xref: /freebsd/sys/kern/kern_descrip.c (revision 13ec1e3155c7e9bf037b12af186351b7fa9b9450)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/poll.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sbuf.h>
71 #include <sys/signalvar.h>
72 #include <sys/kdb.h>
73 #include <sys/smr.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
106 VFS_SMR_DECLARE;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, bool holdleaders, bool audit);
110 static void	export_file_to_kinfo(struct file *fp, int fd,
111 		    cap_rights_t *rightsp, struct kinfo_file *kif,
112 		    struct filedesc *fdp, int flags);
113 static int	fd_first_free(struct filedesc *fdp, int low, int size);
114 static void	fdgrowtable(struct filedesc *fdp, int nfd);
115 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
116 static void	fdunused(struct filedesc *fdp, int fd);
117 static void	fdused(struct filedesc *fdp, int fd);
118 static int	fget_unlocked_seq(struct thread *td, int fd,
119 		    cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp);
120 static int	getmaxfd(struct thread *td);
121 static u_long	*filecaps_copy_prep(const struct filecaps *src);
122 static void	filecaps_copy_finish(const struct filecaps *src,
123 		    struct filecaps *dst, u_long *ioctls);
124 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
125 static void	filecaps_free_finish(u_long *ioctls);
126 
127 static struct pwd *pwd_alloc(void);
128 
129 /*
130  * Each process has:
131  *
132  * - An array of open file descriptors (fd_ofiles)
133  * - An array of file flags (fd_ofileflags)
134  * - A bitmap recording which descriptors are in use (fd_map)
135  *
136  * A process starts out with NDFILE descriptors.  The value of NDFILE has
137  * been selected based the historical limit of 20 open files, and an
138  * assumption that the majority of processes, especially short-lived
139  * processes like shells, will never need more.
140  *
141  * If this initial allocation is exhausted, a larger descriptor table and
142  * map are allocated dynamically, and the pointers in the process's struct
143  * filedesc are updated to point to those.  This is repeated every time
144  * the process runs out of file descriptors (provided it hasn't hit its
145  * resource limit).
146  *
147  * Since threads may hold references to individual descriptor table
148  * entries, the tables are never freed.  Instead, they are placed on a
149  * linked list and freed only when the struct filedesc is released.
150  */
151 #define NDFILE		20
152 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
153 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
154 #define NDSLOT(x)	((x) / NDENTRIES)
155 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
156 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
157 
158 /*
159  * SLIST entry used to keep track of ofiles which must be reclaimed when
160  * the process exits.
161  */
162 struct freetable {
163 	struct fdescenttbl *ft_table;
164 	SLIST_ENTRY(freetable) ft_next;
165 };
166 
167 /*
168  * Initial allocation: a filedesc structure + the head of SLIST used to
169  * keep track of old ofiles + enough space for NDFILE descriptors.
170  */
171 
172 struct fdescenttbl0 {
173 	int	fdt_nfiles;
174 	struct	filedescent fdt_ofiles[NDFILE];
175 };
176 
177 struct filedesc0 {
178 	struct filedesc fd_fd;
179 	SLIST_HEAD(, freetable) fd_free;
180 	struct	fdescenttbl0 fd_dfiles;
181 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
182 };
183 
184 /*
185  * Descriptor management.
186  */
187 static int __exclusive_cache_line openfiles; /* actual number of open files */
188 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
189 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
190 
191 /*
192  * If low >= size, just return low. Otherwise find the first zero bit in the
193  * given bitmap, starting at low and not exceeding size - 1. Return size if
194  * not found.
195  */
196 static int
197 fd_first_free(struct filedesc *fdp, int low, int size)
198 {
199 	NDSLOTTYPE *map = fdp->fd_map;
200 	NDSLOTTYPE mask;
201 	int off, maxoff;
202 
203 	if (low >= size)
204 		return (low);
205 
206 	off = NDSLOT(low);
207 	if (low % NDENTRIES) {
208 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
209 		if ((mask &= ~map[off]) != 0UL)
210 			return (off * NDENTRIES + ffsl(mask) - 1);
211 		++off;
212 	}
213 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
214 		if (map[off] != ~0UL)
215 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
216 	return (size);
217 }
218 
219 /*
220  * Find the last used fd.
