xref: /freebsd/sys/kern/kern_descrip.c (revision a0b9e2e854027e6ff61fb075a1309dbc71c42b54)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, int holdleaders);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seqc_write_begin(&fde->fde_seqc);
313 #endif
314 	fde->fde_file = NULL;
315 #ifdef CAPABILITIES
316 	seqc_write_end(&fde->fde_seqc);
317 #endif
318 	fdefree_last(fde);
319 	fdunused(fdp, fd);
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 #ifdef	RACCT
335 	uint64_t lim;
336 #endif
337 
338 	td->td_retval[0] = getmaxfd(td);
339 #ifdef	RACCT
340 	PROC_LOCK(td->td_proc);
341 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 	PROC_UNLOCK(td->td_proc);
343 	if (lim < td->td_retval[0])
344 		td->td_retval[0] = lim;
345 #endif
346 	return (0);
347 }
348 
349 /*
350  * Duplicate a file descriptor to a particular value.
351  *
352  * Note: keep in mind that a potential race condition exists when closing
353  * descriptors from a shared descriptor table (via rfork).
354  */
355 #ifndef _SYS_SYSPROTO_H_
356 struct dup2_args {
357 	u_int	from;
358 	u_int	to;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_dup2(struct thread *td, struct dup2_args *uap)
364 {
365 
366 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
367 }
368 
369 /*
370  * Duplicate a file descriptor.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup_args {
374 	u_int	fd;
375 };
376 #endif
377 /* ARGSUSED */
378 int
379 sys_dup(struct thread *td, struct dup_args *uap)
380 {
381 
382 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
383 }
384 
385 /*
386  * The file control system call.
387  */
388 #ifndef _SYS_SYSPROTO_H_
389 struct fcntl_args {
390 	int	fd;
391 	int	cmd;
392 	long	arg;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 {
399 
400 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
401 }
402 
403 int
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
405 {
406 	struct flock fl;
407 	struct __oflock ofl;
408 	intptr_t arg1;
409 	int error, newcmd;
410 
411 	error = 0;
412 	newcmd = cmd;
413 	switch (cmd) {
414 	case F_OGETLK:
415 	case F_OSETLK:
416 	case F_OSETLKW:
417 		/*
418 		 * Convert old flock structure to new.
419 		 */
420 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 		fl.l_start = ofl.l_start;
422 		fl.l_len = ofl.l_len;
423 		fl.l_pid = ofl.l_pid;
424 		fl.l_type = ofl.l_type;
425 		fl.l_whence = ofl.l_whence;
426 		fl.l_sysid = 0;
427 
428 		switch (cmd) {
429 		case F_OGETLK:
430 			newcmd = F_GETLK;
431 			break;
432 		case F_OSETLK:
433 			newcmd = F_SETLK;
434 			break;
435 		case F_OSETLKW:
436 			newcmd = F_SETLKW;
437 			break;
438 		}
439 		arg1 = (intptr_t)&fl;
440 		break;
441 	case F_GETLK:
442 	case F_SETLK:
443 	case F_SETLKW:
444 	case F_SETLK_REMOTE:
445 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 		arg1 = (intptr_t)&fl;
447 		break;
448 	default:
449 		arg1 = arg;
450 		break;
451 	}
452 	if (error)
453 		return (error);
454 	error = kern_fcntl(td, fd, newcmd, arg1);
455 	if (error)
456 		return (error);
457 	if (cmd == F_OGETLK) {
458 		ofl.l_start = fl.l_start;
459 		ofl.l_len = fl.l_len;
460 		ofl.l_pid = fl.l_pid;
461 		ofl.l_type = fl.l_type;
462 		ofl.l_whence = fl.l_whence;
463 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 	} else if (cmd == F_GETLK) {
465 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
466 	}
467 	return (error);
468 }
469 
470 int
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
472 {
473 	struct filedesc *fdp;
474 	struct flock *flp;
475 	struct file *fp, *fp2;
476 	struct filedescent *fde;
477 	struct proc *p;
478 	struct vnode *vp;
479 	struct mount *mp;
480 	int error, flg, seals, tmp;
481 	uint64_t bsize;
482 	off_t foffset;
483 
484 	error = 0;
485 	flg = F_POSIX;
486 	p = td->td_proc;
487 	fdp = p->p_fd;
488 
489 	AUDIT_ARG_FD(cmd);
490 	AUDIT_ARG_CMD(cmd);
491 	switch (cmd) {
492 	case F_DUPFD:
493 		tmp = arg;
494 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
495 		break;
496 
497 	case F_DUPFD_CLOEXEC:
498 		tmp = arg;
499 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD:
503 		tmp = arg;
504 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
505 		break;
506 
507 	case F_DUP2FD_CLOEXEC:
508 		tmp = arg;
509 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 		break;
511 
512 	case F_GETFD:
513 		error = EBADF;
514 		FILEDESC_SLOCK(fdp);
515 		fde = fdeget_locked(fdp, fd);
516 		if (fde != NULL) {
517 			td->td_retval[0] =
518 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
519 			error = 0;
520 		}
521 		FILEDESC_SUNLOCK(fdp);
522 		break;
523 
524 	case F_SETFD:
525 		error = EBADF;
526 		FILEDESC_XLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
531 			error = 0;
532 		}
533 		FILEDESC_XUNLOCK(fdp);
534 		break;
535 
536 	case F_GETFL:
537 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
538 		if (error != 0)
539 			break;
540 		td->td_retval[0] = OFLAGS(fp->f_flag);
541 		fdrop(fp, td);
542 		break;
543 
544 	case F_SETFL:
545 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
546 		if (error != 0)
547 			break;
548 		do {
549 			tmp = flg = fp->f_flag;
550 			tmp &= ~FCNTLFLAGS;
551 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
552 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
553 		tmp = fp->f_flag & FNONBLOCK;
554 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
555 		if (error != 0) {
556 			fdrop(fp, td);
557 			break;
558 		}
559 		tmp = fp->f_flag & FASYNC;
560 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
561 		if (error == 0) {
562 			fdrop(fp, td);
563 			break;
564 		}
565 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
566 		tmp = 0;
567 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 		fdrop(fp, td);
569 		break;
570 
571 	case F_GETOWN:
572 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
573 		if (error != 0)
574 			break;
575 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
576 		if (error == 0)
577 			td->td_retval[0] = tmp;
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_SETOWN:
582 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
583 		if (error != 0)
584 			break;
585 		tmp = arg;
586 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
587 		fdrop(fp, td);
588 		break;
589 
590 	case F_SETLK_REMOTE:
591 		error = priv_check(td, PRIV_NFS_LOCKD);
592 		if (error != 0)
593 			return (error);
594 		flg = F_REMOTE;
595 		goto do_setlk;
596 
597 	case F_SETLKW:
598 		flg |= F_WAIT;
599 		/* FALLTHROUGH F_SETLK */
600 
601 	case F_SETLK:
602 	do_setlk:
603 		flp = (struct flock *)arg;
604 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
605 			error = EINVAL;
606 			break;
607 		}
608 
609 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
610 		if (error != 0)
611 			break;
612 		if (fp->f_type != DTYPE_VNODE) {
613 			error = EBADF;
614 			fdrop(fp, td);
615 			break;
616 		}
617 
618 		if (flp->l_whence == SEEK_CUR) {
619 			foffset = foffset_get(fp);
620 			if (foffset < 0 ||
621 			    (flp->l_start > 0 &&
622 			     foffset > OFF_MAX - flp->l_start)) {
623 				error = EOVERFLOW;
624 				fdrop(fp, td);
625 				break;
626 			}
627 			flp->l_start += foffset;
628 		}
629 
630 		vp = fp->f_vnode;
631 		switch (flp->l_type) {
632 		case F_RDLCK:
633 			if ((fp->f_flag & FREAD) == 0) {
634 				error = EBADF;
635 				break;
636 			}
637 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
638 				PROC_LOCK(p->p_leader);
639 				p->p_leader->p_flag |= P_ADVLOCK;
640 				PROC_UNLOCK(p->p_leader);
641 			}
642 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
643 			    flp, flg);
644 			break;
645 		case F_WRLCK:
646 			if ((fp->f_flag & FWRITE) == 0) {
647 				error = EBADF;
648 				break;
649 			}
650 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 				PROC_LOCK(p->p_leader);
652 				p->p_leader->p_flag |= P_ADVLOCK;
653 				PROC_UNLOCK(p->p_leader);
654 			}
655 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
656 			    flp, flg);
657 			break;
658 		case F_UNLCK:
659 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
660 			    flp, flg);
661 			break;
662 		case F_UNLCKSYS:
663 			if (flg != F_REMOTE) {
664 				error = EINVAL;
665 				break;
666 			}
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
668 			    F_UNLCKSYS, flp, flg);
669 			break;
670 		default:
671 			error = EINVAL;
672 			break;
673 		}
674 		if (error != 0 || flp->l_type == F_UNLCK ||
675 		    flp->l_type == F_UNLCKSYS) {
676 			fdrop(fp, td);
677 			break;
678 		}
679 
680 		/*
681 		 * Check for a race with close.
682 		 *
683 		 * The vnode is now advisory locked (or unlocked, but this case
684 		 * is not really important) as the caller requested.
685 		 * We had to drop the filedesc lock, so we need to recheck if
686 		 * the descriptor is still valid, because if it was closed
687 		 * in the meantime we need to remove advisory lock from the
688 		 * vnode - close on any descriptor leading to an advisory
689 		 * locked vnode, removes that lock.
690 		 * We will return 0 on purpose in that case, as the result of
691 		 * successful advisory lock might have been externally visible
692 		 * already. This is fine - effectively we pretend to the caller
693 		 * that the closing thread was a bit slower and that the
694 		 * advisory lock succeeded before the close.
695 		 */
696 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
697 		if (error != 0) {
698 			fdrop(fp, td);
699 			break;
700 		}
701 		if (fp != fp2) {
702 			flp->l_whence = SEEK_SET;
703 			flp->l_start = 0;
704 			flp->l_len = 0;
705 			flp->l_type = F_UNLCK;
706 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
707 			    F_UNLCK, flp, F_POSIX);
708 		}
709 		fdrop(fp, td);
710 		fdrop(fp2, td);
711 		break;
712 
713 	case F_GETLK:
714 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
715 		if (error != 0)
716 			break;
717 		if (fp->f_type != DTYPE_VNODE) {
718 			error = EBADF;
719 			fdrop(fp, td);
720 			break;
721 		}
722 		flp = (struct flock *)arg;
723 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
724 		    flp->l_type != F_UNLCK) {
725 			error = EINVAL;
726 			fdrop(fp, td);
727 			break;
728 		}
729 		if (flp->l_whence == SEEK_CUR) {
730 			foffset = foffset_get(fp);
731 			if ((flp->l_start > 0 &&
732 			    foffset > OFF_MAX - flp->l_start) ||
733 			    (flp->l_start < 0 &&
734 			    foffset < OFF_MIN - flp->l_start)) {
735 				error = EOVERFLOW;
736 				fdrop(fp, td);
737 				break;
738 			}
739 			flp->l_start += foffset;
740 		}
741 		vp = fp->f_vnode;
742 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
743 		    F_POSIX);
744 		fdrop(fp, td);
745 		break;
746 
747 	case F_ADD_SEALS:
748 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
749 		if (error != 0)
750 			break;
751 		error = fo_add_seals(fp, arg);
752 		fdrop(fp, td);
753 		break;
754 
755 	case F_GET_SEALS:
756 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
757 		if (error != 0)
758 			break;
759 		if (fo_get_seals(fp, &seals) == 0)
760 			td->td_retval[0] = seals;
761 		else
762 			error = EINVAL;
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_RDAHEAD:
767 		arg = arg ? 128 * 1024: 0;
768 		/* FALLTHROUGH */
769 	case F_READAHEAD:
770 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
771 		if (error != 0)
772 			break;
773 		if (fp->f_type != DTYPE_VNODE) {
774 			fdrop(fp, td);
775 			error = EBADF;
776 			break;
777 		}
778 		vp = fp->f_vnode;
779 		if (vp->v_type != VREG) {
780 			fdrop(fp, td);
781 			error = ENOTTY;
782 			break;
783 		}
784 
785 		/*
786 		 * Exclusive lock synchronizes against f_seqcount reads and
787 		 * writes in sequential_heuristic().
788 		 */
789 		error = vn_lock(vp, LK_EXCLUSIVE);
790 		if (error != 0) {
791 			fdrop(fp, td);
792 			break;
793 		}
794 		if (arg >= 0) {
795 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
796 			arg = MIN(arg, INT_MAX - bsize + 1);
797 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
798 			    (arg + bsize - 1) / bsize);
799 			atomic_set_int(&fp->f_flag, FRDAHEAD);
800 		} else {
801 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
802 		}
803 		VOP_UNLOCK(vp);
804 		fdrop(fp, td);
805 		break;
806 
807 	case F_ISUNIONSTACK:
808 		/*
809 		 * Check if the vnode is part of a union stack (either the
810 		 * "union" flag from mount(2) or unionfs).
811 		 *
812 		 * Prior to introduction of this op libc's readdir would call
813 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
814 		 * data just to check fs name and a mount flag.
815 		 *
816 		 * Fixing the code to handle everything in the kernel instead
817 		 * is a non-trivial endeavor and has low priority, thus this
818 		 * horrible kludge facilitates the current behavior in a much
819 		 * cheaper manner until someone(tm) sorts this out.
820 		 */
821 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
822 		if (error != 0)
823 			break;
824 		if (fp->f_type != DTYPE_VNODE) {
825 			fdrop(fp, td);
826 			error = EBADF;
827 			break;
828 		}
829 		vp = fp->f_vnode;
830 		/*
831 		 * Since we don't prevent dooming the vnode even non-null mp
832 		 * found can become immediately stale. This is tolerable since
833 		 * mount points are type-stable (providing safe memory access)
834 		 * and any vfs op on this vnode going forward will return an
835 		 * error (meaning return value in this case is meaningless).
836 		 */
837 		mp = atomic_load_ptr(&vp->v_mount);
838 		if (__predict_false(mp == NULL)) {
839 			fdrop(fp, td);
840 			error = EBADF;
841 			break;
842 		}
843 		td->td_retval[0] = 0;
844 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
845 		    mp->mnt_flag & MNT_UNION)
846 			td->td_retval[0] = 1;
847 		fdrop(fp, td);
848 		break;
849 
850 	default:
851 		error = EINVAL;
852 		break;
853 	}
854 	return (error);
855 }
856 
857 static int
858 getmaxfd(struct thread *td)
859 {
860 
861 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
862 }
863 
864 /*
865  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
866  */
867 int
868 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
869 {
870 	struct filedesc *fdp;
871 	struct filedescent *oldfde, *newfde;
872 	struct proc *p;
873 	struct file *delfp, *oldfp;
874 	u_long *oioctls, *nioctls;
875 	int error, maxfd;
876 
877 	p = td->td_proc;
878 	fdp = p->p_fd;
879 	oioctls = NULL;
880 
881 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
882 	MPASS(mode < FDDUP_LASTMODE);
883 
884 	AUDIT_ARG_FD(old);
885 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
886 
887 	/*
888 	 * Verify we have a valid descriptor to dup from and possibly to
889 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
890 	 * return EINVAL when the new descriptor is out of bounds.
891 	 */
892 	if (old < 0)
893 		return (EBADF);
894 	if (new < 0)
895 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
896 	maxfd = getmaxfd(td);
897 	if (new >= maxfd)
898 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
899 
900 	error = EBADF;
901 	FILEDESC_XLOCK(fdp);
902 	if (fget_locked(fdp, old) == NULL)
903 		goto unlock;
904 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
905 		td->td_retval[0] = new;
906 		if (flags & FDDUP_FLAG_CLOEXEC)
907 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
908 		error = 0;
909 		goto unlock;
910 	}
911 
912 	oldfde = &fdp->fd_ofiles[old];
913 	oldfp = oldfde->fde_file;
914 	if (!fhold(oldfp))
915 		goto unlock;
916 
917 	/*
918 	 * If the caller specified a file descriptor, make sure the file
919 	 * table is large enough to hold it, and grab it.  Otherwise, just
920 	 * allocate a new descriptor the usual way.
921 	 */
922 	switch (mode) {
923 	case FDDUP_NORMAL:
924 	case FDDUP_FCNTL:
925 		if ((error = fdalloc(td, new, &new)) != 0) {
926 			fdrop(oldfp, td);
927 			goto unlock;
928 		}
929 		break;
930 	case FDDUP_MUSTREPLACE:
931 		/* Target file descriptor must exist. */
932 		if (fget_locked(fdp, new) == NULL) {
933 			fdrop(oldfp, td);
934 			goto unlock;
935 		}
936 		break;
937 	case FDDUP_FIXED:
938 		if (new >= fdp->fd_nfiles) {
939 			/*
940 			 * The resource limits are here instead of e.g.
941 			 * fdalloc(), because the file descriptor table may be
942 			 * shared between processes, so we can't really use
943 			 * racct_add()/racct_sub().  Instead of counting the
944 			 * number of actually allocated descriptors, just put
945 			 * the limit on the size of the file descriptor table.
