xref: /freebsd/sys/kern/kern_descrip.c (revision 9097e3cbcac455eb0dedd097d8d5548c72568d0a)
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, bool holdleaders, bool audit);
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 	FILEDESC_XLOCK_ASSERT(fdp);
311 	fde = &fdp->fd_ofiles[fd];
312 #ifdef CAPABILITIES
313 	seqc_write_begin(&fde->fde_seqc);
314 #endif
315 	fde->fde_file = NULL;
316 #ifdef CAPABILITIES
317 	seqc_write_end(&fde->fde_seqc);
318 #endif
319 	fdefree_last(fde);
320 	fdunused(fdp, fd);
321 }
322 
323 /*
324  * System calls on descriptors.
325  */
326 #ifndef _SYS_SYSPROTO_H_
327 struct getdtablesize_args {
328 	int	dummy;
329 };
330 #endif
331 /* ARGSUSED */
332 int
333 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 {
335 #ifdef	RACCT
336 	uint64_t lim;
337 #endif
338 
339 	td->td_retval[0] = getmaxfd(td);
340 #ifdef	RACCT
341 	PROC_LOCK(td->td_proc);
342 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
343 	PROC_UNLOCK(td->td_proc);
344 	if (lim < td->td_retval[0])
345 		td->td_retval[0] = lim;
346 #endif
347 	return (0);
348 }
349 
350 /*
351  * Duplicate a file descriptor to a particular value.
352  *
353  * Note: keep in mind that a potential race condition exists when closing
354  * descriptors from a shared descriptor table (via rfork).
355  */
356 #ifndef _SYS_SYSPROTO_H_
357 struct dup2_args {
358 	u_int	from;
359 	u_int	to;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup2(struct thread *td, struct dup2_args *uap)
365 {
366 
367 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
368 }
369 
370 /*
371  * Duplicate a file descriptor.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct dup_args {
375 	u_int	fd;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup(struct thread *td, struct dup_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
384 }
385 
386 /*
387  * The file control system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct fcntl_args {
391 	int	fd;
392 	int	cmd;
393 	long	arg;
394 };
395 #endif
396 /* ARGSUSED */
397 int
398 sys_fcntl(struct thread *td, struct fcntl_args *uap)
399 {
400 
401 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 }
403 
404 int
405 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
406 {
407 	struct flock fl;
408 	struct __oflock ofl;
409 	intptr_t arg1;
410 	int error, newcmd;
411 
412 	error = 0;
413 	newcmd = cmd;
414 	switch (cmd) {
415 	case F_OGETLK:
416 	case F_OSETLK:
417 	case F_OSETLKW:
418 		/*
419 		 * Convert old flock structure to new.
420 		 */
421 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
422 		fl.l_start = ofl.l_start;
423 		fl.l_len = ofl.l_len;
424 		fl.l_pid = ofl.l_pid;
425 		fl.l_type = ofl.l_type;
426 		fl.l_whence = ofl.l_whence;
427 		fl.l_sysid = 0;
428 
429 		switch (cmd) {
430 		case F_OGETLK:
431 			newcmd = F_GETLK;
432 			break;
433 		case F_OSETLK:
434 			newcmd = F_SETLK;
435 			break;
436 		case F_OSETLKW:
437 			newcmd = F_SETLKW;
438 			break;
439 		}
440 		arg1 = (intptr_t)&fl;
441 		break;
442 	case F_GETLK:
443 	case F_SETLK:
444 	case F_SETLKW:
445 	case F_SETLK_REMOTE:
446 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	default:
450 		arg1 = arg;
451 		break;
452 	}
453 	if (error)
454 		return (error);
455 	error = kern_fcntl(td, fd, newcmd, arg1);
456 	if (error)
457 		return (error);
458 	if (cmd == F_OGETLK) {
459 		ofl.l_start = fl.l_start;
460 		ofl.l_len = fl.l_len;
461 		ofl.l_pid = fl.l_pid;
462 		ofl.l_type = fl.l_type;
463 		ofl.l_whence = fl.l_whence;
464 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
465 	} else if (cmd == F_GETLK) {
466 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
467 	}
468 	return (error);
469 }
470 
471 int
472 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 {
474 	struct filedesc *fdp;
475 	struct flock *flp;
476 	struct file *fp, *fp2;
477 	struct filedescent *fde;
478 	struct proc *p;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	int error, flg, seals, tmp;
482 	uint64_t bsize;
483 	off_t foffset;
484 
485 	error = 0;
486 	flg = F_POSIX;
487 	p = td->td_proc;
488 	fdp = p->p_fd;
489 
490 	AUDIT_ARG_FD(cmd);
491 	AUDIT_ARG_CMD(cmd);
492 	switch (cmd) {
493 	case F_DUPFD:
494 		tmp = arg;
495 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
496 		break;
497 
498 	case F_DUPFD_CLOEXEC:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
501 		break;
502 
503 	case F_DUP2FD:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD_CLOEXEC:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
511 		break;
512 
513 	case F_GETFD:
514 		error = EBADF;
515 		FILEDESC_SLOCK(fdp);
516 		fde = fdeget_locked(fdp, fd);
517 		if (fde != NULL) {
518 			td->td_retval[0] =
519 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
520 			error = 0;
521 		}
522 		FILEDESC_SUNLOCK(fdp);
523 		break;
524 
525 	case F_SETFD:
526 		error = EBADF;
527 		FILEDESC_XLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
531 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
532 			error = 0;
533 		}
534 		FILEDESC_XUNLOCK(fdp);
535 		break;
536 
537 	case F_GETFL:
538 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
539 		if (error != 0)
540 			break;
541 		td->td_retval[0] = OFLAGS(fp->f_flag);
542 		fdrop(fp, td);
543 		break;
544 
545 	case F_SETFL:
546 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
547 		if (error != 0)
548 			break;
549 		do {
550 			tmp = flg = fp->f_flag;
551 			tmp &= ~FCNTLFLAGS;
552 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
553 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
554 		tmp = fp->f_flag & FNONBLOCK;
555 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
556 		if (error != 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		tmp = fp->f_flag & FASYNC;
561 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
562 		if (error == 0) {
563 			fdrop(fp, td);
564 			break;
565 		}
566 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 		tmp = 0;
568 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 		fdrop(fp, td);
570 		break;
571 
572 	case F_GETOWN:
573 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
574 		if (error != 0)
575 			break;
576 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 		if (error == 0)
578 			td->td_retval[0] = tmp;
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_SETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		tmp = arg;
587 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
588 		fdrop(fp, td);
589 		break;
590 
591 	case F_SETLK_REMOTE:
592 		error = priv_check(td, PRIV_NFS_LOCKD);
593 		if (error != 0)
594 			return (error);
595 		flg = F_REMOTE;
596 		goto do_setlk;
597 
598 	case F_SETLKW:
599 		flg |= F_WAIT;
600 		/* FALLTHROUGH F_SETLK */
601 
602 	case F_SETLK:
603 	do_setlk:
604 		flp = (struct flock *)arg;
605 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
606 			error = EINVAL;
607 			break;
608 		}
609 
610 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
611 		if (error != 0)
612 			break;
613 		if (fp->f_type != DTYPE_VNODE) {
614 			error = EBADF;
615 			fdrop(fp, td);
616 			break;
617 		}
618 
619 		if (flp->l_whence == SEEK_CUR) {
620 			foffset = foffset_get(fp);
621 			if (foffset < 0 ||
622 			    (flp->l_start > 0 &&
623 			     foffset > OFF_MAX - flp->l_start)) {
624 				error = EOVERFLOW;
625 				fdrop(fp, td);
626 				break;
627 			}
628 			flp->l_start += foffset;
629 		}
630 
631 		vp = fp->f_vnode;
632 		switch (flp->l_type) {
633 		case F_RDLCK:
634 			if ((fp->f_flag & FREAD) == 0) {
635 				error = EBADF;
636 				break;
637 			}
638 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
639 				PROC_LOCK(p->p_leader);
640 				p->p_leader->p_flag |= P_ADVLOCK;
641 				PROC_UNLOCK(p->p_leader);
642 			}
643 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
644 			    flp, flg);
645 			break;
646 		case F_WRLCK:
647 			if ((fp->f_flag & FWRITE) == 0) {
648 				error = EBADF;
649 				break;
650 			}
651 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
652 				PROC_LOCK(p->p_leader);
653 				p->p_leader->p_flag |= P_ADVLOCK;
654 				PROC_UNLOCK(p->p_leader);
655 			}
656 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 			    flp, flg);
658 			break;
659 		case F_UNLCK:
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCKSYS:
664 			if (flg != F_REMOTE) {
665 				error = EINVAL;
666 				break;
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
669 			    F_UNLCKSYS, flp, flg);
670 			break;
671 		default:
672 			error = EINVAL;
673 			break;
674 		}
675 		if (error != 0 || flp->l_type == F_UNLCK ||
676 		    flp->l_type == F_UNLCKSYS) {
677 			fdrop(fp, td);
678 			break;
679 		}
680 
681 		/*
682 		 * Check for a race with close.
683 		 *
684 		 * The vnode is now advisory locked (or unlocked, but this case
685 		 * is not really important) as the caller requested.
686 		 * We had to drop the filedesc lock, so we need to recheck if
687 		 * the descriptor is still valid, because if it was closed
688 		 * in the meantime we need to remove advisory lock from the
689 		 * vnode - close on any descriptor leading to an advisory
690 		 * locked vnode, removes that lock.
691 		 * We will return 0 on purpose in that case, as the result of
692 		 * successful advisory lock might have been externally visible
693 		 * already. This is fine - effectively we pretend to the caller
694 		 * that the closing thread was a bit slower and that the
695 		 * advisory lock succeeded before the close.
696 		 */
697 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
698 		if (error != 0) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 		if (fp != fp2) {
703 			flp->l_whence = SEEK_SET;
704 			flp->l_start = 0;
705 			flp->l_len = 0;
706 			flp->l_type = F_UNLCK;
707 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
708 			    F_UNLCK, flp, F_POSIX);
709 		}
710 		fdrop(fp, td);
711 		fdrop(fp2, td);
712 		break;
713 
714 	case F_GETLK:
715 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
716 		if (error != 0)
717 			break;
718 		if (fp->f_type != DTYPE_VNODE) {
719 			error = EBADF;
720 			fdrop(fp, td);
721 			break;
722 		}
723 		flp = (struct flock *)arg;
724 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
725 		    flp->l_type != F_UNLCK) {
726 			error = EINVAL;
727 			fdrop(fp, td);
728 			break;
729 		}
730 		if (flp->l_whence == SEEK_CUR) {
731 			foffset = foffset_get(fp);
732 			if ((flp->l_start > 0 &&
733 			    foffset > OFF_MAX - flp->l_start) ||
734 			    (flp->l_start < 0 &&
735 			    foffset < OFF_MIN - flp->l_start)) {
736 				error = EOVERFLOW;
737 				fdrop(fp, td);
738 				break;
739 			}
740 			flp->l_start += foffset;
741 		}
742 		vp = fp->f_vnode;
743 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
744 		    F_POSIX);
745 		fdrop(fp, td);
746 		break;
747 
748 	case F_ADD_SEALS:
749 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
750 		if (error != 0)
751 			break;
752 		error = fo_add_seals(fp, arg);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_GET_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		if (fo_get_seals(fp, &seals) == 0)
761 			td->td_retval[0] = seals;
762 		else
763 			error = EINVAL;
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_RDAHEAD:
768 		arg = arg ? 128 * 1024: 0;
769 		/* FALLTHROUGH */
770 	case F_READAHEAD:
771 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		if (fp->f_type != DTYPE_VNODE) {
775 			fdrop(fp, td);
776 			error = EBADF;
777 			break;
778 		}
779 		vp = fp->f_vnode;
780 		if (vp->v_type != VREG) {
781 			fdrop(fp, td);
782 			error = ENOTTY;
783 			break;
784 		}
785 
786 		/*
787 		 * Exclusive lock synchronizes against f_seqcount reads and
788 		 * writes in sequential_heuristic().
789 		 */
790 		error = vn_lock(vp, LK_EXCLUSIVE);
791 		if (error != 0) {
792 			fdrop(fp, td);
793 			break;
794 		}
795 		if (arg >= 0) {
796 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
797 			arg = MIN(arg, INT_MAX - bsize + 1);
798 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
799 			    (arg + bsize - 1) / bsize);
800 			atomic_set_int(&fp->f_flag, FRDAHEAD);
801 		} else {
802 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 		}
804 		VOP_UNLOCK(vp);
805 		fdrop(fp, td);
806 		break;
807 
808 	case F_ISUNIONSTACK:
809 		/*
810 		 * Check if the vnode is part of a union stack (either the
811 		 * "union" flag from mount(2) or unionfs).
812 		 *
813 		 * Prior to introduction of this op libc's readdir would call
814 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
815 		 * data just to check fs name and a mount flag.
816 		 *
817 		 * Fixing the code to handle everything in the kernel instead
818 		 * is a non-trivial endeavor and has low priority, thus this
819 		 * horrible kludge facilitates the current behavior in a much
820 		 * cheaper manner until someone(tm) sorts this out.
821 		 */
822 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
823 		if (error != 0)
824 			break;
825 		if (fp->f_type != DTYPE_VNODE) {
826 			fdrop(fp, td);
827 			error = EBADF;
828 			break;
829 		}
830 		vp = fp->f_vnode;
831 		/*
832 		 * Since we don't prevent dooming the vnode even non-null mp
833 		 * found can become immediately stale. This is tolerable since
834 		 * mount points are type-stable (providing safe memory access)
835 		 * and any vfs op on this vnode going forward will return an
836 		 * error (meaning return value in this case is meaningless).
837 		 */
838 		mp = atomic_load_ptr(&vp->v_mount);
839 		if (__predict_false(mp == NULL)) {
840 			fdrop(fp, td);
841 			error = EBADF;
842 			break;
843 		}
844 		td->td_retval[0] = 0;
845 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
846 		    mp->mnt_flag & MNT_UNION)
847 			td->td_retval[0] = 1;
848 		fdrop(fp, td);
849 		break;
850 
851 	default:
852 		error = EINVAL;
853 		break;
854 	}
855 	return (error);
856 }
857 
858 static int
859 getmaxfd(struct thread *td)
860 {
861 
862 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
863 }
864 
865 /*
866  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
867  */
868 int
869 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 {
871 	struct filedesc *fdp;
872 	struct filedescent *oldfde, *newfde;
873 	struct proc *p;
874 	struct file *delfp, *oldfp;
875 	u_long *oioctls, *nioctls;
876 	int error, maxfd;
877 
878 	p = td->td_proc;
879 	fdp = p->p_fd;
880 	oioctls = NULL;
881 
882 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
883 	MPASS(mode < FDDUP_LASTMODE);
884 
885 	AUDIT_ARG_FD(old);
886 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
887 
888 	/*
889 	 * Verify we have a valid descriptor to dup from and possibly to
890 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
891 	 * return EINVAL when the new descriptor is out of bounds.
892 	 */
893 	if (old < 0)
894 		return (EBADF);
895 	if (new < 0)
896 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
897 	maxfd = getmaxfd(td);
898 	if (new >= maxfd)
899 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
900 
901 	error = EBADF;
902 	FILEDESC_XLOCK(fdp);
903 	if (fget_locked(fdp, old) == NULL)
904 		goto unlock;
905 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
906 		td->td_retval[0] = new;
907 		if (flags & FDDUP_FLAG_CLOEXEC)
908 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
909 		error = 0;
910 		goto unlock;
911 	}
912 
913 	oldfde = &fdp->fd_ofiles[old];
914 	oldfp = oldfde->fde_file;
915 	if (!fhold(oldfp))
916 		goto unlock;
917 
918 	/*
919 	 * If the caller specified a file descriptor, make sure the file
920 	 * table is large enough to hold it, and grab it.  Otherwise, just
921 	 * allocate a new descriptor the usual way.
922 	 */
923 	switch (mode) {
924 	case FDDUP_NORMAL:
925 	case FDDUP_FCNTL:
926 		if ((error = fdalloc(td, new, &new)) != 0) {
927 			fdrop(oldfp, td);
928 			goto unlock;
929 		}
930 		break;
931 	case FDDUP_MUSTREPLACE:
932 		/* Target file descriptor must exist. */
933 		if (fget_locked(fdp, new) == NULL) {
934 			fdrop(oldfp, td);
935 			goto unlock;
936 		}
937 		break;
938 	case FDDUP_FIXED:
939 		if (new >= fdp->fd_nfiles) {
940 			/*
941 			 * The resource limits are here instead of e.g.
942 			 * fdalloc(), because the file descriptor table may be
943 			 * shared between processes, so we can't really use
944 			 * racct_add()/racct_sub().  Instead of counting the
945 			 * number of actually allocated descriptors, just put
946 			 * the limit on the size of the file descriptor table.
947 			 */
948 #ifdef RACCT
949 			if (RACCT_ENABLED()) {
950 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
951 				if (error != 0) {
952 					error = EMFILE;
953 					fdrop(oldfp, td);
954 					goto unlock;
955 				}
956 			}
957 #endif
958 			fdgrowtable_exp(fdp, new + 1);
959 		}
960 		if (!fdisused(fdp, new))
961 			fdused(fdp, new);
962 		break;
963 	default:
964 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
965 	}
966 
967 	KASSERT(old != new, ("new fd is same as old"));
968 
969 	/* Refetch oldfde because the table may have grown and old one freed. */
970 	oldfde = &fdp->fd_ofiles[old];
971 	KASSERT(oldfp == oldfde->fde_file,
972 	    ("fdt_ofiles shift from growth observed at fd %d",
973 	    old));
974 
975 	newfde = &fdp->fd_ofiles[new];
976 	delfp = newfde->fde_file;
977 
978 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
979 
980 	/*
981 	 * Duplicate the source descriptor.