221  *
222  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
223  * Otherwise use fdlastfile.
224  */
225 int
226 fdlastfile_single(struct filedesc *fdp)
227 {
228 	NDSLOTTYPE *map = fdp->fd_map;
229 	int off, minoff;
230 
231 	off = NDSLOT(fdp->fd_nfiles - 1);
232 	for (minoff = NDSLOT(0); off >= minoff; --off)
233 		if (map[off] != 0)
234 			return (off * NDENTRIES + flsl(map[off]) - 1);
235 	return (-1);
236 }
237 
238 int
239 fdlastfile(struct filedesc *fdp)
240 {
241 
242 	FILEDESC_LOCK_ASSERT(fdp);
243 	return (fdlastfile_single(fdp));
244 }
245 
246 static int
247 fdisused(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
251 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
252 
253 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
254 }
255 
256 /*
257  * Mark a file descriptor as used.
258  */
259 static void
260 fdused_init(struct filedesc *fdp, int fd)
261 {
262 
263 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
264 
265 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
266 }
267 
268 static void
269 fdused(struct filedesc *fdp, int fd)
270 {
271 
272 	FILEDESC_XLOCK_ASSERT(fdp);
273 
274 	fdused_init(fdp, fd);
275 	if (fd == fdp->fd_freefile)
276 		fdp->fd_freefile++;
277 }
278 
279 /*
280  * Mark a file descriptor as unused.
281  */
282 static void
283 fdunused(struct filedesc *fdp, int fd)
284 {
285 
286 	FILEDESC_XLOCK_ASSERT(fdp);
287 
288 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
289 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
290 	    ("fd=%d is still in use", fd));
291 
292 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
293 	if (fd < fdp->fd_freefile)
294 		fdp->fd_freefile = fd;
295 }
296 
297 /*
298  * Free a file descriptor.
299  *
300  * Avoid some work if fdp is about to be destroyed.
301  */
302 static inline void
303 fdefree_last(struct filedescent *fde)
304 {
305 
306 	filecaps_free(&fde->fde_caps);
307 }
308 
309 static inline void
310 fdfree(struct filedesc *fdp, int fd)
311 {
312 	struct filedescent *fde;
313 
314 	FILEDESC_XLOCK_ASSERT(fdp);
315 	fde = &fdp->fd_ofiles[fd];
316 #ifdef CAPABILITIES
317 	seqc_write_begin(&fde->fde_seqc);
318 #endif
319 	fde->fde_file = NULL;
320 #ifdef CAPABILITIES
321 	seqc_write_end(&fde->fde_seqc);
322 #endif
323 	fdefree_last(fde);
324 	fdunused(fdp, fd);
325 }
326 
327 /*
328  * System calls on descriptors.
329  */
330 #ifndef _SYS_SYSPROTO_H_
331 struct getdtablesize_args {
332 	int	dummy;
333 };
334 #endif
335 /* ARGSUSED */
336 int
337 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 {
339 #ifdef	RACCT
340 	uint64_t lim;
341 #endif
342 
343 	td->td_retval[0] = getmaxfd(td);
344 #ifdef	RACCT
345 	PROC_LOCK(td->td_proc);
346 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
347 	PROC_UNLOCK(td->td_proc);
348 	if (lim < td->td_retval[0])
349 		td->td_retval[0] = lim;
350 #endif
351 	return (0);
352 }
353 
354 /*
355  * Duplicate a file descriptor to a particular value.
356  *
357  * Note: keep in mind that a potential race condition exists when closing
358  * descriptors from a shared descriptor table (via rfork).
359  */
360 #ifndef _SYS_SYSPROTO_H_
361 struct dup2_args {
362 	u_int	from;
363 	u_int	to;
364 };
365 #endif
366 /* ARGSUSED */
367 int
368 sys_dup2(struct thread *td, struct dup2_args *uap)
369 {
370 
371 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
372 }
373 
374 /*
375  * Duplicate a file descriptor.
