3790 	}
3791 	vfs_smr_exit();
3792 	PWDDESC_XLOCK(pdp);
3793 	pwd = pwd_hold_pwddesc(pdp);
3794 	MPASS(pwd != NULL);
3795 	PWDDESC_XUNLOCK(pdp);
3796 	return (pwd);
3797 }
3798 
3799 struct pwd *
3800 pwd_hold_proc(struct proc *p)
3801 {
3802 	struct pwddesc *pdp;
3803 	struct pwd *pwd;
3804 
3805 	PROC_ASSERT_HELD(p);
3806 	PROC_LOCK(p);
3807 	pdp = pdhold(p);
3808 	MPASS(pdp != NULL);
3809 	PROC_UNLOCK(p);
3810 
3811 	PWDDESC_XLOCK(pdp);
3812 	pwd = pwd_hold_pwddesc(pdp);
3813 	MPASS(pwd != NULL);
3814 	PWDDESC_XUNLOCK(pdp);
3815 	pddrop(pdp);
3816 	return (pwd);
3817 }
3818 
3819 static struct pwd *
3820 pwd_alloc(void)
3821 {
3822 	struct pwd *pwd;
3823 
3824 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3825 	bzero(pwd, sizeof(*pwd));
3826 	refcount_init(&pwd->pwd_refcount, 1);
3827 	return (pwd);
3828 }
3829 
3830 void
3831 pwd_drop(struct pwd *pwd)
3832 {
3833 
3834 	if (!refcount_release(&pwd->pwd_refcount))
3835 		return;
3836 
3837 	if (pwd->pwd_cdir != NULL)
3838 		vrele(pwd->pwd_cdir);
3839 	if (pwd->pwd_rdir != NULL)
3840 		vrele(pwd->pwd_rdir);
3841 	if (pwd->pwd_jdir != NULL)
3842 		vrele(pwd->pwd_jdir);
3843 	uma_zfree_smr(pwd_zone, pwd);
3844 }
3845 
3846 /*
3847 * The caller is responsible for invoking priv_check() and
3848 * mac_vnode_check_chroot() to authorize this operation.
3849 */
3850 int
3851 pwd_chroot(struct thread *td, struct vnode *vp)
3852 {
3853 	struct pwddesc *pdp;
3854 	struct filedesc *fdp;
3855 	struct pwd *newpwd, *oldpwd;
3856 	int error;
3857 
3858 	fdp = td->td_proc->p_fd;
3859 	pdp = td->td_proc->p_pd;
3860 	newpwd = pwd_alloc();
3861 	FILEDESC_SLOCK(fdp);
3862 	PWDDESC_XLOCK(pdp);
3863 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3864 	if (chroot_allow_open_directories == 0 ||
3865 	    (chroot_allow_open_directories == 1 &&
3866 	    oldpwd->pwd_rdir != rootvnode)) {
3867 		error = chroot_refuse_vdir_fds(fdp);
3868 		FILEDESC_SUNLOCK(fdp);
3869 		if (error != 0) {
3870 			PWDDESC_XUNLOCK(pdp);
3871 			pwd_drop(newpwd);
3872 			return (error);
3873 		}
3874 	} else {
3875 		FILEDESC_SUNLOCK(fdp);
3876 	}
3877 
3878 	vrefact(vp);
3879 	newpwd->pwd_rdir = vp;
3880 	if (oldpwd->pwd_jdir == NULL) {
3881 		vrefact(vp);
3882 		newpwd->pwd_jdir = vp;
3883 	}
3884 	pwd_fill(oldpwd, newpwd);
3885 	pwd_set(pdp, newpwd);
3886 	PWDDESC_XUNLOCK(pdp);
3887 	pwd_drop(oldpwd);
3888 	return (0);
3889 }
3890 
3891 void
3892 pwd_chdir(struct thread *td, struct vnode *vp)
3893 {
3894 	struct pwddesc *pdp;
3895 	struct pwd *newpwd, *oldpwd;
3896 
3897 	VNPASS(vp->v_usecount > 0, vp);
3898 
3899 	newpwd = pwd_alloc();
3900 	pdp = td->td_proc->p_pd;
3901 	PWDDESC_XLOCK(pdp);
3902 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3903 	newpwd->pwd_cdir = vp;
3904 	pwd_fill(oldpwd, newpwd);
3905 	pwd_set(pdp, newpwd);
3906 	PWDDESC_XUNLOCK(pdp);
3907 	pwd_drop(oldpwd);
3908 }
3909 
3910 /*
3911  * jail_attach(2) changes both root and working directories.