982 	 */
983 #ifdef CAPABILITIES
984 	seqc_write_begin(&newfde->fde_seqc);
985 #endif
986 	oioctls = filecaps_free_prep(&newfde->fde_caps);
987 	memcpy(newfde, oldfde, fde_change_size);
988 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
989 	    nioctls);
990 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
991 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
992 	else
993 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
994 #ifdef CAPABILITIES
995 	seqc_write_end(&newfde->fde_seqc);
996 #endif
997 	td->td_retval[0] = new;
998 
999 	error = 0;
1000 
1001 	if (delfp != NULL) {
1002 		(void) closefp(fdp, new, delfp, td, true, false);
1003 		FILEDESC_UNLOCK_ASSERT(fdp);
1004 	} else {
1005 unlock:
1006 		FILEDESC_XUNLOCK(fdp);
1007 	}
1008 
1009 	filecaps_free_finish(oioctls);
1010 	return (error);
1011 }
1012 
1013 static void
1014 sigiofree(struct sigio *sigio)
1015 {
1016 	crfree(sigio->sio_ucred);
1017 	free(sigio, M_SIGIO);
1018 }
1019 
1020 static struct sigio *
1021 funsetown_locked(struct sigio *sigio)
1022 {
1023 	struct proc *p;
1024 	struct pgrp *pg;
1025 
1026 	SIGIO_ASSERT_LOCKED();
1027 
1028 	if (sigio == NULL)
1029 		return (NULL);
1030 	*(sigio->sio_myref) = NULL;
1031 	if (sigio->sio_pgid < 0) {
1032 		pg = sigio->sio_pgrp;
1033 		PGRP_LOCK(pg);
1034 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1035 		    sigio, sio_pgsigio);
1036 		PGRP_UNLOCK(pg);
1037 	} else {
1038 		p = sigio->sio_proc;
1039 		PROC_LOCK(p);
1040 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1041 		    sigio, sio_pgsigio);
1042 		PROC_UNLOCK(p);
1043 	}
1044 	return (sigio);
1045 }
1046 
1047 /*
1048  * If sigio is on the list associated with a process or process group,
1049  * disable signalling from the device, remove sigio from the list and
1050  * free sigio.
1051  */
1052 void
1053 funsetown(struct sigio **sigiop)
1054 {
1055 	struct sigio *sigio;
1056 
1057 	/* Racy check, consumers must provide synchronization. */
1058 	if (*sigiop == NULL)
1059 		return;
1060 
1061 	SIGIO_LOCK();
1062 	sigio = funsetown_locked(*sigiop);
1063 	SIGIO_UNLOCK();
1064 	if (sigio != NULL)
1065 		sigiofree(sigio);
1066 }
1067 
1068 /*
1069  * Free a list of sigio structures.  The caller must ensure that new sigio
1070  * structures cannot be added after this point.  For process groups this is
1071  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1072  * as an interlock.
1073  */
1074 void
1075 funsetownlst(struct sigiolst *sigiolst)
1076 {
1077 	struct proc *p;
1078 	struct pgrp *pg;
1079 	struct sigio *sigio, *tmp;
1080 
1081 	/* Racy check. */
1082 	sigio = SLIST_FIRST(sigiolst);
1083 	if (sigio == NULL)
1084 		return;
1085 
1086 	p = NULL;
1087 	pg = NULL;
1088 
1089 	SIGIO_LOCK();
1090 	sigio = SLIST_FIRST(sigiolst);
1091 	if (sigio == NULL) {
1092 		SIGIO_UNLOCK();
1093 		return;
1094 	}
1095 
1096 	/*
1097 	 * Every entry of the list should belong to a single proc or pgrp.
1098 	 */
1099 	if (sigio->sio_pgid < 0) {
1100 		pg = sigio->sio_pgrp;
1101 		sx_assert(&proctree_lock, SX_XLOCKED);
1102 		PGRP_LOCK(pg);
1103 	} else /* if (sigio->sio_pgid > 0) */ {
1104 		p = sigio->sio_proc;
1105 		PROC_LOCK(p);
1106 		KASSERT((p->p_flag & P_WEXIT) != 0,
1107 		    ("%s: process %p is not exiting", __func__, p));
1108 	}
1109 
1110 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1111 		*sigio->sio_myref = NULL;
1112 		if (pg != NULL) {
1113 			KASSERT(sigio->sio_pgid < 0,
1114 			    ("Proc sigio in pgrp sigio list"));
1115 			KASSERT(sigio->sio_pgrp == pg,
1116 			    ("Bogus pgrp in sigio list"));
1117 		} else /* if (p != NULL) */ {
1118 			KASSERT(sigio->sio_pgid > 0,
1119 			    ("Pgrp sigio in proc sigio list"));
1120 			KASSERT(sigio->sio_proc == p,
1121 			    ("Bogus proc in sigio list"));
1122 		}
1123 	}
1124 
1125 	if (pg != NULL)
1126 		PGRP_UNLOCK(pg);
1127 	else
1128 		PROC_UNLOCK(p);
1129 	SIGIO_UNLOCK();
1130 
1131 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1132 		sigiofree(sigio);
1133 }
1134 
1135 /*
1136  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1137  *
1138  * After permission checking, add a sigio structure to the sigio list for
1139  * the process or process group.
1140  */
1141 int
1142 fsetown(pid_t pgid, struct sigio **sigiop)
1143 {
1144 	struct proc *proc;
1145 	struct pgrp *pgrp;
1146 	struct sigio *osigio, *sigio;
1147 	int ret;
1148 
1149 	if (pgid == 0) {
1150 		funsetown(sigiop);
1151 		return (0);
1152 	}
1153 
1154 	ret = 0;
1155 
1156 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 	sigio->sio_pgid = pgid;
1158 	sigio->sio_ucred = crhold(curthread->td_ucred);
1159 	sigio->sio_myref = sigiop;
1160 
1161 	sx_slock(&proctree_lock);
1162 	SIGIO_LOCK();
1163 	osigio = funsetown_locked(*sigiop);
1164 	if (pgid > 0) {
1165 		proc = pfind(pgid);
1166 		if (proc == NULL) {
1167 			ret = ESRCH;
1168 			goto fail;
1169 		}
1170 
1171 		/*
1172 		 * Policy - Don't allow a process to FSETOWN a process
1173 		 * in another session.
1174 		 *
1175 		 * Remove this test to allow maximum flexibility or
1176 		 * restrict FSETOWN to the current process or process
1177 		 * group for maximum safety.
1178 		 */
1179 		if (proc->p_session != curthread->td_proc->p_session) {
1180 			PROC_UNLOCK(proc);
1181 			ret = EPERM;
1182 			goto fail;
1183 		}
1184 
1185 		sigio->sio_proc = proc;
1186 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1187 		PROC_UNLOCK(proc);
1188 	} else /* if (pgid < 0) */ {
1189 		pgrp = pgfind(-pgid);
1190 		if (pgrp == NULL) {
1191 			ret = ESRCH;
1192 			goto fail;
1193 		}
1194 
1195 		/*
1196 		 * Policy - Don't allow a process to FSETOWN a process
1197 		 * in another session.
1198 		 *
1199 		 * Remove this test to allow maximum flexibility or
1200 		 * restrict FSETOWN to the current process or process
1201 		 * group for maximum safety.
1202 		 */
1203 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1204 			PGRP_UNLOCK(pgrp);
1205 			ret = EPERM;
1206 			goto fail;
1207 		}
1208 
1209 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1210 		sigio->sio_pgrp = pgrp;
1211 		PGRP_UNLOCK(pgrp);
1212 	}
1213 	sx_sunlock(&proctree_lock);
1214 	*sigiop = sigio;
1215 	SIGIO_UNLOCK();
1216 	if (osigio != NULL)
1217 		sigiofree(osigio);
1218 	return (0);
1219 
1220 fail:
1221 	SIGIO_UNLOCK();
1222 	sx_sunlock(&proctree_lock);
1223 	sigiofree(sigio);
1224 	if (osigio != NULL)
1225 		sigiofree(osigio);
1226 	return (ret);
1227 }
1228 
1229 /*
1230  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1231  */
1232 pid_t
1233 fgetown(struct sigio **sigiop)
1234 {
1235 	pid_t pgid;
1236 
1237 	SIGIO_LOCK();
1238 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 	SIGIO_UNLOCK();
1240 	return (pgid);
1241 }
1242 
1243 static int
1244 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1245     bool audit)
1246 {
1247 	int error;
1248 
1249 	FILEDESC_XLOCK_ASSERT(fdp);
1250 
1251 	/*
1252 	 * We now hold the fp reference that used to be owned by the
1253 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1254 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1255 	 * added, and deleteing a knote for the new fd.
1256 	 */
1257 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1258 		knote_fdclose(td, fd);
1259 
1260 	/*
1261 	 * We need to notify mqueue if the object is of type mqueue.
1262 	 */
1263 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1264 		mq_fdclose(td, fd, fp);
1265 	FILEDESC_XUNLOCK(fdp);
1266 
1267 #ifdef AUDIT
1268 	if (AUDITING_TD(td) && audit)
1269 		audit_sysclose(td, fd, fp);
1270 #endif
1271 	error = closef(fp, td);
1272 
1273 	/*
1274 	 * All paths leading up to closefp() will have already removed or
1275 	 * replaced the fd in the filedesc table, so a restart would not
1276 	 * operate on the same file.
1277 	 */
1278 	if (error == ERESTART)
1279 		error = EINTR;
1280 
1281 	return (error);
1282 }
1283 
1284 static int
1285 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1286     bool holdleaders, bool audit)
1287 {
1288 	int error;
1289 
1290 	FILEDESC_XLOCK_ASSERT(fdp);
1291 
1292 	if (holdleaders) {
1293 		if (td->td_proc->p_fdtol != NULL) {
1294 			/*
1295 			 * Ask fdfree() to sleep to ensure that all relevant
1296 			 * process leaders can be traversed in closef().
1297 			 */
1298 			fdp->fd_holdleaderscount++;
1299 		} else {
1300 			holdleaders = false;
1301 		}
1302 	}
1303 
1304 	error = closefp_impl(fdp, fd, fp, td, audit);
1305 	if (holdleaders) {
1306 		FILEDESC_XLOCK(fdp);
1307 		fdp->fd_holdleaderscount--;
1308 		if (fdp->fd_holdleaderscount == 0 &&
1309 		    fdp->fd_holdleaderswakeup != 0) {
1310 			fdp->fd_holdleaderswakeup = 0;
1311 			wakeup(&fdp->fd_holdleaderscount);
1312 		}
1313 		FILEDESC_XUNLOCK(fdp);
1314 	}
1315 	return (error);
1316 }
1317 
1318 static int
1319 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1320     bool holdleaders, bool audit)
1321 {
1322 
1323 	FILEDESC_XLOCK_ASSERT(fdp);
1324 
1325 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1326 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1327 	} else {
1328 		return (closefp_impl(fdp, fd, fp, td, audit));
1329 	}
1330 }
1331 
1332 /*
1333  * Close a file descriptor.
1334  */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct close_args {
1337 	int     fd;
1338 };
1339 #endif
1340 /* ARGSUSED */
1341 int
1342 sys_close(struct thread *td, struct close_args *uap)
1343 {
1344 
1345 	return (kern_close(td, uap->fd));
1346 }
1347 
1348 int
1349 kern_close(struct thread *td, int fd)
1350 {
1351 	struct filedesc *fdp;
1352 	struct file *fp;
1353 
1354 	fdp = td->td_proc->p_fd;
1355 
1356 	FILEDESC_XLOCK(fdp);
1357 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1358 		FILEDESC_XUNLOCK(fdp);
1359 		return (EBADF);
1360 	}
1361 	fdfree(fdp, fd);
1362 
1363 	/* closefp() drops the FILEDESC lock for us. */
1364 	return (closefp(fdp, fd, fp, td, true, true));
1365 }
1366 
1367 int
1368 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1369 {
1370 	struct filedesc *fdp;
1371 	const struct fdescenttbl *fdt;
1372 	struct file *fp;
1373 	int fd;
1374 
1375 	/*
1376 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1377 	 * open should not be a usage error.  From a close_range() perspective,
1378 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1379 	 * be a usage error as all fd above 3 are in-fact already closed.
1380 	 */
1381 	if (highfd < lowfd) {
1382 		return (EINVAL);
1383 	}
1384 
1385 	fdp = td->td_proc->p_fd;
1386 	FILEDESC_XLOCK(fdp);
1387 	fdt = atomic_load_ptr(&fdp->fd_files);
1388 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1389 	fd = lowfd;
1390 	if (__predict_false(fd > highfd)) {
1391 		goto out_locked;
1392 	}
1393 	for (;;) {
1394 		fp = fdt->fdt_ofiles[fd].fde_file;
1395 		if (fp == NULL) {
1396 			if (fd == highfd)
1397 				goto out_locked;
1398 		} else {
1399 			fdfree(fdp, fd);
1400 			(void) closefp(fdp, fd, fp, td, true, true);
1401 			if (fd == highfd)
1402 				goto out_unlocked;
1403 			FILEDESC_XLOCK(fdp);
1404 			fdt = atomic_load_ptr(&fdp->fd_files);
1405 		}
1406 		fd++;
1407 	}
1408 out_locked:
1409 	FILEDESC_XUNLOCK(fdp);
1410 out_unlocked:
1411 	return (0);
1412 }
1413 
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct close_range_args {
1416 	u_int	lowfd;
1417 	u_int	highfd;
1418 	int	flags;
1419 };
1420 #endif
1421 int
1422 sys_close_range(struct thread *td, struct close_range_args *uap)
1423 {
1424 
1425 	AUDIT_ARG_FD(uap->lowfd);
1426 	AUDIT_ARG_CMD(uap->highfd);
1427 	AUDIT_ARG_FFLAGS(uap->flags);
1428 
1429 	/* No flags currently defined */
1430 	if (uap->flags != 0)
1431 		return (EINVAL);
1432 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1433 }
1434 
1435 #ifdef COMPAT_FREEBSD12
1436 /*
1437  * Close open file descriptors.
1438  */
1439 #ifndef _SYS_SYSPROTO_H_
1440 struct freebsd12_closefrom_args {
1441 	int	lowfd;
1442 };
1443 #endif
1444 /* ARGSUSED */
1445 int
1446 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1447 {
1448 	u_int lowfd;
1449 
1450 	AUDIT_ARG_FD(uap->lowfd);
1451 
1452 	/*
1453 	 * Treat negative starting file descriptor values identical to
1454 	 * closefrom(0) which closes all files.
1455 	 */
1456 	lowfd = MAX(0, uap->lowfd);
1457 	return (kern_close_range(td, lowfd, ~0U));
1458 }
1459 #endif	/* COMPAT_FREEBSD12 */
1460 
1461 #if defined(COMPAT_43)
1462 /*
1463  * Return status information about a file descriptor.
1464  */
1465 #ifndef _SYS_SYSPROTO_H_
1466 struct ofstat_args {
1467 	int	fd;
1468 	struct	ostat *sb;
1469 };
1470 #endif
1471 /* ARGSUSED */
1472 int
1473 ofstat(struct thread *td, struct ofstat_args *uap)
1474 {
1475 	struct ostat oub;
1476 	struct stat ub;
1477 	int error;
1478 
1479 	error = kern_fstat(td, uap->fd, &ub);
1480 	if (error == 0) {
1481 		cvtstat(&ub, &oub);
1482 		error = copyout(&oub, uap->sb, sizeof(oub));
1483 	}
1484 	return (error);
1485 }
1486 #endif /* COMPAT_43 */
1487 
1488 #if defined(COMPAT_FREEBSD11)
1489 int
1490 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1491 {
1492 	struct stat sb;
1493 	struct freebsd11_stat osb;
1494 	int error;
1495 
1496 	error = kern_fstat(td, uap->fd, &sb);
1497 	if (error != 0)
1498 		return (error);
1499 	error = freebsd11_cvtstat(&sb, &osb);
1500 	if (error == 0)
1501 		error = copyout(&osb, uap->sb, sizeof(osb));
1502 	return (error);
1503 }
1504 #endif	/* COMPAT_FREEBSD11 */
1505 
1506 /*
1507  * Return status information about a file descriptor.
1508  */
1509 #ifndef _SYS_SYSPROTO_H_
1510 struct fstat_args {
1511 	int	fd;
1512 	struct	stat *sb;
1513 };
1514 #endif
1515 /* ARGSUSED */
1516 int
1517 sys_fstat(struct thread *td, struct fstat_args *uap)
1518 {
1519 	struct stat ub;
1520 	int error;
1521 
1522 	error = kern_fstat(td, uap->fd, &ub);
1523 	if (error == 0)
1524 		error = copyout(&ub, uap->sb, sizeof(ub));
1525 	return (error);
1526 }
1527 
1528 int
1529 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1530 {
1531 	struct file *fp;
1532 	int error;
1533 
1534 	AUDIT_ARG_FD(fd);
1535 
1536 	error = fget(td, fd, &cap_fstat_rights, &fp);
1537 	if (__predict_false(error != 0))
1538 		return (error);
1539 
1540 	AUDIT_ARG_FILE(td->td_proc, fp);
1541 
1542 	error = fo_stat(fp, sbp, td->td_ucred, td);
1543 	fdrop(fp, td);
1544 #ifdef __STAT_TIME_T_EXT
1545 	sbp->st_atim_ext = 0;
1546 	sbp->st_mtim_ext = 0;
1547 	sbp->st_ctim_ext = 0;
1548 	sbp->st_btim_ext = 0;
1549 #endif
1550 #ifdef KTRACE
1551 	if (KTRPOINT(td, KTR_STRUCT))
1552 		ktrstat_error(sbp, error);
1553 #endif
1554 	return (error);
1555 }
1556 
1557 #if defined(COMPAT_FREEBSD11)
1558 /*
1559  * Return status information about a file descriptor.