946 			 */
947 #ifdef RACCT
948 			if (RACCT_ENABLED()) {
949 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
950 				if (error != 0) {
951 					error = EMFILE;
952 					fdrop(oldfp, td);
953 					goto unlock;
954 				}
955 			}
956 #endif
957 			fdgrowtable_exp(fdp, new + 1);
958 		}
959 		if (!fdisused(fdp, new))
960 			fdused(fdp, new);
961 		break;
962 	default:
963 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
964 	}
965 
966 	KASSERT(old != new, ("new fd is same as old"));
967 
968 	/* Refetch oldfde because the table may have grown and old one freed. */
969 	oldfde = &fdp->fd_ofiles[old];
970 	KASSERT(oldfp == oldfde->fde_file,
971 	    ("fdt_ofiles shift from growth observed at fd %d",
972 	    old));
973 
974 	newfde = &fdp->fd_ofiles[new];
975 	delfp = newfde->fde_file;
976 
977 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
978 
979 	/*
980 	 * Duplicate the source descriptor.
981 	 */
982 #ifdef CAPABILITIES
983 	seqc_write_begin(&newfde->fde_seqc);
984 #endif
985 	oioctls = filecaps_free_prep(&newfde->fde_caps);
986 	memcpy(newfde, oldfde, fde_change_size);
987 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
988 	    nioctls);
989 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
990 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
991 	else
992 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
993 #ifdef CAPABILITIES
994 	seqc_write_end(&newfde->fde_seqc);
995 #endif
996 	td->td_retval[0] = new;
997 
998 	error = 0;
999 
1000 	if (delfp != NULL) {
1001 		(void) closefp(fdp, new, delfp, td, 1);
1002 		FILEDESC_UNLOCK_ASSERT(fdp);
1003 	} else {
1004 unlock:
1005 		FILEDESC_XUNLOCK(fdp);
1006 	}
1007 
1008 	filecaps_free_finish(oioctls);
1009 	return (error);
1010 }
1011 
1012 static void
1013 sigiofree(struct sigio *sigio)
1014 {
1015 	crfree(sigio->sio_ucred);
1016 	free(sigio, M_SIGIO);
1017 }
1018 
1019 static struct sigio *
1020 funsetown_locked(struct sigio *sigio)
1021 {
1022 	struct proc *p;
1023 	struct pgrp *pg;
1024 
1025 	SIGIO_ASSERT_LOCKED();
1026 
1027 	if (sigio == NULL)
1028 		return (NULL);
1029 	*(sigio->sio_myref) = NULL;
1030 	if (sigio->sio_pgid < 0) {
1031 		pg = sigio->sio_pgrp;
1032 		PGRP_LOCK(pg);
1033 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1034 		    sigio, sio_pgsigio);
1035 		PGRP_UNLOCK(pg);
1036 	} else {
1037 		p = sigio->sio_proc;
1038 		PROC_LOCK(p);
1039 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1040 		    sigio, sio_pgsigio);
1041 		PROC_UNLOCK(p);
1042 	}
1043 	return (sigio);
1044 }
1045 
1046 /*
1047  * If sigio is on the list associated with a process or process group,
1048  * disable signalling from the device, remove sigio from the list and
1049  * free sigio.
1050  */
1051 void
1052 funsetown(struct sigio **sigiop)
1053 {
1054 	struct sigio *sigio;
1055 
1056 	/* Racy check, consumers must provide synchronization. */
1057 	if (*sigiop == NULL)
1058 		return;
1059 
1060 	SIGIO_LOCK();
1061 	sigio = funsetown_locked(*sigiop);
1062 	SIGIO_UNLOCK();
1063 	if (sigio != NULL)
1064 		sigiofree(sigio);
1065 }
1066 
1067 /*
1068  * Free a list of sigio structures.  The caller must ensure that new sigio
1069  * structures cannot be added after this point.  For process groups this is
1070  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1071  * as an interlock.
1072  */
1073 void
1074 funsetownlst(struct sigiolst *sigiolst)
1075 {
1076 	struct proc *p;
1077 	struct pgrp *pg;
1078 	struct sigio *sigio, *tmp;
1079 
1080 	/* Racy check. */
1081 	sigio = SLIST_FIRST(sigiolst);
1082 	if (sigio == NULL)
1083 		return;
1084 
1085 	p = NULL;
1086 	pg = NULL;
1087 
1088 	SIGIO_LOCK();
1089 	sigio = SLIST_FIRST(sigiolst);
1090 	if (sigio == NULL) {
1091 		SIGIO_UNLOCK();
1092 		return;
1093 	}
1094 
1095 	/*
1096 	 * Every entry of the list should belong to a single proc or pgrp.
1097 	 */
1098 	if (sigio->sio_pgid < 0) {
1099 		pg = sigio->sio_pgrp;
1100 		sx_assert(&proctree_lock, SX_XLOCKED);
1101 		PGRP_LOCK(pg);
1102 	} else /* if (sigio->sio_pgid > 0) */ {
1103 		p = sigio->sio_proc;
1104 		PROC_LOCK(p);
1105 		KASSERT((p->p_flag & P_WEXIT) != 0,
1106 		    ("%s: process %p is not exiting", __func__, p));
1107 	}
1108 
1109 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1110 		*sigio->sio_myref = NULL;
1111 		if (pg != NULL) {
1112 			KASSERT(sigio->sio_pgid < 0,
1113 			    ("Proc sigio in pgrp sigio list"));
1114 			KASSERT(sigio->sio_pgrp == pg,
1115 			    ("Bogus pgrp in sigio list"));
1116 		} else /* if (p != NULL) */ {
1117 			KASSERT(sigio->sio_pgid > 0,
1118 			    ("Pgrp sigio in proc sigio list"));
1119 			KASSERT(sigio->sio_proc == p,
1120 			    ("Bogus proc in sigio list"));
1121 		}
1122 	}
1123 
1124 	if (pg != NULL)
1125 		PGRP_UNLOCK(pg);
1126 	else
1127 		PROC_UNLOCK(p);
1128 	SIGIO_UNLOCK();
1129 
1130 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1131 		sigiofree(sigio);
1132 }
1133 
1134 /*
1135  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1136  *
1137  * After permission checking, add a sigio structure to the sigio list for
1138  * the process or process group.
1139  */
1140 int
1141 fsetown(pid_t pgid, struct sigio **sigiop)
1142 {
1143 	struct proc *proc;
1144 	struct pgrp *pgrp;
1145 	struct sigio *osigio, *sigio;
1146 	int ret;
1147 
1148 	if (pgid == 0) {
1149 		funsetown(sigiop);
1150 		return (0);
1151 	}
1152 
1153 	ret = 0;
1154 
1155 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1156 	sigio->sio_pgid = pgid;
1157 	sigio->sio_ucred = crhold(curthread->td_ucred);
1158 	sigio->sio_myref = sigiop;
1159 
1160 	sx_slock(&proctree_lock);
1161 	SIGIO_LOCK();
1162 	osigio = funsetown_locked(*sigiop);
1163 	if (pgid > 0) {
1164 		proc = pfind(pgid);
1165 		if (proc == NULL) {
1166 			ret = ESRCH;
1167 			goto fail;
1168 		}
1169 
1170 		/*
1171 		 * Policy - Don't allow a process to FSETOWN a process
1172 		 * in another session.
1173 		 *
1174 		 * Remove this test to allow maximum flexibility or
1175 		 * restrict FSETOWN to the current process or process
1176 		 * group for maximum safety.
1177 		 */
1178 		if (proc->p_session != curthread->td_proc->p_session) {
1179 			PROC_UNLOCK(proc);
1180 			ret = EPERM;
1181 			goto fail;
1182 		}
1183 
1184 		sigio->sio_proc = proc;
1185 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1186 		PROC_UNLOCK(proc);
1187 	} else /* if (pgid < 0) */ {
1188 		pgrp = pgfind(-pgid);
1189 		if (pgrp == NULL) {
1190 			ret = ESRCH;
1191 			goto fail;
1192 		}
1193 
1194 		/*
1195 		 * Policy - Don't allow a process to FSETOWN a process
1196 		 * in another session.
1197 		 *
1198 		 * Remove this test to allow maximum flexibility or
1199 		 * restrict FSETOWN to the current process or process
1200 		 * group for maximum safety.
1201 		 */
1202 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1203 			PGRP_UNLOCK(pgrp);
1204 			ret = EPERM;
1205 			goto fail;
1206 		}
1207 
1208 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1209 		sigio->sio_pgrp = pgrp;
1210 		PGRP_UNLOCK(pgrp);
1211 	}
1212 	sx_sunlock(&proctree_lock);
1213 	*sigiop = sigio;
1214 	SIGIO_UNLOCK();
1215 	if (osigio != NULL)
1216 		sigiofree(osigio);
1217 	return (0);
1218 
1219 fail:
1220 	SIGIO_UNLOCK();
1221 	sx_sunlock(&proctree_lock);
1222 	sigiofree(sigio);
1223 	if (osigio != NULL)
1224 		sigiofree(osigio);
1225 	return (ret);
1226 }
1227 
1228 /*
1229  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1230  */
1231 pid_t
1232 fgetown(struct sigio **sigiop)
1233 {
1234 	pid_t pgid;
1235 
1236 	SIGIO_LOCK();
1237 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1238 	SIGIO_UNLOCK();
1239 	return (pgid);
1240 }
1241 
1242 /*
1243  * Function drops the filedesc lock on return.
1244  */
1245 static int
1246 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1247     int holdleaders)
1248 {
1249 	int error;
1250 
1251 	FILEDESC_XLOCK_ASSERT(fdp);
1252 
1253 	if (holdleaders) {
1254 		if (td->td_proc->p_fdtol != NULL) {
1255 			/*
1256 			 * Ask fdfree() to sleep to ensure that all relevant
1257 			 * process leaders can be traversed in closef().
1258 			 */
1259 			fdp->fd_holdleaderscount++;
1260 		} else {
1261 			holdleaders = 0;
1262 		}
1263 	}
1264 
1265 	/*
1266 	 * We now hold the fp reference that used to be owned by the
1267 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1268 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1269 	 * added, and deleteing a knote for the new fd.
1270 	 */
1271 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1272 		knote_fdclose(td, fd);
1273 
1274 	/*
1275 	 * We need to notify mqueue if the object is of type mqueue.
1276 	 */
1277 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1278 		mq_fdclose(td, fd, fp);
1279 	FILEDESC_XUNLOCK(fdp);
1280 
1281 	error = closef(fp, td);
1282 
1283 	/*
1284 	 * All paths leading up to closefp() will have already removed or
1285 	 * replaced the fd in the filedesc table, so a restart would not
1286 	 * operate on the same file.
1287 	 */
1288 	if (error == ERESTART)
1289 		error = EINTR;
1290 
1291 	if (holdleaders) {
1292 		FILEDESC_XLOCK(fdp);
1293 		fdp->fd_holdleaderscount--;
1294 		if (fdp->fd_holdleaderscount == 0 &&
1295 		    fdp->fd_holdleaderswakeup != 0) {
1296 			fdp->fd_holdleaderswakeup = 0;
1297 			wakeup(&fdp->fd_holdleaderscount);
1298 		}
1299 		FILEDESC_XUNLOCK(fdp);
1300 	}
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Close a file descriptor.
1306  */
1307 #ifndef _SYS_SYSPROTO_H_
1308 struct close_args {
1309 	int     fd;
1310 };
1311 #endif
1312 /* ARGSUSED */
1313 int
1314 sys_close(struct thread *td, struct close_args *uap)
1315 {
1316 
1317 	return (kern_close(td, uap->fd));
1318 }
1319 
1320 int
1321 kern_close(struct thread *td, int fd)
1322 {
1323 	struct filedesc *fdp;
1324 	struct file *fp;
1325 
1326 	fdp = td->td_proc->p_fd;
1327 
1328 	AUDIT_SYSCLOSE(td, fd);
1329 
1330 	FILEDESC_XLOCK(fdp);
1331 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1332 		FILEDESC_XUNLOCK(fdp);
1333 		return (EBADF);
1334 	}
1335 	fdfree(fdp, fd);
1336 
1337 	/* closefp() drops the FILEDESC lock for us. */
1338 	return (closefp(fdp, fd, fp, td, 1));
1339 }
1340 
1341 int
1342 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1343 {
1344 	struct filedesc *fdp;
1345 	int fd, ret, lastfile;
1346 
1347 	ret = 0;
1348 	fdp = td->td_proc->p_fd;
1349 	FILEDESC_SLOCK(fdp);
1350 
1351 	/*
1352 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1353 	 * open should not be a usage error.  From a close_range() perspective,
1354 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1355 	 * be a usage error as all fd above 3 are in-fact already closed.
1356 	 */
1357 	if (highfd < lowfd) {
1358 		ret = EINVAL;
1359 		goto out;
1360 	}
1361 
1362 	/*
1363 	 * If lastfile == -1, we're dealing with either a fresh file
1364 	 * table or one in which every fd has been closed.  Just return
1365 	 * successful; there's nothing left to do.
1366 	 */
1367 	lastfile = fdlastfile(fdp);
1368 	if (lastfile == -1)
1369 		goto out;
1370 	/* Clamped to [lowfd, lastfile] */
1371 	highfd = MIN(highfd, lastfile);
1372 	for (fd = lowfd; fd <= highfd; fd++) {
1373 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1374 			FILEDESC_SUNLOCK(fdp);
1375 			(void)kern_close(td, fd);
1376 			FILEDESC_SLOCK(fdp);
1377 		}
1378 	}
1379 out:
1380 	FILEDESC_SUNLOCK(fdp);
1381 	return (ret);
1382 }
1383 
1384 #ifndef _SYS_SYSPROTO_H_
1385 struct close_range_args {
1386 	u_int	lowfd;
1387 	u_int	highfd;
1388 	int	flags;
1389 };
1390 #endif
1391 int
1392 sys_close_range(struct thread *td, struct close_range_args *uap)
1393 {
1394 
1395 	/* No flags currently defined */
1396 	if (uap->flags != 0)
1397 		return (EINVAL);
1398 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1399 }
1400 
1401 #ifdef COMPAT_FREEBSD12
1402 /*
1403  * Close open file descriptors.
1404  */
1405 #ifndef _SYS_SYSPROTO_H_
1406 struct freebsd12_closefrom_args {
1407 	int	lowfd;
1408 };
1409 #endif
1410 /* ARGSUSED */
1411 int
1412 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1413 {
1414 	u_int lowfd;
1415 
1416 	AUDIT_ARG_FD(uap->lowfd);
1417 
1418 	/*
1419 	 * Treat negative starting file descriptor values identical to
1420 	 * closefrom(0) which closes all files.
1421 	 */
1422 	lowfd = MAX(0, uap->lowfd);
1423 	return (kern_close_range(td, lowfd, ~0U));
1424 }
1425 #endif	/* COMPAT_FREEBSD12 */
1426 
1427 #if defined(COMPAT_43)
1428 /*
1429  * Return status information about a file descriptor.
1430  */
1431 #ifndef _SYS_SYSPROTO_H_
1432 struct ofstat_args {
1433 	int	fd;
1434 	struct	ostat *sb;
1435 };
1436 #endif
1437 /* ARGSUSED */
1438 int
1439 ofstat(struct thread *td, struct ofstat_args *uap)
1440 {
1441 	struct ostat oub;
1442 	struct stat ub;
1443 	int error;
1444 
1445 	error = kern_fstat(td, uap->fd, &ub);
1446 	if (error == 0) {
1447 		cvtstat(&ub, &oub);
1448 		error = copyout(&oub, uap->sb, sizeof(oub));
1449 	}
1450 	return (error);
1451 }
1452 #endif /* COMPAT_43 */
1453 
1454 #if defined(COMPAT_FREEBSD11)
1455 int
1456 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1457 {
1458 	struct stat sb;
1459 	struct freebsd11_stat osb;
1460 	int error;
1461 
1462 	error = kern_fstat(td, uap->fd, &sb);
1463 	if (error != 0)
1464 		return (error);
1465 	error = freebsd11_cvtstat(&sb, &osb);
1466 	if (error == 0)
1467 		error = copyout(&osb, uap->sb, sizeof(osb));
1468 	return (error);
1469 }
1470 #endif	/* COMPAT_FREEBSD11 */
1471 
1472 /*
1473  * Return status information about a file descriptor.
1474  */
1475 #ifndef _SYS_SYSPROTO_H_
1476 struct fstat_args {
1477 	int	fd;
1478 	struct	stat *sb;
1479 };
1480 #endif
1481 /* ARGSUSED */
1482 int
1483 sys_fstat(struct thread *td, struct fstat_args *uap)
1484 {
1485 	struct stat ub;
1486 	int error;
1487 
1488 	error = kern_fstat(td, uap->fd, &ub);
1489 	if (error == 0)
1490 		error = copyout(&ub, uap->sb, sizeof(ub));
1491 	return (error);
1492 }
1493 
1494 int
1495 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1496 {
1497 	struct file *fp;
1498 	int error;
1499 
1500 	AUDIT_ARG_FD(fd);
1501 
1502 	error = fget(td, fd, &cap_fstat_rights, &fp);
1503 	if (__predict_false(error != 0))
1504 		return (error);
1505 
1506 	AUDIT_ARG_FILE(td->td_proc, fp);
1507 
1508 	error = fo_stat(fp, sbp, td->td_ucred, td);
1509 	fdrop(fp, td);
1510 #ifdef __STAT_TIME_T_EXT
1511 	sbp->st_atim_ext = 0;
1512 	sbp->st_mtim_ext = 0;
1513 	sbp->st_ctim_ext = 0;
1514 	sbp->st_btim_ext = 0;
1515 #endif
1516 #ifdef KTRACE
1517 	if (KTRPOINT(td, KTR_STRUCT))
1518 		ktrstat_error(sbp, error);
1519 #endif
1520 	return (error);
1521 }
1522 
1523 #if defined(COMPAT_FREEBSD11)
1524 /*
1525  * Return status information about a file descriptor.