3912  */
3913 int
3914 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3915 {
3916 	struct pwddesc *pdp;
3917 	struct filedesc *fdp;
3918 	struct pwd *newpwd, *oldpwd;
3919 	int error;
3920 
3921 	fdp = td->td_proc->p_fd;
3922 	pdp = td->td_proc->p_pd;
3923 	newpwd = pwd_alloc();
3924 	FILEDESC_SLOCK(fdp);
3925 	PWDDESC_XLOCK(pdp);
3926 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3927 	error = chroot_refuse_vdir_fds(fdp);
3928 	FILEDESC_SUNLOCK(fdp);
3929 	if (error != 0) {
3930 		PWDDESC_XUNLOCK(pdp);
3931 		pwd_drop(newpwd);
3932 		return (error);
3933 	}
3934 
3935 	vrefact(vp);
3936 	newpwd->pwd_rdir = vp;
3937 	vrefact(vp);
3938 	newpwd->pwd_cdir = vp;
3939 	if (oldpwd->pwd_jdir == NULL) {
3940 		vrefact(vp);
3941 		newpwd->pwd_jdir = vp;
3942 	}
3943 	pwd_fill(oldpwd, newpwd);
3944 	pwd_set(pdp, newpwd);
3945 	PWDDESC_XUNLOCK(pdp);
3946 	pwd_drop(oldpwd);
3947 	return (0);
3948 }
3949 
3950 void
3951 pwd_ensure_dirs(void)
3952 {
3953 	struct pwddesc *pdp;
3954 	struct pwd *oldpwd, *newpwd;
3955 
3956 	pdp = curproc->p_pd;
3957 	PWDDESC_XLOCK(pdp);
3958 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3959 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3960 		PWDDESC_XUNLOCK(pdp);
3961 		return;
3962 	}
3963 	PWDDESC_XUNLOCK(pdp);
3964 
3965 	newpwd = pwd_alloc();
3966 	PWDDESC_XLOCK(pdp);
3967 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3968 	pwd_fill(oldpwd, newpwd);
3969 	if (newpwd->pwd_cdir == NULL) {
3970 		vrefact(rootvnode);
3971 		newpwd->pwd_cdir = rootvnode;
3972 	}
3973 	if (newpwd->pwd_rdir == NULL) {
3974 		vrefact(rootvnode);
3975 		newpwd->pwd_rdir = rootvnode;
3976 	}
3977 	pwd_set(pdp, newpwd);
3978 	PWDDESC_XUNLOCK(pdp);
3979 	pwd_drop(oldpwd);
3980 }
3981 
3982 void
3983 pwd_set_rootvnode(void)
3984 {
3985 	struct pwddesc *pdp;
3986 	struct pwd *oldpwd, *newpwd;
3987 
3988 	pdp = curproc->p_pd;
3989 
3990 	newpwd = pwd_alloc();
3991 	PWDDESC_XLOCK(pdp);
3992 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3993 	vrefact(rootvnode);
3994 	newpwd->pwd_cdir = rootvnode;
3995 	vrefact(rootvnode);
3996 	newpwd->pwd_rdir = rootvnode;
3997 	pwd_fill(oldpwd, newpwd);
3998 	pwd_set(pdp, newpwd);
3999 	PWDDESC_XUNLOCK(pdp);
4000 	pwd_drop(oldpwd);
4001 }
4002 
4003 /*
4004  * Scan all active processes and prisons to see if any of them have a current
4005  * or root directory of `olddp'. If so, replace them with the new mount point.
4006  */
4007 void
4008 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4009 {
4010 	struct pwddesc *pdp;
4011 	struct pwd *newpwd, *oldpwd;
4012 	struct prison *pr;
4013 	struct proc *p;
4014 	int nrele;
4015 
4016 	if (vrefcnt(olddp) == 1)
4017 		return;
4018 	nrele = 0;
4019 	newpwd = pwd_alloc();
4020 	sx_slock(&allproc_lock);
4021 	FOREACH_PROC_IN_SYSTEM(p) {
4022 		PROC_LOCK(p);
4023 		pdp = pdhold(p);
4024 		PROC_UNLOCK(p);
4025 		if (pdp == NULL)
4026 			continue;
4027 		PWDDESC_XLOCK(pdp);
4028 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4029 		if (oldpwd == NULL ||
4030 		    (oldpwd->pwd_cdir != olddp &&
4031 		    oldpwd->pwd_rdir != olddp &&
4032 		    oldpwd->pwd_jdir != olddp)) {
4033 			PWDDESC_XUNLOCK(pdp);
4034 			pddrop(pdp);
4035 			continue;
4036 		}
4037 		if (oldpwd->pwd_cdir == olddp) {
4038 			vrefact(newdp);
4039 			newpwd->pwd_cdir = newdp;
4040 		}
4041 		if (oldpwd->pwd_rdir == olddp) {
4042 			vrefact(newdp);
4043 			newpwd->pwd_rdir = newdp;
4044 		}
4045 		if (oldpwd->pwd_jdir == olddp) {
4046 			vrefact(newdp);
4047 			newpwd->pwd_jdir = newdp;
4048 		}
4049 		pwd_fill(oldpwd, newpwd);
4050 		pwd_set(pdp, newpwd);
4051 		PWDDESC_XUNLOCK(pdp);
4052 		pwd_drop(oldpwd);
4053 		pddrop(pdp);
4054 		newpwd = pwd_alloc();
4055 	}
4056 	sx_sunlock(&allproc_lock);
4057 	pwd_drop(newpwd);
4058 	if (rootvnode == olddp) {
4059 		vrefact(newdp);
4060 		rootvnode = newdp;
4061 		nrele++;
4062 	}
4063 	mtx_lock(&prison0.pr_mtx);
4064 	if (prison0.pr_root == olddp) {
4065 		vrefact(newdp);
4066 		prison0.pr_root = newdp;
4067 		nrele++;
4068 	}
4069 	mtx_unlock(&prison0.pr_mtx);
4070 	sx_slock(&allprison_lock);
4071 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4072 		mtx_lock(&pr->pr_mtx);
4073 		if (pr->pr_root == olddp) {
4074 			vrefact(newdp);
4075 			pr->pr_root = newdp;
4076 			nrele++;
4077 		}
4078 		mtx_unlock(&pr->pr_mtx);
4079 	}