1560  */
1561 #ifndef _SYS_SYSPROTO_H_
1562 struct freebsd11_nfstat_args {
1563 	int	fd;
1564 	struct	nstat *sb;
1565 };
1566 #endif
1567 /* ARGSUSED */
1568 int
1569 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1570 {
1571 	struct nstat nub;
1572 	struct stat ub;
1573 	int error;
1574 
1575 	error = kern_fstat(td, uap->fd, &ub);
1576 	if (error == 0) {
1577 		freebsd11_cvtnstat(&ub, &nub);
1578 		error = copyout(&nub, uap->sb, sizeof(nub));
1579 	}
1580 	return (error);
1581 }
1582 #endif /* COMPAT_FREEBSD11 */
1583 
1584 /*
1585  * Return pathconf information about a file descriptor.
1586  */
1587 #ifndef _SYS_SYSPROTO_H_
1588 struct fpathconf_args {
1589 	int	fd;
1590 	int	name;
1591 };
1592 #endif
1593 /* ARGSUSED */
1594 int
1595 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1596 {
1597 	long value;
1598 	int error;
1599 
1600 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1601 	if (error == 0)
1602 		td->td_retval[0] = value;
1603 	return (error);
1604 }
1605 
1606 int
1607 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1608 {
1609 	struct file *fp;
1610 	struct vnode *vp;
1611 	int error;
1612 
1613 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1614 	if (error != 0)
1615 		return (error);
1616 
1617 	if (name == _PC_ASYNC_IO) {
1618 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1619 		goto out;
1620 	}
1621 	vp = fp->f_vnode;
1622 	if (vp != NULL) {
1623 		vn_lock(vp, LK_SHARED | LK_RETRY);
1624 		error = VOP_PATHCONF(vp, name, valuep);
1625 		VOP_UNLOCK(vp);
1626 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1627 		if (name != _PC_PIPE_BUF) {
1628 			error = EINVAL;
1629 		} else {
1630 			*valuep = PIPE_BUF;
1631 			error = 0;
1632 		}
1633 	} else {
1634 		error = EOPNOTSUPP;
1635 	}
1636 out:
1637 	fdrop(fp, td);
1638 	return (error);
1639 }
1640 
1641 /*
1642  * Copy filecaps structure allocating memory for ioctls array if needed.
1643  *
1644  * The last parameter indicates whether the fdtable is locked. If it is not and
1645  * ioctls are encountered, copying fails and the caller must lock the table.
1646  *
1647  * Note that if the table was not locked, the caller has to check the relevant
1648  * sequence counter to determine whether the operation was successful.
1649  */
1650 bool
1651 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1652 {
1653 	size_t size;
1654 
1655 	if (src->fc_ioctls != NULL && !locked)
1656 		return (false);
1657 	memcpy(dst, src, sizeof(*src));
1658 	if (src->fc_ioctls == NULL)
1659 		return (true);
1660 
1661 	KASSERT(src->fc_nioctls > 0,
1662 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1663 
1664 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1665 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1666 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1667 	return (true);
1668 }
1669 
1670 static u_long *
1671 filecaps_copy_prep(const struct filecaps *src)
1672 {
1673 	u_long *ioctls;
1674 	size_t size;
1675 
1676 	if (__predict_true(src->fc_ioctls == NULL))
1677 		return (NULL);
1678 
1679 	KASSERT(src->fc_nioctls > 0,
1680 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1681 
1682 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1683 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1684 	return (ioctls);
1685 }
1686 
1687 static void
1688 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1689     u_long *ioctls)
1690 {
1691 	size_t size;
1692 
1693 	*dst = *src;
1694 	if (__predict_true(src->fc_ioctls == NULL)) {
1695 		MPASS(ioctls == NULL);
1696 		return;
1697 	}
1698 
1699 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1700 	dst->fc_ioctls = ioctls;
1701 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1702 }
1703 
1704 /*
1705  * Move filecaps structure to the new place and clear the old place.
1706  */
1707 void
1708 filecaps_move(struct filecaps *src, struct filecaps *dst)
1709 {
1710 
1711 	*dst = *src;
1712 	bzero(src, sizeof(*src));
1713 }
1714 
1715 /*
1716  * Fill the given filecaps structure with full rights.
1717  */
1718 static void
1719 filecaps_fill(struct filecaps *fcaps)
1720 {
1721 
1722 	CAP_ALL(&fcaps->fc_rights);
1723 	fcaps->fc_ioctls = NULL;
1724 	fcaps->fc_nioctls = -1;
1725 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1726 }
1727 
1728 /*
1729  * Free memory allocated within filecaps structure.
1730  */
1731 void
1732 filecaps_free(struct filecaps *fcaps)
1733 {
1734 
1735 	free(fcaps->fc_ioctls, M_FILECAPS);
1736 	bzero(fcaps, sizeof(*fcaps));
1737 }
1738 
1739 static u_long *
1740 filecaps_free_prep(struct filecaps *fcaps)
1741 {
1742 	u_long *ioctls;
1743 
1744 	ioctls = fcaps->fc_ioctls;
1745 	bzero(fcaps, sizeof(*fcaps));
1746 	return (ioctls);
1747 }
1748 
1749 static void
1750 filecaps_free_finish(u_long *ioctls)
1751 {
1752 
1753 	free(ioctls, M_FILECAPS);
1754 }
1755 
1756 /*
1757  * Validate the given filecaps structure.
1758  */
1759 static void
1760 filecaps_validate(const struct filecaps *fcaps, const char *func)
1761 {
1762 
1763 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1764 	    ("%s: invalid rights", func));
1765 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1766 	    ("%s: invalid fcntls", func));
1767 	KASSERT(fcaps->fc_fcntls == 0 ||
1768 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1769 	    ("%s: fcntls without CAP_FCNTL", func));
1770 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1771 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1772 	    ("%s: invalid ioctls", func));
1773 	KASSERT(fcaps->fc_nioctls == 0 ||
1774 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1775 	    ("%s: ioctls without CAP_IOCTL", func));
1776 }
1777 
1778 static void
1779 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1780 {
1781 	int nfd1;
1782 
1783 	FILEDESC_XLOCK_ASSERT(fdp);
1784 
1785 	nfd1 = fdp->fd_nfiles * 2;
1786 	if (nfd1 < nfd)
1787 		nfd1 = nfd;
1788 	fdgrowtable(fdp, nfd1);
1789 }
1790 
1791 /*
1792  * Grow the file table to accommodate (at least) nfd descriptors.
1793  */
1794 static void
1795 fdgrowtable(struct filedesc *fdp, int nfd)
1796 {
1797 	struct filedesc0 *fdp0;
1798 	struct freetable *ft;
1799 	struct fdescenttbl *ntable;
1800 	struct fdescenttbl *otable;
1801 	int nnfiles, onfiles;
1802 	NDSLOTTYPE *nmap, *omap;
1803 
1804 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1805 
1806 	/* save old values */
1807 	onfiles = fdp->fd_nfiles;
1808 	otable = fdp->fd_files;
1809 	omap = fdp->fd_map;
1810 
1811 	/* compute the size of the new table */
1812 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1813 	if (nnfiles <= onfiles)
1814 		/* the table is already large enough */
1815 		return;
1816 
1817 	/*
1818 	 * Allocate a new table.  We need enough space for the number of
1819 	 * entries, file entries themselves and the struct freetable we will use
1820 	 * when we decommission the table and place it on the freelist.
1821 	 * We place the struct freetable in the middle so we don't have
1822 	 * to worry about padding.
1823 	 */
1824 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1825 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1826 	    sizeof(struct freetable),
1827 	    M_FILEDESC, M_ZERO | M_WAITOK);
1828 	/* copy the old data */
1829 	ntable->fdt_nfiles = nnfiles;
1830 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1831 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1832 
1833 	/*
1834 	 * Allocate a new map only if the old is not large enough.  It will
1835 	 * grow at a slower rate than the table as it can map more
1836 	 * entries than the table can hold.
1837 	 */
1838 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1839 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1840 		    M_ZERO | M_WAITOK);
1841 		/* copy over the old data and update the pointer */
1842 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1843 		fdp->fd_map = nmap;
1844 	}
1845 
1846 	/*
1847 	 * Make sure that ntable is correctly initialized before we replace
1848 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1849 	 * data.
1850 	 */
1851 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1852 
1853 	/*
1854 	 * Free the old file table when not shared by other threads or processes.
1855 	 * The old file table is considered to be shared when either are true:
1856 	 * - The process has more than one thread.
1857 	 * - The file descriptor table has been shared via fdshare().
1858 	 *
1859 	 * When shared, the old file table will be placed on a freelist
1860 	 * which will be processed when the struct filedesc is released.
1861 	 *
1862 	 * Note that if onfiles == NDFILE, we're dealing with the original
1863 	 * static allocation contained within (struct filedesc0 *)fdp,
1864 	 * which must not be freed.
1865 	 */
1866 	if (onfiles > NDFILE) {
1867 		/*
1868 		 * Note we may be called here from fdinit while allocating a
1869 		 * table for a new process in which case ->p_fd points
1870 		 * elsewhere.
1871 		 */
1872 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1873 			free(otable, M_FILEDESC);
1874 		} else {
1875 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1876 			fdp0 = (struct filedesc0 *)fdp;
1877 			ft->ft_table = otable;
1878 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1879 		}
1880 	}
1881 	/*
1882 	 * The map does not have the same possibility of threads still
1883 	 * holding references to it.  So always free it as long as it
1884 	 * does not reference the original static allocation.
1885 	 */
1886 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1887 		free(omap, M_FILEDESC);
1888 }
1889 
1890 /*
1891  * Allocate a file descriptor for the process.
1892  */
1893 int
1894 fdalloc(struct thread *td, int minfd, int *result)
1895 {
1896 	struct proc *p = td->td_proc;
1897 	struct filedesc *fdp = p->p_fd;
1898 	int fd, maxfd, allocfd;
1899 #ifdef RACCT
1900 	int error;
1901 #endif
1902 
1903 	FILEDESC_XLOCK_ASSERT(fdp);
1904 
1905 	if (fdp->fd_freefile > minfd)
1906 		minfd = fdp->fd_freefile;
1907 
1908 	maxfd = getmaxfd(td);
1909 
1910 	/*
1911 	 * Search the bitmap for a free descriptor starting at minfd.
1912 	 * If none is found, grow the file table.
1913 	 */
1914 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1915 	if (__predict_false(fd >= maxfd))
1916 		return (EMFILE);
1917 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1918 		allocfd = min(fd * 2, maxfd);
1919 #ifdef RACCT
1920 		if (RACCT_ENABLED()) {
1921 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1922 			if (error != 0)
1923 				return (EMFILE);
1924 		}
1925 #endif
1926 		/*
1927 		 * fd is already equal to first free descriptor >= minfd, so
1928 		 * we only need to grow the table and we are done.
1929 		 */
1930 		fdgrowtable_exp(fdp, allocfd);
1931 	}
1932 
1933 	/*
1934 	 * Perform some sanity checks, then mark the file descriptor as
1935 	 * used and return it to the caller.
1936 	 */
1937 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1938 	    ("invalid descriptor %d", fd));
1939 	KASSERT(!fdisused(fdp, fd),
1940 	    ("fd_first_free() returned non-free descriptor"));
1941 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1942 	    ("file descriptor isn't free"));
1943 	fdused(fdp, fd);
1944 	*result = fd;
1945 	return (0);
1946 }
1947 
1948 /*
1949  * Allocate n file descriptors for the process.
1950  */
1951 int
1952 fdallocn(struct thread *td, int minfd, int *fds, int n)
1953 {
1954 	struct proc *p = td->td_proc;
1955 	struct filedesc *fdp = p->p_fd;
1956 	int i;
1957 
1958 	FILEDESC_XLOCK_ASSERT(fdp);
1959 
1960 	for (i = 0; i < n; i++)
1961 		if (fdalloc(td, 0, &fds[i]) != 0)
1962 			break;
1963 
1964 	if (i < n) {
1965 		for (i--; i >= 0; i--)
1966 			fdunused(fdp, fds[i]);
1967 		return (EMFILE);
1968 	}
1969 
1970 	return (0);
1971 }
1972 
1973 /*
1974  * Create a new open file structure and allocate a file descriptor for the
1975  * process that refers to it.  We add one reference to the file for the
1976  * descriptor table and one reference for resultfp. This is to prevent us
1977  * being preempted and the entry in the descriptor table closed after we
1978  * release the FILEDESC lock.
1979  */
1980 int
1981 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1982     struct filecaps *fcaps)
1983 {
1984 	struct file *fp;
1985 	int error, fd;
1986 
1987 	MPASS(resultfp != NULL);
1988 	MPASS(resultfd != NULL);
1989 
1990 	error = _falloc_noinstall(td, &fp, 2);
1991 	if (__predict_false(error != 0)) {
1992 		return (error);
1993 	}
1994 
1995 	error = finstall_refed(td, fp, &fd, flags, fcaps);
1996 	if (__predict_false(error != 0)) {
1997 		falloc_abort(td, fp);
1998 		return (error);
1999 	}
2000 
2001 	*resultfp = fp;
2002 	*resultfd = fd;
2003 
2004 	return (0);
2005 }
2006 
2007 /*
2008  * Create a new open file structure without allocating a file descriptor.
2009  */
2010 int
2011 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2012 {
2013 	struct file *fp;
2014 	int maxuserfiles = maxfiles - (maxfiles / 20);
2015 	int openfiles_new;
2016 	static struct timeval lastfail;
2017 	static int curfail;
2018 
2019 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2020 	MPASS(n > 0);
2021 
2022 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2023 	if ((openfiles_new >= maxuserfiles &&
2024 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2025 	    openfiles_new >= maxfiles) {
2026 		atomic_subtract_int(&openfiles, 1);
2027 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2028 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2029 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2030 		}
2031 		return (ENFILE);
2032 	}
2033 	fp = uma_zalloc(file_zone, M_WAITOK);
2034 	bzero(fp, sizeof(*fp));
2035 	refcount_init(&fp->f_count, n);
2036 	fp->f_cred = crhold(td->td_ucred);
2037 	fp->f_ops = &badfileops;
2038 	*resultfp = fp;
2039 	return (0);
2040 }
2041 
2042 void
2043 falloc_abort(struct thread *td, struct file *fp)
2044 {
2045 
2046 	/*
2047 	 * For assertion purposes.
2048 	 */
2049 	refcount_init(&fp->f_count, 0);
2050 	_fdrop(fp, td);
2051 }
2052 
2053 /*
2054  * Install a file in a file descriptor table.
2055  */
2056 void
2057 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2058     struct filecaps *fcaps)
2059 {
2060 	struct filedescent *fde;
2061 
2062 	MPASS(fp != NULL);
2063 	if (fcaps != NULL)
2064 		filecaps_validate(fcaps, __func__);
2065 	FILEDESC_XLOCK_ASSERT(fdp);
2066 
2067 	fde = &fdp->fd_ofiles[fd];
2068 #ifdef CAPABILITIES
2069 	seqc_write_begin(&fde->fde_seqc);
2070 #endif
2071 	fde->fde_file = fp;
2072 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2073 	if (fcaps != NULL)
2074 		filecaps_move(fcaps, &fde->fde_caps);
2075 	else
2076 		filecaps_fill(&fde->fde_caps);
2077 #ifdef CAPABILITIES
2078 	seqc_write_end(&fde->fde_seqc);
2079 #endif
2080 }
2081 
2082 int
2083 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2084     struct filecaps *fcaps)
2085 {
2086 	struct filedesc *fdp = td->td_proc->p_fd;
2087 	int error;
2088 
2089 	MPASS(fd != NULL);
2090 
2091 	FILEDESC_XLOCK(fdp);
2092 	error = fdalloc(td, 0, fd);
2093 	if (__predict_true(error == 0)) {
2094 		_finstall(fdp, fp, *fd, flags, fcaps);
2095 	}
2096 	FILEDESC_XUNLOCK(fdp);
2097 	return (error);
2098 }
2099 
2100 int
2101 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2102     struct filecaps *fcaps)
2103 {
2104 	int error;
2105 
2106 	MPASS(fd != NULL);
2107 
2108 	if (!fhold(fp))
2109 		return (EBADF);
2110 	error = finstall_refed(td, fp, fd, flags, fcaps);
2111 	if (__predict_false(error != 0)) {
2112 		fdrop(fp, td);
2113 	}
2114 	return (error);
2115 }
2116 
2117 /*
2118  * Build a new filedesc structure from another.
2119  *
2120  * If fdp is not NULL, return with it shared locked.
2121  */
2122 struct filedesc *
2123 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2124 {
2125 	struct filedesc0 *newfdp0;
2126 	struct filedesc *newfdp;
2127 
2128 	if (prepfiles)
2129 		MPASS(lastfile != NULL);
2130 	else
2131 		MPASS(lastfile == NULL);
2132 
2133 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2134 	newfdp = &newfdp0->fd_fd;
2135 
2136 	/* Create the file descriptor table. */
2137 	FILEDESC_LOCK_INIT(newfdp);
2138 	refcount_init(&newfdp->fd_refcnt, 1);
2139 	refcount_init(&newfdp->fd_holdcnt, 1);
2140 	newfdp->fd_map = newfdp0->fd_dmap;
2141 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2142 	newfdp->fd_files->fdt_nfiles = NDFILE;
2143 
2144 	if (fdp == NULL)
2145 		return (newfdp);
2146 
2147 	FILEDESC_SLOCK(fdp);
2148 	if (!prepfiles) {
2149 		FILEDESC_SUNLOCK(fdp);
2150 		return (newfdp);
2151 	}
2152 
2153 	for (;;) {
2154 		*lastfile = fdlastfile(fdp);
2155 		if (*lastfile < newfdp->fd_nfiles)
2156 			break;
2157 		FILEDESC_SUNLOCK(fdp);
2158 		fdgrowtable(newfdp, *lastfile + 1);
2159 		FILEDESC_SLOCK(fdp);
2160 	}
2161 
2162 	return (newfdp);
2163 }
2164 
2165 /*
2166  * Build a pwddesc structure from another.