1526  */
1527 #ifndef _SYS_SYSPROTO_H_
1528 struct freebsd11_nfstat_args {
1529 	int	fd;
1530 	struct	nstat *sb;
1531 };
1532 #endif
1533 /* ARGSUSED */
1534 int
1535 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1536 {
1537 	struct nstat nub;
1538 	struct stat ub;
1539 	int error;
1540 
1541 	error = kern_fstat(td, uap->fd, &ub);
1542 	if (error == 0) {
1543 		freebsd11_cvtnstat(&ub, &nub);
1544 		error = copyout(&nub, uap->sb, sizeof(nub));
1545 	}
1546 	return (error);
1547 }
1548 #endif /* COMPAT_FREEBSD11 */
1549 
1550 /*
1551  * Return pathconf information about a file descriptor.
1552  */
1553 #ifndef _SYS_SYSPROTO_H_
1554 struct fpathconf_args {
1555 	int	fd;
1556 	int	name;
1557 };
1558 #endif
1559 /* ARGSUSED */
1560 int
1561 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1562 {
1563 	long value;
1564 	int error;
1565 
1566 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1567 	if (error == 0)
1568 		td->td_retval[0] = value;
1569 	return (error);
1570 }
1571 
1572 int
1573 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1574 {
1575 	struct file *fp;
1576 	struct vnode *vp;
1577 	int error;
1578 
1579 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1580 	if (error != 0)
1581 		return (error);
1582 
1583 	if (name == _PC_ASYNC_IO) {
1584 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1585 		goto out;
1586 	}
1587 	vp = fp->f_vnode;
1588 	if (vp != NULL) {
1589 		vn_lock(vp, LK_SHARED | LK_RETRY);
1590 		error = VOP_PATHCONF(vp, name, valuep);
1591 		VOP_UNLOCK(vp);
1592 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1593 		if (name != _PC_PIPE_BUF) {
1594 			error = EINVAL;
1595 		} else {
1596 			*valuep = PIPE_BUF;
1597 			error = 0;
1598 		}
1599 	} else {
1600 		error = EOPNOTSUPP;
1601 	}
1602 out:
1603 	fdrop(fp, td);
1604 	return (error);
1605 }
1606 
1607 /*
1608  * Copy filecaps structure allocating memory for ioctls array if needed.
1609  *
1610  * The last parameter indicates whether the fdtable is locked. If it is not and
1611  * ioctls are encountered, copying fails and the caller must lock the table.
1612  *
1613  * Note that if the table was not locked, the caller has to check the relevant
1614  * sequence counter to determine whether the operation was successful.
1615  */
1616 bool
1617 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1618 {
1619 	size_t size;
1620 
1621 	if (src->fc_ioctls != NULL && !locked)
1622 		return (false);
1623 	memcpy(dst, src, sizeof(*src));
1624 	if (src->fc_ioctls == NULL)
1625 		return (true);
1626 
1627 	KASSERT(src->fc_nioctls > 0,
1628 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1629 
1630 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1631 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1632 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1633 	return (true);
1634 }
1635 
1636 static u_long *
1637 filecaps_copy_prep(const struct filecaps *src)
1638 {
1639 	u_long *ioctls;
1640 	size_t size;
1641 
1642 	if (__predict_true(src->fc_ioctls == NULL))
1643 		return (NULL);
1644 
1645 	KASSERT(src->fc_nioctls > 0,
1646 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1647 
1648 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1649 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1650 	return (ioctls);
1651 }
1652 
1653 static void
1654 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1655     u_long *ioctls)
1656 {
1657 	size_t size;
1658 
1659 	*dst = *src;
1660 	if (__predict_true(src->fc_ioctls == NULL)) {
1661 		MPASS(ioctls == NULL);
1662 		return;
1663 	}
1664 
1665 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1666 	dst->fc_ioctls = ioctls;
1667 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1668 }
1669 
1670 /*
1671  * Move filecaps structure to the new place and clear the old place.
1672  */
1673 void
1674 filecaps_move(struct filecaps *src, struct filecaps *dst)
1675 {
1676 
1677 	*dst = *src;
1678 	bzero(src, sizeof(*src));
1679 }
1680 
1681 /*
1682  * Fill the given filecaps structure with full rights.
1683  */
1684 static void
1685 filecaps_fill(struct filecaps *fcaps)
1686 {
1687 
1688 	CAP_ALL(&fcaps->fc_rights);
1689 	fcaps->fc_ioctls = NULL;
1690 	fcaps->fc_nioctls = -1;
1691 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1692 }
1693 
1694 /*
1695  * Free memory allocated within filecaps structure.
1696  */
1697 void
1698 filecaps_free(struct filecaps *fcaps)
1699 {
1700 
1701 	free(fcaps->fc_ioctls, M_FILECAPS);
1702 	bzero(fcaps, sizeof(*fcaps));
1703 }
1704 
1705 static u_long *
1706 filecaps_free_prep(struct filecaps *fcaps)
1707 {
1708 	u_long *ioctls;
1709 
1710 	ioctls = fcaps->fc_ioctls;
1711 	bzero(fcaps, sizeof(*fcaps));
1712 	return (ioctls);
1713 }
1714 
1715 static void
1716 filecaps_free_finish(u_long *ioctls)
1717 {
1718 
1719 	free(ioctls, M_FILECAPS);
1720 }
1721 
1722 /*
1723  * Validate the given filecaps structure.
1724  */
1725 static void
1726 filecaps_validate(const struct filecaps *fcaps, const char *func)
1727 {
1728 
1729 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1730 	    ("%s: invalid rights", func));
1731 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1732 	    ("%s: invalid fcntls", func));
1733 	KASSERT(fcaps->fc_fcntls == 0 ||
1734 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1735 	    ("%s: fcntls without CAP_FCNTL", func));
1736 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1737 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1738 	    ("%s: invalid ioctls", func));
1739 	KASSERT(fcaps->fc_nioctls == 0 ||
1740 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1741 	    ("%s: ioctls without CAP_IOCTL", func));
1742 }
1743 
1744 static void
1745 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1746 {
1747 	int nfd1;
1748 
1749 	FILEDESC_XLOCK_ASSERT(fdp);
1750 
1751 	nfd1 = fdp->fd_nfiles * 2;
1752 	if (nfd1 < nfd)
1753 		nfd1 = nfd;
1754 	fdgrowtable(fdp, nfd1);
1755 }
1756 
1757 /*
1758  * Grow the file table to accommodate (at least) nfd descriptors.
1759  */
1760 static void
1761 fdgrowtable(struct filedesc *fdp, int nfd)
1762 {
1763 	struct filedesc0 *fdp0;
1764 	struct freetable *ft;
1765 	struct fdescenttbl *ntable;
1766 	struct fdescenttbl *otable;
1767 	int nnfiles, onfiles;
1768 	NDSLOTTYPE *nmap, *omap;
1769 
1770 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1771 
1772 	/* save old values */
1773 	onfiles = fdp->fd_nfiles;
1774 	otable = fdp->fd_files;
1775 	omap = fdp->fd_map;
1776 
1777 	/* compute the size of the new table */
1778 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1779 	if (nnfiles <= onfiles)
1780 		/* the table is already large enough */
1781 		return;
1782 
1783 	/*
1784 	 * Allocate a new table.  We need enough space for the number of
1785 	 * entries, file entries themselves and the struct freetable we will use
1786 	 * when we decommission the table and place it on the freelist.
1787 	 * We place the struct freetable in the middle so we don't have
1788 	 * to worry about padding.
1789 	 */
1790 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1791 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1792 	    sizeof(struct freetable),
1793 	    M_FILEDESC, M_ZERO | M_WAITOK);
1794 	/* copy the old data */
1795 	ntable->fdt_nfiles = nnfiles;
1796 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1797 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1798 
1799 	/*
1800 	 * Allocate a new map only if the old is not large enough.  It will
1801 	 * grow at a slower rate than the table as it can map more
1802 	 * entries than the table can hold.
1803 	 */
1804 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1805 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1806 		    M_ZERO | M_WAITOK);
1807 		/* copy over the old data and update the pointer */
1808 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1809 		fdp->fd_map = nmap;
1810 	}
1811 
1812 	/*
1813 	 * Make sure that ntable is correctly initialized before we replace
1814 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1815 	 * data.
1816 	 */
1817 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1818 
1819 	/*
1820 	 * Free the old file table when not shared by other threads or processes.
1821 	 * The old file table is considered to be shared when either are true:
1822 	 * - The process has more than one thread.
1823 	 * - The file descriptor table has been shared via fdshare().
1824 	 *
1825 	 * When shared, the old file table will be placed on a freelist
1826 	 * which will be processed when the struct filedesc is released.
1827 	 *
1828 	 * Note that if onfiles == NDFILE, we're dealing with the original
1829 	 * static allocation contained within (struct filedesc0 *)fdp,
1830 	 * which must not be freed.
1831 	 */
1832 	if (onfiles > NDFILE) {
1833 		if (curproc->p_numthreads == 1 && fdp->fd_refcnt == 1)
1834 			free(otable, M_FILEDESC);
1835 		else {
1836 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1837 			fdp0 = (struct filedesc0 *)fdp;
1838 			ft->ft_table = otable;
1839 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1840 		}
1841 	}
1842 	/*
1843 	 * The map does not have the same possibility of threads still
1844 	 * holding references to it.  So always free it as long as it
1845 	 * does not reference the original static allocation.
1846 	 */
1847 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1848 		free(omap, M_FILEDESC);
1849 }
1850 
1851 /*
1852  * Allocate a file descriptor for the process.
1853  */
1854 int
1855 fdalloc(struct thread *td, int minfd, int *result)
1856 {
1857 	struct proc *p = td->td_proc;
1858 	struct filedesc *fdp = p->p_fd;
1859 	int fd, maxfd, allocfd;
1860 #ifdef RACCT
1861 	int error;
1862 #endif
1863 
1864 	FILEDESC_XLOCK_ASSERT(fdp);
1865 
1866 	if (fdp->fd_freefile > minfd)
1867 		minfd = fdp->fd_freefile;
1868 
1869 	maxfd = getmaxfd(td);
1870 
1871 	/*
1872 	 * Search the bitmap for a free descriptor starting at minfd.
1873 	 * If none is found, grow the file table.
1874 	 */
1875 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1876 	if (__predict_false(fd >= maxfd))
1877 		return (EMFILE);
1878 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1879 		allocfd = min(fd * 2, maxfd);
1880 #ifdef RACCT
1881 		if (RACCT_ENABLED()) {
1882 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1883 			if (error != 0)
1884 				return (EMFILE);
1885 		}
1886 #endif
1887 		/*
1888 		 * fd is already equal to first free descriptor >= minfd, so
1889 		 * we only need to grow the table and we are done.
1890 		 */
1891 		fdgrowtable_exp(fdp, allocfd);
1892 	}
1893 
1894 	/*
1895 	 * Perform some sanity checks, then mark the file descriptor as
1896 	 * used and return it to the caller.
1897 	 */
1898 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1899 	    ("invalid descriptor %d", fd));
1900 	KASSERT(!fdisused(fdp, fd),
1901 	    ("fd_first_free() returned non-free descriptor"));
1902 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1903 	    ("file descriptor isn't free"));
1904 	fdused(fdp, fd);
1905 	*result = fd;
1906 	return (0);
1907 }
1908 
1909 /*
1910  * Allocate n file descriptors for the process.
1911  */
1912 int
1913 fdallocn(struct thread *td, int minfd, int *fds, int n)
1914 {
1915 	struct proc *p = td->td_proc;
1916 	struct filedesc *fdp = p->p_fd;
1917 	int i;
1918 
1919 	FILEDESC_XLOCK_ASSERT(fdp);
1920 
1921 	for (i = 0; i < n; i++)
1922 		if (fdalloc(td, 0, &fds[i]) != 0)
1923 			break;
1924 
1925 	if (i < n) {
1926 		for (i--; i >= 0; i--)
1927 			fdunused(fdp, fds[i]);
1928 		return (EMFILE);
1929 	}
1930 
1931 	return (0);
1932 }
1933 
1934 /*
1935  * Create a new open file structure and allocate a file descriptor for the
1936  * process that refers to it.  We add one reference to the file for the
1937  * descriptor table and one reference for resultfp. This is to prevent us
1938  * being preempted and the entry in the descriptor table closed after we
1939  * release the FILEDESC lock.
1940  */
1941 int
1942 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1943     struct filecaps *fcaps)
1944 {
1945 	struct file *fp;
1946 	int error, fd;
1947 
1948 	error = falloc_noinstall(td, &fp);
1949 	if (error)
1950 		return (error);		/* no reference held on error */
1951 
1952 	error = finstall(td, fp, &fd, flags, fcaps);
1953 	if (error) {
1954 		fdrop(fp, td);		/* one reference (fp only) */
1955 		return (error);
1956 	}
1957 
1958 	if (resultfp != NULL)
1959 		*resultfp = fp;		/* copy out result */
1960 	else
1961 		fdrop(fp, td);		/* release local reference */
1962 
1963 	if (resultfd != NULL)
1964 		*resultfd = fd;
1965 
1966 	return (0);
1967 }
1968 
1969 /*
1970  * Create a new open file structure without allocating a file descriptor.
1971  */
1972 int
1973 falloc_noinstall(struct thread *td, struct file **resultfp)
1974 {
1975 	struct file *fp;
1976 	int maxuserfiles = maxfiles - (maxfiles / 20);
1977 	int openfiles_new;
1978 	static struct timeval lastfail;
1979 	static int curfail;
1980 
1981 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1982 
1983 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1984 	if ((openfiles_new >= maxuserfiles &&
1985 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1986 	    openfiles_new >= maxfiles) {
1987 		atomic_subtract_int(&openfiles, 1);
1988 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1989 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1990 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1991 		}
1992 		return (ENFILE);
1993 	}
1994 	fp = uma_zalloc(file_zone, M_WAITOK);
1995 	bzero(fp, sizeof(*fp));
1996 	refcount_init(&fp->f_count, 1);
1997 	fp->f_cred = crhold(td->td_ucred);
1998 	fp->f_ops = &badfileops;
1999 	*resultfp = fp;
2000 	return (0);
2001 }
2002 
2003 /*
2004  * Install a file in a file descriptor table.
2005  */
2006 void
2007 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2008     struct filecaps *fcaps)
2009 {
2010 	struct filedescent *fde;
2011 
2012 	MPASS(fp != NULL);
2013 	if (fcaps != NULL)
2014 		filecaps_validate(fcaps, __func__);
2015 	FILEDESC_XLOCK_ASSERT(fdp);
2016 
2017 	fde = &fdp->fd_ofiles[fd];
2018 #ifdef CAPABILITIES
2019 	seqc_write_begin(&fde->fde_seqc);
2020 #endif
2021 	fde->fde_file = fp;
2022 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2023 	if (fcaps != NULL)
2024 		filecaps_move(fcaps, &fde->fde_caps);
2025 	else
2026 		filecaps_fill(&fde->fde_caps);
2027 #ifdef CAPABILITIES
2028 	seqc_write_end(&fde->fde_seqc);
2029 #endif
2030 }
2031 
2032 int
2033 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2034     struct filecaps *fcaps)
2035 {
2036 	struct filedesc *fdp = td->td_proc->p_fd;
2037 	int error;
2038 
2039 	MPASS(fd != NULL);
2040 
2041 	if (!fhold(fp))
2042 		return (EBADF);
2043 	FILEDESC_XLOCK(fdp);
2044 	error = fdalloc(td, 0, fd);
2045 	if (__predict_false(error != 0)) {
2046 		FILEDESC_XUNLOCK(fdp);
2047 		fdrop(fp, td);
2048 		return (error);
2049 	}
2050 	_finstall(fdp, fp, *fd, flags, fcaps);
2051 	FILEDESC_XUNLOCK(fdp);
2052 	return (0);
2053 }
2054 
2055 /*
2056  * Build a new filedesc structure from another.
2057  *
2058  * If fdp is not NULL, return with it shared locked.