4080 	sx_sunlock(&allprison_lock);
4081 	while (nrele--)
4082 		vrele(olddp);
4083 }
4084 
4085 struct filedesc_to_leader *
4086 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4087 {
4088 	struct filedesc_to_leader *fdtol;
4089 
4090 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4091 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4092 	fdtol->fdl_refcount = 1;
4093 	fdtol->fdl_holdcount = 0;
4094 	fdtol->fdl_wakeup = 0;
4095 	fdtol->fdl_leader = leader;
4096 	if (old != NULL) {
4097 		FILEDESC_XLOCK(fdp);
4098 		fdtol->fdl_next = old->fdl_next;
4099 		fdtol->fdl_prev = old;
4100 		old->fdl_next = fdtol;
4101 		fdtol->fdl_next->fdl_prev = fdtol;
4102 		FILEDESC_XUNLOCK(fdp);
4103 	} else {
4104 		fdtol->fdl_next = fdtol;
4105 		fdtol->fdl_prev = fdtol;
4106 	}
4107 	return (fdtol);
4108 }
4109 
4110 static int
4111 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4112 {
4113 	NDSLOTTYPE *map;
4114 	struct filedesc *fdp;
4115 	u_int namelen;
4116 	int count, off, minoff;
4117 
4118 	namelen = arg2;
4119 	if (namelen != 1)
4120 		return (EINVAL);
4121 
4122 	if (*(int *)arg1 != 0)
4123 		return (EINVAL);
4124 
4125 	fdp = curproc->p_fd;
4126 	count = 0;
4127 	FILEDESC_SLOCK(fdp);
4128 	map = fdp->fd_map;
4129 	off = NDSLOT(fdp->fd_nfiles - 1);
4130 	for (minoff = NDSLOT(0); off >= minoff; --off)
4131 		count += bitcountl(map[off]);
4132 	FILEDESC_SUNLOCK(fdp);
4133 
4134 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4135 }
4136 
4137 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4138     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4139     "Number of open file descriptors");
4140 
4141 /*
4142  * Get file structures globally.
4143  */
4144 static int
4145 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4146 {
4147 	struct xfile xf;
4148 	struct filedesc *fdp;
4149 	struct file *fp;
4150 	struct proc *p;
4151 	int error, n;
4152 
4153 	error = sysctl_wire_old_buffer(req, 0);
4154 	if (error != 0)
4155 		return (error);
4156 	if (req->oldptr == NULL) {
4157 		n = 0;
4158 		sx_slock(&allproc_lock);
4159 		FOREACH_PROC_IN_SYSTEM(p) {
4160 			PROC_LOCK(p);
4161 			if (p->p_state == PRS_NEW) {
4162 				PROC_UNLOCK(p);
4163 				continue;
4164 			}
4165 			fdp = fdhold(p);
4166 			PROC_UNLOCK(p);
4167 			if (fdp == NULL)
4168 				continue;
4169 			/* overestimates sparse tables. */
4170 			n += fdp->fd_nfiles;
4171 			fddrop(fdp);
4172 		}
4173 		sx_sunlock(&allproc_lock);
4174 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4175 	}
4176 	error = 0;
4177 	bzero(&xf, sizeof(xf));
4178 	xf.xf_size = sizeof(xf);
4179 	sx_slock(&allproc_lock);
4180 	FOREACH_PROC_IN_SYSTEM(p) {
4181 		PROC_LOCK(p);
4182 		if (p->p_state == PRS_NEW) {
4183 			PROC_UNLOCK(p);
4184 			continue;
4185 		}
4186 		if (p_cansee(req->td, p) != 0) {
4187 			PROC_UNLOCK(p);
4188 			continue;
4189 		}
4190 		xf.xf_pid = p->p_pid;
4191 		xf.xf_uid = p->p_ucred->cr_uid;
4192 		fdp = fdhold(p);
4193 		PROC_UNLOCK(p);
4194 		if (fdp == NULL)
4195 			continue;
4196 		FILEDESC_SLOCK(fdp);
4197 		FILEDESC_FOREACH_FP(fdp, n, fp) {
4198 			if (refcount_load(&fdp->fd_refcnt) == 0)
4199 				break;
4200 			xf.xf_fd = n;
4201 			xf.xf_file = (uintptr_t)fp;
4202 			xf.xf_data = (uintptr_t)fp->f_data;
4203 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4204 			xf.xf_type = (uintptr_t)fp->f_type;
4205 			xf.xf_count = refcount_load(&fp->f_count);
4206 			xf.xf_msgcount = 0;
4207 			xf.xf_offset = foffset_get(fp);
4208 			xf.xf_flag = fp->f_flag;
4209 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4210 			if (error)
4211 				break;
4212 		}
4213 		FILEDESC_SUNLOCK(fdp);
4214 		fddrop(fdp);
4215 		if (error)
4216 			break;
4217 	}
4218 	sx_sunlock(&allproc_lock);
4219 	return (error);
4220 }
4221 
4222 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4223     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4224 
4225 #ifdef KINFO_FILE_SIZE
4226 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4227 #endif
4228 
4229 static int
4230 xlate_fflags(int fflags)
4231 {
4232 	static const struct {
4233 		int	fflag;
4234 		int	kf_fflag;
4235 	} fflags_table[] = {
4236 		{ FAPPEND, KF_FLAG_APPEND },
4237 		{ FASYNC, KF_FLAG_ASYNC },
4238 		{ FFSYNC, KF_FLAG_FSYNC },
4239 		{ FHASLOCK, KF_FLAG_HASLOCK },
4240 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4241 		{ FREAD, KF_FLAG_READ },