2167  * Copy the current, root, and jail root vnode references.
2168  *
2169  * If pdp is not NULL, return with it shared locked.
2170  */
2171 struct pwddesc *
2172 pdinit(struct pwddesc *pdp, bool keeplock)
2173 {
2174 	struct pwddesc *newpdp;
2175 	struct pwd *newpwd;
2176 
2177 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2178 
2179 	PWDDESC_LOCK_INIT(newpdp);
2180 	refcount_init(&newpdp->pd_refcount, 1);
2181 	newpdp->pd_cmask = CMASK;
2182 
2183 	if (pdp == NULL) {
2184 		newpwd = pwd_alloc();
2185 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2186 		return (newpdp);
2187 	}
2188 
2189 	PWDDESC_XLOCK(pdp);
2190 	newpwd = pwd_hold_pwddesc(pdp);
2191 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2192 	if (!keeplock)
2193 		PWDDESC_XUNLOCK(pdp);
2194 	return (newpdp);
2195 }
2196 
2197 static struct filedesc *
2198 fdhold(struct proc *p)
2199 {
2200 	struct filedesc *fdp;
2201 
2202 	PROC_LOCK_ASSERT(p, MA_OWNED);
2203 	fdp = p->p_fd;
2204 	if (fdp != NULL)
2205 		refcount_acquire(&fdp->fd_holdcnt);
2206 	return (fdp);
2207 }
2208 
2209 static struct pwddesc *
2210 pdhold(struct proc *p)
2211 {
2212 	struct pwddesc *pdp;
2213 
2214 	PROC_LOCK_ASSERT(p, MA_OWNED);
2215 	pdp = p->p_pd;
2216 	if (pdp != NULL)
2217 		refcount_acquire(&pdp->pd_refcount);
2218 	return (pdp);
2219 }
2220 
2221 static void
2222 fddrop(struct filedesc *fdp)
2223 {
2224 
2225 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2226 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2227 			return;
2228 	}
2229 
2230 	FILEDESC_LOCK_DESTROY(fdp);
2231 	uma_zfree(filedesc0_zone, fdp);
2232 }
2233 
2234 static void
2235 pddrop(struct pwddesc *pdp)
2236 {
2237 	struct pwd *pwd;
2238 
2239 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2240 		return;
2241 
2242 	PWDDESC_XLOCK(pdp);
2243 	if (refcount_release(&pdp->pd_refcount) == 0) {
2244 		PWDDESC_XUNLOCK(pdp);
2245 		return;
2246 	}
2247 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2248 	pwd_set(pdp, NULL);
2249 	PWDDESC_XUNLOCK(pdp);
2250 	pwd_drop(pwd);
2251 
2252 	PWDDESC_LOCK_DESTROY(pdp);
2253 	free(pdp, M_PWDDESC);
2254 }
2255 
2256 /*
2257  * Share a filedesc structure.
2258  */
2259 struct filedesc *
2260 fdshare(struct filedesc *fdp)
2261 {
2262 
2263 	refcount_acquire(&fdp->fd_refcnt);
2264 	return (fdp);
2265 }
2266 
2267 /*
2268  * Share a pwddesc structure.
2269  */
2270 struct pwddesc *
2271 pdshare(struct pwddesc *pdp)
2272 {
2273 	refcount_acquire(&pdp->pd_refcount);
2274 	return (pdp);
2275 }
2276 
2277 /*
2278  * Unshare a filedesc structure, if necessary by making a copy
2279  */
2280 void
2281 fdunshare(struct thread *td)
2282 {
2283 	struct filedesc *tmp;
2284 	struct proc *p = td->td_proc;
2285 
2286 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2287 		return;
2288 
2289 	tmp = fdcopy(p->p_fd);
2290 	fdescfree(td);
2291 	p->p_fd = tmp;
2292 }
2293 
2294 /*
2295  * Unshare a pwddesc structure.
2296  */
2297 void
2298 pdunshare(struct thread *td)
2299 {
2300 	struct pwddesc *pdp;
2301 	struct proc *p;
2302 
2303 	p = td->td_proc;
2304 	/* Not shared. */
2305 	if (p->p_pd->pd_refcount == 1)
2306 		return;
2307 
2308 	pdp = pdcopy(p->p_pd);
2309 	pdescfree(td);
2310 	p->p_pd = pdp;
2311 }
2312 
2313 void
2314 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2315 {
2316 
2317 	fdescfree(td);
2318 	td->td_proc->p_fd = fdp;
2319 }
2320 
2321 /*
2322  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2323  * this is to ease callers, not catch errors.
2324  */
2325 struct filedesc *
2326 fdcopy(struct filedesc *fdp)
2327 {
2328 	struct filedesc *newfdp;
2329 	struct filedescent *nfde, *ofde;
2330 	int i, lastfile;
2331 
2332 	MPASS(fdp != NULL);
2333 
2334 	newfdp = fdinit(fdp, true, &lastfile);
2335 	/* copy all passable descriptors (i.e. not kqueue) */
2336 	newfdp->fd_freefile = -1;
2337 	for (i = 0; i <= lastfile; ++i) {
2338 		ofde = &fdp->fd_ofiles[i];
2339 		if (ofde->fde_file == NULL ||
2340 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2341 		    !fhold(ofde->fde_file)) {
2342 			if (newfdp->fd_freefile == -1)
2343 				newfdp->fd_freefile = i;
2344 			continue;
2345 		}
2346 		nfde = &newfdp->fd_ofiles[i];
2347 		*nfde = *ofde;
2348 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2349 		fdused_init(newfdp, i);
2350 	}
2351 	if (newfdp->fd_freefile == -1)
2352 		newfdp->fd_freefile = i;
2353 	FILEDESC_SUNLOCK(fdp);
2354 	return (newfdp);
2355 }
2356 
2357 /*
2358  * Copy a pwddesc structure.
2359  */
2360 struct pwddesc *
2361 pdcopy(struct pwddesc *pdp)
2362 {
2363 	struct pwddesc *newpdp;
2364 
2365 	MPASS(pdp != NULL);
2366 
2367 	newpdp = pdinit(pdp, true);
2368 	newpdp->pd_cmask = pdp->pd_cmask;
2369 	PWDDESC_XUNLOCK(pdp);
2370 	return (newpdp);
2371 }
2372 
2373 /*
2374  * Copies a filedesc structure, while remapping all file descriptors
2375  * stored inside using a translation table.
2376  *
2377  * File descriptors are copied over to the new file descriptor table,
2378  * regardless of whether the close-on-exec flag is set.
2379  */
2380 int
2381 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2382     struct filedesc **ret)
2383 {
2384 	struct filedesc *newfdp;
2385 	struct filedescent *nfde, *ofde;
2386 	int error, i, lastfile;
2387 
2388 	MPASS(fdp != NULL);
2389 
2390 	newfdp = fdinit(fdp, true, &lastfile);
2391 	if (nfds > lastfile + 1) {
2392 		/* New table cannot be larger than the old one. */
2393 		error = E2BIG;
2394 		goto bad;
2395 	}
2396 	/* Copy all passable descriptors (i.e. not kqueue). */
2397 	newfdp->fd_freefile = nfds;
2398 	for (i = 0; i < nfds; ++i) {
2399 		if (fds[i] < 0 || fds[i] > lastfile) {
2400 			/* File descriptor out of bounds. */
2401 			error = EBADF;
2402 			goto bad;
2403 		}
2404 		ofde = &fdp->fd_ofiles[fds[i]];
2405 		if (ofde->fde_file == NULL) {
2406 			/* Unused file descriptor. */
2407 			error = EBADF;
2408 			goto bad;
2409 		}
2410 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2411 			/* File descriptor cannot be passed. */
2412 			error = EINVAL;
2413 			goto bad;
2414 		}
2415 		if (!fhold(ofde->fde_file)) {
2416 			error = EBADF;
2417 			goto bad;
2418 		}
2419 		nfde = &newfdp->fd_ofiles[i];
2420 		*nfde = *ofde;
2421 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2422 		fdused_init(newfdp, i);
2423 	}
2424 	FILEDESC_SUNLOCK(fdp);
2425 	*ret = newfdp;
2426 	return (0);
2427 bad:
2428 	FILEDESC_SUNLOCK(fdp);
2429 	fdescfree_remapped(newfdp);
2430 	return (error);
2431 }
2432 
2433 /*
2434  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2435  * one of processes using it exits) and the table used to be shared.
2436  */
2437 static void
2438 fdclearlocks(struct thread *td)
2439 {
2440 	struct filedesc *fdp;
2441 	struct filedesc_to_leader *fdtol;
2442 	struct flock lf;
2443 	struct file *fp;
2444 	struct proc *p;
2445 	struct vnode *vp;
2446 	int i, lastfile;
2447 
2448 	p = td->td_proc;
2449 	fdp = p->p_fd;
2450 	fdtol = p->p_fdtol;
2451 	MPASS(fdtol != NULL);
2452 
2453 	FILEDESC_XLOCK(fdp);
2454 	KASSERT(fdtol->fdl_refcount > 0,
2455 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2456 	    fdtol->fdl_refcount));
2457 	if (fdtol->fdl_refcount == 1 &&
2458 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2459 		lastfile = fdlastfile(fdp);
2460 		for (i = 0; i <= lastfile; i++) {
2461 			fp = fdp->fd_ofiles[i].fde_file;
2462 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2463 			    !fhold(fp))
2464 				continue;
2465 			FILEDESC_XUNLOCK(fdp);
2466 			lf.l_whence = SEEK_SET;
2467 			lf.l_start = 0;
2468 			lf.l_len = 0;
2469 			lf.l_type = F_UNLCK;
2470 			vp = fp->f_vnode;
2471 			(void) VOP_ADVLOCK(vp,
2472 			    (caddr_t)p->p_leader, F_UNLCK,
2473 			    &lf, F_POSIX);
2474 			FILEDESC_XLOCK(fdp);
2475 			fdrop(fp, td);
2476 		}
2477 	}
2478 retry:
2479 	if (fdtol->fdl_refcount == 1) {
2480 		if (fdp->fd_holdleaderscount > 0 &&
2481 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2482 			/*
2483 			 * close() or kern_dup() has cleared a reference
2484 			 * in a shared file descriptor table.
2485 			 */
2486 			fdp->fd_holdleaderswakeup = 1;
2487 			sx_sleep(&fdp->fd_holdleaderscount,
2488 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2489 			goto retry;
2490 		}
2491 		if (fdtol->fdl_holdcount > 0) {
2492 			/*
2493 			 * Ensure that fdtol->fdl_leader remains
2494 			 * valid in closef().
2495 			 */
2496 			fdtol->fdl_wakeup = 1;
2497 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2498 			    "fdlhold", 0);
2499 			goto retry;
2500 		}
2501 	}
2502 	fdtol->fdl_refcount--;
2503 	if (fdtol->fdl_refcount == 0 &&
2504 	    fdtol->fdl_holdcount == 0) {
2505 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2506 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2507 	} else
2508 		fdtol = NULL;
2509 	p->p_fdtol = NULL;
2510 	FILEDESC_XUNLOCK(fdp);
2511 	if (fdtol != NULL)
2512 		free(fdtol, M_FILEDESC_TO_LEADER);
2513 }
2514 
2515 /*
2516  * Release a filedesc structure.
2517  */
2518 static void
2519 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2520 {
2521 	struct filedesc0 *fdp0;
2522 	struct freetable *ft, *tft;
2523 	struct filedescent *fde;
2524 	struct file *fp;
2525 	int i, lastfile;
2526 
2527 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2528 	    ("%s: fd table %p carries references", __func__, fdp));
2529 
2530 	/*
2531 	 * Serialize with threads iterating over the table, if any.
2532 	 */
2533 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2534 		FILEDESC_XLOCK(fdp);
2535 		FILEDESC_XUNLOCK(fdp);
2536 	}
2537 
2538 	lastfile = fdlastfile_single(fdp);
2539 	for (i = 0; i <= lastfile; i++) {
2540 		fde = &fdp->fd_ofiles[i];
2541 		fp = fde->fde_file;
2542 		if (fp != NULL) {
2543 			fdefree_last(fde);
2544 			if (needclose)
2545 				(void) closef(fp, td);
2546 			else
2547 				fdrop(fp, td);
2548 		}
2549 	}
2550 
2551 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2552 		free(fdp->fd_map, M_FILEDESC);
2553 	if (fdp->fd_nfiles > NDFILE)
2554 		free(fdp->fd_files, M_FILEDESC);
2555 
2556 	fdp0 = (struct filedesc0 *)fdp;
2557 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2558 		free(ft->ft_table, M_FILEDESC);
2559 
2560 	fddrop(fdp);
2561 }
2562 
2563 void
2564 fdescfree(struct thread *td)
2565 {
2566 	struct proc *p;
2567 	struct filedesc *fdp;
2568 
2569 	p = td->td_proc;
2570 	fdp = p->p_fd;
2571 	MPASS(fdp != NULL);
2572 
2573 #ifdef RACCT
2574 	if (RACCT_ENABLED())
2575 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2576 #endif
2577 
2578 	if (p->p_fdtol != NULL)
2579 		fdclearlocks(td);
2580 
2581 	PROC_LOCK(p);
2582 	p->p_fd = NULL;
2583 	PROC_UNLOCK(p);
2584 
2585 	if (refcount_release(&fdp->fd_refcnt) == 0)
2586 		return;
2587 
2588 	fdescfree_fds(td, fdp, 1);
2589 }
2590 
2591 void
2592 pdescfree(struct thread *td)
2593 {
2594 	struct proc *p;
2595 	struct pwddesc *pdp;
2596 
2597 	p = td->td_proc;
2598 	pdp = p->p_pd;
2599 	MPASS(pdp != NULL);
2600 
2601 	PROC_LOCK(p);
2602 	p->p_pd = NULL;
2603 	PROC_UNLOCK(p);
2604 
2605 	pddrop(pdp);
2606 }
2607 
2608 void
2609 fdescfree_remapped(struct filedesc *fdp)
2610 {
2611 #ifdef INVARIANTS
2612 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2613 	if (!refcount_release(&fdp->fd_refcnt))
2614 		panic("%s: fd table %p has extra references", __func__, fdp);
2615 #endif
2616 	fdescfree_fds(curthread, fdp, 0);
2617 }
2618 
2619 /*
2620  * For setugid programs, we don't want to people to use that setugidness
2621  * to generate error messages which write to a file which otherwise would
2622  * otherwise be off-limits to the process.  We check for filesystems where
2623  * the vnode can change out from under us after execve (like [lin]procfs).
2624  *
2625  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2626  * sufficient.  We also don't check for setugidness since we know we are.
2627  */
2628 static bool
2629 is_unsafe(struct file *fp)
2630 {
2631 	struct vnode *vp;
2632 
2633 	if (fp->f_type != DTYPE_VNODE)
2634 		return (false);
2635 
2636 	vp = fp->f_vnode;
2637 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2638 }
2639 
2640 /*
2641  * Make this setguid thing safe, if at all possible.
2642  */
2643 void
2644 fdsetugidsafety(struct thread *td)
2645 {
2646 	struct filedesc *fdp;
2647 	struct file *fp;
2648 	int i;
2649 
2650 	fdp = td->td_proc->p_fd;
2651 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2652 	    ("the fdtable should not be shared"));
2653 	MPASS(fdp->fd_nfiles >= 3);
2654 	for (i = 0; i <= 2; i++) {
2655 		fp = fdp->fd_ofiles[i].fde_file;
2656 		if (fp != NULL && is_unsafe(fp)) {
2657 			FILEDESC_XLOCK(fdp);
2658 			knote_fdclose(td, i);
2659 			/*
2660 			 * NULL-out descriptor prior to close to avoid
2661 			 * a race while close blocks.
2662 			 */
2663 			fdfree(fdp, i);
2664 			FILEDESC_XUNLOCK(fdp);
2665 			(void) closef(fp, td);
2666 		}
2667 	}
2668 }
2669 
2670 /*
2671  * If a specific file object occupies a specific file descriptor, close the
2672  * file descriptor entry and drop a reference on the file object.  This is a
2673  * convenience function to handle a subsequent error in a function that calls
2674  * falloc() that handles the race that another thread might have closed the
2675  * file descriptor out from under the thread creating the file object.
2676  */
2677 void
2678 fdclose(struct thread *td, struct file *fp, int idx)
2679 {
2680 	struct filedesc *fdp = td->td_proc->p_fd;
2681 
2682 	FILEDESC_XLOCK(fdp);
2683 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2684 		fdfree(fdp, idx);
2685 		FILEDESC_XUNLOCK(fdp);
2686 		fdrop(fp, td);
2687 	} else
2688 		FILEDESC_XUNLOCK(fdp);
2689 }
2690 
2691 /*
2692  * Close any files on exec?