376  */
377 #ifndef _SYS_SYSPROTO_H_
378 struct dup_args {
379 	u_int	fd;
380 };
381 #endif
382 /* ARGSUSED */
383 int
384 sys_dup(struct thread *td, struct dup_args *uap)
385 {
386 
387 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
388 }
389 
390 /*
391  * The file control system call.
392  */
393 #ifndef _SYS_SYSPROTO_H_
394 struct fcntl_args {
395 	int	fd;
396 	int	cmd;
397 	long	arg;
398 };
399 #endif
400 /* ARGSUSED */
401 int
402 sys_fcntl(struct thread *td, struct fcntl_args *uap)
403 {
404 
405 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
406 }
407 
408 int
409 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
410 {
411 	struct flock fl;
412 	struct __oflock ofl;
413 	intptr_t arg1;
414 	int error, newcmd;
415 
416 	error = 0;
417 	newcmd = cmd;
418 	switch (cmd) {
419 	case F_OGETLK:
420 	case F_OSETLK:
421 	case F_OSETLKW:
422 		/*
423 		 * Convert old flock structure to new.
424 		 */
425 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
426 		fl.l_start = ofl.l_start;
427 		fl.l_len = ofl.l_len;
428 		fl.l_pid = ofl.l_pid;
429 		fl.l_type = ofl.l_type;
430 		fl.l_whence = ofl.l_whence;
431 		fl.l_sysid = 0;
432 
433 		switch (cmd) {
434 		case F_OGETLK:
435 			newcmd = F_GETLK;
436 			break;
437 		case F_OSETLK:
438 			newcmd = F_SETLK;
439 			break;
440 		case F_OSETLKW:
441 			newcmd = F_SETLKW;
442 			break;
443 		}
444 		arg1 = (intptr_t)&fl;
445 		break;
446 	case F_GETLK:
447 	case F_SETLK:
448 	case F_SETLKW:
449 	case F_SETLK_REMOTE:
450 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
451 		arg1 = (intptr_t)&fl;
452 		break;
453 	default:
454 		arg1 = arg;
455 		break;
456 	}
457 	if (error)
458 		return (error);
459 	error = kern_fcntl(td, fd, newcmd, arg1);
460 	if (error)
461 		return (error);
462 	if (cmd == F_OGETLK) {
463 		ofl.l_start = fl.l_start;
464 		ofl.l_len = fl.l_len;
465 		ofl.l_pid = fl.l_pid;
466 		ofl.l_type = fl.l_type;
467 		ofl.l_whence = fl.l_whence;
468 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
469 	} else if (cmd == F_GETLK) {
470 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 	}
472 	return (error);
473 }
474 
475 int
476 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
477 {
478 	struct filedesc *fdp;
479 	struct flock *flp;
480 	struct file *fp, *fp2;
481 	struct filedescent *fde;
482 	struct proc *p;
483 	struct vnode *vp;
484 	struct mount *mp;
485 	struct kinfo_file *kif;
486 	int error, flg, kif_sz, seals, tmp;
487 	uint64_t bsize;
488 	off_t foffset;
489 
490 	error = 0;
491 	flg = F_POSIX;
492 	p = td->td_proc;
493 	fdp = p->p_fd;
494 
495 	AUDIT_ARG_FD(cmd);
496 	AUDIT_ARG_CMD(cmd);
497 	switch (cmd) {
498 	case F_DUPFD:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
501 		break;
502 
503 	case F_DUPFD_CLOEXEC:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
511 		break;
512 
513 	case F_DUP2FD_CLOEXEC:
514 		tmp = arg;
515 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
516 		break;
517 
518 	case F_GETFD:
519 		error = EBADF;
520 		FILEDESC_SLOCK(fdp);
521 		fde = fdeget_locked(fdp, fd);
522 		if (fde != NULL) {
523 			td->td_retval[0] =
524 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
525 			error = 0;
526 		}
527 		FILEDESC_SUNLOCK(fdp);
528 		break;
529 
530 	case F_SETFD:
531 		error = EBADF;
532 		FILEDESC_XLOCK(fdp);
533 		fde = fdeget_locked(fdp, fd);
534 		if (fde != NULL) {
535 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
536 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
537 			error = 0;
538 		}
539 		FILEDESC_XUNLOCK(fdp);
540 		break;
541 
542 	case F_GETFL:
543 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