4242 		{ FWRITE, KF_FLAG_WRITE },
4243 		{ O_CREAT, KF_FLAG_CREAT },
4244 		{ O_DIRECT, KF_FLAG_DIRECT },
4245 		{ O_EXCL, KF_FLAG_EXCL },
4246 		{ O_EXEC, KF_FLAG_EXEC },
4247 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4248 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4249 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4250 		{ O_TRUNC, KF_FLAG_TRUNC }
4251 	};
4252 	unsigned int i;
4253 	int kflags;
4254 
4255 	kflags = 0;
4256 	for (i = 0; i < nitems(fflags_table); i++)
4257 		if (fflags & fflags_table[i].fflag)
4258 			kflags |=  fflags_table[i].kf_fflag;
4259 	return (kflags);
4260 }
4261 
4262 /* Trim unused data from kf_path by truncating the structure size. */
4263 void
4264 pack_kinfo(struct kinfo_file *kif)
4265 {
4266 
4267 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4268 	    strlen(kif->kf_path) + 1;
4269 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4270 }
4271 
4272 static void
4273 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4274     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4275 {
4276 	int error;
4277 
4278 	bzero(kif, sizeof(*kif));
4279 
4280 	/* Set a default type to allow for empty fill_kinfo() methods. */
4281 	kif->kf_type = KF_TYPE_UNKNOWN;
4282 	kif->kf_flags = xlate_fflags(fp->f_flag);
4283 	if (rightsp != NULL)
4284 		kif->kf_cap_rights = *rightsp;
4285 	else
4286 		cap_rights_init_zero(&kif->kf_cap_rights);
4287 	kif->kf_fd = fd;
4288 	kif->kf_ref_count = refcount_load(&fp->f_count);
4289 	kif->kf_offset = foffset_get(fp);
4290 
4291 	/*
4292 	 * This may drop the filedesc lock, so the 'fp' cannot be
4293 	 * accessed after this call.
4294 	 */
4295 	error = fo_fill_kinfo(fp, kif, fdp);
4296 	if (error == 0)
4297 		kif->kf_status |= KF_ATTR_VALID;
4298 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4299 		pack_kinfo(kif);
4300 	else
4301 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4302 }
4303 
4304 static void
4305 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4306     struct kinfo_file *kif, int flags)
4307 {
4308 	int error;
4309 
4310 	bzero(kif, sizeof(*kif));
4311 
4312 	kif->kf_type = KF_TYPE_VNODE;
4313 	error = vn_fill_kinfo_vnode(vp, kif);
4314 	if (error == 0)
4315 		kif->kf_status |= KF_ATTR_VALID;
4316 	kif->kf_flags = xlate_fflags(fflags);
4317 	cap_rights_init_zero(&kif->kf_cap_rights);
4318 	kif->kf_fd = fd;
4319 	kif->kf_ref_count = -1;
4320 	kif->kf_offset = -1;
4321 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4322 		pack_kinfo(kif);
4323 	else
4324 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4325 	vrele(vp);
4326 }
4327 
4328 struct export_fd_buf {
4329 	struct filedesc		*fdp;
4330 	struct pwddesc	*pdp;
4331 	struct sbuf 		*sb;
4332 	ssize_t			remainder;
4333 	struct kinfo_file	kif;
4334 	int			flags;
4335 };
4336 
4337 static int
4338 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4339 {
4340 	struct kinfo_file *kif;
4341 
4342 	kif = &efbuf->kif;
4343 	if (efbuf->remainder != -1) {
4344 		if (efbuf->remainder < kif->kf_structsize)
4345 			return (ENOMEM);
4346 		efbuf->remainder -= kif->kf_structsize;
4347 	}
4348 	if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4349 		return (sbuf_error(efbuf->sb));
4350 	return (0);
4351 }
4352 
4353 static int
4354 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4355     struct export_fd_buf *efbuf)
4356 {
4357 	int error;
4358 
4359 	if (efbuf->remainder == 0)
4360 		return (ENOMEM);
4361 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4362 	    efbuf->flags);
4363 	FILEDESC_SUNLOCK(efbuf->fdp);
4364 	error = export_kinfo_to_sb(efbuf);
4365 	FILEDESC_SLOCK(efbuf->fdp);
4366 	return (error);
4367 }
4368 
4369 static int
4370 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4371     struct export_fd_buf *efbuf)
4372 {
4373 	int error;
4374 
4375 	if (efbuf->remainder == 0)
4376 		return (ENOMEM);
4377 	if (efbuf->pdp != NULL)
4378 		PWDDESC_XUNLOCK(efbuf->pdp);
4379 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4380 	error = export_kinfo_to_sb(efbuf);
4381 	if (efbuf->pdp != NULL)
4382 		PWDDESC_XLOCK(efbuf->pdp);
4383 	return (error);
4384 }
4385 
4386 /*
4387  * Store a process file descriptor information to sbuf.
4388  *
4389  * Takes a locked proc as argument, and returns with the proc unlocked.