2693  */
2694 void
2695 fdcloseexec(struct thread *td)
2696 {
2697 	struct filedesc *fdp;
2698 	struct filedescent *fde;
2699 	struct file *fp;
2700 	int i, lastfile;
2701 
2702 	fdp = td->td_proc->p_fd;
2703 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2704 	    ("the fdtable should not be shared"));
2705 	lastfile = fdlastfile_single(fdp);
2706 	for (i = 0; i <= lastfile; i++) {
2707 		fde = &fdp->fd_ofiles[i];
2708 		fp = fde->fde_file;
2709 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2710 		    (fde->fde_flags & UF_EXCLOSE))) {
2711 			FILEDESC_XLOCK(fdp);
2712 			fdfree(fdp, i);
2713 			(void) closefp(fdp, i, fp, td, false, false);
2714 			FILEDESC_UNLOCK_ASSERT(fdp);
2715 		}
2716 	}
2717 }
2718 
2719 /*
2720  * It is unsafe for set[ug]id processes to be started with file
2721  * descriptors 0..2 closed, as these descriptors are given implicit
2722  * significance in the Standard C library.  fdcheckstd() will create a
2723  * descriptor referencing /dev/null for each of stdin, stdout, and
2724  * stderr that is not already open.
2725  */
2726 int
2727 fdcheckstd(struct thread *td)
2728 {
2729 	struct filedesc *fdp;
2730 	register_t save;
2731 	int i, error, devnull;
2732 
2733 	fdp = td->td_proc->p_fd;
2734 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2735 	    ("the fdtable should not be shared"));
2736 	MPASS(fdp->fd_nfiles >= 3);
2737 	devnull = -1;
2738 	for (i = 0; i <= 2; i++) {
2739 		if (fdp->fd_ofiles[i].fde_file != NULL)
2740 			continue;
2741 
2742 		save = td->td_retval[0];
2743 		if (devnull != -1) {
2744 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2745 		} else {
2746 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2747 			    UIO_SYSSPACE, O_RDWR, 0);
2748 			if (error == 0) {
2749 				devnull = td->td_retval[0];
2750 				KASSERT(devnull == i, ("we didn't get our fd"));
2751 			}
2752 		}
2753 		td->td_retval[0] = save;
2754 		if (error != 0)
2755 			return (error);
2756 	}
2757 	return (0);
2758 }
2759 
2760 /*
2761  * Internal form of close.  Decrement reference count on file structure.
2762  * Note: td may be NULL when closing a file that was being passed in a
2763  * message.
2764  */
2765 int
2766 closef(struct file *fp, struct thread *td)
2767 {
2768 	struct vnode *vp;
2769 	struct flock lf;
2770 	struct filedesc_to_leader *fdtol;
2771 	struct filedesc *fdp;
2772 
2773 	MPASS(td != NULL);
2774 
2775 	/*
2776 	 * POSIX record locking dictates that any close releases ALL
2777 	 * locks owned by this process.  This is handled by setting
2778 	 * a flag in the unlock to free ONLY locks obeying POSIX
2779 	 * semantics, and not to free BSD-style file locks.
2780 	 * If the descriptor was in a message, POSIX-style locks
2781 	 * aren't passed with the descriptor, and the thread pointer
2782 	 * will be NULL.  Callers should be careful only to pass a
2783 	 * NULL thread pointer when there really is no owning
2784 	 * context that might have locks, or the locks will be
2785 	 * leaked.
2786 	 */
2787 	if (fp->f_type == DTYPE_VNODE) {
2788 		vp = fp->f_vnode;
2789 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2790 			lf.l_whence = SEEK_SET;
2791 			lf.l_start = 0;
2792 			lf.l_len = 0;
2793 			lf.l_type = F_UNLCK;
2794 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2795 			    F_UNLCK, &lf, F_POSIX);
2796 		}
2797 		fdtol = td->td_proc->p_fdtol;
2798 		if (fdtol != NULL) {
2799 			/*
2800 			 * Handle special case where file descriptor table is
2801 			 * shared between multiple process leaders.
2802 			 */
2803 			fdp = td->td_proc->p_fd;
2804 			FILEDESC_XLOCK(fdp);
2805 			for (fdtol = fdtol->fdl_next;
2806 			    fdtol != td->td_proc->p_fdtol;
2807 			    fdtol = fdtol->fdl_next) {
2808 				if ((fdtol->fdl_leader->p_flag &
2809 				    P_ADVLOCK) == 0)
2810 					continue;
2811 				fdtol->fdl_holdcount++;
2812 				FILEDESC_XUNLOCK(fdp);
2813 				lf.l_whence = SEEK_SET;
2814 				lf.l_start = 0;
2815 				lf.l_len = 0;
2816 				lf.l_type = F_UNLCK;
2817 				vp = fp->f_vnode;
2818 				(void) VOP_ADVLOCK(vp,
2819 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2820 				    F_POSIX);
2821 				FILEDESC_XLOCK(fdp);
2822 				fdtol->fdl_holdcount--;
2823 				if (fdtol->fdl_holdcount == 0 &&
2824 				    fdtol->fdl_wakeup != 0) {
2825 					fdtol->fdl_wakeup = 0;
2826 					wakeup(fdtol);
2827 				}
2828 			}
2829 			FILEDESC_XUNLOCK(fdp);
2830 		}
2831 	}
2832 	return (fdrop_close(fp, td));
2833 }
2834 
2835 /*
2836  * Hack for file descriptor passing code.
2837  */
2838 void
2839 closef_nothread(struct file *fp)
2840 {
2841 
2842 	fdrop(fp, NULL);
2843 }
2844 
2845 /*
2846  * Initialize the file pointer with the specified properties.
2847  *
2848  * The ops are set with release semantics to be certain that the flags, type,
2849  * and data are visible when ops is.  This is to prevent ops methods from being
2850  * called with bad data.
2851  */
2852 void
2853 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2854 {
2855 	fp->f_data = data;
2856 	fp->f_flag = flag;
2857 	fp->f_type = type;
2858 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2859 }
2860 
2861 void
2862 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2863 {
2864 	fp->f_seqcount[UIO_READ] = 1;
2865 	fp->f_seqcount[UIO_WRITE] = 1;
2866 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2867 	    data, ops);
2868 }
2869 
2870 int
2871 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2872     struct file **fpp, struct filecaps *havecapsp)
2873 {
2874 	struct filedescent *fde;
2875 	int error;
2876 
2877 	FILEDESC_LOCK_ASSERT(fdp);
2878 
2879 	fde = fdeget_locked(fdp, fd);
2880 	if (fde == NULL) {
2881 		error = EBADF;
2882 		goto out;
2883 	}
2884 
2885 #ifdef CAPABILITIES
2886 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2887 	if (error != 0)
2888 		goto out;
2889 #endif
2890 
2891 	if (havecapsp != NULL)
2892 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2893 
2894 	*fpp = fde->fde_file;
2895 
2896 	error = 0;
2897 out:
2898 	return (error);
2899 }
2900 
2901 int
2902 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2903     struct file **fpp, struct filecaps *havecapsp)
2904 {
2905 	struct filedesc *fdp = td->td_proc->p_fd;
2906 	int error;
2907 #ifndef CAPABILITIES
2908 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2909 	if (havecapsp != NULL && error == 0)
2910 		filecaps_fill(havecapsp);
2911 #else
2912 	struct file *fp;
2913 	seqc_t seq;
2914 
2915 	*fpp = NULL;
2916 	for (;;) {
2917 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2918 		if (error != 0)
2919 			return (error);
2920 
2921 		if (havecapsp != NULL) {
2922 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2923 			    havecapsp, false)) {
2924 				fdrop(fp, td);
2925 				goto get_locked;
2926 			}
2927 		}
2928 
2929 		if (!fd_modified(fdp, fd, seq))
2930 			break;
2931 		fdrop(fp, td);
2932 	}
2933 
2934 	*fpp = fp;
2935 	return (0);
2936 
2937 get_locked:
2938 	FILEDESC_SLOCK(fdp);
2939 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2940 	if (error == 0 && !fhold(*fpp))
2941 		error = EBADF;
2942 	FILEDESC_SUNLOCK(fdp);
2943 #endif
2944 	return (error);
2945 }
2946 
2947 #ifdef CAPABILITIES
2948 int
2949 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2950 {
2951 	const struct filedescent *fde;
2952 	const struct fdescenttbl *fdt;
2953 	struct filedesc *fdp;
2954 	struct file *fp;
2955 	struct vnode *vp;
2956 	const cap_rights_t *haverights;
2957 	cap_rights_t rights;
2958 	seqc_t seq;
2959 
2960 	VFS_SMR_ASSERT_ENTERED();
2961 
2962 	rights = *ndp->ni_rightsneeded;
2963 	cap_rights_set_one(&rights, CAP_LOOKUP);
2964 
2965 	fdp = curproc->p_fd;
2966 	fdt = fdp->fd_files;
2967 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2968 		return (EBADF);
2969 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2970 	fde = &fdt->fdt_ofiles[fd];
2971 	haverights = cap_rights_fde_inline(fde);
2972 	fp = fde->fde_file;
2973 	if (__predict_false(fp == NULL))
2974 		return (EAGAIN);
2975 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2976 		return (EAGAIN);
2977 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2978 	vp = fp->f_vnode;
2979 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2980 		return (EAGAIN);
2981 	}
2982 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2983 		return (EAGAIN);
2984 	}
2985 	/*
2986 	 * Use an acquire barrier to force re-reading of fdt so it is
2987 	 * refreshed for verification.
2988 	 */
2989 	atomic_thread_fence_acq();
2990 	fdt = fdp->fd_files;
2991 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2992 		return (EAGAIN);
2993 	/*
2994 	 * If file descriptor doesn't have all rights,
2995 	 * all lookups relative to it must also be
2996 	 * strictly relative.
2997 	 *
2998 	 * Not yet supported by fast path.
2999 	 */
3000 	CAP_ALL(&rights);
3001 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3002 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3003 	    ndp->ni_filecaps.fc_nioctls != -1) {
3004 #ifdef notyet
3005 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3006 #else
3007 		return (EAGAIN);
3008 #endif
3009 	}
3010 	*vpp = vp;
3011 	return (0);
3012 }
3013 #else
3014 int
3015 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3016 {
3017 	const struct fdescenttbl *fdt;
3018 	struct filedesc *fdp;
3019 	struct file *fp;
3020 	struct vnode *vp;
3021 
3022 	VFS_SMR_ASSERT_ENTERED();
3023 
3024 	fdp = curproc->p_fd;
3025 	fdt = fdp->fd_files;
3026 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3027 		return (EBADF);
3028 	fp = fdt->fdt_ofiles[fd].fde_file;
3029 	if (__predict_false(fp == NULL))
3030 		return (EAGAIN);
3031 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3032 	vp = fp->f_vnode;
3033 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3034 		return (EAGAIN);
3035 	}
3036 	/*
3037 	 * Use an acquire barrier to force re-reading of fdt so it is
3038 	 * refreshed for verification.
3039 	 */
3040 	atomic_thread_fence_acq();
3041 	fdt = fdp->fd_files;
3042 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3043 		return (EAGAIN);
3044 	filecaps_fill(&ndp->ni_filecaps);
3045 	*vpp = vp;
3046 	return (0);
3047 }
3048 #endif
3049 
3050 int
3051 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3052     struct file **fpp, seqc_t *seqp)
3053 {
3054 #ifdef CAPABILITIES
3055 	const struct filedescent *fde;
3056 #endif
3057 	const struct fdescenttbl *fdt;
3058 	struct file *fp;
3059 #ifdef CAPABILITIES
3060 	seqc_t seq;
3061 	cap_rights_t haverights;
3062 	int error;
3063 #endif
3064 
3065 	fdt = fdp->fd_files;
3066 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3067 		return (EBADF);
3068 	/*
3069 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3070 	 * never raising a refcount above 0.  To accomplish this we have
3071 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3072 	 * must be re-verified once we acquire a reference to be certain
3073 	 * that the identity is still correct and we did not lose a race
3074 	 * due to preemption.
3075 	 */
3076 	for (;;) {
3077 #ifdef CAPABILITIES
3078 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3079 		fde = &fdt->fdt_ofiles[fd];
3080 		haverights = *cap_rights_fde_inline(fde);
3081 		fp = fde->fde_file;
3082 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3083 			continue;
3084 #else
3085 		fp = fdt->fdt_ofiles[fd].fde_file;
3086 #endif
3087 		if (fp == NULL)
3088 			return (EBADF);
3089 #ifdef CAPABILITIES
3090 		error = cap_check_inline(&haverights, needrightsp);
3091 		if (error != 0)
3092 			return (error);
3093 #endif
3094 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3095 			/*
3096 			 * Force a reload. Other thread could reallocate the
3097 			 * table before this fd was closed, so it is possible
3098 			 * that there is a stale fp pointer in cached version.
3099 			 */
3100 			fdt = atomic_load_ptr(&fdp->fd_files);
3101 			continue;
3102 		}
3103 		/*
3104 		 * Use an acquire barrier to force re-reading of fdt so it is
3105 		 * refreshed for verification.
3106 		 */
3107 		atomic_thread_fence_acq();
3108 		fdt = fdp->fd_files;
3109 #ifdef	CAPABILITIES
3110 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3111 #else
3112 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3113 #endif
3114 			break;
3115 		fdrop(fp, curthread);
3116 	}
3117 	*fpp = fp;
3118 	if (seqp != NULL) {
3119 #ifdef CAPABILITIES
3120 		*seqp = seq;
3121 #endif
3122 	}
3123 	return (0);
3124 }
3125 
3126 /*
3127  * See the comments in fget_unlocked_seq for an explanation of how this works.
3128  *
3129  * This is a simplified variant which bails out to the aforementioned routine
3130  * if anything goes wrong. In practice this only happens when userspace is
3131  * racing with itself.
3132  */
3133 int
3134 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3135     struct file **fpp)
3136 {
3137 #ifdef CAPABILITIES
3138 	const struct filedescent *fde;
3139 #endif
3140 	const struct fdescenttbl *fdt;
3141 	struct file *fp;
3142 #ifdef CAPABILITIES
3143 	seqc_t seq;
3144 	const cap_rights_t *haverights;
3145 #endif
3146 
3147 	fdt = fdp->fd_files;
3148 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3149 		return (EBADF);
3150 #ifdef CAPABILITIES
3151 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3152 	fde = &fdt->fdt_ofiles[fd];
3153 	haverights = cap_rights_fde_inline(fde);
3154 	fp = fde->fde_file;
3155 #else
3156 	fp = fdt->fdt_ofiles[fd].fde_file;
3157 #endif
3158 	if (__predict_false(fp == NULL))
3159 		goto out_fallback;
3160 #ifdef CAPABILITIES
3161 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3162 		goto out_fallback;
3163 #endif
3164 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3165 		goto out_fallback;
3166 
3167 	/*
3168 	 * Use an acquire barrier to force re-reading of fdt so it is
3169 	 * refreshed for verification.
3170 	 */
3171 	atomic_thread_fence_acq();
3172 	fdt = fdp->fd_files;
3173 #ifdef	CAPABILITIES
3174 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3175 #else
3176 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3177 #endif
3178 		goto out_fdrop;
3179 	*fpp = fp;
3180 	return (0);
3181 out_fdrop:
3182 	fdrop(fp, curthread);
3183 out_fallback:
3184 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3185 }
3186 
3187 /*
3188  * Translate fd -> file when the caller guarantees the file descriptor table
3189  * can't be changed by others.
3190  *
3191  * Note this does not mean the file object itself is only visible to the caller,
3192  * merely that it wont disappear without having to be referenced.
3193  *
3194  * Must be paired with fput_only_user.
3195  */
3196 #ifdef	CAPABILITIES
3197 int
3198 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3199     struct file **fpp)
3200 {
3201 	const struct filedescent *fde;
3202 	const struct fdescenttbl *fdt;
3203 	const cap_rights_t *haverights;
3204 	struct file *fp;
3205 	int error;
3206 
3207 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3208 
3209 	if (__predict_false(fd >= fdp->fd_nfiles))
3210 		return (EBADF);
3211 
3212 	fdt = fdp->fd_files;
3213 	fde = &fdt->fdt_ofiles[fd];
3214 	fp = fde->fde_file;
3215 	if (__predict_false(fp == NULL))
3216 		return (EBADF);
3217 	MPASS(refcount_load(&fp->f_count) > 0);
3218 	haverights = cap_rights_fde_inline(fde);
3219 	error = cap_check_inline(haverights, needrightsp);
3220 	if (__predict_false(error != 0))
3221 		return (error);
3222 	*fpp = fp;
3223 	return (0);
3224 }
3225 #else
3226 int
3227 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3228     struct file **fpp)
3229 {
3230 	struct file *fp;
3231 
3232 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3233 
3234 	if (__predict_false(fd >= fdp->fd_nfiles))
3235 		return (EBADF);
3236 
3237 	fp = fdp->fd_ofiles[fd].fde_file;
3238 	if (__predict_false(fp == NULL))
3239 		return (EBADF);
3240 
3241 	MPASS(refcount_load(&fp->f_count) > 0);
3242 	*fpp = fp;
3243 	return (0);
3244 }
3245 #endif
3246 
3247 /*
3248  * Extract the file pointer associated with the specified descriptor for the
3249  * current user process.
3250  *
3251  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3252  * returned.
3253  *
3254  * File's rights will be checked against the capability rights mask.
3255  *
3256  * If an error occurred the non-zero error is returned and *fpp is set to
3257  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3258  * responsible for fdrop().