544 		if (error != 0)
545 			break;
546 		td->td_retval[0] = OFLAGS(fp->f_flag);
547 		fdrop(fp, td);
548 		break;
549 
550 	case F_SETFL:
551 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
552 		if (error != 0)
553 			break;
554 		if (fp->f_ops == &path_fileops) {
555 			fdrop(fp, td);
556 			error = EBADF;
557 			break;
558 		}
559 		do {
560 			tmp = flg = fp->f_flag;
561 			tmp &= ~FCNTLFLAGS;
562 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
563 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
564 		tmp = fp->f_flag & FNONBLOCK;
565 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
566 		if (error != 0) {
567 			fdrop(fp, td);
568 			break;
569 		}
570 		tmp = fp->f_flag & FASYNC;
571 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
572 		if (error == 0) {
573 			fdrop(fp, td);
574 			break;
575 		}
576 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
577 		tmp = 0;
578 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_GETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
587 		if (error == 0)
588 			td->td_retval[0] = tmp;
589 		fdrop(fp, td);
590 		break;
591 
592 	case F_SETOWN:
593 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
594 		if (error != 0)
595 			break;
596 		tmp = arg;
597 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
598 		fdrop(fp, td);
599 		break;
600 
601 	case F_SETLK_REMOTE:
602 		error = priv_check(td, PRIV_NFS_LOCKD);
603 		if (error != 0)
604 			return (error);
605 		flg = F_REMOTE;
606 		goto do_setlk;
607 
608 	case F_SETLKW:
609 		flg |= F_WAIT;
610 		/* FALLTHROUGH F_SETLK */
611 
612 	case F_SETLK:
613 	do_setlk:
614 		flp = (struct flock *)arg;
615 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
616 			error = EINVAL;
617 			break;
618 		}
619 
620 		error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
621 		if (error != 0)
622 			break;
623 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
624 			error = EBADF;
625 			fdrop(fp, td);
626 			break;
627 		}
628 
629 		if (flp->l_whence == SEEK_CUR) {
630 			foffset = foffset_get(fp);
631 			if (foffset < 0 ||
632 			    (flp->l_start > 0 &&
633 			     foffset > OFF_MAX - flp->l_start)) {
634 				error = EOVERFLOW;
635 				fdrop(fp, td);
636 				break;
637 			}
638 			flp->l_start += foffset;
639 		}
640 
641 		vp = fp->f_vnode;
642 		switch (flp->l_type) {
643 		case F_RDLCK:
644 			if ((fp->f_flag & FREAD) == 0) {
645 				error = EBADF;
646 				break;
647 			}
648 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
649 				PROC_LOCK(p->p_leader);
650 				p->p_leader->p_flag |= P_ADVLOCK;
651 				PROC_UNLOCK(p->p_leader);
652 			}
653 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 			    flp, flg);
655 			break;
656 		case F_WRLCK:
657 			if ((fp->f_flag & FWRITE) == 0) {
658 				error = EBADF;
659 				break;
660 			}
661 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
662 				PROC_LOCK(p->p_leader);
663 				p->p_leader->p_flag |= P_ADVLOCK;
664 				PROC_UNLOCK(p->p_leader);
665 			}
666 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
667 			    flp, flg);
668 			break;
669 		case F_UNLCK:
670 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
671 			    flp, flg);
672 			break;
673 		case F_UNLCKSYS:
674 			if (flg != F_REMOTE) {
675 				error = EINVAL;
676 				break;
677 			}
678 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
679 			    F_UNLCKSYS, flp, flg);
680 			break;
681 		default:
682 			error = EINVAL;
683 			break;
684 		}
685 		if (error != 0 || flp->l_type == F_UNLCK ||
686 		    flp->l_type == F_UNLCKSYS) {
687 			fdrop(fp, td);
688 			break;
689 		}
690 
691 		/*
692 		 * Check for a race with close.