4390  */
4391 int
4392 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4393     int flags)
4394 {
4395 	struct file *fp;
4396 	struct filedesc *fdp;
4397 	struct pwddesc *pdp;
4398 	struct export_fd_buf *efbuf;
4399 	struct vnode *cttyvp, *textvp, *tracevp;
4400 	struct pwd *pwd;
4401 	int error, i;
4402 	cap_rights_t rights;
4403 
4404 	PROC_LOCK_ASSERT(p, MA_OWNED);
4405 
4406 	/* ktrace vnode */
4407 	tracevp = ktr_get_tracevp(p, true);
4408 	/* text vnode */
4409 	textvp = p->p_textvp;
4410 	if (textvp != NULL)
4411 		vrefact(textvp);
4412 	/* Controlling tty. */
4413 	cttyvp = NULL;
4414 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4415 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4416 		if (cttyvp != NULL)
4417 			vrefact(cttyvp);
4418 	}
4419 	fdp = fdhold(p);
4420 	pdp = pdhold(p);
4421 	PROC_UNLOCK(p);
4422 
4423 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4424 	efbuf->fdp = NULL;
4425 	efbuf->pdp = NULL;
4426 	efbuf->sb = sb;
4427 	efbuf->remainder = maxlen;
4428 	efbuf->flags = flags;
4429 
4430 	error = 0;
4431 	if (tracevp != NULL)
4432 		error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4433 		    FREAD | FWRITE, efbuf);
4434 	if (error == 0 && textvp != NULL)
4435 		error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4436 		    efbuf);
4437 	if (error == 0 && cttyvp != NULL)
4438 		error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4439 		    FREAD | FWRITE, efbuf);
4440 	if (error != 0 || pdp == NULL || fdp == NULL)
4441 		goto fail;
4442 	efbuf->fdp = fdp;
4443 	efbuf->pdp = pdp;
4444 	PWDDESC_XLOCK(pdp);
4445 	pwd = pwd_hold_pwddesc(pdp);
4446 	if (pwd != NULL) {
4447 		/* working directory */
4448 		if (pwd->pwd_cdir != NULL) {
4449 			vrefact(pwd->pwd_cdir);
4450 			error = export_vnode_to_sb(pwd->pwd_cdir,
4451 			    KF_FD_TYPE_CWD, FREAD, efbuf);
4452 		}
4453 		/* root directory */
4454 		if (error == 0 && pwd->pwd_rdir != NULL) {
4455 			vrefact(pwd->pwd_rdir);
4456 			error = export_vnode_to_sb(pwd->pwd_rdir,
4457 			    KF_FD_TYPE_ROOT, FREAD, efbuf);
4458 		}
4459 		/* jail directory */
4460 		if (error == 0 && pwd->pwd_jdir != NULL) {
4461 			vrefact(pwd->pwd_jdir);
4462 			error = export_vnode_to_sb(pwd->pwd_jdir,
4463 			    KF_FD_TYPE_JAIL, FREAD, efbuf);
4464 		}
4465 	}
4466 	PWDDESC_XUNLOCK(pdp);
4467 	if (error != 0)
4468 		goto fail;
4469 	if (pwd != NULL)
4470 		pwd_drop(pwd);
4471 	FILEDESC_SLOCK(fdp);
4472 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4473 		if (refcount_load(&fdp->fd_refcnt) == 0)
4474 			break;
4475 #ifdef CAPABILITIES
4476 		rights = *cap_rights(fdp, i);
4477 #else /* !CAPABILITIES */
4478 		rights = cap_no_rights;
4479 #endif
4480 		/*
4481 		 * Create sysctl entry.  It is OK to drop the filedesc
4482 		 * lock inside of export_file_to_sb() as we will
4483 		 * re-validate and re-evaluate its properties when the
4484 		 * loop continues.
4485 		 */
4486 		error = export_file_to_sb(fp, i, &rights, efbuf);
4487 		if (error != 0)
4488 			break;
4489 	}
4490 	FILEDESC_SUNLOCK(fdp);
4491 fail:
4492 	if (fdp != NULL)
4493 		fddrop(fdp);
4494 	if (pdp != NULL)
4495 		pddrop(pdp);
4496 	free(efbuf, M_TEMP);
4497 	return (error);
4498 }
4499 
4500 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4501 
4502 /*
4503  * Get per-process file descriptors for use by procstat(1), et al.
4504  */
4505 static int
4506 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4507 {
4508 	struct sbuf sb;
4509 	struct proc *p;
4510 	ssize_t maxlen;
4511 	u_int namelen;
4512 	int error, error2, *name;
4513 
4514 	namelen = arg2;
4515 	if (namelen != 1)
4516 		return (EINVAL);
4517 
4518 	name = (int *)arg1;
4519 
4520 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4521 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4522 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4523 	if (error != 0) {
4524 		sbuf_delete(&sb);
4525 		return (error);
4526 	}
4527 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4528 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4529 	    KERN_FILEDESC_PACK_KINFO);
4530 	error2 = sbuf_finish(&sb);
4531 	sbuf_delete(&sb);
4532 	return (error != 0 ? error : error2);
4533 }
4534 
4535 #ifdef COMPAT_FREEBSD7
4536 #ifdef KINFO_OFILE_SIZE
4537 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4538 #endif
4539 
4540 static void
4541 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4542 {
4543 
4544 	okif->kf_structsize = sizeof(*okif);
4545 	okif->kf_type = kif->kf_type;
4546 	okif->kf_fd = kif->kf_fd;
4547 	okif->kf_ref_count = kif->kf_ref_count;
4548 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4549 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4550 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4551 	okif->kf_offset = kif->kf_offset;
4552 	if (kif->kf_type == KF_TYPE_VNODE)
4553 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4554 	else
4555 		okif->kf_vnode_type = KF_VTYPE_VNON;
4556 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4557 	if (kif->kf_type == KF_TYPE_SOCKET) {
4558 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4559 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4560 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4561 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4562 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4563 	} else {
4564 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4565 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4566 	}
4567 }
4568 
4569 static int
4570 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4571     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4572 {
4573 	int error;
4574 
4575 	vrefact(vp);
4576 	PWDDESC_XUNLOCK(pdp);
4577 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4578 	kinfo_to_okinfo(kif, okif);
4579 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4580 	PWDDESC_XLOCK(pdp);
4581 	return (error);
4582 }
4583 
4584 /*
4585  * Get per-process file descriptors for use by procstat(1), et al.