3259  */
3260 static __inline int
3261 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3262     cap_rights_t *needrightsp)
3263 {
3264 	struct filedesc *fdp;
3265 	struct file *fp;
3266 	int error;
3267 
3268 	*fpp = NULL;
3269 	fdp = td->td_proc->p_fd;
3270 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3271 	if (__predict_false(error != 0))
3272 		return (error);
3273 	if (__predict_false(fp->f_ops == &badfileops)) {
3274 		fdrop(fp, td);
3275 		return (EBADF);
3276 	}
3277 
3278 	/*
3279 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3280 	 */
3281 	error = 0;
3282 	switch (flags) {
3283 	case FREAD:
3284 	case FWRITE:
3285 		if ((fp->f_flag & flags) == 0)
3286 			error = EBADF;
3287 		break;
3288 	case FEXEC:
3289 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3290 		    ((fp->f_flag & FWRITE) != 0))
3291 			error = EBADF;
3292 		break;
3293 	case 0:
3294 		break;
3295 	default:
3296 		KASSERT(0, ("wrong flags"));
3297 	}
3298 
3299 	if (error != 0) {
3300 		fdrop(fp, td);
3301 		return (error);
3302 	}
3303 
3304 	*fpp = fp;
3305 	return (0);
3306 }
3307 
3308 int
3309 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3310 {
3311 
3312 	return (_fget(td, fd, fpp, 0, rightsp));
3313 }
3314 
3315 int
3316 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3317     struct file **fpp)
3318 {
3319 	int error;
3320 #ifndef CAPABILITIES
3321 	error = _fget(td, fd, fpp, 0, rightsp);
3322 	if (maxprotp != NULL)
3323 		*maxprotp = VM_PROT_ALL;
3324 	return (error);
3325 #else
3326 	cap_rights_t fdrights;
3327 	struct filedesc *fdp;
3328 	struct file *fp;
3329 	seqc_t seq;
3330 
3331 	*fpp = NULL;
3332 	fdp = td->td_proc->p_fd;
3333 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3334 	for (;;) {
3335 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3336 		if (__predict_false(error != 0))
3337 			return (error);
3338 		if (__predict_false(fp->f_ops == &badfileops)) {
3339 			fdrop(fp, td);
3340 			return (EBADF);
3341 		}
3342 		if (maxprotp != NULL)
3343 			fdrights = *cap_rights(fdp, fd);
3344 		if (!fd_modified(fdp, fd, seq))
3345 			break;
3346 		fdrop(fp, td);
3347 	}
3348 
3349 	/*
3350 	 * If requested, convert capability rights to access flags.
3351 	 */
3352 	if (maxprotp != NULL)
3353 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3354 	*fpp = fp;
3355 	return (0);
3356 #endif
3357 }
3358 
3359 int
3360 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3361 {
3362 
3363 	return (_fget(td, fd, fpp, FREAD, rightsp));
3364 }
3365 
3366 int
3367 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3368 {
3369 
3370 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3371 }
3372 
3373 int
3374 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3375     struct file **fpp)
3376 {
3377 	struct filedesc *fdp = td->td_proc->p_fd;
3378 #ifndef CAPABILITIES
3379 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3380 #else
3381 	struct file *fp;
3382 	int error;
3383 	seqc_t seq;
3384 
3385 	*fpp = NULL;
3386 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3387 	for (;;) {
3388 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3389 		if (error != 0)
3390 			return (error);
3391 		error = cap_fcntl_check(fdp, fd, needfcntl);
3392 		if (!fd_modified(fdp, fd, seq))
3393 			break;
3394 		fdrop(fp, td);
3395 	}
3396 	if (error != 0) {
3397 		fdrop(fp, td);
3398 		return (error);
3399 	}
3400 	*fpp = fp;
3401 	return (0);
3402 #endif
3403 }
3404 
3405 /*
3406  * Like fget() but loads the underlying vnode, or returns an error if the
3407  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3408  * never have VM objects.  The returned vnode will be vref()'d.
3409  *
3410  * XXX: what about the unused flags ?
3411  */
3412 static __inline int
3413 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3414     struct vnode **vpp)
3415 {
3416 	struct file *fp;
3417 	int error;
3418 
3419 	*vpp = NULL;
3420 	error = _fget(td, fd, &fp, flags, needrightsp);
3421 	if (error != 0)
3422 		return (error);
3423 	if (fp->f_vnode == NULL) {
3424 		error = EINVAL;
3425 	} else {
3426 		*vpp = fp->f_vnode;
3427 		vrefact(*vpp);
3428 	}
3429 	fdrop(fp, td);
3430 
3431 	return (error);
3432 }
3433 
3434 int
3435 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3436 {
3437 
3438 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3439 }
3440 
3441 int
3442 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3443     struct filecaps *havecaps, struct vnode **vpp)
3444 {
3445 	struct filecaps caps;
3446 	struct file *fp;
3447 	int error;
3448 
3449 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3450 	if (error != 0)
3451 		return (error);
3452 	if (fp->f_ops == &badfileops) {
3453 		error = EBADF;
3454 		goto out;
3455 	}
3456 	if (fp->f_vnode == NULL) {
3457 		error = EINVAL;
3458 		goto out;
3459 	}
3460 
3461 	*havecaps = caps;
3462 	*vpp = fp->f_vnode;
3463 	vrefact(*vpp);
3464 	fdrop(fp, td);
3465 
3466 	return (0);
3467 out:
3468 	filecaps_free(&caps);
3469 	fdrop(fp, td);
3470 	return (error);
3471 }
3472 
3473 int
3474 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3475 {
3476 
3477 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3478 }
3479 
3480 int
3481 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3482 {
3483 
3484 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3485 }
3486 
3487 #ifdef notyet
3488 int
3489 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3490     struct vnode **vpp)
3491 {
3492 
3493 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3494 }
3495 #endif
3496 
3497 /*
3498  * Handle the last reference to a file being closed.
3499  *
3500  * Without the noinline attribute clang keeps inlining the func thorough this
3501  * file when fdrop is used.
3502  */
3503 int __noinline
3504 _fdrop(struct file *fp, struct thread *td)
3505 {
3506 	int error;
3507 #ifdef INVARIANTS
3508 	int count;
3509 
3510 	count = refcount_load(&fp->f_count);
3511 	if (count != 0)
3512 		panic("fdrop: fp %p count %d", fp, count);
3513 #endif
3514 	error = fo_close(fp, td);
3515 	atomic_subtract_int(&openfiles, 1);
3516 	crfree(fp->f_cred);
3517 	free(fp->f_advice, M_FADVISE);
3518 	uma_zfree(file_zone, fp);
3519 
3520 	return (error);
3521 }
3522 
3523 /*
3524  * Apply an advisory lock on a file descriptor.
3525  *
3526  * Just attempt to get a record lock of the requested type on the entire file
3527  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3528  */
3529 #ifndef _SYS_SYSPROTO_H_
3530 struct flock_args {
3531 	int	fd;
3532 	int	how;
3533 };
3534 #endif
3535 /* ARGSUSED */
3536 int
3537 sys_flock(struct thread *td, struct flock_args *uap)
3538 {
3539 	struct file *fp;
3540 	struct vnode *vp;
3541 	struct flock lf;
3542 	int error;
3543 
3544 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3545 	if (error != 0)
3546 		return (error);
3547 	if (fp->f_type != DTYPE_VNODE) {
3548 		fdrop(fp, td);
3549 		return (EOPNOTSUPP);
3550 	}
3551 
3552 	vp = fp->f_vnode;
3553 	lf.l_whence = SEEK_SET;
3554 	lf.l_start = 0;
3555 	lf.l_len = 0;
3556 	if (uap->how & LOCK_UN) {
3557 		lf.l_type = F_UNLCK;
3558 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3559 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3560 		goto done2;
3561 	}
3562 	if (uap->how & LOCK_EX)
3563 		lf.l_type = F_WRLCK;
3564 	else if (uap->how & LOCK_SH)
3565 		lf.l_type = F_RDLCK;
3566 	else {
3567 		error = EBADF;
3568 		goto done2;
3569 	}
3570 	atomic_set_int(&fp->f_flag, FHASLOCK);
3571 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3572 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3573 done2:
3574 	fdrop(fp, td);
3575 	return (error);
3576 }
3577 /*
3578  * Duplicate the specified descriptor to a free descriptor.
3579  */
3580 int
3581 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3582     int openerror, int *indxp)
3583 {
3584 	struct filedescent *newfde, *oldfde;
3585 	struct file *fp;
3586 	u_long *ioctls;
3587 	int error, indx;
3588 
3589 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3590 	    ("unexpected error %d in %s", openerror, __func__));
3591 
3592 	/*
3593 	 * If the to-be-dup'd fd number is greater than the allowed number
3594 	 * of file descriptors, or the fd to be dup'd has already been
3595 	 * closed, then reject.
3596 	 */
3597 	FILEDESC_XLOCK(fdp);
3598 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3599 		FILEDESC_XUNLOCK(fdp);
3600 		return (EBADF);
3601 	}
3602 
3603 	error = fdalloc(td, 0, &indx);
3604 	if (error != 0) {
3605 		FILEDESC_XUNLOCK(fdp);
3606 		return (error);
3607 	}
3608 
3609 	/*
3610 	 * There are two cases of interest here.
3611 	 *
3612 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3613 	 *
3614 	 * For ENXIO steal away the file structure from (dfd) and store it in
3615 	 * (indx).  (dfd) is effectively closed by this operation.
3616 	 */
3617 	switch (openerror) {
3618 	case ENODEV:
3619 		/*
3620 		 * Check that the mode the file is being opened for is a
3621 		 * subset of the mode of the existing descriptor.
3622 		 */
3623 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3624 			fdunused(fdp, indx);
3625 			FILEDESC_XUNLOCK(fdp);
3626 			return (EACCES);
3627 		}
3628 		if (!fhold(fp)) {
3629 			fdunused(fdp, indx);
3630 			FILEDESC_XUNLOCK(fdp);
3631 			return (EBADF);
3632 		}
3633 		newfde = &fdp->fd_ofiles[indx];
3634 		oldfde = &fdp->fd_ofiles[dfd];
3635 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3636 #ifdef CAPABILITIES
3637 		seqc_write_begin(&newfde->fde_seqc);
3638 #endif
3639 		memcpy(newfde, oldfde, fde_change_size);
3640 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3641 		    ioctls);
3642 #ifdef CAPABILITIES
3643 		seqc_write_end(&newfde->fde_seqc);
3644 #endif
3645 		break;
3646 	case ENXIO:
3647 		/*
3648 		 * Steal away the file pointer from dfd and stuff it into indx.
3649 		 */
3650 		newfde = &fdp->fd_ofiles[indx];
3651 		oldfde = &fdp->fd_ofiles[dfd];
3652 #ifdef CAPABILITIES
3653 		seqc_write_begin(&newfde->fde_seqc);
3654 #endif
3655 		memcpy(newfde, oldfde, fde_change_size);
3656 		oldfde->fde_file = NULL;
3657 		fdunused(fdp, dfd);
3658 #ifdef CAPABILITIES
3659 		seqc_write_end(&newfde->fde_seqc);
3660 #endif
3661 		break;
3662 	}
3663 	FILEDESC_XUNLOCK(fdp);
3664 	*indxp = indx;
3665 	return (0);
3666 }
3667 
3668 /*
3669  * This sysctl determines if we will allow a process to chroot(2) if it
3670  * has a directory open:
3671  *	0: disallowed for all processes.
3672  *	1: allowed for processes that were not already chroot(2)'ed.
3673  *	2: allowed for all processes.
3674  */
3675 
3676 static int chroot_allow_open_directories = 1;
3677 
3678 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3679     &chroot_allow_open_directories, 0,
3680     "Allow a process to chroot(2) if it has a directory open");
3681 
3682 /*
3683  * Helper function for raised chroot(2) security function:  Refuse if
3684  * any filedescriptors are open directories.
3685  */
3686 static int
3687 chroot_refuse_vdir_fds(struct filedesc *fdp)
3688 {
3689 	struct vnode *vp;
3690 	struct file *fp;
3691 	int fd, lastfile;
3692 
3693 	FILEDESC_LOCK_ASSERT(fdp);
3694 
3695 	lastfile = fdlastfile(fdp);
3696 	for (fd = 0; fd <= lastfile; fd++) {
3697 		fp = fget_locked(fdp, fd);
3698 		if (fp == NULL)
3699 			continue;
3700 		if (fp->f_type == DTYPE_VNODE) {
3701 			vp = fp->f_vnode;
3702 			if (vp->v_type == VDIR)
3703 				return (EPERM);
3704 		}
3705 	}
3706 	return (0);
3707 }
3708 
3709 static void
3710 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3711 {
3712 
3713 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3714 		vrefact(oldpwd->pwd_cdir);
3715 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3716 	}
3717 
3718 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3719 		vrefact(oldpwd->pwd_rdir);
3720 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3721 	}
3722 
3723 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3724 		vrefact(oldpwd->pwd_jdir);
3725 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3726 	}
3727 }
3728 
3729 struct pwd *
3730 pwd_hold_pwddesc(struct pwddesc *pdp)
3731 {
3732 	struct pwd *pwd;
3733 
3734 	PWDDESC_ASSERT_XLOCKED(pdp);
3735 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3736 	if (pwd != NULL)
3737 		refcount_acquire(&pwd->pwd_refcount);
3738 	return (pwd);
3739 }
3740 
3741 bool
3742 pwd_hold_smr(struct pwd *pwd)
3743 {
3744 
3745 	MPASS(pwd != NULL);
3746 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3747 		return (true);
3748 	}
3749 	return (false);
3750 }
3751 
3752 struct pwd *
3753 pwd_hold(struct thread *td)
3754 {
3755 	struct pwddesc *pdp;
3756 	struct pwd *pwd;
3757 
3758 	pdp = td->td_proc->p_pd;
3759 
3760 	vfs_smr_enter();
3761 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3762 	if (pwd_hold_smr(pwd)) {
3763 		vfs_smr_exit();
3764 		return (pwd);
3765 	}
3766 	vfs_smr_exit();
3767 	PWDDESC_XLOCK(pdp);
3768 	pwd = pwd_hold_pwddesc(pdp);
3769 	MPASS(pwd != NULL);
3770 	PWDDESC_XUNLOCK(pdp);
3771 	return (pwd);
3772 }
3773 
3774 static struct pwd *
3775 pwd_alloc(void)
3776 {
3777 	struct pwd *pwd;
3778 
3779 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3780 	bzero(pwd, sizeof(*pwd));
3781 	refcount_init(&pwd->pwd_refcount, 1);
3782 	return (pwd);
3783 }
3784 
3785 void
3786 pwd_drop(struct pwd *pwd)
3787 {
3788 
3789 	if (!refcount_release(&pwd->pwd_refcount))
3790 		return;
3791 
3792 	if (pwd->pwd_cdir != NULL)
3793 		vrele(pwd->pwd_cdir);
3794 	if (pwd->pwd_rdir != NULL)
3795 		vrele(pwd->pwd_rdir);
3796 	if (pwd->pwd_jdir != NULL)
3797 		vrele(pwd->pwd_jdir);
3798 	uma_zfree_smr(pwd_zone, pwd);
3799 }
3800 
3801 /*
3802 * The caller is responsible for invoking priv_check() and
3803 * mac_vnode_check_chroot() to authorize this operation.
3804 */
3805 int
3806 pwd_chroot(struct thread *td, struct vnode *vp)
3807 {
3808 	struct pwddesc *pdp;
3809 	struct filedesc *fdp;
3810 	struct pwd *newpwd, *oldpwd;
3811 	int error;
3812 
3813 	fdp = td->td_proc->p_fd;
3814 	pdp = td->td_proc->p_pd;
3815 	newpwd = pwd_alloc();
3816 	FILEDESC_SLOCK(fdp);
3817 	PWDDESC_XLOCK(pdp);
3818 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3819 	if (chroot_allow_open_directories == 0 ||
3820 	    (chroot_allow_open_directories == 1 &&
3821 	    oldpwd->pwd_rdir != rootvnode)) {
3822 		error = chroot_refuse_vdir_fds(fdp);
3823 		FILEDESC_SUNLOCK(fdp);
3824 		if (error != 0) {
3825 			PWDDESC_XUNLOCK(pdp);
3826 			pwd_drop(newpwd);
3827 			return (error);
3828 		}
3829 	} else {
3830 		FILEDESC_SUNLOCK(fdp);
3831 	}
3832 
3833 	vrefact(vp);
3834 	newpwd->pwd_rdir = vp;
3835 	if (oldpwd->pwd_jdir == NULL) {
3836 		vrefact(vp);
3837 		newpwd->pwd_jdir = vp;
3838 	}
3839 	pwd_fill(oldpwd, newpwd);
3840 	pwd_set(pdp, newpwd);
3841 	PWDDESC_XUNLOCK(pdp);
3842 	pwd_drop(oldpwd);
3843 	return (0);
3844 }
3845 
3846 void
3847 pwd_chdir(struct thread *td, struct vnode *vp)
3848 {
3849 	struct pwddesc *pdp;
3850 	struct pwd *newpwd, *oldpwd;
3851 
3852 	VNPASS(vp->v_usecount > 0, vp);
3853 
3854 	newpwd = pwd_alloc();
3855 	pdp = td->td_proc->p_pd;
3856 	PWDDESC_XLOCK(pdp);
3857 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3858 	newpwd->pwd_cdir = vp;
3859 	pwd_fill(oldpwd, newpwd);
3860 	pwd_set(pdp, newpwd);
3861 	PWDDESC_XUNLOCK(pdp);
3862 	pwd_drop(oldpwd);
3863 }
3864 
3865 /*
3866  * jail_attach(2) changes both root and working directories.