2059  */
2060 struct filedesc *
2061 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2062 {
2063 	struct filedesc0 *newfdp0;
2064 	struct filedesc *newfdp;
2065 
2066 	if (prepfiles)
2067 		MPASS(lastfile != NULL);
2068 	else
2069 		MPASS(lastfile == NULL);
2070 
2071 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2072 	newfdp = &newfdp0->fd_fd;
2073 
2074 	/* Create the file descriptor table. */
2075 	FILEDESC_LOCK_INIT(newfdp);
2076 	refcount_init(&newfdp->fd_refcnt, 1);
2077 	refcount_init(&newfdp->fd_holdcnt, 1);
2078 	newfdp->fd_map = newfdp0->fd_dmap;
2079 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2080 	newfdp->fd_files->fdt_nfiles = NDFILE;
2081 
2082 	if (fdp == NULL)
2083 		return (newfdp);
2084 
2085 	FILEDESC_SLOCK(fdp);
2086 	if (!prepfiles) {
2087 		FILEDESC_SUNLOCK(fdp);
2088 		return (newfdp);
2089 	}
2090 
2091 	for (;;) {
2092 		*lastfile = fdlastfile(fdp);
2093 		if (*lastfile < newfdp->fd_nfiles)
2094 			break;
2095 		FILEDESC_SUNLOCK(fdp);
2096 		fdgrowtable(newfdp, *lastfile + 1);
2097 		FILEDESC_SLOCK(fdp);
2098 	}
2099 
2100 	return (newfdp);
2101 }
2102 
2103 /*
2104  * Build a pwddesc structure from another.
2105  * Copy the current, root, and jail root vnode references.
2106  *
2107  * If pdp is not NULL, return with it shared locked.
2108  */
2109 struct pwddesc *
2110 pdinit(struct pwddesc *pdp, bool keeplock)
2111 {
2112 	struct pwddesc *newpdp;
2113 	struct pwd *newpwd;
2114 
2115 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2116 
2117 	PWDDESC_LOCK_INIT(newpdp);
2118 	refcount_init(&newpdp->pd_refcount, 1);
2119 	newpdp->pd_cmask = CMASK;
2120 
2121 	if (pdp == NULL) {
2122 		newpwd = pwd_alloc();
2123 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2124 		return (newpdp);
2125 	}
2126 
2127 	PWDDESC_XLOCK(pdp);
2128 	newpwd = pwd_hold_pwddesc(pdp);
2129 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2130 	if (!keeplock)
2131 		PWDDESC_XUNLOCK(pdp);
2132 	return (newpdp);
2133 }
2134 
2135 static struct filedesc *
2136 fdhold(struct proc *p)
2137 {
2138 	struct filedesc *fdp;
2139 
2140 	PROC_LOCK_ASSERT(p, MA_OWNED);
2141 	fdp = p->p_fd;
2142 	if (fdp != NULL)
2143 		refcount_acquire(&fdp->fd_holdcnt);
2144 	return (fdp);
2145 }
2146 
2147 static struct pwddesc *
2148 pdhold(struct proc *p)
2149 {
2150 	struct pwddesc *pdp;
2151 
2152 	PROC_LOCK_ASSERT(p, MA_OWNED);
2153 	pdp = p->p_pd;
2154 	if (pdp != NULL)
2155 		refcount_acquire(&pdp->pd_refcount);
2156 	return (pdp);
2157 }
2158 
2159 static void
2160 fddrop(struct filedesc *fdp)
2161 {
2162 
2163 	if (fdp->fd_holdcnt > 1) {
2164 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2165 			return;
2166 	}
2167 
2168 	FILEDESC_LOCK_DESTROY(fdp);
2169 	uma_zfree(filedesc0_zone, fdp);
2170 }
2171 
2172 static void
2173 pddrop(struct pwddesc *pdp)
2174 {
2175 	struct pwd *pwd;
2176 
2177 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2178 		return;
2179 
2180 	PWDDESC_XLOCK(pdp);
2181 	if (refcount_release(&pdp->pd_refcount) == 0) {
2182 		PWDDESC_XUNLOCK(pdp);
2183 		return;
2184 	}
2185 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2186 	pwd_set(pdp, NULL);
2187 	PWDDESC_XUNLOCK(pdp);
2188 	pwd_drop(pwd);
2189 
2190 	PWDDESC_LOCK_DESTROY(pdp);
2191 	free(pdp, M_PWDDESC);
2192 }
2193 
2194 /*
2195  * Share a filedesc structure.
2196  */
2197 struct filedesc *
2198 fdshare(struct filedesc *fdp)
2199 {
2200 
2201 	refcount_acquire(&fdp->fd_refcnt);
2202 	return (fdp);
2203 }
2204 
2205 /*
2206  * Share a pwddesc structure.
2207  */
2208 struct pwddesc *
2209 pdshare(struct pwddesc *pdp)
2210 {
2211 	refcount_acquire(&pdp->pd_refcount);
2212 	return (pdp);
2213 }
2214 
2215 /*
2216  * Unshare a filedesc structure, if necessary by making a copy
2217  */
2218 void
2219 fdunshare(struct thread *td)
2220 {
2221 	struct filedesc *tmp;
2222 	struct proc *p = td->td_proc;
2223 
2224 	if (p->p_fd->fd_refcnt == 1)
2225 		return;
2226 
2227 	tmp = fdcopy(p->p_fd);
2228 	fdescfree(td);
2229 	p->p_fd = tmp;
2230 }
2231 
2232 /*
2233  * Unshare a pwddesc structure.
2234  */
2235 void
2236 pdunshare(struct thread *td)
2237 {
2238 	struct pwddesc *pdp;
2239 	struct proc *p;
2240 
2241 	p = td->td_proc;
2242 	/* Not shared. */
2243 	if (p->p_pd->pd_refcount == 1)
2244 		return;
2245 
2246 	pdp = pdcopy(p->p_pd);
2247 	pdescfree(td);
2248 	p->p_pd = pdp;
2249 }
2250 
2251 void
2252 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2253 {
2254 
2255 	fdescfree(td);
2256 	td->td_proc->p_fd = fdp;
2257 }
2258 
2259 /*
2260  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2261  * this is to ease callers, not catch errors.
2262  */
2263 struct filedesc *
2264 fdcopy(struct filedesc *fdp)
2265 {
2266 	struct filedesc *newfdp;
2267 	struct filedescent *nfde, *ofde;
2268 	int i, lastfile;
2269 
2270 	MPASS(fdp != NULL);
2271 
2272 	newfdp = fdinit(fdp, true, &lastfile);
2273 	/* copy all passable descriptors (i.e. not kqueue) */
2274 	newfdp->fd_freefile = -1;
2275 	for (i = 0; i <= lastfile; ++i) {
2276 		ofde = &fdp->fd_ofiles[i];
2277 		if (ofde->fde_file == NULL ||
2278 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2279 		    !fhold(ofde->fde_file)) {
2280 			if (newfdp->fd_freefile == -1)
2281 				newfdp->fd_freefile = i;
2282 			continue;
2283 		}
2284 		nfde = &newfdp->fd_ofiles[i];
2285 		*nfde = *ofde;
2286 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2287 		fdused_init(newfdp, i);
2288 	}
2289 	if (newfdp->fd_freefile == -1)
2290 		newfdp->fd_freefile = i;
2291 	FILEDESC_SUNLOCK(fdp);
2292 	return (newfdp);
2293 }
2294 
2295 /*
2296  * Copy a pwddesc structure.
2297  */
2298 struct pwddesc *
2299 pdcopy(struct pwddesc *pdp)
2300 {
2301 	struct pwddesc *newpdp;
2302 
2303 	MPASS(pdp != NULL);
2304 
2305 	newpdp = pdinit(pdp, true);
2306 	newpdp->pd_cmask = pdp->pd_cmask;
2307 	PWDDESC_XUNLOCK(pdp);
2308 	return (newpdp);
2309 }
2310 
2311 /*
2312  * Copies a filedesc structure, while remapping all file descriptors
2313  * stored inside using a translation table.
2314  *
2315  * File descriptors are copied over to the new file descriptor table,
2316  * regardless of whether the close-on-exec flag is set.
2317  */
2318 int
2319 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2320     struct filedesc **ret)
2321 {
2322 	struct filedesc *newfdp;
2323 	struct filedescent *nfde, *ofde;
2324 	int error, i, lastfile;
2325 
2326 	MPASS(fdp != NULL);
2327 
2328 	newfdp = fdinit(fdp, true, &lastfile);
2329 	if (nfds > lastfile + 1) {
2330 		/* New table cannot be larger than the old one. */
2331 		error = E2BIG;
2332 		goto bad;
2333 	}
2334 	/* Copy all passable descriptors (i.e. not kqueue). */
2335 	newfdp->fd_freefile = nfds;
2336 	for (i = 0; i < nfds; ++i) {
2337 		if (fds[i] < 0 || fds[i] > lastfile) {
2338 			/* File descriptor out of bounds. */
2339 			error = EBADF;
2340 			goto bad;
2341 		}
2342 		ofde = &fdp->fd_ofiles[fds[i]];
2343 		if (ofde->fde_file == NULL) {
2344 			/* Unused file descriptor. */
2345 			error = EBADF;
2346 			goto bad;
2347 		}
2348 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2349 			/* File descriptor cannot be passed. */
2350 			error = EINVAL;
2351 			goto bad;
2352 		}
2353 		if (!fhold(ofde->fde_file)) {
2354 			error = EBADF;
2355 			goto bad;
2356 		}
2357 		nfde = &newfdp->fd_ofiles[i];
2358 		*nfde = *ofde;
2359 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2360 		fdused_init(newfdp, i);
2361 	}
2362 	FILEDESC_SUNLOCK(fdp);
2363 	*ret = newfdp;
2364 	return (0);
2365 bad:
2366 	FILEDESC_SUNLOCK(fdp);
2367 	fdescfree_remapped(newfdp);
2368 	return (error);
2369 }
2370 
2371 /*
2372  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2373  * one of processes using it exits) and the table used to be shared.
2374  */
2375 static void
2376 fdclearlocks(struct thread *td)
2377 {
2378 	struct filedesc *fdp;
2379 	struct filedesc_to_leader *fdtol;
2380 	struct flock lf;
2381 	struct file *fp;
2382 	struct proc *p;
2383 	struct vnode *vp;
2384 	int i, lastfile;
2385 
2386 	p = td->td_proc;
2387 	fdp = p->p_fd;
2388 	fdtol = p->p_fdtol;
2389 	MPASS(fdtol != NULL);
2390 
2391 	FILEDESC_XLOCK(fdp);
2392 	KASSERT(fdtol->fdl_refcount > 0,
2393 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2394 	    fdtol->fdl_refcount));
2395 	if (fdtol->fdl_refcount == 1 &&
2396 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2397 		lastfile = fdlastfile(fdp);
2398 		for (i = 0; i <= lastfile; i++) {
2399 			fp = fdp->fd_ofiles[i].fde_file;
2400 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2401 			    !fhold(fp))
2402 				continue;
2403 			FILEDESC_XUNLOCK(fdp);
2404 			lf.l_whence = SEEK_SET;
2405 			lf.l_start = 0;
2406 			lf.l_len = 0;
2407 			lf.l_type = F_UNLCK;
2408 			vp = fp->f_vnode;
2409 			(void) VOP_ADVLOCK(vp,
2410 			    (caddr_t)p->p_leader, F_UNLCK,
2411 			    &lf, F_POSIX);
2412 			FILEDESC_XLOCK(fdp);
2413 			fdrop(fp, td);
2414 		}
2415 	}
2416 retry:
2417 	if (fdtol->fdl_refcount == 1) {
2418 		if (fdp->fd_holdleaderscount > 0 &&
2419 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2420 			/*
2421 			 * close() or kern_dup() has cleared a reference
2422 			 * in a shared file descriptor table.
2423 			 */
2424 			fdp->fd_holdleaderswakeup = 1;
2425 			sx_sleep(&fdp->fd_holdleaderscount,
2426 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2427 			goto retry;
2428 		}
2429 		if (fdtol->fdl_holdcount > 0) {
2430 			/*
2431 			 * Ensure that fdtol->fdl_leader remains
2432 			 * valid in closef().
2433 			 */
2434 			fdtol->fdl_wakeup = 1;
2435 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2436 			    "fdlhold", 0);
2437 			goto retry;
2438 		}
2439 	}
2440 	fdtol->fdl_refcount--;
2441 	if (fdtol->fdl_refcount == 0 &&
2442 	    fdtol->fdl_holdcount == 0) {
2443 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2444 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2445 	} else
2446 		fdtol = NULL;
2447 	p->p_fdtol = NULL;
2448 	FILEDESC_XUNLOCK(fdp);
2449 	if (fdtol != NULL)
2450 		free(fdtol, M_FILEDESC_TO_LEADER);
2451 }
2452 
2453 /*
2454  * Release a filedesc structure.
2455  */
2456 static void
2457 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2458 {
2459 	struct filedesc0 *fdp0;
2460 	struct freetable *ft, *tft;
2461 	struct filedescent *fde;
2462 	struct file *fp;
2463 	int i, lastfile;
2464 
2465 	lastfile = fdlastfile_single(fdp);
2466 	for (i = 0; i <= lastfile; i++) {
2467 		fde = &fdp->fd_ofiles[i];
2468 		fp = fde->fde_file;
2469 		if (fp != NULL) {
2470 			fdefree_last(fde);
2471 			if (needclose)
2472 				(void) closef(fp, td);
2473 			else
2474 				fdrop(fp, td);
2475 		}
2476 	}
2477 
2478 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2479 		free(fdp->fd_map, M_FILEDESC);
2480 	if (fdp->fd_nfiles > NDFILE)
2481 		free(fdp->fd_files, M_FILEDESC);
2482 
2483 	fdp0 = (struct filedesc0 *)fdp;
2484 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2485 		free(ft->ft_table, M_FILEDESC);
2486 
2487 	fddrop(fdp);
2488 }
2489 
2490 void
2491 fdescfree(struct thread *td)
2492 {
2493 	struct proc *p;
2494 	struct filedesc *fdp;
2495 
2496 	p = td->td_proc;
2497 	fdp = p->p_fd;
2498 	MPASS(fdp != NULL);
2499 
2500 #ifdef RACCT
2501 	if (RACCT_ENABLED())
2502 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2503 #endif
2504 
2505 	if (p->p_fdtol != NULL)
2506 		fdclearlocks(td);
2507 
2508 	PROC_LOCK(p);
2509 	p->p_fd = NULL;
2510 	PROC_UNLOCK(p);
2511 
2512 	if (refcount_release(&fdp->fd_refcnt) == 0)
2513 		return;
2514 
2515 	fdescfree_fds(td, fdp, 1);
2516 }
2517 
2518 void
2519 pdescfree(struct thread *td)
2520 {
2521 	struct proc *p;
2522 	struct pwddesc *pdp;
2523 
2524 	p = td->td_proc;
2525 	pdp = p->p_pd;
2526 	MPASS(pdp != NULL);
2527 
2528 	PROC_LOCK(p);
2529 	p->p_pd = NULL;
2530 	PROC_UNLOCK(p);
2531 
2532 	pddrop(pdp);
2533 }
2534 
2535 void
2536 fdescfree_remapped(struct filedesc *fdp)
2537 {
2538 	fdescfree_fds(curthread, fdp, 0);
2539 }
2540 
2541 /*
2542  * For setugid programs, we don't want to people to use that setugidness
2543  * to generate error messages which write to a file which otherwise would
2544  * otherwise be off-limits to the process.  We check for filesystems where
2545  * the vnode can change out from under us after execve (like [lin]procfs).
2546  *
2547  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2548  * sufficient.  We also don't check for setugidness since we know we are.
2549  */
2550 static bool
2551 is_unsafe(struct file *fp)
2552 {
2553 	struct vnode *vp;
2554 
2555 	if (fp->f_type != DTYPE_VNODE)
2556 		return (false);
2557 
2558 	vp = fp->f_vnode;
2559 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2560 }
2561 
2562 /*
2563  * Make this setguid thing safe, if at all possible.
2564  */
2565 void
2566 fdsetugidsafety(struct thread *td)
2567 {
2568 	struct filedesc *fdp;
2569 	struct file *fp;
2570 	int i;
2571 
2572 	fdp = td->td_proc->p_fd;
2573 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2574 	MPASS(fdp->fd_nfiles >= 3);
2575 	for (i = 0; i <= 2; i++) {
2576 		fp = fdp->fd_ofiles[i].fde_file;
2577 		if (fp != NULL && is_unsafe(fp)) {
2578 			FILEDESC_XLOCK(fdp);
2579 			knote_fdclose(td, i);
2580 			/*
2581 			 * NULL-out descriptor prior to close to avoid
2582 			 * a race while close blocks.
2583 			 */
2584 			fdfree(fdp, i);
2585 			FILEDESC_XUNLOCK(fdp);
2586 			(void) closef(fp, td);
2587 		}
2588 	}
2589 }
2590 
2591 /*
2592  * If a specific file object occupies a specific file descriptor, close the
2593  * file descriptor entry and drop a reference on the file object.  This is a
2594  * convenience function to handle a subsequent error in a function that calls
2595  * falloc() that handles the race that another thread might have closed the
2596  * file descriptor out from under the thread creating the file object.
2597  */
2598 void
2599 fdclose(struct thread *td, struct file *fp, int idx)
2600 {
2601 	struct filedesc *fdp = td->td_proc->p_fd;
2602 
2603 	FILEDESC_XLOCK(fdp);
2604 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2605 		fdfree(fdp, idx);
2606 		FILEDESC_XUNLOCK(fdp);
2607 		fdrop(fp, td);
2608 	} else
2609 		FILEDESC_XUNLOCK(fdp);
2610 }
2611 
2612 /*
2613  * Close any files on exec?