4586  */
4587 static int
4588 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4589 {
4590 	struct kinfo_ofile *okif;
4591 	struct kinfo_file *kif;
4592 	struct filedesc *fdp;
4593 	struct pwddesc *pdp;
4594 	struct pwd *pwd;
4595 	u_int namelen;
4596 	int error, i, *name;
4597 	struct file *fp;
4598 	struct proc *p;
4599 
4600 	namelen = arg2;
4601 	if (namelen != 1)
4602 		return (EINVAL);
4603 
4604 	name = (int *)arg1;
4605 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4606 	if (error != 0)
4607 		return (error);
4608 	fdp = fdhold(p);
4609 	if (fdp != NULL)
4610 		pdp = pdhold(p);
4611 	PROC_UNLOCK(p);
4612 	if (fdp == NULL || pdp == NULL) {
4613 		if (fdp != NULL)
4614 			fddrop(fdp);
4615 		return (ENOENT);
4616 	}
4617 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4618 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4619 	PWDDESC_XLOCK(pdp);
4620 	pwd = pwd_hold_pwddesc(pdp);
4621 	if (pwd != NULL) {
4622 		if (pwd->pwd_cdir != NULL)
4623 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4624 			    okif, pdp, req);
4625 		if (pwd->pwd_rdir != NULL)
4626 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4627 			    okif, pdp, req);
4628 		if (pwd->pwd_jdir != NULL)
4629 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4630 			    okif, pdp, req);
4631 	}
4632 	PWDDESC_XUNLOCK(pdp);
4633 	if (pwd != NULL)
4634 		pwd_drop(pwd);
4635 	FILEDESC_SLOCK(fdp);
4636 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4637 		if (refcount_load(&fdp->fd_refcnt) == 0)
4638 			break;
4639 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4640 		    KERN_FILEDESC_PACK_KINFO);
4641 		FILEDESC_SUNLOCK(fdp);
4642 		kinfo_to_okinfo(kif, okif);
4643 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4644 		FILEDESC_SLOCK(fdp);
4645 		if (error)
4646 			break;
4647 	}
4648 	FILEDESC_SUNLOCK(fdp);
4649 	fddrop(fdp);
4650 	pddrop(pdp);
4651 	free(kif, M_TEMP);
4652 	free(okif, M_TEMP);
4653 	return (0);
4654 }
4655 
4656 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4657     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4658     "Process ofiledesc entries");
4659 #endif	/* COMPAT_FREEBSD7 */
4660 
4661 int
4662 vntype_to_kinfo(int vtype)
4663 {
4664 	struct {
4665 		int	vtype;
4666 		int	kf_vtype;
4667 	} vtypes_table[] = {
4668 		{ VBAD, KF_VTYPE_VBAD },
4669 		{ VBLK, KF_VTYPE_VBLK },
4670 		{ VCHR, KF_VTYPE_VCHR },
4671 		{ VDIR, KF_VTYPE_VDIR },
4672 		{ VFIFO, KF_VTYPE_VFIFO },
4673 		{ VLNK, KF_VTYPE_VLNK },
4674 		{ VNON, KF_VTYPE_VNON },
4675 		{ VREG, KF_VTYPE_VREG },
4676 		{ VSOCK, KF_VTYPE_VSOCK }
4677 	};
4678 	unsigned int i;
4679 
4680 	/*
4681 	 * Perform vtype translation.
4682 	 */
4683 	for (i = 0; i < nitems(vtypes_table); i++)
4684 		if (vtypes_table[i].vtype == vtype)
4685 			return (vtypes_table[i].kf_vtype);
4686 
4687 	return (KF_VTYPE_UNKNOWN);
4688 }
4689 
4690 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4691     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4692     "Process filedesc entries");
4693 
4694 /*
4695  * Store a process current working directory information to sbuf.
4696  *
4697  * Takes a locked proc as argument, and returns with the proc unlocked.
4698  */
4699 int
4700 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4701 {
4702 	struct pwddesc *pdp;
4703 	struct pwd *pwd;
4704 	struct export_fd_buf *efbuf;
4705 	struct vnode *cdir;
4706 	int error;
4707 
4708 	PROC_LOCK_ASSERT(p, MA_OWNED);
4709 
4710 	pdp = pdhold(p);
4711 	PROC_UNLOCK(p);
4712 	if (pdp == NULL)
4713 		return (EINVAL);
4714 
4715 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4716 	efbuf->fdp = NULL;
4717 	efbuf->pdp = pdp;
4718 	efbuf->sb = sb;
4719 	efbuf->remainder = maxlen;
4720 	efbuf->flags = 0;
4721 
4722 	PWDDESC_XLOCK(pdp);
4723 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4724 	cdir = pwd->pwd_cdir;
4725 	if (cdir == NULL) {
4726 		error = EINVAL;
4727 	} else {
4728 		vrefact(cdir);
4729 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4730 	}
4731 	PWDDESC_XUNLOCK(pdp);
4732 	pddrop(pdp);
4733 	free(efbuf, M_TEMP);
4734 	return (error);
4735 }
4736 
4737 /*
4738  * Get per-process current working directory.