3867  */
3868 int
3869 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3870 {
3871 	struct pwddesc *pdp;
3872 	struct filedesc *fdp;
3873 	struct pwd *newpwd, *oldpwd;
3874 	int error;
3875 
3876 	fdp = td->td_proc->p_fd;
3877 	pdp = td->td_proc->p_pd;
3878 	newpwd = pwd_alloc();
3879 	FILEDESC_SLOCK(fdp);
3880 	PWDDESC_XLOCK(pdp);
3881 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3882 	error = chroot_refuse_vdir_fds(fdp);
3883 	FILEDESC_SUNLOCK(fdp);
3884 	if (error != 0) {
3885 		PWDDESC_XUNLOCK(pdp);
3886 		pwd_drop(newpwd);
3887 		return (error);
3888 	}
3889 
3890 	vrefact(vp);
3891 	newpwd->pwd_rdir = vp;
3892 	vrefact(vp);
3893 	newpwd->pwd_cdir = vp;
3894 	if (oldpwd->pwd_jdir == NULL) {
3895 		vrefact(vp);
3896 		newpwd->pwd_jdir = vp;
3897 	}
3898 	pwd_fill(oldpwd, newpwd);
3899 	pwd_set(pdp, newpwd);
3900 	PWDDESC_XUNLOCK(pdp);
3901 	pwd_drop(oldpwd);
3902 	return (0);
3903 }
3904 
3905 void
3906 pwd_ensure_dirs(void)
3907 {
3908 	struct pwddesc *pdp;
3909 	struct pwd *oldpwd, *newpwd;
3910 
3911 	pdp = curproc->p_pd;
3912 	PWDDESC_XLOCK(pdp);
3913 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3914 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3915 		PWDDESC_XUNLOCK(pdp);
3916 		return;
3917 	}
3918 	PWDDESC_XUNLOCK(pdp);
3919 
3920 	newpwd = pwd_alloc();
3921 	PWDDESC_XLOCK(pdp);
3922 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3923 	pwd_fill(oldpwd, newpwd);
3924 	if (newpwd->pwd_cdir == NULL) {
3925 		vrefact(rootvnode);
3926 		newpwd->pwd_cdir = rootvnode;
3927 	}
3928 	if (newpwd->pwd_rdir == NULL) {
3929 		vrefact(rootvnode);
3930 		newpwd->pwd_rdir = rootvnode;
3931 	}
3932 	pwd_set(pdp, newpwd);
3933 	PWDDESC_XUNLOCK(pdp);
3934 	pwd_drop(oldpwd);
3935 }
3936 
3937 void
3938 pwd_set_rootvnode(void)
3939 {
3940 	struct pwddesc *pdp;
3941 	struct pwd *oldpwd, *newpwd;
3942 
3943 	pdp = curproc->p_pd;
3944 
3945 	newpwd = pwd_alloc();
3946 	PWDDESC_XLOCK(pdp);
3947 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3948 	vrefact(rootvnode);
3949 	newpwd->pwd_cdir = rootvnode;
3950 	vrefact(rootvnode);
3951 	newpwd->pwd_rdir = rootvnode;
3952 	pwd_fill(oldpwd, newpwd);
3953 	pwd_set(pdp, newpwd);
3954 	PWDDESC_XUNLOCK(pdp);
3955 	pwd_drop(oldpwd);
3956 }
3957 
3958 /*
3959  * Scan all active processes and prisons to see if any of them have a current
3960  * or root directory of `olddp'. If so, replace them with the new mount point.
3961  */
3962 void
3963 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3964 {
3965 	struct pwddesc *pdp;
3966 	struct pwd *newpwd, *oldpwd;
3967 	struct prison *pr;
3968 	struct proc *p;
3969 	int nrele;
3970 
3971 	if (vrefcnt(olddp) == 1)
3972 		return;
3973 	nrele = 0;
3974 	newpwd = pwd_alloc();
3975 	sx_slock(&allproc_lock);
3976 	FOREACH_PROC_IN_SYSTEM(p) {
3977 		PROC_LOCK(p);
3978 		pdp = pdhold(p);
3979 		PROC_UNLOCK(p);
3980 		if (pdp == NULL)
3981 			continue;
3982 		PWDDESC_XLOCK(pdp);
3983 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3984 		if (oldpwd == NULL ||
3985 		    (oldpwd->pwd_cdir != olddp &&
3986 		    oldpwd->pwd_rdir != olddp &&
3987 		    oldpwd->pwd_jdir != olddp)) {
3988 			PWDDESC_XUNLOCK(pdp);
3989 			pddrop(pdp);
3990 			continue;
3991 		}
3992 		if (oldpwd->pwd_cdir == olddp) {
3993 			vrefact(newdp);
3994 			newpwd->pwd_cdir = newdp;
3995 		}
3996 		if (oldpwd->pwd_rdir == olddp) {
3997 			vrefact(newdp);
3998 			newpwd->pwd_rdir = newdp;
3999 		}
4000 		if (oldpwd->pwd_jdir == olddp) {
4001 			vrefact(newdp);
4002 			newpwd->pwd_jdir = newdp;
4003 		}
4004 		pwd_fill(oldpwd, newpwd);
4005 		pwd_set(pdp, newpwd);
4006 		PWDDESC_XUNLOCK(pdp);
4007 		pwd_drop(oldpwd);
4008 		pddrop(pdp);
4009 		newpwd = pwd_alloc();
4010 	}
4011 	sx_sunlock(&allproc_lock);
4012 	pwd_drop(newpwd);
4013 	if (rootvnode == olddp) {
4014 		vrefact(newdp);
4015 		rootvnode = newdp;
4016 		nrele++;
4017 	}
4018 	mtx_lock(&prison0.pr_mtx);
4019 	if (prison0.pr_root == olddp) {
4020 		vrefact(newdp);
4021 		prison0.pr_root = newdp;
4022 		nrele++;
4023 	}
4024 	mtx_unlock(&prison0.pr_mtx);
4025 	sx_slock(&allprison_lock);
4026 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4027 		mtx_lock(&pr->pr_mtx);
4028 		if (pr->pr_root == olddp) {
4029 			vrefact(newdp);
4030 			pr->pr_root = newdp;
4031 			nrele++;
4032 		}
4033 		mtx_unlock(&pr->pr_mtx);
4034 	}
4035 	sx_sunlock(&allprison_lock);
4036 	while (nrele--)
4037 		vrele(olddp);
4038 }
4039 
4040 struct filedesc_to_leader *
4041 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4042 {
4043 	struct filedesc_to_leader *fdtol;
4044 
4045 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4046 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4047 	fdtol->fdl_refcount = 1;
4048 	fdtol->fdl_holdcount = 0;
4049 	fdtol->fdl_wakeup = 0;
4050 	fdtol->fdl_leader = leader;
4051 	if (old != NULL) {
4052 		FILEDESC_XLOCK(fdp);
4053 		fdtol->fdl_next = old->fdl_next;
4054 		fdtol->fdl_prev = old;
4055 		old->fdl_next = fdtol;
4056 		fdtol->fdl_next->fdl_prev = fdtol;
4057 		FILEDESC_XUNLOCK(fdp);
4058 	} else {
4059 		fdtol->fdl_next = fdtol;
4060 		fdtol->fdl_prev = fdtol;
4061 	}
4062 	return (fdtol);
4063 }
4064 
4065 static int
4066 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4067 {
4068 	NDSLOTTYPE *map;
4069 	struct filedesc *fdp;
4070 	int count, off, minoff;
4071 
4072 	if (*(int *)arg1 != 0)
4073 		return (EINVAL);
4074 
4075 	fdp = curproc->p_fd;
4076 	count = 0;
4077 	FILEDESC_SLOCK(fdp);
4078 	map = fdp->fd_map;
4079 	off = NDSLOT(fdp->fd_nfiles - 1);
4080 	for (minoff = NDSLOT(0); off >= minoff; --off)
4081 		count += bitcountl(map[off]);
4082 	FILEDESC_SUNLOCK(fdp);
4083 
4084 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4085 }
4086 
4087 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4088     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4089     "Number of open file descriptors");
4090 
4091 /*
4092  * Get file structures globally.
4093  */
4094 static int
4095 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4096 {
4097 	struct xfile xf;
4098 	struct filedesc *fdp;
4099 	struct file *fp;
4100 	struct proc *p;
4101 	int error, n, lastfile;
4102 
4103 	error = sysctl_wire_old_buffer(req, 0);
4104 	if (error != 0)
4105 		return (error);
4106 	if (req->oldptr == NULL) {
4107 		n = 0;
4108 		sx_slock(&allproc_lock);
4109 		FOREACH_PROC_IN_SYSTEM(p) {
4110 			PROC_LOCK(p);
4111 			if (p->p_state == PRS_NEW) {
4112 				PROC_UNLOCK(p);
4113 				continue;
4114 			}
4115 			fdp = fdhold(p);
4116 			PROC_UNLOCK(p);
4117 			if (fdp == NULL)
4118 				continue;
4119 			/* overestimates sparse tables. */
4120 			n += fdp->fd_nfiles;
4121 			fddrop(fdp);
4122 		}
4123 		sx_sunlock(&allproc_lock);
4124 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4125 	}
4126 	error = 0;
4127 	bzero(&xf, sizeof(xf));
4128 	xf.xf_size = sizeof(xf);
4129 	sx_slock(&allproc_lock);
4130 	FOREACH_PROC_IN_SYSTEM(p) {
4131 		PROC_LOCK(p);
4132 		if (p->p_state == PRS_NEW) {
4133 			PROC_UNLOCK(p);
4134 			continue;
4135 		}
4136 		if (p_cansee(req->td, p) != 0) {
4137 			PROC_UNLOCK(p);
4138 			continue;
4139 		}
4140 		xf.xf_pid = p->p_pid;
4141 		xf.xf_uid = p->p_ucred->cr_uid;
4142 		fdp = fdhold(p);
4143 		PROC_UNLOCK(p);
4144 		if (fdp == NULL)
4145 			continue;
4146 		FILEDESC_SLOCK(fdp);
4147 		lastfile = fdlastfile(fdp);
4148 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4149 		    n++) {
4150 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4151 				continue;
4152 			xf.xf_fd = n;
4153 			xf.xf_file = (uintptr_t)fp;
4154 			xf.xf_data = (uintptr_t)fp->f_data;
4155 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4156 			xf.xf_type = (uintptr_t)fp->f_type;
4157 			xf.xf_count = refcount_load(&fp->f_count);
4158 			xf.xf_msgcount = 0;
4159 			xf.xf_offset = foffset_get(fp);
4160 			xf.xf_flag = fp->f_flag;
4161 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4162 			if (error)
4163 				break;
4164 		}
4165 		FILEDESC_SUNLOCK(fdp);
4166 		fddrop(fdp);
4167 		if (error)
4168 			break;
4169 	}
4170 	sx_sunlock(&allproc_lock);
4171 	return (error);
4172 }
4173 
4174 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4175     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4176 
4177 #ifdef KINFO_FILE_SIZE
4178 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4179 #endif
4180 
4181 static int
4182 xlate_fflags(int fflags)
4183 {
4184 	static const struct {
4185 		int	fflag;
4186 		int	kf_fflag;
4187 	} fflags_table[] = {
4188 		{ FAPPEND, KF_FLAG_APPEND },
4189 		{ FASYNC, KF_FLAG_ASYNC },
4190 		{ FFSYNC, KF_FLAG_FSYNC },
4191 		{ FHASLOCK, KF_FLAG_HASLOCK },
4192 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4193 		{ FREAD, KF_FLAG_READ },
4194 		{ FWRITE, KF_FLAG_WRITE },
4195 		{ O_CREAT, KF_FLAG_CREAT },
4196 		{ O_DIRECT, KF_FLAG_DIRECT },
4197 		{ O_EXCL, KF_FLAG_EXCL },
4198 		{ O_EXEC, KF_FLAG_EXEC },
4199 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4200 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4201 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4202 		{ O_TRUNC, KF_FLAG_TRUNC }
4203 	};
4204 	unsigned int i;
4205 	int kflags;
4206 
4207 	kflags = 0;
4208 	for (i = 0; i < nitems(fflags_table); i++)
4209 		if (fflags & fflags_table[i].fflag)
4210 			kflags |=  fflags_table[i].kf_fflag;
4211 	return (kflags);
4212 }
4213 
4214 /* Trim unused data from kf_path by truncating the structure size. */
4215 void
4216 pack_kinfo(struct kinfo_file *kif)
4217 {
4218 
4219 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4220 	    strlen(kif->kf_path) + 1;
4221 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4222 }
4223 
4224 static void
4225 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4226     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4227 {
4228 	int error;
4229 
4230 	bzero(kif, sizeof(*kif));
4231 
4232 	/* Set a default type to allow for empty fill_kinfo() methods. */
4233 	kif->kf_type = KF_TYPE_UNKNOWN;
4234 	kif->kf_flags = xlate_fflags(fp->f_flag);
4235 	if (rightsp != NULL)
4236 		kif->kf_cap_rights = *rightsp;
4237 	else
4238 		cap_rights_init_zero(&kif->kf_cap_rights);
4239 	kif->kf_fd = fd;
4240 	kif->kf_ref_count = refcount_load(&fp->f_count);
4241 	kif->kf_offset = foffset_get(fp);
4242 
4243 	/*
4244 	 * This may drop the filedesc lock, so the 'fp' cannot be
4245 	 * accessed after this call.
4246 	 */
4247 	error = fo_fill_kinfo(fp, kif, fdp);
4248 	if (error == 0)
4249 		kif->kf_status |= KF_ATTR_VALID;
4250 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4251 		pack_kinfo(kif);
4252 	else
4253 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4254 }
4255 
4256 static void
4257 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4258     struct kinfo_file *kif, int flags)
4259 {
4260 	int error;
4261 
4262 	bzero(kif, sizeof(*kif));
4263 
4264 	kif->kf_type = KF_TYPE_VNODE;
4265 	error = vn_fill_kinfo_vnode(vp, kif);
4266 	if (error == 0)
4267 		kif->kf_status |= KF_ATTR_VALID;
4268 	kif->kf_flags = xlate_fflags(fflags);
4269 	cap_rights_init_zero(&kif->kf_cap_rights);
4270 	kif->kf_fd = fd;
4271 	kif->kf_ref_count = -1;
4272 	kif->kf_offset = -1;
4273 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4274 		pack_kinfo(kif);
4275 	else
4276 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4277 	vrele(vp);
4278 }
4279 
4280 struct export_fd_buf {
4281 	struct filedesc		*fdp;
4282 	struct pwddesc	*pdp;
4283 	struct sbuf 		*sb;
4284 	ssize_t			remainder;
4285 	struct kinfo_file	kif;
4286 	int			flags;
4287 };
4288 
4289 static int
4290 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4291 {
4292 	struct kinfo_file *kif;
4293 
4294 	kif = &efbuf->kif;
4295 	if (efbuf->remainder != -1) {
4296 		if (efbuf->remainder < kif->kf_structsize) {
4297 			/* Terminate export. */
4298 			efbuf->remainder = 0;
4299 			return (0);
4300 		}
4301 		efbuf->remainder -= kif->kf_structsize;
4302 	}
4303 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4304 }
4305 
4306 static int
4307 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4308     struct export_fd_buf *efbuf)
4309 {
4310 	int error;
4311 
4312 	if (efbuf->remainder == 0)
4313 		return (0);
4314 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4315 	    efbuf->flags);
4316 	FILEDESC_SUNLOCK(efbuf->fdp);
4317 	error = export_kinfo_to_sb(efbuf);
4318 	FILEDESC_SLOCK(efbuf->fdp);
4319 	return (error);
4320 }
4321 
4322 static int
4323 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4324     struct export_fd_buf *efbuf)
4325 {
4326 	int error;
4327 
4328 	if (efbuf->remainder == 0)
4329 		return (0);
4330 	if (efbuf->pdp != NULL)
4331 		PWDDESC_XUNLOCK(efbuf->pdp);
4332 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4333 	error = export_kinfo_to_sb(efbuf);
4334 	if (efbuf->pdp != NULL)
4335 		PWDDESC_XLOCK(efbuf->pdp);
4336 	return (error);
4337 }
4338 
4339 /*
4340  * Store a process file descriptor information to sbuf.
4341  *
4342  * Takes a locked proc as argument, and returns with the proc unlocked.
4343  */
4344 int
4345 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4346     int flags)
4347 {
4348 	struct file *fp;
4349 	struct filedesc *fdp;
4350 	struct pwddesc *pdp;
4351 	struct export_fd_buf *efbuf;
4352 	struct vnode *cttyvp, *textvp, *tracevp;
4353 	struct pwd *pwd;
4354 	int error, i, lastfile;
4355 	cap_rights_t rights;
4356 
4357 	PROC_LOCK_ASSERT(p, MA_OWNED);
4358 
4359 	/* ktrace vnode */
4360 	tracevp = p->p_tracevp;
4361 	if (tracevp != NULL)
4362 		vrefact(tracevp);
4363 	/* text vnode */
4364 	textvp = p->p_textvp;
4365 	if (textvp != NULL)
4366 		vrefact(textvp);
4367 	/* Controlling tty. */
4368 	cttyvp = NULL;
4369 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4370 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4371 		if (cttyvp != NULL)
4372 			vrefact(cttyvp);
4373 	}
4374 	fdp = fdhold(p);
4375 	pdp = pdhold(p);
4376 	PROC_UNLOCK(p);
4377 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4378 	efbuf->fdp = NULL;
4379 	efbuf->pdp = NULL;
4380 	efbuf->sb = sb;
4381 	efbuf->remainder = maxlen;
4382 	efbuf->flags = flags;
4383 	if (tracevp != NULL)
4384 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4385 		    efbuf);
4386 	if (textvp != NULL)
4387 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4388 	if (cttyvp != NULL)
4389 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4390 		    efbuf);
4391 	error = 0;
4392 	if (pdp == NULL || fdp == NULL)
4393 		goto fail;
4394 	efbuf->fdp = fdp;
4395 	efbuf->pdp = pdp;
4396 	PWDDESC_XLOCK(pdp);
4397 	pwd = pwd_hold_pwddesc(pdp);
4398 	if (pwd != NULL) {
4399 		/* working directory */
4400 		if (pwd->pwd_cdir != NULL) {
4401 			vrefact(pwd->pwd_cdir);
4402 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4403 		}
4404 		/* root directory */
4405 		if (pwd->pwd_rdir != NULL) {
4406 			vrefact(pwd->pwd_rdir);
4407 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4408 		}
4409 		/* jail directory */
4410 		if (pwd->pwd_jdir != NULL) {
4411 			vrefact(pwd->pwd_jdir);
4412 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4413 		}
4414 	}
4415 	PWDDESC_XUNLOCK(pdp);
4416 	if (pwd != NULL)
4417 		pwd_drop(pwd);
4418 	FILEDESC_SLOCK(fdp);
4419 	lastfile = fdlastfile(fdp);
4420 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4421 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4422 			continue;
4423 #ifdef CAPABILITIES
4424 		rights = *cap_rights(fdp, i);
4425 #else /* !CAPABILITIES */
4426 		rights = cap_no_rights;
4427 #endif
4428 		/*
4429 		 * Create sysctl entry.  It is OK to drop the filedesc
4430 		 * lock inside of export_file_to_sb() as we will
4431 		 * re-validate and re-evaluate its properties when the
4432 		 * loop continues.