2614  */
2615 void
2616 fdcloseexec(struct thread *td)
2617 {
2618 	struct filedesc *fdp;
2619 	struct filedescent *fde;
2620 	struct file *fp;
2621 	int i, lastfile;
2622 
2623 	fdp = td->td_proc->p_fd;
2624 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2625 	lastfile = fdlastfile_single(fdp);
2626 	for (i = 0; i <= lastfile; i++) {
2627 		fde = &fdp->fd_ofiles[i];
2628 		fp = fde->fde_file;
2629 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2630 		    (fde->fde_flags & UF_EXCLOSE))) {
2631 			FILEDESC_XLOCK(fdp);
2632 			fdfree(fdp, i);
2633 			(void) closefp(fdp, i, fp, td, 0);
2634 			FILEDESC_UNLOCK_ASSERT(fdp);
2635 		}
2636 	}
2637 }
2638 
2639 /*
2640  * It is unsafe for set[ug]id processes to be started with file
2641  * descriptors 0..2 closed, as these descriptors are given implicit
2642  * significance in the Standard C library.  fdcheckstd() will create a
2643  * descriptor referencing /dev/null for each of stdin, stdout, and
2644  * stderr that is not already open.
2645  */
2646 int
2647 fdcheckstd(struct thread *td)
2648 {
2649 	struct filedesc *fdp;
2650 	register_t save;
2651 	int i, error, devnull;
2652 
2653 	fdp = td->td_proc->p_fd;
2654 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2655 	MPASS(fdp->fd_nfiles >= 3);
2656 	devnull = -1;
2657 	for (i = 0; i <= 2; i++) {
2658 		if (fdp->fd_ofiles[i].fde_file != NULL)
2659 			continue;
2660 
2661 		save = td->td_retval[0];
2662 		if (devnull != -1) {
2663 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2664 		} else {
2665 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2666 			    UIO_SYSSPACE, O_RDWR, 0);
2667 			if (error == 0) {
2668 				devnull = td->td_retval[0];
2669 				KASSERT(devnull == i, ("we didn't get our fd"));
2670 			}
2671 		}
2672 		td->td_retval[0] = save;
2673 		if (error != 0)
2674 			return (error);
2675 	}
2676 	return (0);
2677 }
2678 
2679 /*
2680  * Internal form of close.  Decrement reference count on file structure.
2681  * Note: td may be NULL when closing a file that was being passed in a
2682  * message.
2683  */
2684 int
2685 closef(struct file *fp, struct thread *td)
2686 {
2687 	struct vnode *vp;
2688 	struct flock lf;
2689 	struct filedesc_to_leader *fdtol;
2690 	struct filedesc *fdp;
2691 
2692 	/*
2693 	 * POSIX record locking dictates that any close releases ALL
2694 	 * locks owned by this process.  This is handled by setting
2695 	 * a flag in the unlock to free ONLY locks obeying POSIX
2696 	 * semantics, and not to free BSD-style file locks.
2697 	 * If the descriptor was in a message, POSIX-style locks
2698 	 * aren't passed with the descriptor, and the thread pointer
2699 	 * will be NULL.  Callers should be careful only to pass a
2700 	 * NULL thread pointer when there really is no owning
2701 	 * context that might have locks, or the locks will be
2702 	 * leaked.
2703 	 */
2704 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2705 		vp = fp->f_vnode;
2706 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2707 			lf.l_whence = SEEK_SET;
2708 			lf.l_start = 0;
2709 			lf.l_len = 0;
2710 			lf.l_type = F_UNLCK;
2711 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2712 			    F_UNLCK, &lf, F_POSIX);
2713 		}
2714 		fdtol = td->td_proc->p_fdtol;
2715 		if (fdtol != NULL) {
2716 			/*
2717 			 * Handle special case where file descriptor table is
2718 			 * shared between multiple process leaders.
2719 			 */
2720 			fdp = td->td_proc->p_fd;
2721 			FILEDESC_XLOCK(fdp);
2722 			for (fdtol = fdtol->fdl_next;
2723 			    fdtol != td->td_proc->p_fdtol;
2724 			    fdtol = fdtol->fdl_next) {
2725 				if ((fdtol->fdl_leader->p_flag &
2726 				    P_ADVLOCK) == 0)
2727 					continue;
2728 				fdtol->fdl_holdcount++;
2729 				FILEDESC_XUNLOCK(fdp);
2730 				lf.l_whence = SEEK_SET;
2731 				lf.l_start = 0;
2732 				lf.l_len = 0;
2733 				lf.l_type = F_UNLCK;
2734 				vp = fp->f_vnode;
2735 				(void) VOP_ADVLOCK(vp,
2736 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2737 				    F_POSIX);
2738 				FILEDESC_XLOCK(fdp);
2739 				fdtol->fdl_holdcount--;
2740 				if (fdtol->fdl_holdcount == 0 &&
2741 				    fdtol->fdl_wakeup != 0) {
2742 					fdtol->fdl_wakeup = 0;
2743 					wakeup(fdtol);
2744 				}
2745 			}
2746 			FILEDESC_XUNLOCK(fdp);
2747 		}
2748 	}
2749 	return (fdrop(fp, td));
2750 }
2751 
2752 /*
2753  * Initialize the file pointer with the specified properties.
2754  *
2755  * The ops are set with release semantics to be certain that the flags, type,
2756  * and data are visible when ops is.  This is to prevent ops methods from being
2757  * called with bad data.
2758  */
2759 void
2760 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2761 {
2762 	fp->f_data = data;
2763 	fp->f_flag = flag;
2764 	fp->f_type = type;
2765 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2766 }
2767 
2768 void
2769 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2770 {
2771 	fp->f_seqcount[UIO_READ] = 1;
2772 	fp->f_seqcount[UIO_WRITE] = 1;
2773 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2774 	    data, ops);
2775 }
2776 
2777 int
2778 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2779     struct file **fpp, struct filecaps *havecapsp)
2780 {
2781 	struct filedescent *fde;
2782 	int error;
2783 
2784 	FILEDESC_LOCK_ASSERT(fdp);
2785 
2786 	fde = fdeget_locked(fdp, fd);
2787 	if (fde == NULL) {
2788 		error = EBADF;
2789 		goto out;
2790 	}
2791 
2792 #ifdef CAPABILITIES
2793 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2794 	if (error != 0)
2795 		goto out;
2796 #endif
2797 
2798 	if (havecapsp != NULL)
2799 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2800 
2801 	*fpp = fde->fde_file;
2802 
2803 	error = 0;
2804 out:
2805 	return (error);
2806 }
2807 
2808 int
2809 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2810     struct file **fpp, struct filecaps *havecapsp)
2811 {
2812 	struct filedesc *fdp = td->td_proc->p_fd;
2813 	int error;
2814 #ifndef CAPABILITIES
2815 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2816 	if (havecapsp != NULL && error == 0)
2817 		filecaps_fill(havecapsp);
2818 #else
2819 	struct file *fp;
2820 	seqc_t seq;
2821 
2822 	*fpp = NULL;
2823 	for (;;) {
2824 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2825 		if (error != 0)
2826 			return (error);
2827 
2828 		if (havecapsp != NULL) {
2829 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2830 			    havecapsp, false)) {
2831 				fdrop(fp, td);
2832 				goto get_locked;
2833 			}
2834 		}
2835 
2836 		if (!fd_modified(fdp, fd, seq))
2837 			break;
2838 		fdrop(fp, td);
2839 	}
2840 
2841 	*fpp = fp;
2842 	return (0);
2843 
2844 get_locked:
2845 	FILEDESC_SLOCK(fdp);
2846 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2847 	if (error == 0 && !fhold(*fpp))
2848 		error = EBADF;
2849 	FILEDESC_SUNLOCK(fdp);
2850 #endif
2851 	return (error);
2852 }
2853 
2854 #ifdef CAPABILITIES
2855 int
2856 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2857 {
2858 	const struct filedescent *fde;
2859 	const struct fdescenttbl *fdt;
2860 	struct filedesc *fdp;
2861 	struct file *fp;
2862 	struct vnode *vp;
2863 	const cap_rights_t *haverights;
2864 	cap_rights_t rights;
2865 	seqc_t seq;
2866 
2867 	VFS_SMR_ASSERT_ENTERED();
2868 
2869 	rights = *ndp->ni_rightsneeded;
2870 	cap_rights_set_one(&rights, CAP_LOOKUP);
2871 
2872 	fdp = curproc->p_fd;
2873 	fdt = fdp->fd_files;
2874 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2875 		return (EBADF);
2876 	seq = seqc_read_any(fd_seqc(fdt, fd));
2877 	if (__predict_false(seqc_in_modify(seq)))
2878 		return (EAGAIN);
2879 	fde = &fdt->fdt_ofiles[fd];
2880 	haverights = cap_rights_fde_inline(fde);
2881 	fp = fde->fde_file;
2882 	if (__predict_false(fp == NULL))
2883 		return (EAGAIN);
2884 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2885 		return (EAGAIN);
2886 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2887 	vp = fp->f_vnode;
2888 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2889 		return (EAGAIN);
2890 	}
2891 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2892 		return (EAGAIN);
2893 	}
2894 	/*
2895 	 * Use an acquire barrier to force re-reading of fdt so it is
2896 	 * refreshed for verification.
2897 	 */
2898 	atomic_thread_fence_acq();
2899 	fdt = fdp->fd_files;
2900 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2901 		return (EAGAIN);
2902 	/*
2903 	 * If file descriptor doesn't have all rights,
2904 	 * all lookups relative to it must also be
2905 	 * strictly relative.
2906 	 *
2907 	 * Not yet supported by fast path.
2908 	 */
2909 	CAP_ALL(&rights);
2910 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2911 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2912 	    ndp->ni_filecaps.fc_nioctls != -1) {
2913 #ifdef notyet
2914 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2915 #else
2916 		return (EAGAIN);
2917 #endif
2918 	}
2919 	*vpp = vp;
2920 	return (0);
2921 }
2922 #else
2923 int
2924 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2925 {
2926 	const struct fdescenttbl *fdt;
2927 	struct filedesc *fdp;
2928 	struct file *fp;
2929 	struct vnode *vp;
2930 
2931 	VFS_SMR_ASSERT_ENTERED();
2932 
2933 	fdp = curproc->p_fd;
2934 	fdt = fdp->fd_files;
2935 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2936 		return (EBADF);
2937 	fp = fdt->fdt_ofiles[fd].fde_file;
2938 	if (__predict_false(fp == NULL))
2939 		return (EAGAIN);
2940 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2941 	vp = fp->f_vnode;
2942 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2943 		return (EAGAIN);
2944 	}
2945 	/*
2946 	 * Use an acquire barrier to force re-reading of fdt so it is
2947 	 * refreshed for verification.
2948 	 */
2949 	atomic_thread_fence_acq();
2950 	fdt = fdp->fd_files;
2951 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2952 		return (EAGAIN);
2953 	filecaps_fill(&ndp->ni_filecaps);
2954 	*vpp = vp;
2955 	return (0);
2956 }
2957 #endif
2958 
2959 int
2960 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2961     struct file **fpp, seqc_t *seqp)
2962 {
2963 #ifdef CAPABILITIES
2964 	const struct filedescent *fde;
2965 #endif
2966 	const struct fdescenttbl *fdt;
2967 	struct file *fp;
2968 #ifdef CAPABILITIES
2969 	seqc_t seq;
2970 	cap_rights_t haverights;
2971 	int error;
2972 #endif
2973 
2974 	fdt = fdp->fd_files;
2975 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2976 		return (EBADF);
2977 	/*
2978 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2979 	 * never raising a refcount above 0.  To accomplish this we have
2980 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2981 	 * must be re-verified once we acquire a reference to be certain
2982 	 * that the identity is still correct and we did not lose a race
2983 	 * due to preemption.
2984 	 */
2985 	for (;;) {
2986 #ifdef CAPABILITIES
2987 		seq = seqc_read(fd_seqc(fdt, fd));
2988 		fde = &fdt->fdt_ofiles[fd];
2989 		haverights = *cap_rights_fde_inline(fde);
2990 		fp = fde->fde_file;
2991 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2992 			continue;
2993 #else
2994 		fp = fdt->fdt_ofiles[fd].fde_file;
2995 #endif
2996 		if (fp == NULL)
2997 			return (EBADF);
2998 #ifdef CAPABILITIES
2999 		error = cap_check_inline(&haverights, needrightsp);
3000 		if (error != 0)
3001 			return (error);
3002 #endif
3003 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3004 			/*
3005 			 * The count was found either saturated or zero.
3006 			 * This re-read is not any more racy than using the
3007 			 * return value from fcmpset.
3008 			 */
3009 			if (refcount_load(&fp->f_count) != 0)
3010 				return (EBADF);
3011 			/*
3012 			 * Force a reload. Other thread could reallocate the
3013 			 * table before this fd was closed, so it is possible
3014 			 * that there is a stale fp pointer in cached version.
3015 			 */
3016 			fdt = atomic_load_ptr(&fdp->fd_files);
3017 			continue;
3018 		}
3019 		/*
3020 		 * Use an acquire barrier to force re-reading of fdt so it is
3021 		 * refreshed for verification.
3022 		 */
3023 		atomic_thread_fence_acq();
3024 		fdt = fdp->fd_files;
3025 #ifdef	CAPABILITIES
3026 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3027 #else
3028 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3029 #endif
3030 			break;
3031 		fdrop(fp, curthread);
3032 	}
3033 	*fpp = fp;
3034 	if (seqp != NULL) {
3035 #ifdef CAPABILITIES
3036 		*seqp = seq;
3037 #endif
3038 	}
3039 	return (0);
3040 }
3041 
3042 /*
3043  * See the comments in fget_unlocked_seq for an explanation of how this works.
3044  *
3045  * This is a simplified variant which bails out to the aforementioned routine
3046  * if anything goes wrong. In practice this only happens when userspace is
3047  * racing with itself.
3048  */
3049 int
3050 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3051     struct file **fpp)
3052 {
3053 #ifdef CAPABILITIES
3054 	const struct filedescent *fde;
3055 #endif
3056 	const struct fdescenttbl *fdt;
3057 	struct file *fp;
3058 #ifdef CAPABILITIES
3059 	seqc_t seq;
3060 	const cap_rights_t *haverights;
3061 #endif
3062 
3063 	fdt = fdp->fd_files;
3064 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3065 		return (EBADF);
3066 #ifdef CAPABILITIES
3067 	seq = seqc_read_any(fd_seqc(fdt, fd));
3068 	if (__predict_false(seqc_in_modify(seq)))
3069 		goto out_fallback;
3070 	fde = &fdt->fdt_ofiles[fd];
3071 	haverights = cap_rights_fde_inline(fde);
3072 	fp = fde->fde_file;
3073 #else
3074 	fp = fdt->fdt_ofiles[fd].fde_file;
3075 #endif
3076 	if (__predict_false(fp == NULL))
3077 		goto out_fallback;
3078 #ifdef CAPABILITIES
3079 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3080 		goto out_fallback;
3081 #endif
3082 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3083 		goto out_fallback;
3084 
3085 	/*
3086 	 * Use an acquire barrier to force re-reading of fdt so it is
3087 	 * refreshed for verification.
3088 	 */
3089 	atomic_thread_fence_acq();
3090 	fdt = fdp->fd_files;
3091 #ifdef	CAPABILITIES
3092 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3093 #else
3094 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3095 #endif
3096 		goto out_fdrop;
3097 	*fpp = fp;
3098 	return (0);
3099 out_fdrop:
3100 	fdrop(fp, curthread);
3101 out_fallback:
3102 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3103 }
3104 
3105 /*
3106  * Extract the file pointer associated with the specified descriptor for the
3107  * current user process.
3108  *
3109  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3110  * returned.
3111  *
3112  * File's rights will be checked against the capability rights mask.
3113  *
3114  * If an error occurred the non-zero error is returned and *fpp is set to
3115  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3116  * responsible for fdrop().
3117  */
3118 static __inline int
3119 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3120     cap_rights_t *needrightsp)
3121 {
3122 	struct filedesc *fdp;
3123 	struct file *fp;
3124 	int error;
3125 
3126 	*fpp = NULL;
3127 	fdp = td->td_proc->p_fd;
3128 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3129 	if (__predict_false(error != 0))
3130 		return (error);
3131 	if (__predict_false(fp->f_ops == &badfileops)) {
3132 		fdrop(fp, td);
3133 		return (EBADF);
3134 	}
3135 
3136 	/*
3137 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3138 	 */
3139 	error = 0;
3140 	switch (flags) {
3141 	case FREAD:
3142 	case FWRITE:
3143 		if ((fp->f_flag & flags) == 0)
3144 			error = EBADF;
3145 		break;
3146 	case FEXEC:
3147 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3148 		    ((fp->f_flag & FWRITE) != 0))
3149 			error = EBADF;
3150 		break;
3151 	case 0:
3152 		break;
3153 	default:
3154 		KASSERT(0, ("wrong flags"));
3155 	}
3156 
3157 	if (error != 0) {
3158 		fdrop(fp, td);
3159 		return (error);
3160 	}
3161 
3162 	*fpp = fp;
3163 	return (0);
3164 }
3165 
3166 int
3167 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3168 {
3169 
3170 	return (_fget(td, fd, fpp, 0, rightsp));
3171 }
3172 
3173 int
3174 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3175     struct file **fpp)
3176 {
3177 	int error;
3178 #ifndef CAPABILITIES
3179 	error = _fget(td, fd, fpp, 0, rightsp);
3180 	if (maxprotp != NULL)
3181 		*maxprotp = VM_PROT_ALL;
3182 	return (error);
3183 #else
3184 	cap_rights_t fdrights;
3185 	struct filedesc *fdp;
3186 	struct file *fp;
3187 	seqc_t seq;
3188 
3189 	*fpp = NULL;
3190 	fdp = td->td_proc->p_fd;
3191 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3192 	for (;;) {
3193 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3194 		if (__predict_false(error != 0))
3195 			return (error);
3196 		if (__predict_false(fp->f_ops == &badfileops)) {
3197 			fdrop(fp, td);
3198 			return (EBADF);
3199 		}
3200 		if (maxprotp != NULL)
3201 			fdrights = *cap_rights(fdp, fd);
3202 		if (!fd_modified(fdp, fd, seq))
3203 			break;
3204 		fdrop(fp, td);
3205 	}
3206 
3207 	/*
3208 	 * If requested, convert capability rights to access flags.