4739  */
4740 static int
4741 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4742 {
4743 	struct sbuf sb;
4744 	struct proc *p;
4745 	ssize_t maxlen;
4746 	u_int namelen;
4747 	int error, error2, *name;
4748 
4749 	namelen = arg2;
4750 	if (namelen != 1)
4751 		return (EINVAL);
4752 
4753 	name = (int *)arg1;
4754 
4755 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4756 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4757 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4758 	if (error != 0) {
4759 		sbuf_delete(&sb);
4760 		return (error);
4761 	}
4762 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4763 	error = kern_proc_cwd_out(p, &sb, maxlen);
4764 	error2 = sbuf_finish(&sb);
4765 	sbuf_delete(&sb);
4766 	return (error != 0 ? error : error2);
4767 }
4768 
4769 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4770     sysctl_kern_proc_cwd, "Process current working directory");
4771 
4772 #ifdef DDB
4773 /*
4774  * For the purposes of debugging, generate a human-readable string for the
4775  * file type.
4776  */
4777 static const char *
4778 file_type_to_name(short type)
4779 {
4780 
4781 	switch (type) {
4782 	case 0:
4783 		return ("zero");
4784 	case DTYPE_VNODE:
4785 		return ("vnode");
4786 	case DTYPE_SOCKET:
4787 		return ("socket");
4788 	case DTYPE_PIPE:
4789 		return ("pipe");
4790 	case DTYPE_FIFO:
4791 		return ("fifo");
4792 	case DTYPE_KQUEUE:
4793 		return ("kqueue");
4794 	case DTYPE_CRYPTO:
4795 		return ("crypto");
4796 	case DTYPE_MQUEUE:
4797 		return ("mqueue");
4798 	case DTYPE_SHM:
4799 		return ("shm");
4800 	case DTYPE_SEM:
4801 		return ("ksem");
4802 	case DTYPE_PTS:
4803 		return ("pts");
4804 	case DTYPE_DEV:
4805 		return ("dev");
4806 	case DTYPE_PROCDESC:
4807 		return ("proc");
4808 	case DTYPE_EVENTFD:
4809 		return ("eventfd");
4810 	case DTYPE_LINUXTFD:
4811 		return ("ltimer");
4812 	default:
4813 		return ("unkn");
4814 	}
4815 }
4816 
4817 /*
4818  * For the purposes of debugging, identify a process (if any, perhaps one of
4819  * many) that references the passed file in its file descriptor array. Return
4820  * NULL if none.
4821  */
4822 static struct proc *
4823 file_to_first_proc(struct file *fp)
4824 {
4825 	struct filedesc *fdp;
4826 	struct proc *p;
4827 	int n;
4828 
4829 	FOREACH_PROC_IN_SYSTEM(p) {
4830 		if (p->p_state == PRS_NEW)
4831 			continue;
4832 		fdp = p->p_fd;
4833 		if (fdp == NULL)
4834 			continue;
4835 		for (n = 0; n < fdp->fd_nfiles; n++) {
4836 			if (fp == fdp->fd_ofiles[n].fde_file)
4837 				return (p);
4838 		}
4839 	}
4840 	return (NULL);
4841 }
4842 
4843 static void
4844 db_print_file(struct file *fp, int header)
4845 {
4846 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4847 	struct proc *p;
4848 
4849 	if (header)
4850 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4851 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4852 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4853 		    "FCmd");
4854 	p = file_to_first_proc(fp);
4855 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4856 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4857 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4858 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4859 
4860 #undef XPTRWIDTH
4861 }
4862 
4863 DB_SHOW_COMMAND(file, db_show_file)
4864 {
4865 	struct file *fp;
4866 
4867 	if (!have_addr) {
4868 		db_printf("usage: show file <addr>\n");
4869 		return;
4870 	}
4871 	fp = (struct file *)addr;
4872 	db_print_file(fp, 1);
4873 }
4874 
4875 DB_SHOW_COMMAND(files, db_show_files)
4876 {
4877 	struct filedesc *fdp;
4878 	struct file *fp;
4879 	struct proc *p;
4880 	int header;
4881 	int n;
4882 
4883 	header = 1;
4884 	FOREACH_PROC_IN_SYSTEM(p) {
4885 		if (p->p_state == PRS_NEW)
4886 			continue;
4887 		if ((fdp = p->p_fd) == NULL)
4888 			continue;
4889 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4890 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4891 				continue;
4892 			db_print_file(fp, header);
4893 			header = 0;
4894 		}
4895 	}
4896 }
4897 #endif
4898 
4899 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4900     &maxfilesperproc, 0, "Maximum files allowed open per process");
4901 
4902 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4903     &maxfiles, 0, "Maximum number of files");
4904 
4905 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4906     &openfiles, 0, "System-wide number of open files");
4907 
4908 /* ARGSUSED*/
4909 static void
4910 filelistinit(void *dummy)
4911 {
4912 
4913 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4914 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4915 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4916 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4917 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4918 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4919 	/*
4920 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4921 	 * the vnode zone.