4433 		 */
4434 		error = export_file_to_sb(fp, i, &rights, efbuf);
4435 		if (error != 0 || efbuf->remainder == 0)
4436 			break;
4437 	}
4438 	FILEDESC_SUNLOCK(fdp);
4439 fail:
4440 	if (fdp != NULL)
4441 		fddrop(fdp);
4442 	if (pdp != NULL)
4443 		pddrop(pdp);
4444 	free(efbuf, M_TEMP);
4445 	return (error);
4446 }
4447 
4448 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4449 
4450 /*
4451  * Get per-process file descriptors for use by procstat(1), et al.
4452  */
4453 static int
4454 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4455 {
4456 	struct sbuf sb;
4457 	struct proc *p;
4458 	ssize_t maxlen;
4459 	int error, error2, *name;
4460 
4461 	name = (int *)arg1;
4462 
4463 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4464 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4465 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4466 	if (error != 0) {
4467 		sbuf_delete(&sb);
4468 		return (error);
4469 	}
4470 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4471 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4472 	    KERN_FILEDESC_PACK_KINFO);
4473 	error2 = sbuf_finish(&sb);
4474 	sbuf_delete(&sb);
4475 	return (error != 0 ? error : error2);
4476 }
4477 
4478 #ifdef COMPAT_FREEBSD7
4479 #ifdef KINFO_OFILE_SIZE
4480 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4481 #endif
4482 
4483 static void
4484 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4485 {
4486 
4487 	okif->kf_structsize = sizeof(*okif);
4488 	okif->kf_type = kif->kf_type;
4489 	okif->kf_fd = kif->kf_fd;
4490 	okif->kf_ref_count = kif->kf_ref_count;
4491 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4492 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4493 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4494 	okif->kf_offset = kif->kf_offset;
4495 	if (kif->kf_type == KF_TYPE_VNODE)
4496 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4497 	else
4498 		okif->kf_vnode_type = KF_VTYPE_VNON;
4499 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4500 	if (kif->kf_type == KF_TYPE_SOCKET) {
4501 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4502 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4503 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4504 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4505 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4506 	} else {
4507 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4508 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4509 	}
4510 }
4511 
4512 static int
4513 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4514     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4515 {
4516 	int error;
4517 
4518 	vrefact(vp);
4519 	PWDDESC_XUNLOCK(pdp);
4520 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4521 	kinfo_to_okinfo(kif, okif);
4522 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4523 	PWDDESC_XLOCK(pdp);
4524 	return (error);
4525 }
4526 
4527 /*
4528  * Get per-process file descriptors for use by procstat(1), et al.
4529  */
4530 static int
4531 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4532 {
4533 	struct kinfo_ofile *okif;
4534 	struct kinfo_file *kif;
4535 	struct filedesc *fdp;
4536 	struct pwddesc *pdp;
4537 	struct pwd *pwd;
4538 	int error, i, lastfile, *name;
4539 	struct file *fp;
4540 	struct proc *p;
4541 
4542 	name = (int *)arg1;
4543 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4544 	if (error != 0)
4545 		return (error);
4546 	fdp = fdhold(p);
4547 	if (fdp != NULL)
4548 		pdp = pdhold(p);
4549 	PROC_UNLOCK(p);
4550 	if (fdp == NULL || pdp == NULL) {
4551 		if (fdp != NULL)
4552 			fddrop(fdp);
4553 		return (ENOENT);
4554 	}
4555 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4556 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4557 	PWDDESC_XLOCK(pdp);
4558 	pwd = pwd_hold_pwddesc(pdp);
4559 	if (pwd != NULL) {
4560 		if (pwd->pwd_cdir != NULL)
4561 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4562 			    okif, pdp, req);
4563 		if (pwd->pwd_rdir != NULL)
4564 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4565 			    okif, pdp, req);
4566 		if (pwd->pwd_jdir != NULL)
4567 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4568 			    okif, pdp, req);
4569 	}
4570 	PWDDESC_XUNLOCK(pdp);
4571 	if (pwd != NULL)
4572 		pwd_drop(pwd);
4573 	FILEDESC_SLOCK(fdp);
4574 	lastfile = fdlastfile(fdp);
4575 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4576 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4577 			continue;
4578 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4579 		    KERN_FILEDESC_PACK_KINFO);
4580 		FILEDESC_SUNLOCK(fdp);
4581 		kinfo_to_okinfo(kif, okif);
4582 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4583 		FILEDESC_SLOCK(fdp);
4584 		if (error)
4585 			break;
4586 	}
4587 	FILEDESC_SUNLOCK(fdp);
4588 	fddrop(fdp);
4589 	pddrop(pdp);
4590 	free(kif, M_TEMP);
4591 	free(okif, M_TEMP);
4592 	return (0);
4593 }
4594 
4595 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4596     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4597     "Process ofiledesc entries");
4598 #endif	/* COMPAT_FREEBSD7 */
4599 
4600 int
4601 vntype_to_kinfo(int vtype)
4602 {
4603 	struct {
4604 		int	vtype;
4605 		int	kf_vtype;
4606 	} vtypes_table[] = {
4607 		{ VBAD, KF_VTYPE_VBAD },
4608 		{ VBLK, KF_VTYPE_VBLK },
4609 		{ VCHR, KF_VTYPE_VCHR },
4610 		{ VDIR, KF_VTYPE_VDIR },
4611 		{ VFIFO, KF_VTYPE_VFIFO },
4612 		{ VLNK, KF_VTYPE_VLNK },
4613 		{ VNON, KF_VTYPE_VNON },
4614 		{ VREG, KF_VTYPE_VREG },
4615 		{ VSOCK, KF_VTYPE_VSOCK }
4616 	};
4617 	unsigned int i;
4618 
4619 	/*
4620 	 * Perform vtype translation.
4621 	 */
4622 	for (i = 0; i < nitems(vtypes_table); i++)
4623 		if (vtypes_table[i].vtype == vtype)
4624 			return (vtypes_table[i].kf_vtype);
4625 
4626 	return (KF_VTYPE_UNKNOWN);
4627 }
4628 
4629 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4630     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4631     "Process filedesc entries");
4632 
4633 /*
4634  * Store a process current working directory information to sbuf.
4635  *
4636  * Takes a locked proc as argument, and returns with the proc unlocked.
4637  */
4638 int
4639 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4640 {
4641 	struct pwddesc *pdp;
4642 	struct pwd *pwd;
4643 	struct export_fd_buf *efbuf;
4644 	struct vnode *cdir;
4645 	int error;
4646 
4647 	PROC_LOCK_ASSERT(p, MA_OWNED);
4648 
4649 	pdp = pdhold(p);
4650 	PROC_UNLOCK(p);
4651 	if (pdp == NULL)
4652 		return (EINVAL);
4653 
4654 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4655 	efbuf->pdp = pdp;
4656 	efbuf->sb = sb;
4657 	efbuf->remainder = maxlen;
4658 
4659 	PWDDESC_XLOCK(pdp);
4660 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4661 	cdir = pwd->pwd_cdir;
4662 	if (cdir == NULL) {
4663 		error = EINVAL;
4664 	} else {
4665 		vrefact(cdir);
4666 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4667 	}
4668 	PWDDESC_XUNLOCK(pdp);
4669 	pddrop(pdp);
4670 	free(efbuf, M_TEMP);
4671 	return (error);
4672 }
4673 
4674 /*
4675  * Get per-process current working directory.
4676  */
4677 static int
4678 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4679 {
4680 	struct sbuf sb;
4681 	struct proc *p;
4682 	ssize_t maxlen;
4683 	int error, error2, *name;
4684 
4685 	name = (int *)arg1;
4686 
4687 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4688 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4689 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4690 	if (error != 0) {
4691 		sbuf_delete(&sb);
4692 		return (error);
4693 	}
4694 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4695 	error = kern_proc_cwd_out(p, &sb, maxlen);
4696 	error2 = sbuf_finish(&sb);
4697 	sbuf_delete(&sb);
4698 	return (error != 0 ? error : error2);
4699 }
4700 
4701 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4702     sysctl_kern_proc_cwd, "Process current working directory");
4703 
4704 #ifdef DDB
4705 /*
4706  * For the purposes of debugging, generate a human-readable string for the
4707  * file type.
4708  */
4709 static const char *
4710 file_type_to_name(short type)
4711 {
4712 
4713 	switch (type) {
4714 	case 0:
4715 		return ("zero");
4716 	case DTYPE_VNODE:
4717 		return ("vnode");
4718 	case DTYPE_SOCKET:
4719 		return ("socket");
4720 	case DTYPE_PIPE:
4721 		return ("pipe");
4722 	case DTYPE_FIFO:
4723 		return ("fifo");
4724 	case DTYPE_KQUEUE:
4725 		return ("kqueue");
4726 	case DTYPE_CRYPTO:
4727 		return ("crypto");
4728 	case DTYPE_MQUEUE:
4729 		return ("mqueue");
4730 	case DTYPE_SHM:
4731 		return ("shm");
4732 	case DTYPE_SEM:
4733 		return ("ksem");
4734 	case DTYPE_PTS:
4735 		return ("pts");
4736 	case DTYPE_DEV:
4737 		return ("dev");
4738 	case DTYPE_PROCDESC:
4739 		return ("proc");
4740 	case DTYPE_EVENTFD:
4741 		return ("eventfd");
4742 	case DTYPE_LINUXTFD:
4743 		return ("ltimer");
4744 	default:
4745 		return ("unkn");
4746 	}
4747 }
4748 
4749 /*
4750  * For the purposes of debugging, identify a process (if any, perhaps one of
4751  * many) that references the passed file in its file descriptor array. Return
4752  * NULL if none.
4753  */
4754 static struct proc *
4755 file_to_first_proc(struct file *fp)
4756 {
4757 	struct filedesc *fdp;
4758 	struct proc *p;
4759 	int n;
4760 
4761 	FOREACH_PROC_IN_SYSTEM(p) {
4762 		if (p->p_state == PRS_NEW)
4763 			continue;
4764 		fdp = p->p_fd;
4765 		if (fdp == NULL)
4766 			continue;
4767 		for (n = 0; n < fdp->fd_nfiles; n++) {
4768 			if (fp == fdp->fd_ofiles[n].fde_file)
4769 				return (p);
4770 		}
4771 	}
4772 	return (NULL);
4773 }
4774 
4775 static void
4776 db_print_file(struct file *fp, int header)
4777 {
4778 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4779 	struct proc *p;
4780 
4781 	if (header)
4782 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4783 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4784 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4785 		    "FCmd");
4786 	p = file_to_first_proc(fp);
4787 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4788 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4789 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4790 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4791 
4792 #undef XPTRWIDTH
4793 }
4794 
4795 DB_SHOW_COMMAND(file, db_show_file)
4796 {
4797 	struct file *fp;
4798 
4799 	if (!have_addr) {
4800 		db_printf("usage: show file <addr>\n");
4801 		return;
4802 	}
4803 	fp = (struct file *)addr;
4804 	db_print_file(fp, 1);
4805 }
4806 
4807 DB_SHOW_COMMAND(files, db_show_files)
4808 {
4809 	struct filedesc *fdp;
4810 	struct file *fp;
4811 	struct proc *p;
4812 	int header;
4813 	int n;
4814 
4815 	header = 1;
4816 	FOREACH_PROC_IN_SYSTEM(p) {
4817 		if (p->p_state == PRS_NEW)
4818 			continue;
4819 		if ((fdp = p->p_fd) == NULL)
4820 			continue;
4821 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4822 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4823 				continue;
4824 			db_print_file(fp, header);
4825 			header = 0;
4826 		}
4827 	}
4828 }
4829 #endif
4830 
4831 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4832     &maxfilesperproc, 0, "Maximum files allowed open per process");
4833 
4834 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4835     &maxfiles, 0, "Maximum number of files");
4836 
4837 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4838     &openfiles, 0, "System-wide number of open files");
4839 
4840 /* ARGSUSED*/
4841 static void
4842 filelistinit(void *dummy)
4843 {
4844 
4845 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4846 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4847 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4848 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4849 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4850 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4851 	/*
4852 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4853 	 * the vnode zone.
4854 	 */
4855 	vfs_smr = uma_zone_get_smr(pwd_zone);
4856 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4857 }
4858 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4859 
4860 /*-------------------------------------------------------------------*/
4861 
4862 static int
4863 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4864     int flags, struct thread *td)
4865 {
4866 
4867 	return (EBADF);
4868 }
4869 
4870 static int
4871 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4872     struct thread *td)
4873 {
4874 
4875 	return (EINVAL);
4876 }
4877 
4878 static int
4879 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4880     struct thread *td)
4881 {
4882 
4883 	return (EBADF);
4884 }
4885 
4886 static int
4887 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4888     struct thread *td)
4889 {
4890 
4891 	return (0);
4892 }
4893 
4894 static int
4895 badfo_kqfilter(struct file *fp, struct knote *kn)
4896 {
4897 
4898 	return (EBADF);
4899 }
4900 
4901 static int
4902 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4903     struct thread *td)
4904 {
4905 
4906 	return (EBADF);
4907 }
4908 
4909 static int
4910 badfo_close(struct file *fp, struct thread *td)
4911 {
4912 
4913 	return (0);
4914 }
4915 
4916 static int
4917 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4918     struct thread *td)
4919 {
4920 
4921 	return (EBADF);
4922 }
4923 
4924 static int
4925 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4926     struct thread *td)
4927 {
4928 
4929 	return (EBADF);
4930 }
4931 
4932 static int
4933 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4934     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4935     struct thread *td)
4936 {
4937 
4938 	return (EBADF);
4939 }
4940 
4941 static int
4942 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4943 {
4944 
4945 	return (0);
4946 }
4947 
4948 struct fileops badfileops = {
4949 	.fo_read = badfo_readwrite,
4950 	.fo_write = badfo_readwrite,
4951 	.fo_truncate = badfo_truncate,
4952 	.fo_ioctl = badfo_ioctl,
4953 	.fo_poll = badfo_poll,
4954 	.fo_kqfilter = badfo_kqfilter,
4955 	.fo_stat = badfo_stat,
4956 	.fo_close = badfo_close,
4957 	.fo_chmod = badfo_chmod,
4958 	.fo_chown = badfo_chown,
4959 	.fo_sendfile = badfo_sendfile,
4960 	.fo_fill_kinfo = badfo_fill_kinfo,
4961 };
4962 
4963 int
4964 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4965     int flags, struct thread *td)
4966 {
4967 
4968 	return (EOPNOTSUPP);
4969 }
4970 
4971 int
4972 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4973     struct thread *td)
4974 {
4975 
4976 	return (EINVAL);
4977 }
4978 
4979 int
4980 invfo_ioctl(struct file *fp, u_long com, void *data,
4981     struct ucred *active_cred, struct thread *td)
4982 {
4983 
4984 	return (ENOTTY);
4985 }
4986 
4987 int
4988 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4989     struct thread *td)
4990 {
4991 
4992 	return (poll_no_poll(events));
4993 }
4994 
4995 int
4996 invfo_kqfilter(struct file *fp, struct knote *kn)
4997 {
4998 
4999 	return (EINVAL);
5000 }
5001 
5002 int
5003 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5004     struct thread *td)
5005 {
5006 
5007 	return (EINVAL);
5008 }
5009 
5010 int
5011 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5012     struct thread *td)
5013 {
5014 
5015 	return (EINVAL);
5016 }
5017 
5018 int
5019 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5020     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5021     struct thread *td)
5022 {
5023 
5024 	return (EINVAL);
5025 }
5026 
5027 /*-------------------------------------------------------------------*/
5028 
5029 /*
5030  * File Descriptor pseudo-device driver (/dev/fd/).
5031  *
5032  * Opening minor device N dup()s the file (if any) connected to file
5033  * descriptor N belonging to the calling process.  Note that this driver
5034  * consists of only the ``open()'' routine, because all subsequent
5035  * references to this file will be direct to the other driver.
5036  *
5037  * XXX: we could give this one a cloning event handler if necessary.
5038  */
5039 
5040 /* ARGSUSED */
5041 static int
5042 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5043 {
5044 
5045 	/*
5046 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5047 	 * the file descriptor being sought for duplication. The error
5048 	 * return ensures that the vnode for this device will be released
5049 	 * by vn_open. Open will detect this special error and take the
5050 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5051 	 * will simply report the error.
5052 	 */
5053 	td->td_dupfd = dev2unit(dev);
5054 	return (ENODEV);
5055 }
5056 
5057 static struct cdevsw fildesc_cdevsw = {
5058 	.d_version =	D_VERSION,
5059 	.d_open =	fdopen,
5060 	.d_name =	"FD",
5061 };
5062 
5063 static void
5064 fildesc_drvinit(void *unused)
5065 {
5066 	struct cdev *dev;
5067 
5068 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5069 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5070 	make_dev_alias(dev, "stdin");
5071 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5072 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5073 	make_dev_alias(dev, "stdout");
5074 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5075 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5076 	make_dev_alias(dev, "stderr");
5077 }
5078 
5079 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5080