3209 	 */
3210 	if (maxprotp != NULL)
3211 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3212 	*fpp = fp;
3213 	return (0);
3214 #endif
3215 }
3216 
3217 int
3218 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3219 {
3220 
3221 	return (_fget(td, fd, fpp, FREAD, rightsp));
3222 }
3223 
3224 int
3225 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3226 {
3227 
3228 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3229 }
3230 
3231 int
3232 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3233     struct file **fpp)
3234 {
3235 	struct filedesc *fdp = td->td_proc->p_fd;
3236 #ifndef CAPABILITIES
3237 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3238 #else
3239 	struct file *fp;
3240 	int error;
3241 	seqc_t seq;
3242 
3243 	*fpp = NULL;
3244 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3245 	for (;;) {
3246 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3247 		if (error != 0)
3248 			return (error);
3249 		error = cap_fcntl_check(fdp, fd, needfcntl);
3250 		if (!fd_modified(fdp, fd, seq))
3251 			break;
3252 		fdrop(fp, td);
3253 	}
3254 	if (error != 0) {
3255 		fdrop(fp, td);
3256 		return (error);
3257 	}
3258 	*fpp = fp;
3259 	return (0);
3260 #endif
3261 }
3262 
3263 /*
3264  * Like fget() but loads the underlying vnode, or returns an error if the
3265  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3266  * never have VM objects.  The returned vnode will be vref()'d.
3267  *
3268  * XXX: what about the unused flags ?
3269  */
3270 static __inline int
3271 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3272     struct vnode **vpp)
3273 {
3274 	struct file *fp;
3275 	int error;
3276 
3277 	*vpp = NULL;
3278 	error = _fget(td, fd, &fp, flags, needrightsp);
3279 	if (error != 0)
3280 		return (error);
3281 	if (fp->f_vnode == NULL) {
3282 		error = EINVAL;
3283 	} else {
3284 		*vpp = fp->f_vnode;
3285 		vrefact(*vpp);
3286 	}
3287 	fdrop(fp, td);
3288 
3289 	return (error);
3290 }
3291 
3292 int
3293 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3294 {
3295 
3296 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3297 }
3298 
3299 int
3300 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3301     struct filecaps *havecaps, struct vnode **vpp)
3302 {
3303 	struct filecaps caps;
3304 	struct file *fp;
3305 	int error;
3306 
3307 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3308 	if (error != 0)
3309 		return (error);
3310 	if (fp->f_ops == &badfileops) {
3311 		error = EBADF;
3312 		goto out;
3313 	}
3314 	if (fp->f_vnode == NULL) {
3315 		error = EINVAL;
3316 		goto out;
3317 	}
3318 
3319 	*havecaps = caps;
3320 	*vpp = fp->f_vnode;
3321 	vrefact(*vpp);
3322 	fdrop(fp, td);
3323 
3324 	return (0);
3325 out:
3326 	filecaps_free(&caps);
3327 	fdrop(fp, td);
3328 	return (error);
3329 }
3330 
3331 int
3332 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3333 {
3334 
3335 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3336 }
3337 
3338 int
3339 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3340 {
3341 
3342 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3343 }
3344 
3345 #ifdef notyet
3346 int
3347 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3348     struct vnode **vpp)
3349 {
3350 
3351 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3352 }
3353 #endif
3354 
3355 /*
3356  * Handle the last reference to a file being closed.
3357  *
3358  * Without the noinline attribute clang keeps inlining the func thorough this
3359  * file when fdrop is used.
3360  */
3361 int __noinline
3362 _fdrop(struct file *fp, struct thread *td)
3363 {
3364 	int error;
3365 #ifdef INVARIANTS
3366 	int count;
3367 
3368 	count = refcount_load(&fp->f_count);
3369 	if (count != 0)
3370 		panic("fdrop: fp %p count %d", fp, count);
3371 #endif
3372 	error = fo_close(fp, td);
3373 	atomic_subtract_int(&openfiles, 1);
3374 	crfree(fp->f_cred);
3375 	free(fp->f_advice, M_FADVISE);
3376 	uma_zfree(file_zone, fp);
3377 
3378 	return (error);
3379 }
3380 
3381 /*
3382  * Apply an advisory lock on a file descriptor.
3383  *
3384  * Just attempt to get a record lock of the requested type on the entire file
3385  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3386  */
3387 #ifndef _SYS_SYSPROTO_H_
3388 struct flock_args {
3389 	int	fd;
3390 	int	how;
3391 };
3392 #endif
3393 /* ARGSUSED */
3394 int
3395 sys_flock(struct thread *td, struct flock_args *uap)
3396 {
3397 	struct file *fp;
3398 	struct vnode *vp;
3399 	struct flock lf;
3400 	int error;
3401 
3402 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3403 	if (error != 0)
3404 		return (error);
3405 	if (fp->f_type != DTYPE_VNODE) {
3406 		fdrop(fp, td);
3407 		return (EOPNOTSUPP);
3408 	}
3409 
3410 	vp = fp->f_vnode;
3411 	lf.l_whence = SEEK_SET;
3412 	lf.l_start = 0;
3413 	lf.l_len = 0;
3414 	if (uap->how & LOCK_UN) {
3415 		lf.l_type = F_UNLCK;
3416 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3417 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3418 		goto done2;
3419 	}
3420 	if (uap->how & LOCK_EX)
3421 		lf.l_type = F_WRLCK;
3422 	else if (uap->how & LOCK_SH)
3423 		lf.l_type = F_RDLCK;
3424 	else {
3425 		error = EBADF;
3426 		goto done2;
3427 	}
3428 	atomic_set_int(&fp->f_flag, FHASLOCK);
3429 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3430 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3431 done2:
3432 	fdrop(fp, td);
3433 	return (error);
3434 }
3435 /*
3436  * Duplicate the specified descriptor to a free descriptor.
3437  */
3438 int
3439 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3440     int openerror, int *indxp)
3441 {
3442 	struct filedescent *newfde, *oldfde;
3443 	struct file *fp;
3444 	u_long *ioctls;
3445 	int error, indx;
3446 
3447 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3448 	    ("unexpected error %d in %s", openerror, __func__));
3449 
3450 	/*
3451 	 * If the to-be-dup'd fd number is greater than the allowed number
3452 	 * of file descriptors, or the fd to be dup'd has already been
3453 	 * closed, then reject.
3454 	 */
3455 	FILEDESC_XLOCK(fdp);
3456 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3457 		FILEDESC_XUNLOCK(fdp);
3458 		return (EBADF);
3459 	}
3460 
3461 	error = fdalloc(td, 0, &indx);
3462 	if (error != 0) {
3463 		FILEDESC_XUNLOCK(fdp);
3464 		return (error);
3465 	}
3466 
3467 	/*
3468 	 * There are two cases of interest here.
3469 	 *
3470 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3471 	 *
3472 	 * For ENXIO steal away the file structure from (dfd) and store it in
3473 	 * (indx).  (dfd) is effectively closed by this operation.
3474 	 */
3475 	switch (openerror) {
3476 	case ENODEV:
3477 		/*
3478 		 * Check that the mode the file is being opened for is a
3479 		 * subset of the mode of the existing descriptor.
3480 		 */
3481 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3482 			fdunused(fdp, indx);
3483 			FILEDESC_XUNLOCK(fdp);
3484 			return (EACCES);
3485 		}
3486 		if (!fhold(fp)) {
3487 			fdunused(fdp, indx);
3488 			FILEDESC_XUNLOCK(fdp);
3489 			return (EBADF);
3490 		}
3491 		newfde = &fdp->fd_ofiles[indx];
3492 		oldfde = &fdp->fd_ofiles[dfd];
3493 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3494 #ifdef CAPABILITIES
3495 		seqc_write_begin(&newfde->fde_seqc);
3496 #endif
3497 		memcpy(newfde, oldfde, fde_change_size);
3498 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3499 		    ioctls);
3500 #ifdef CAPABILITIES
3501 		seqc_write_end(&newfde->fde_seqc);
3502 #endif
3503 		break;
3504 	case ENXIO:
3505 		/*
3506 		 * Steal away the file pointer from dfd and stuff it into indx.
3507 		 */
3508 		newfde = &fdp->fd_ofiles[indx];
3509 		oldfde = &fdp->fd_ofiles[dfd];
3510 #ifdef CAPABILITIES
3511 		seqc_write_begin(&newfde->fde_seqc);
3512 #endif
3513 		memcpy(newfde, oldfde, fde_change_size);
3514 		oldfde->fde_file = NULL;
3515 		fdunused(fdp, dfd);
3516 #ifdef CAPABILITIES
3517 		seqc_write_end(&newfde->fde_seqc);
3518 #endif
3519 		break;
3520 	}
3521 	FILEDESC_XUNLOCK(fdp);
3522 	*indxp = indx;
3523 	return (0);
3524 }
3525 
3526 /*
3527  * This sysctl determines if we will allow a process to chroot(2) if it
3528  * has a directory open:
3529  *	0: disallowed for all processes.
3530  *	1: allowed for processes that were not already chroot(2)'ed.
3531  *	2: allowed for all processes.
3532  */
3533 
3534 static int chroot_allow_open_directories = 1;
3535 
3536 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3537     &chroot_allow_open_directories, 0,
3538     "Allow a process to chroot(2) if it has a directory open");
3539 
3540 /*
3541  * Helper function for raised chroot(2) security function:  Refuse if
3542  * any filedescriptors are open directories.
3543  */
3544 static int
3545 chroot_refuse_vdir_fds(struct filedesc *fdp)
3546 {
3547 	struct vnode *vp;
3548 	struct file *fp;
3549 	int fd, lastfile;
3550 
3551 	FILEDESC_LOCK_ASSERT(fdp);
3552 
3553 	lastfile = fdlastfile(fdp);
3554 	for (fd = 0; fd <= lastfile; fd++) {
3555 		fp = fget_locked(fdp, fd);
3556 		if (fp == NULL)
3557 			continue;
3558 		if (fp->f_type == DTYPE_VNODE) {
3559 			vp = fp->f_vnode;
3560 			if (vp->v_type == VDIR)
3561 				return (EPERM);
3562 		}
3563 	}
3564 	return (0);
3565 }
3566 
3567 static void
3568 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3569 {
3570 
3571 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3572 		vrefact(oldpwd->pwd_cdir);
3573 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3574 	}
3575 
3576 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3577 		vrefact(oldpwd->pwd_rdir);
3578 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3579 	}
3580 
3581 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3582 		vrefact(oldpwd->pwd_jdir);
3583 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3584 	}
3585 }
3586 
3587 struct pwd *
3588 pwd_hold_pwddesc(struct pwddesc *pdp)
3589 {
3590 	struct pwd *pwd;
3591 
3592 	PWDDESC_ASSERT_XLOCKED(pdp);
3593 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3594 	if (pwd != NULL)
3595 		refcount_acquire(&pwd->pwd_refcount);
3596 	return (pwd);
3597 }
3598 
3599 bool
3600 pwd_hold_smr(struct pwd *pwd)
3601 {
3602 
3603 	MPASS(pwd != NULL);
3604 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3605 		return (true);
3606 	}
3607 	return (false);
3608 }
3609 
3610 struct pwd *
3611 pwd_hold(struct thread *td)
3612 {
3613 	struct pwddesc *pdp;
3614 	struct pwd *pwd;
3615 
3616 	pdp = td->td_proc->p_pd;
3617 
3618 	vfs_smr_enter();
3619 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3620 	if (pwd_hold_smr(pwd)) {
3621 		vfs_smr_exit();
3622 		return (pwd);
3623 	}
3624 	vfs_smr_exit();
3625 	PWDDESC_XLOCK(pdp);
3626 	pwd = pwd_hold_pwddesc(pdp);
3627 	MPASS(pwd != NULL);
3628 	PWDDESC_XUNLOCK(pdp);
3629 	return (pwd);
3630 }
3631 
3632 struct pwd *
3633 pwd_get_smr(void)
3634 {
3635 	struct pwd *pwd;
3636 
3637 	pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3638 	MPASS(pwd != NULL);
3639 	return (pwd);
3640 }
3641 
3642 static struct pwd *
3643 pwd_alloc(void)
3644 {
3645 	struct pwd *pwd;
3646 
3647 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3648 	bzero(pwd, sizeof(*pwd));
3649 	refcount_init(&pwd->pwd_refcount, 1);
3650 	return (pwd);
3651 }
3652 
3653 void
3654 pwd_drop(struct pwd *pwd)
3655 {
3656 
3657 	if (!refcount_release(&pwd->pwd_refcount))
3658 		return;
3659 
3660 	if (pwd->pwd_cdir != NULL)
3661 		vrele(pwd->pwd_cdir);
3662 	if (pwd->pwd_rdir != NULL)
3663 		vrele(pwd->pwd_rdir);
3664 	if (pwd->pwd_jdir != NULL)
3665 		vrele(pwd->pwd_jdir);
3666 	uma_zfree_smr(pwd_zone, pwd);
3667 }
3668 
3669 /*
3670 * Common routine for kern_chroot() and jail_attach().  The caller is
3671 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3672 * authorize this operation.
3673 */
3674 int
3675 pwd_chroot(struct thread *td, struct vnode *vp)
3676 {
3677 	struct pwddesc *pdp;
3678 	struct filedesc *fdp;
3679 	struct pwd *newpwd, *oldpwd;
3680 	int error;
3681 
3682 	fdp = td->td_proc->p_fd;
3683 	pdp = td->td_proc->p_pd;
3684 	newpwd = pwd_alloc();
3685 	FILEDESC_SLOCK(fdp);
3686 	PWDDESC_XLOCK(pdp);
3687 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3688 	if (chroot_allow_open_directories == 0 ||
3689 	    (chroot_allow_open_directories == 1 &&
3690 	    oldpwd->pwd_rdir != rootvnode)) {
3691 		error = chroot_refuse_vdir_fds(fdp);
3692 		FILEDESC_SUNLOCK(fdp);
3693 		if (error != 0) {
3694 			PWDDESC_XUNLOCK(pdp);
3695 			pwd_drop(newpwd);
3696 			return (error);
3697 		}
3698 	} else {
3699 		FILEDESC_SUNLOCK(fdp);
3700 	}
3701 
3702 	vrefact(vp);
3703 	newpwd->pwd_rdir = vp;
3704 	if (oldpwd->pwd_jdir == NULL) {
3705 		vrefact(vp);
3706 		newpwd->pwd_jdir = vp;
3707 	}
3708 	pwd_fill(oldpwd, newpwd);
3709 	pwd_set(pdp, newpwd);
3710 	PWDDESC_XUNLOCK(pdp);
3711 	pwd_drop(oldpwd);
3712 	return (0);
3713 }
3714 
3715 void
3716 pwd_chdir(struct thread *td, struct vnode *vp)
3717 {
3718 	struct pwddesc *pdp;
3719 	struct pwd *newpwd, *oldpwd;
3720 
3721 	VNPASS(vp->v_usecount > 0, vp);
3722 
3723 	newpwd = pwd_alloc();
3724 	pdp = td->td_proc->p_pd;
3725 	PWDDESC_XLOCK(pdp);
3726 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3727 	newpwd->pwd_cdir = vp;
3728 	pwd_fill(oldpwd, newpwd);
3729 	pwd_set(pdp, newpwd);
3730 	PWDDESC_XUNLOCK(pdp);
3731 	pwd_drop(oldpwd);
3732 }
3733 
3734 void
3735 pwd_ensure_dirs(void)
3736 {
3737 	struct pwddesc *pdp;
3738 	struct pwd *oldpwd, *newpwd;
3739 
3740 	pdp = curproc->p_pd;
3741 	PWDDESC_XLOCK(pdp);
3742 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3743 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3744 		PWDDESC_XUNLOCK(pdp);
3745 		return;
3746 	}
3747 	PWDDESC_XUNLOCK(pdp);
3748 
3749 	newpwd = pwd_alloc();
3750 	PWDDESC_XLOCK(pdp);
3751 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3752 	pwd_fill(oldpwd, newpwd);
3753 	if (newpwd->pwd_cdir == NULL) {
3754 		vrefact(rootvnode);
3755 		newpwd->pwd_cdir = rootvnode;
3756 	}
3757 	if (newpwd->pwd_rdir == NULL) {
3758 		vrefact(rootvnode);
3759 		newpwd->pwd_rdir = rootvnode;
3760 	}
3761 	pwd_set(pdp, newpwd);
3762 	PWDDESC_XUNLOCK(pdp);
3763 	pwd_drop(oldpwd);
3764 }
3765 
3766 void
3767 pwd_set_rootvnode(void)
3768 {
3769 	struct pwddesc *pdp;
3770 	struct pwd *oldpwd, *newpwd;
3771 
3772 	pdp = curproc->p_pd;
3773 
3774 	newpwd = pwd_alloc();
3775 	PWDDESC_XLOCK(pdp);
3776 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3777 	vrefact(rootvnode);
3778 	newpwd->pwd_cdir = rootvnode;
3779 	vrefact(rootvnode);
3780 	newpwd->pwd_rdir = rootvnode;
3781 	pwd_fill(oldpwd, newpwd);
3782 	pwd_set(pdp, newpwd);
3783 	PWDDESC_XUNLOCK(pdp);
3784 	pwd_drop(oldpwd);
3785 }
3786 
3787 /*
3788  * Scan all active processes and prisons to see if any of them have a current
3789  * or root directory of `olddp'. If so, replace them with the new mount point.