4922 	 */
4923 	vfs_smr = uma_zone_get_smr(pwd_zone);
4924 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4925 }
4926 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4927 
4928 /*-------------------------------------------------------------------*/
4929 
4930 static int
4931 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4932     int flags, struct thread *td)
4933 {
4934 
4935 	return (EBADF);
4936 }
4937 
4938 static int
4939 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4940     struct thread *td)
4941 {
4942 
4943 	return (EINVAL);
4944 }
4945 
4946 static int
4947 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4948     struct thread *td)
4949 {
4950 
4951 	return (EBADF);
4952 }
4953 
4954 static int
4955 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4956     struct thread *td)
4957 {
4958 
4959 	return (0);
4960 }
4961 
4962 static int
4963 badfo_kqfilter(struct file *fp, struct knote *kn)
4964 {
4965 
4966 	return (EBADF);
4967 }
4968 
4969 static int
4970 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4971 {
4972 
4973 	return (EBADF);
4974 }
4975 
4976 static int
4977 badfo_close(struct file *fp, struct thread *td)
4978 {
4979 
4980 	return (0);
4981 }
4982 
4983 static int
4984 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4985     struct thread *td)
4986 {
4987 
4988 	return (EBADF);
4989 }
4990 
4991 static int
4992 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4993     struct thread *td)
4994 {
4995 
4996 	return (EBADF);
4997 }
4998 
4999 static int
5000 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5001     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5002     struct thread *td)
5003 {
5004 
5005 	return (EBADF);
5006 }
5007 
5008 static int
5009 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5010 {
5011 
5012 	return (0);
5013 }
5014 
5015 struct fileops badfileops = {
5016 	.fo_read = badfo_readwrite,
5017 	.fo_write = badfo_readwrite,
5018 	.fo_truncate = badfo_truncate,
5019 	.fo_ioctl = badfo_ioctl,
5020 	.fo_poll = badfo_poll,
5021 	.fo_kqfilter = badfo_kqfilter,
5022 	.fo_stat = badfo_stat,
5023 	.fo_close = badfo_close,
5024 	.fo_chmod = badfo_chmod,
5025 	.fo_chown = badfo_chown,
5026 	.fo_sendfile = badfo_sendfile,
5027 	.fo_fill_kinfo = badfo_fill_kinfo,
5028 };
5029 
5030 static int
5031 path_poll(struct file *fp, int events, struct ucred *active_cred,
5032     struct thread *td)
5033 {
5034 	return (POLLNVAL);
5035 }
5036 
5037 static int
5038 path_close(struct file *fp, struct thread *td)
5039 {
5040 	MPASS(fp->f_type == DTYPE_VNODE);
5041 	fp->f_ops = &badfileops;
5042 	vdrop(fp->f_vnode);
5043 	return (0);
5044 }
5045 
5046 struct fileops path_fileops = {
5047 	.fo_read = badfo_readwrite,
5048 	.fo_write = badfo_readwrite,
5049 	.fo_truncate = badfo_truncate,
5050 	.fo_ioctl = badfo_ioctl,
5051 	.fo_poll = path_poll,
5052 	.fo_kqfilter = vn_kqfilter_opath,
5053 	.fo_stat = vn_statfile,
5054 	.fo_close = path_close,
5055 	.fo_chmod = badfo_chmod,
5056 	.fo_chown = badfo_chown,
5057 	.fo_sendfile = badfo_sendfile,
5058 	.fo_fill_kinfo = vn_fill_kinfo,
5059 	.fo_flags = DFLAG_PASSABLE,
5060 };
5061 
5062 int
5063 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5064     int flags, struct thread *td)
5065 {
5066 
5067 	return (EOPNOTSUPP);
5068 }
5069 
5070 int
5071 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5072     struct thread *td)
5073 {
5074 
5075 	return (EINVAL);
5076 }
5077 
5078 int
5079 invfo_ioctl(struct file *fp, u_long com, void *data,
5080     struct ucred *active_cred, struct thread *td)
5081 {
5082 
5083 	return (ENOTTY);
5084 }
5085 
5086 int
5087 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5088     struct thread *td)
5089 {
5090 
5091 	return (poll_no_poll(events));
5092 }
5093 
5094 int
5095 invfo_kqfilter(struct file *fp, struct knote *kn)
5096 {
5097 
5098 	return (EINVAL);
5099 }
5100 
5101 int
5102 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5103     struct thread *td)
5104 {
5105 
5106 	return (EINVAL);
5107 }
5108 
5109 int
5110 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5111     struct thread *td)
5112 {
5113 
5114 	return (EINVAL);
5115 }
5116 
5117 int
5118 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5119     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5120     struct thread *td)
5121 {
5122 
5123 	return (EINVAL);
5124 }
5125 
5126 /*-------------------------------------------------------------------*/
5127 
5128 /*
5129  * File Descriptor pseudo-device driver (/dev/fd/).
5130  *
5131  * Opening minor device N dup()s the file (if any) connected to file
5132  * descriptor N belonging to the calling process.  Note that this driver
5133  * consists of only the ``open()'' routine, because all subsequent
5134  * references to this file will be direct to the other driver.
5135  *
5136  * XXX: we could give this one a cloning event handler if necessary.
5137  */
5138 
5139 /* ARGSUSED */
5140 static int
5141 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5142 {
5143 
5144 	/*
5145 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5146 	 * the file descriptor being sought for duplication. The error
5147 	 * return ensures that the vnode for this device will be released
5148 	 * by vn_open. Open will detect this special error and take the
5149 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5150 	 * will simply report the error.
5151 	 */
5152 	td->td_dupfd = dev2unit(dev);
5153 	return (ENODEV);
5154 }
5155 
5156 static struct cdevsw fildesc_cdevsw = {
5157 	.d_version =	D_VERSION,
5158 	.d_open =	fdopen,
5159 	.d_name =	"FD",
5160 };
5161 
5162 static void
5163 fildesc_drvinit(void *unused)
5164 {
5165 	struct cdev *dev;
5166 
5167 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5168 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5169 	make_dev_alias(dev, "stdin");
5170 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5171 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5172 	make_dev_alias(dev, "stdout");
5173 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5174 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5175 	make_dev_alias(dev, "stderr");
5176 }
5177 
5178 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5179