693 		 *
694 		 * The vnode is now advisory locked (or unlocked, but this case
695 		 * is not really important) as the caller requested.
696 		 * We had to drop the filedesc lock, so we need to recheck if
697 		 * the descriptor is still valid, because if it was closed
698 		 * in the meantime we need to remove advisory lock from the
699 		 * vnode - close on any descriptor leading to an advisory
700 		 * locked vnode, removes that lock.
701 		 * We will return 0 on purpose in that case, as the result of
702 		 * successful advisory lock might have been externally visible
703 		 * already. This is fine - effectively we pretend to the caller
704 		 * that the closing thread was a bit slower and that the
705 		 * advisory lock succeeded before the close.
706 		 */
707 		error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
708 		if (error != 0) {
709 			fdrop(fp, td);
710 			break;
711 		}
712 		if (fp != fp2) {
713 			flp->l_whence = SEEK_SET;
714 			flp->l_start = 0;
715 			flp->l_len = 0;
716 			flp->l_type = F_UNLCK;
717 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
718 			    F_UNLCK, flp, F_POSIX);
719 		}
720 		fdrop(fp, td);
721 		fdrop(fp2, td);
722 		break;
723 
724 	case F_GETLK:
725 		error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
726 		if (error != 0)
727 			break;
728 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
729 			error = EBADF;
730 			fdrop(fp, td);
731 			break;
732 		}
733 		flp = (struct flock *)arg;
734 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
735 		    flp->l_type != F_UNLCK) {
736 			error = EINVAL;
737 			fdrop(fp, td);
738 			break;
739 		}
740 		if (flp->l_whence == SEEK_CUR) {
741 			foffset = foffset_get(fp);
742 			if ((flp->l_start > 0 &&
743 			    foffset > OFF_MAX - flp->l_start) ||
744 			    (flp->l_start < 0 &&
745 			    foffset < OFF_MIN - flp->l_start)) {
746 				error = EOVERFLOW;
747 				fdrop(fp, td);
748 				break;
749 			}
750 			flp->l_start += foffset;
751 		}
752 		vp = fp->f_vnode;
753 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
754 		    F_POSIX);
755 		fdrop(fp, td);
756 		break;
757 
758 	case F_ADD_SEALS:
759 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
760 		if (error != 0)
761 			break;
762 		error = fo_add_seals(fp, arg);
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_GET_SEALS:
767 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
768 		if (error != 0)
769 			break;
770 		if (fo_get_seals(fp, &seals) == 0)
771 			td->td_retval[0] = seals;
772 		else
773 			error = EINVAL;
774 		fdrop(fp, td);
775 		break;
776 
777 	case F_RDAHEAD:
778 		arg = arg ? 128 * 1024: 0;
779 		/* FALLTHROUGH */
780 	case F_READAHEAD:
781 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
782 		if (error != 0)
783 			break;
784 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
785 			fdrop(fp, td);
786 			error = EBADF;
787 			break;
788 		}
789 		vp = fp->f_vnode;
790 		if (vp->v_type != VREG) {
791 			fdrop(fp, td);
792 			error = ENOTTY;
793 			break;
794 		}
795 
796 		/*
797 		 * Exclusive lock synchronizes against f_seqcount reads and
798 		 * writes in sequential_heuristic().
799 		 */
800 		error = vn_lock(vp, LK_EXCLUSIVE);
801 		if (error != 0) {
802 			fdrop(fp, td);
803 			break;
804 		}
805 		if (arg >= 0) {
806 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
807 			arg = MIN(arg, INT_MAX - bsize + 1);
808 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
809 			    (arg + bsize - 1) / bsize);
810 			atomic_set_int(&fp->f_flag, FRDAHEAD);
811 		} else {
812 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
813 		}
814 		VOP_UNLOCK(vp);
815 		fdrop(fp, td);
816 		break;
817 
818 	case F_ISUNIONSTACK:
819 		/*
820 		 * Check if the vnode is part of a union stack (either the
821 		 * "union" flag from mount(2) or unionfs).
822 		 *
823 		 * Prior to introduction of this op libc's readdir would call
824 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
825 		 * data just to check fs name and a mount flag.