3790  */
3791 void
3792 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3793 {
3794 	struct pwddesc *pdp;
3795 	struct pwd *newpwd, *oldpwd;
3796 	struct prison *pr;
3797 	struct proc *p;
3798 	int nrele;
3799 
3800 	if (vrefcnt(olddp) == 1)
3801 		return;
3802 	nrele = 0;
3803 	newpwd = pwd_alloc();
3804 	sx_slock(&allproc_lock);
3805 	FOREACH_PROC_IN_SYSTEM(p) {
3806 		PROC_LOCK(p);
3807 		pdp = pdhold(p);
3808 		PROC_UNLOCK(p);
3809 		if (pdp == NULL)
3810 			continue;
3811 		PWDDESC_XLOCK(pdp);
3812 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3813 		if (oldpwd == NULL ||
3814 		    (oldpwd->pwd_cdir != olddp &&
3815 		    oldpwd->pwd_rdir != olddp &&
3816 		    oldpwd->pwd_jdir != olddp)) {
3817 			PWDDESC_XUNLOCK(pdp);
3818 			pddrop(pdp);
3819 			continue;
3820 		}
3821 		if (oldpwd->pwd_cdir == olddp) {
3822 			vrefact(newdp);
3823 			newpwd->pwd_cdir = newdp;
3824 		}
3825 		if (oldpwd->pwd_rdir == olddp) {
3826 			vrefact(newdp);
3827 			newpwd->pwd_rdir = newdp;
3828 		}
3829 		if (oldpwd->pwd_jdir == olddp) {
3830 			vrefact(newdp);
3831 			newpwd->pwd_jdir = newdp;
3832 		}
3833 		pwd_fill(oldpwd, newpwd);
3834 		pwd_set(pdp, newpwd);
3835 		PWDDESC_XUNLOCK(pdp);
3836 		pwd_drop(oldpwd);
3837 		pddrop(pdp);
3838 		newpwd = pwd_alloc();
3839 	}
3840 	sx_sunlock(&allproc_lock);
3841 	pwd_drop(newpwd);
3842 	if (rootvnode == olddp) {
3843 		vrefact(newdp);
3844 		rootvnode = newdp;
3845 		nrele++;
3846 	}
3847 	mtx_lock(&prison0.pr_mtx);
3848 	if (prison0.pr_root == olddp) {
3849 		vrefact(newdp);
3850 		prison0.pr_root = newdp;
3851 		nrele++;
3852 	}
3853 	mtx_unlock(&prison0.pr_mtx);
3854 	sx_slock(&allprison_lock);
3855 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3856 		mtx_lock(&pr->pr_mtx);
3857 		if (pr->pr_root == olddp) {
3858 			vrefact(newdp);
3859 			pr->pr_root = newdp;
3860 			nrele++;
3861 		}
3862 		mtx_unlock(&pr->pr_mtx);
3863 	}
3864 	sx_sunlock(&allprison_lock);
3865 	while (nrele--)
3866 		vrele(olddp);
3867 }
3868 
3869 struct filedesc_to_leader *
3870 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3871 {
3872 	struct filedesc_to_leader *fdtol;
3873 
3874 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3875 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3876 	fdtol->fdl_refcount = 1;
3877 	fdtol->fdl_holdcount = 0;
3878 	fdtol->fdl_wakeup = 0;
3879 	fdtol->fdl_leader = leader;
3880 	if (old != NULL) {
3881 		FILEDESC_XLOCK(fdp);
3882 		fdtol->fdl_next = old->fdl_next;
3883 		fdtol->fdl_prev = old;
3884 		old->fdl_next = fdtol;
3885 		fdtol->fdl_next->fdl_prev = fdtol;
3886 		FILEDESC_XUNLOCK(fdp);
3887 	} else {
3888 		fdtol->fdl_next = fdtol;
3889 		fdtol->fdl_prev = fdtol;
3890 	}
3891 	return (fdtol);
3892 }
3893 
3894 static int
3895 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3896 {
3897 	NDSLOTTYPE *map;
3898 	struct filedesc *fdp;
3899 	int count, off, minoff;
3900 
3901 	if (*(int *)arg1 != 0)
3902 		return (EINVAL);
3903 
3904 	fdp = curproc->p_fd;
3905 	count = 0;
3906 	FILEDESC_SLOCK(fdp);
3907 	map = fdp->fd_map;
3908 	off = NDSLOT(fdp->fd_nfiles - 1);
3909 	for (minoff = NDSLOT(0); off >= minoff; --off)
3910 		count += bitcountl(map[off]);
3911 	FILEDESC_SUNLOCK(fdp);
3912 
3913 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3914 }
3915 
3916 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3917     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3918     "Number of open file descriptors");
3919 
3920 /*
3921  * Get file structures globally.
3922  */
3923 static int
3924 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3925 {
3926 	struct xfile xf;
3927 	struct filedesc *fdp;
3928 	struct file *fp;
3929 	struct proc *p;
3930 	int error, n, lastfile;
3931 
3932 	error = sysctl_wire_old_buffer(req, 0);
3933 	if (error != 0)
3934 		return (error);
3935 	if (req->oldptr == NULL) {
3936 		n = 0;
3937 		sx_slock(&allproc_lock);
3938 		FOREACH_PROC_IN_SYSTEM(p) {
3939 			PROC_LOCK(p);
3940 			if (p->p_state == PRS_NEW) {
3941 				PROC_UNLOCK(p);
3942 				continue;
3943 			}
3944 			fdp = fdhold(p);
3945 			PROC_UNLOCK(p);
3946 			if (fdp == NULL)
3947 				continue;
3948 			/* overestimates sparse tables. */
3949 			n += fdp->fd_nfiles;
3950 			fddrop(fdp);
3951 		}
3952 		sx_sunlock(&allproc_lock);
3953 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3954 	}
3955 	error = 0;
3956 	bzero(&xf, sizeof(xf));
3957 	xf.xf_size = sizeof(xf);
3958 	sx_slock(&allproc_lock);
3959 	FOREACH_PROC_IN_SYSTEM(p) {
3960 		PROC_LOCK(p);
3961 		if (p->p_state == PRS_NEW) {
3962 			PROC_UNLOCK(p);
3963 			continue;
3964 		}
3965 		if (p_cansee(req->td, p) != 0) {
3966 			PROC_UNLOCK(p);
3967 			continue;
3968 		}
3969 		xf.xf_pid = p->p_pid;
3970 		xf.xf_uid = p->p_ucred->cr_uid;
3971 		fdp = fdhold(p);
3972 		PROC_UNLOCK(p);
3973 		if (fdp == NULL)
3974 			continue;
3975 		FILEDESC_SLOCK(fdp);
3976 		lastfile = fdlastfile(fdp);
3977 		for (n = 0; fdp->fd_refcnt > 0 && n <= lastfile; ++n) {
3978 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3979 				continue;
3980 			xf.xf_fd = n;
3981 			xf.xf_file = (uintptr_t)fp;
3982 			xf.xf_data = (uintptr_t)fp->f_data;
3983 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3984 			xf.xf_type = (uintptr_t)fp->f_type;
3985 			xf.xf_count = refcount_load(&fp->f_count);
3986 			xf.xf_msgcount = 0;
3987 			xf.xf_offset = foffset_get(fp);
3988 			xf.xf_flag = fp->f_flag;
3989 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3990 			if (error)
3991 				break;
3992 		}
3993 		FILEDESC_SUNLOCK(fdp);
3994 		fddrop(fdp);
3995 		if (error)
3996 			break;
3997 	}
3998 	sx_sunlock(&allproc_lock);
3999 	return (error);
4000 }
4001 
4002 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4003     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4004 
4005 #ifdef KINFO_FILE_SIZE
4006 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4007 #endif
4008 
4009 static int
4010 xlate_fflags(int fflags)
4011 {
4012 	static const struct {
4013 		int	fflag;
4014 		int	kf_fflag;
4015 	} fflags_table[] = {
4016 		{ FAPPEND, KF_FLAG_APPEND },
4017 		{ FASYNC, KF_FLAG_ASYNC },
4018 		{ FFSYNC, KF_FLAG_FSYNC },
4019 		{ FHASLOCK, KF_FLAG_HASLOCK },
4020 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4021 		{ FREAD, KF_FLAG_READ },
4022 		{ FWRITE, KF_FLAG_WRITE },
4023 		{ O_CREAT, KF_FLAG_CREAT },
4024 		{ O_DIRECT, KF_FLAG_DIRECT },
4025 		{ O_EXCL, KF_FLAG_EXCL },
4026 		{ O_EXEC, KF_FLAG_EXEC },
4027 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4028 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4029 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4030 		{ O_TRUNC, KF_FLAG_TRUNC }
4031 	};
4032 	unsigned int i;
4033 	int kflags;
4034 
4035 	kflags = 0;
4036 	for (i = 0; i < nitems(fflags_table); i++)
4037 		if (fflags & fflags_table[i].fflag)
4038 			kflags |=  fflags_table[i].kf_fflag;
4039 	return (kflags);
4040 }
4041 
4042 /* Trim unused data from kf_path by truncating the structure size. */
4043 void
4044 pack_kinfo(struct kinfo_file *kif)
4045 {
4046 
4047 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4048 	    strlen(kif->kf_path) + 1;
4049 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4050 }
4051 
4052 static void
4053 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4054     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4055 {
4056 	int error;
4057 
4058 	bzero(kif, sizeof(*kif));
4059 
4060 	/* Set a default type to allow for empty fill_kinfo() methods. */
4061 	kif->kf_type = KF_TYPE_UNKNOWN;
4062 	kif->kf_flags = xlate_fflags(fp->f_flag);
4063 	if (rightsp != NULL)
4064 		kif->kf_cap_rights = *rightsp;
4065 	else
4066 		cap_rights_init_zero(&kif->kf_cap_rights);
4067 	kif->kf_fd = fd;
4068 	kif->kf_ref_count = refcount_load(&fp->f_count);
4069 	kif->kf_offset = foffset_get(fp);
4070 
4071 	/*
4072 	 * This may drop the filedesc lock, so the 'fp' cannot be
4073 	 * accessed after this call.
4074 	 */
4075 	error = fo_fill_kinfo(fp, kif, fdp);
4076 	if (error == 0)
4077 		kif->kf_status |= KF_ATTR_VALID;
4078 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4079 		pack_kinfo(kif);
4080 	else
4081 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4082 }
4083 
4084 static void
4085 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4086     struct kinfo_file *kif, int flags)
4087 {
4088 	int error;
4089 
4090 	bzero(kif, sizeof(*kif));
4091 
4092 	kif->kf_type = KF_TYPE_VNODE;
4093 	error = vn_fill_kinfo_vnode(vp, kif);
4094 	if (error == 0)
4095 		kif->kf_status |= KF_ATTR_VALID;
4096 	kif->kf_flags = xlate_fflags(fflags);
4097 	cap_rights_init_zero(&kif->kf_cap_rights);
4098 	kif->kf_fd = fd;
4099 	kif->kf_ref_count = -1;
4100 	kif->kf_offset = -1;
4101 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4102 		pack_kinfo(kif);
4103 	else
4104 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4105 	vrele(vp);
4106 }
4107 
4108 struct export_fd_buf {
4109 	struct filedesc		*fdp;
4110 	struct pwddesc	*pdp;
4111 	struct sbuf 		*sb;
4112 	ssize_t			remainder;
4113 	struct kinfo_file	kif;
4114 	int			flags;
4115 };
4116 
4117 static int
4118 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4119 {
4120 	struct kinfo_file *kif;
4121 
4122 	kif = &efbuf->kif;
4123 	if (efbuf->remainder != -1) {
4124 		if (efbuf->remainder < kif->kf_structsize) {
4125 			/* Terminate export. */
4126 			efbuf->remainder = 0;
4127 			return (0);
4128 		}
4129 		efbuf->remainder -= kif->kf_structsize;
4130 	}
4131 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4132 }
4133 
4134 static int
4135 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4136     struct export_fd_buf *efbuf)
4137 {
4138 	int error;
4139 
4140 	if (efbuf->remainder == 0)
4141 		return (0);
4142 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4143 	    efbuf->flags);
4144 	FILEDESC_SUNLOCK(efbuf->fdp);
4145 	error = export_kinfo_to_sb(efbuf);
4146 	FILEDESC_SLOCK(efbuf->fdp);
4147 	return (error);
4148 }
4149 
4150 static int
4151 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4152     struct export_fd_buf *efbuf)
4153 {
4154 	int error;
4155 
4156 	if (efbuf->remainder == 0)
4157 		return (0);
4158 	if (efbuf->pdp != NULL)
4159 		PWDDESC_XUNLOCK(efbuf->pdp);
4160 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4161 	error = export_kinfo_to_sb(efbuf);
4162 	if (efbuf->pdp != NULL)
4163 		PWDDESC_XLOCK(efbuf->pdp);
4164 	return (error);
4165 }
4166 
4167 /*
4168  * Store a process file descriptor information to sbuf.
4169  *
4170  * Takes a locked proc as argument, and returns with the proc unlocked.
4171  */
4172 int
4173 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4174     int flags)
4175 {
4176 	struct file *fp;
4177 	struct filedesc *fdp;
4178 	struct pwddesc *pdp;
4179 	struct export_fd_buf *efbuf;
4180 	struct vnode *cttyvp, *textvp, *tracevp;
4181 	struct pwd *pwd;
4182 	int error, i, lastfile;
4183 	cap_rights_t rights;
4184 
4185 	PROC_LOCK_ASSERT(p, MA_OWNED);
4186 
4187 	/* ktrace vnode */
4188 	tracevp = p->p_tracevp;
4189 	if (tracevp != NULL)
4190 		vrefact(tracevp);
4191 	/* text vnode */
4192 	textvp = p->p_textvp;
4193 	if (textvp != NULL)
4194 		vrefact(textvp);
4195 	/* Controlling tty. */
4196 	cttyvp = NULL;
4197 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4198 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4199 		if (cttyvp != NULL)
4200 			vrefact(cttyvp);
4201 	}
4202 	fdp = fdhold(p);
4203 	pdp = pdhold(p);
4204 	PROC_UNLOCK(p);
4205 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4206 	efbuf->fdp = NULL;
4207 	efbuf->pdp = NULL;
4208 	efbuf->sb = sb;
4209 	efbuf->remainder = maxlen;
4210 	efbuf->flags = flags;
4211 	if (tracevp != NULL)
4212 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4213 		    efbuf);
4214 	if (textvp != NULL)
4215 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4216 	if (cttyvp != NULL)
4217 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4218 		    efbuf);
4219 	error = 0;
4220 	if (pdp == NULL || fdp == NULL)
4221 		goto fail;
4222 	efbuf->fdp = fdp;
4223 	efbuf->pdp = pdp;
4224 	PWDDESC_XLOCK(pdp);
4225 	pwd = pwd_hold_pwddesc(pdp);
4226 	if (pwd != NULL) {
4227 		/* working directory */
4228 		if (pwd->pwd_cdir != NULL) {
4229 			vrefact(pwd->pwd_cdir);
4230 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4231 		}
4232 		/* root directory */
4233 		if (pwd->pwd_rdir != NULL) {
4234 			vrefact(pwd->pwd_rdir);
4235 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4236 		}
4237 		/* jail directory */
4238 		if (pwd->pwd_jdir != NULL) {
4239 			vrefact(pwd->pwd_jdir);
4240 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4241 		}
4242 	}
4243 	PWDDESC_XUNLOCK(pdp);
4244 	if (pwd != NULL)
4245 		pwd_drop(pwd);
4246 	FILEDESC_SLOCK(fdp);
4247 	lastfile = fdlastfile(fdp);
4248 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4249 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4250 			continue;
4251 #ifdef CAPABILITIES
4252 		rights = *cap_rights(fdp, i);
4253 #else /* !CAPABILITIES */
4254 		rights = cap_no_rights;
4255 #endif
4256 		/*
4257 		 * Create sysctl entry.  It is OK to drop the filedesc
4258 		 * lock inside of export_file_to_sb() as we will
4259 		 * re-validate and re-evaluate its properties when the
4260 		 * loop continues.