826 		 *
827 		 * Fixing the code to handle everything in the kernel instead
828 		 * is a non-trivial endeavor and has low priority, thus this
829 		 * horrible kludge facilitates the current behavior in a much
830 		 * cheaper manner until someone(tm) sorts this out.
831 		 */
832 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
833 		if (error != 0)
834 			break;
835 		if (fp->f_type != DTYPE_VNODE) {
836 			fdrop(fp, td);
837 			error = EBADF;
838 			break;
839 		}
840 		vp = fp->f_vnode;
841 		/*
842 		 * Since we don't prevent dooming the vnode even non-null mp
843 		 * found can become immediately stale. This is tolerable since
844 		 * mount points are type-stable (providing safe memory access)
845 		 * and any vfs op on this vnode going forward will return an
846 		 * error (meaning return value in this case is meaningless).
847 		 */
848 		mp = atomic_load_ptr(&vp->v_mount);
849 		if (__predict_false(mp == NULL)) {
850 			fdrop(fp, td);
851 			error = EBADF;
852 			break;
853 		}
854 		td->td_retval[0] = 0;
855 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
856 		    mp->mnt_flag & MNT_UNION)
857 			td->td_retval[0] = 1;
858 		fdrop(fp, td);
859 		break;
860 
861 	case F_KINFO:
862 #ifdef CAPABILITY_MODE
863 		if (IN_CAPABILITY_MODE(td)) {
864 			error = ECAPMODE;
865 			break;
866 		}
867 #endif
868 		error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
869 		if (error != 0)
870 			break;
871 		if (kif_sz != sizeof(*kif)) {
872 			error = EINVAL;
873 			break;
874 		}
875 		kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
876 		FILEDESC_SLOCK(fdp);
877 		error = fget_cap_locked(fdp, fd, &cap_fcntl_rights, &fp, NULL);
878 		if (error == 0 && fhold(fp)) {
879 			export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
880 			FILEDESC_SUNLOCK(fdp);
881 			fdrop(fp, td);
882 			if ((kif->kf_status & KF_ATTR_VALID) != 0) {
883 				kif->kf_structsize = sizeof(*kif);
884 				error = copyout(kif, (void *)arg, sizeof(*kif));
885 			} else {
886 				error = EBADF;
887 			}
888 		} else {
889 			FILEDESC_SUNLOCK(fdp);
890 			if (error == 0)
891 				error = EBADF;
892 		}
893 		free(kif, M_TEMP);
894 		break;
895 
896 	default:
897 		error = EINVAL;
898 		break;
899 	}
900 	return (error);
901 }
902 
903 static int
904 getmaxfd(struct thread *td)
905 {
906 
907 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
908 }
909 
910 /*
911  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
912  */
913 int
914 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
915 {
916 	struct filedesc *fdp;
917 	struct filedescent *oldfde, *newfde;
918 	struct proc *p;
919 	struct file *delfp, *oldfp;
920 	u_long *oioctls, *nioctls;
921 	int error, maxfd;
922 
923 	p = td->td_proc;
924 	fdp = p->p_fd;
925 	oioctls = NULL;
926 
927 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
928 	MPASS(mode < FDDUP_LASTMODE);
929 
930 	AUDIT_ARG_FD(old);
931 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
932 
933 	/*
934 	 * Verify we have a valid descriptor to dup from and possibly to
935 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
936 	 * return EINVAL when the new descriptor is out of bounds.
937 	 */
938 	if (old < 0)
939 		return (EBADF);
940 	if (new < 0)
941 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
942 	maxfd = getmaxfd(td);
943 	if (new >= maxfd)
944 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
945 
946 	error = EBADF;
947 	FILEDESC_XLOCK(fdp);
948 	if (fget_locked(fdp, old) == NULL)
949 		goto unlock;
950 	if (mode == FDDUP_FIXED && old == new) {
951 		td->td_retval[0] = new;
952 		if (flags & FDDUP_FLAG_CLOEXEC)
953 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
954 		error = 0;
955 		goto unlock;
956 	}
957 
958 	oldfde = &fdp->fd_ofiles[old];
959 	oldfp = oldfde->fde_file;
960 	if (!fhold(oldfp))
961 		goto unlock;
962 
963 	/*
964 	 * If the caller specified a file descriptor, make sure the file
965 	 * table is large enough to hold it, and grab it.  Otherwise, just
966 	 * allocate a new descriptor the usual way.