4261 		 */
4262 		error = export_file_to_sb(fp, i, &rights, efbuf);
4263 		if (error != 0 || efbuf->remainder == 0)
4264 			break;
4265 	}
4266 	FILEDESC_SUNLOCK(fdp);
4267 fail:
4268 	if (fdp != NULL)
4269 		fddrop(fdp);
4270 	if (pdp != NULL)
4271 		pddrop(pdp);
4272 	free(efbuf, M_TEMP);
4273 	return (error);
4274 }
4275 
4276 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4277 
4278 /*
4279  * Get per-process file descriptors for use by procstat(1), et al.
4280  */
4281 static int
4282 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4283 {
4284 	struct sbuf sb;
4285 	struct proc *p;
4286 	ssize_t maxlen;
4287 	int error, error2, *name;
4288 
4289 	name = (int *)arg1;
4290 
4291 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4292 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4293 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4294 	if (error != 0) {
4295 		sbuf_delete(&sb);
4296 		return (error);
4297 	}
4298 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4299 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4300 	    KERN_FILEDESC_PACK_KINFO);
4301 	error2 = sbuf_finish(&sb);
4302 	sbuf_delete(&sb);
4303 	return (error != 0 ? error : error2);
4304 }
4305 
4306 #ifdef COMPAT_FREEBSD7
4307 #ifdef KINFO_OFILE_SIZE
4308 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4309 #endif
4310 
4311 static void
4312 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4313 {
4314 
4315 	okif->kf_structsize = sizeof(*okif);
4316 	okif->kf_type = kif->kf_type;
4317 	okif->kf_fd = kif->kf_fd;
4318 	okif->kf_ref_count = kif->kf_ref_count;
4319 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4320 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4321 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4322 	okif->kf_offset = kif->kf_offset;
4323 	if (kif->kf_type == KF_TYPE_VNODE)
4324 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4325 	else
4326 		okif->kf_vnode_type = KF_VTYPE_VNON;
4327 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4328 	if (kif->kf_type == KF_TYPE_SOCKET) {
4329 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4330 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4331 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4332 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4333 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4334 	} else {
4335 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4336 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4337 	}
4338 }
4339 
4340 static int
4341 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4342     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4343 {
4344 	int error;
4345 
4346 	vrefact(vp);
4347 	PWDDESC_XUNLOCK(pdp);
4348 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4349 	kinfo_to_okinfo(kif, okif);
4350 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4351 	PWDDESC_XLOCK(pdp);
4352 	return (error);
4353 }
4354 
4355 /*
4356  * Get per-process file descriptors for use by procstat(1), et al.
4357  */
4358 static int
4359 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4360 {
4361 	struct kinfo_ofile *okif;
4362 	struct kinfo_file *kif;
4363 	struct filedesc *fdp;
4364 	struct pwddesc *pdp;
4365 	struct pwd *pwd;
4366 	int error, i, lastfile, *name;
4367 	struct file *fp;
4368 	struct proc *p;
4369 
4370 	name = (int *)arg1;
4371 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4372 	if (error != 0)
4373 		return (error);
4374 	fdp = fdhold(p);
4375 	if (fdp != NULL)
4376 		pdp = pdhold(p);
4377 	PROC_UNLOCK(p);
4378 	if (fdp == NULL || pdp == NULL) {
4379 		if (fdp != NULL)
4380 			fddrop(fdp);
4381 		return (ENOENT);
4382 	}
4383 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4384 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4385 	PWDDESC_XLOCK(pdp);
4386 	pwd = pwd_hold_pwddesc(pdp);
4387 	if (pwd != NULL) {
4388 		if (pwd->pwd_cdir != NULL)
4389 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4390 			    okif, pdp, req);
4391 		if (pwd->pwd_rdir != NULL)
4392 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4393 			    okif, pdp, req);
4394 		if (pwd->pwd_jdir != NULL)
4395 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4396 			    okif, pdp, req);
4397 	}
4398 	PWDDESC_XUNLOCK(pdp);
4399 	if (pwd != NULL)
4400 		pwd_drop(pwd);
4401 	FILEDESC_SLOCK(fdp);
4402 	lastfile = fdlastfile(fdp);
4403 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4404 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4405 			continue;
4406 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4407 		    KERN_FILEDESC_PACK_KINFO);
4408 		FILEDESC_SUNLOCK(fdp);
4409 		kinfo_to_okinfo(kif, okif);
4410 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4411 		FILEDESC_SLOCK(fdp);
4412 		if (error)
4413 			break;
4414 	}
4415 	FILEDESC_SUNLOCK(fdp);
4416 	fddrop(fdp);
4417 	pddrop(pdp);
4418 	free(kif, M_TEMP);
4419 	free(okif, M_TEMP);
4420 	return (0);
4421 }
4422 
4423 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4424     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4425     "Process ofiledesc entries");
4426 #endif	/* COMPAT_FREEBSD7 */
4427 
4428 int
4429 vntype_to_kinfo(int vtype)
4430 {
4431 	struct {
4432 		int	vtype;
4433 		int	kf_vtype;
4434 	} vtypes_table[] = {
4435 		{ VBAD, KF_VTYPE_VBAD },
4436 		{ VBLK, KF_VTYPE_VBLK },
4437 		{ VCHR, KF_VTYPE_VCHR },
4438 		{ VDIR, KF_VTYPE_VDIR },
4439 		{ VFIFO, KF_VTYPE_VFIFO },
4440 		{ VLNK, KF_VTYPE_VLNK },
4441 		{ VNON, KF_VTYPE_VNON },
4442 		{ VREG, KF_VTYPE_VREG },
4443 		{ VSOCK, KF_VTYPE_VSOCK }
4444 	};
4445 	unsigned int i;
4446 
4447 	/*
4448 	 * Perform vtype translation.
4449 	 */
4450 	for (i = 0; i < nitems(vtypes_table); i++)
4451 		if (vtypes_table[i].vtype == vtype)
4452 			return (vtypes_table[i].kf_vtype);
4453 
4454 	return (KF_VTYPE_UNKNOWN);
4455 }
4456 
4457 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4458     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4459     "Process filedesc entries");
4460 
4461 /*
4462  * Store a process current working directory information to sbuf.
4463  *
4464  * Takes a locked proc as argument, and returns with the proc unlocked.
4465  */
4466 int
4467 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4468 {
4469 	struct pwddesc *pdp;
4470 	struct pwd *pwd;
4471 	struct export_fd_buf *efbuf;
4472 	struct vnode *cdir;
4473 	int error;
4474 
4475 	PROC_LOCK_ASSERT(p, MA_OWNED);
4476 
4477 	pdp = pdhold(p);
4478 	PROC_UNLOCK(p);
4479 	if (pdp == NULL)
4480 		return (EINVAL);
4481 
4482 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4483 	efbuf->pdp = pdp;
4484 	efbuf->sb = sb;
4485 	efbuf->remainder = maxlen;
4486 
4487 	PWDDESC_XLOCK(pdp);
4488 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4489 	cdir = pwd->pwd_cdir;
4490 	if (cdir == NULL) {
4491 		error = EINVAL;
4492 	} else {
4493 		vrefact(cdir);
4494 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4495 	}
4496 	PWDDESC_XUNLOCK(pdp);
4497 	pddrop(pdp);
4498 	free(efbuf, M_TEMP);
4499 	return (error);
4500 }
4501 
4502 /*
4503  * Get per-process current working directory.
4504  */
4505 static int
4506 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4507 {
4508 	struct sbuf sb;
4509 	struct proc *p;
4510 	ssize_t maxlen;
4511 	int error, error2, *name;
4512 
4513 	name = (int *)arg1;
4514 
4515 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4516 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4517 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4518 	if (error != 0) {
4519 		sbuf_delete(&sb);
4520 		return (error);
4521 	}
4522 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4523 	error = kern_proc_cwd_out(p, &sb, maxlen);
4524 	error2 = sbuf_finish(&sb);
4525 	sbuf_delete(&sb);
4526 	return (error != 0 ? error : error2);
4527 }
4528 
4529 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4530     sysctl_kern_proc_cwd, "Process current working directory");
4531 
4532 #ifdef DDB
4533 /*
4534  * For the purposes of debugging, generate a human-readable string for the
4535  * file type.
4536  */
4537 static const char *
4538 file_type_to_name(short type)
4539 {
4540 
4541 	switch (type) {
4542 	case 0:
4543 		return ("zero");
4544 	case DTYPE_VNODE:
4545 		return ("vnode");
4546 	case DTYPE_SOCKET:
4547 		return ("socket");
4548 	case DTYPE_PIPE:
4549 		return ("pipe");
4550 	case DTYPE_FIFO:
4551 		return ("fifo");
4552 	case DTYPE_KQUEUE:
4553 		return ("kqueue");
4554 	case DTYPE_CRYPTO:
4555 		return ("crypto");
4556 	case DTYPE_MQUEUE:
4557 		return ("mqueue");
4558 	case DTYPE_SHM:
4559 		return ("shm");
4560 	case DTYPE_SEM:
4561 		return ("ksem");
4562 	case DTYPE_PTS:
4563 		return ("pts");
4564 	case DTYPE_DEV:
4565 		return ("dev");
4566 	case DTYPE_PROCDESC:
4567 		return ("proc");
4568 	case DTYPE_LINUXEFD:
4569 		return ("levent");
4570 	case DTYPE_LINUXTFD:
4571 		return ("ltimer");
4572 	default:
4573 		return ("unkn");
4574 	}
4575 }
4576 
4577 /*
4578  * For the purposes of debugging, identify a process (if any, perhaps one of
4579  * many) that references the passed file in its file descriptor array. Return
4580  * NULL if none.
4581  */
4582 static struct proc *
4583 file_to_first_proc(struct file *fp)
4584 {
4585 	struct filedesc *fdp;
4586 	struct proc *p;
4587 	int n;
4588 
4589 	FOREACH_PROC_IN_SYSTEM(p) {
4590 		if (p->p_state == PRS_NEW)
4591 			continue;
4592 		fdp = p->p_fd;
4593 		if (fdp == NULL)
4594 			continue;
4595 		for (n = 0; n < fdp->fd_nfiles; n++) {
4596 			if (fp == fdp->fd_ofiles[n].fde_file)
4597 				return (p);
4598 		}
4599 	}
4600 	return (NULL);
4601 }
4602 
4603 static void
4604 db_print_file(struct file *fp, int header)
4605 {
4606 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4607 	struct proc *p;
4608 
4609 	if (header)
4610 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4611 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4612 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4613 		    "FCmd");
4614 	p = file_to_first_proc(fp);
4615 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4616 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4617 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4618 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4619 
4620 #undef XPTRWIDTH
4621 }
4622 
4623 DB_SHOW_COMMAND(file, db_show_file)
4624 {
4625 	struct file *fp;
4626 
4627 	if (!have_addr) {
4628 		db_printf("usage: show file <addr>\n");
4629 		return;
4630 	}
4631 	fp = (struct file *)addr;
4632 	db_print_file(fp, 1);
4633 }
4634 
4635 DB_SHOW_COMMAND(files, db_show_files)
4636 {
4637 	struct filedesc *fdp;
4638 	struct file *fp;
4639 	struct proc *p;
4640 	int header;
4641 	int n;
4642 
4643 	header = 1;
4644 	FOREACH_PROC_IN_SYSTEM(p) {
4645 		if (p->p_state == PRS_NEW)
4646 			continue;
4647 		if ((fdp = p->p_fd) == NULL)
4648 			continue;
4649 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4650 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4651 				continue;
4652 			db_print_file(fp, header);
4653 			header = 0;
4654 		}
4655 	}
4656 }
4657 #endif
4658 
4659 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4660     &maxfilesperproc, 0, "Maximum files allowed open per process");
4661 
4662 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4663     &maxfiles, 0, "Maximum number of files");
4664 
4665 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4666     &openfiles, 0, "System-wide number of open files");
4667 
4668 /* ARGSUSED*/
4669 static void
4670 filelistinit(void *dummy)
4671 {
4672 
4673 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4674 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4675 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4676 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4677 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4678 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4679 	/*
4680 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4681 	 * the vnode zone.
4682 	 */
4683 	vfs_smr = uma_zone_get_smr(pwd_zone);
4684 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4685 }
4686 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4687 
4688 /*-------------------------------------------------------------------*/
4689 
4690 static int
4691 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4692     int flags, struct thread *td)
4693 {
4694 
4695 	return (EBADF);
4696 }
4697 
4698 static int
4699 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4700     struct thread *td)
4701 {
4702 
4703 	return (EINVAL);
4704 }
4705 
4706 static int
4707 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4708     struct thread *td)
4709 {
4710 
4711 	return (EBADF);
4712 }
4713 
4714 static int
4715 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4716     struct thread *td)
4717 {
4718 
4719 	return (0);
4720 }
4721 
4722 static int
4723 badfo_kqfilter(struct file *fp, struct knote *kn)
4724 {
4725 
4726 	return (EBADF);
4727 }
4728 
4729 static int
4730 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4731     struct thread *td)
4732 {
4733 
4734 	return (EBADF);
4735 }
4736 
4737 static int
4738 badfo_close(struct file *fp, struct thread *td)
4739 {
4740 
4741 	return (0);
4742 }
4743 
4744 static int
4745 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4746     struct thread *td)
4747 {
4748 
4749 	return (EBADF);
4750 }
4751 
4752 static int
4753 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4754     struct thread *td)
4755 {
4756 
4757 	return (EBADF);
4758 }
4759 
4760 static int
4761 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4762     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4763     struct thread *td)
4764 {
4765 
4766 	return (EBADF);
4767 }
4768 
4769 static int
4770 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4771 {
4772 
4773 	return (0);
4774 }
4775 
4776 struct fileops badfileops = {
4777 	.fo_read = badfo_readwrite,
4778 	.fo_write = badfo_readwrite,
4779 	.fo_truncate = badfo_truncate,
4780 	.fo_ioctl = badfo_ioctl,
4781 	.fo_poll = badfo_poll,
4782 	.fo_kqfilter = badfo_kqfilter,
4783 	.fo_stat = badfo_stat,
4784 	.fo_close = badfo_close,
4785 	.fo_chmod = badfo_chmod,
4786 	.fo_chown = badfo_chown,
4787 	.fo_sendfile = badfo_sendfile,
4788 	.fo_fill_kinfo = badfo_fill_kinfo,
4789 };
4790 
4791 int
4792 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4793     int flags, struct thread *td)
4794 {
4795 
4796 	return (EOPNOTSUPP);
4797 }
4798 
4799 int
4800 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4801     struct thread *td)
4802 {
4803 
4804 	return (EINVAL);
4805 }
4806 
4807 int
4808 invfo_ioctl(struct file *fp, u_long com, void *data,
4809     struct ucred *active_cred, struct thread *td)
4810 {
4811 
4812 	return (ENOTTY);
4813 }
4814 
4815 int
4816 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4817     struct thread *td)
4818 {
4819 
4820 	return (poll_no_poll(events));
4821 }
4822 
4823 int
4824 invfo_kqfilter(struct file *fp, struct knote *kn)
4825 {
4826 
4827 	return (EINVAL);
4828 }
4829 
4830 int
4831 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4832     struct thread *td)
4833 {
4834 
4835 	return (EINVAL);
4836 }
4837 
4838 int
4839 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4840     struct thread *td)
4841 {
4842 
4843 	return (EINVAL);
4844 }
4845 
4846 int
4847 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4848     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4849     struct thread *td)
4850 {
4851 
4852 	return (EINVAL);
4853 }
4854 
4855 /*-------------------------------------------------------------------*/
4856 
4857 /*
4858  * File Descriptor pseudo-device driver (/dev/fd/).
4859  *
4860  * Opening minor device N dup()s the file (if any) connected to file
4861  * descriptor N belonging to the calling process.  Note that this driver
4862  * consists of only the ``open()'' routine, because all subsequent
4863  * references to this file will be direct to the other driver.
4864  *
4865  * XXX: we could give this one a cloning event handler if necessary.
4866  */
4867 
4868 /* ARGSUSED */
4869 static int
4870 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4871 {
4872 
4873 	/*
4874 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4875 	 * the file descriptor being sought for duplication. The error
4876 	 * return ensures that the vnode for this device will be released
4877 	 * by vn_open. Open will detect this special error and take the
4878 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4879 	 * will simply report the error.
4880 	 */
4881 	td->td_dupfd = dev2unit(dev);
4882 	return (ENODEV);
4883 }
4884 
4885 static struct cdevsw fildesc_cdevsw = {
4886 	.d_version =	D_VERSION,
4887 	.d_open =	fdopen,
4888 	.d_name =	"FD",
4889 };
4890 
4891 static void
4892 fildesc_drvinit(void *unused)
4893 {
4894 	struct cdev *dev;
4895 
4896 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4897 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4898 	make_dev_alias(dev, "stdin");
4899 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4900 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4901 	make_dev_alias(dev, "stdout");
4902 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4903 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4904 	make_dev_alias(dev, "stderr");
4905 }
4906 
4907 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4908