967 	 */
968 	switch (mode) {
969 	case FDDUP_NORMAL:
970 	case FDDUP_FCNTL:
971 		if ((error = fdalloc(td, new, &new)) != 0) {
972 			fdrop(oldfp, td);
973 			goto unlock;
974 		}
975 		break;
976 	case FDDUP_FIXED:
977 		if (new >= fdp->fd_nfiles) {
978 			/*
979 			 * The resource limits are here instead of e.g.
980 			 * fdalloc(), because the file descriptor table may be
981 			 * shared between processes, so we can't really use
982 			 * racct_add()/racct_sub().  Instead of counting the
983 			 * number of actually allocated descriptors, just put
984 			 * the limit on the size of the file descriptor table.
985 			 */
986 #ifdef RACCT
987 			if (RACCT_ENABLED()) {
988 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
989 				if (error != 0) {
990 					error = EMFILE;
991 					fdrop(oldfp, td);
992 					goto unlock;
993 				}
994 			}
995 #endif
996 			fdgrowtable_exp(fdp, new + 1);
997 		}
998 		if (!fdisused(fdp, new))
999 			fdused(fdp, new);
1000 		break;
1001 	default:
1002 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1003 	}
1004 
1005 	KASSERT(old != new, ("new fd is same as old"));
1006 
1007 	/* Refetch oldfde because the table may have grown and old one freed. */
1008 	oldfde = &fdp->fd_ofiles[old];
1009 	KASSERT(oldfp == oldfde->fde_file,
1010 	    ("fdt_ofiles shift from growth observed at fd %d",
1011 	    old));
1012 
1013 	newfde = &fdp->fd_ofiles[new];
1014 	delfp = newfde->fde_file;
1015 
1016 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1017 
1018 	/*
1019 	 * Duplicate the source descriptor.
1020 	 */
1021 #ifdef CAPABILITIES
1022 	seqc_write_begin(&newfde->fde_seqc);
1023 #endif
1024 	oioctls = filecaps_free_prep(&newfde->fde_caps);
1025 	fde_copy(oldfde, newfde);
1026 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1027 	    nioctls);
1028 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1029 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1030 	else
1031 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1032 #ifdef CAPABILITIES
1033 	seqc_write_end(&newfde->fde_seqc);
1034 #endif
1035 	td->td_retval[0] = new;
1036 
1037 	error = 0;
1038 
1039 	if (delfp != NULL) {
1040 		(void) closefp(fdp, new, delfp, td, true, false);
1041 		FILEDESC_UNLOCK_ASSERT(fdp);
1042 	} else {
1043 unlock:
1044 		FILEDESC_XUNLOCK(fdp);
1045 	}
1046 
1047 	filecaps_free_finish(oioctls);
1048 	return (error);
1049 }
1050 
1051 static void
1052 sigiofree(struct sigio *sigio)
1053 {
1054 	crfree(sigio->sio_ucred);
1055 	free(sigio, M_SIGIO);
1056 }
1057 
1058 static struct sigio *
1059 funsetown_locked(struct sigio *sigio)
1060 {
1061 	struct proc *p;
1062 	struct pgrp *pg;
1063 
1064 	SIGIO_ASSERT_LOCKED();
1065 
1066 	if (sigio == NULL)
1067 		return (NULL);
1068 	*sigio->sio_myref = NULL;
1069 	if (sigio->sio_pgid < 0) {
1070 		pg = sigio->sio_pgrp;
1071 		PGRP_LOCK(pg);
1072 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1073 		PGRP_UNLOCK(pg);
1074 	} else {
1075 		p = sigio->sio_proc;
1076 		PROC_LOCK(p);
1077 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1078 		PROC_UNLOCK(p);
1079 	}
1080 	return (sigio);
1081 }
1082 
1083 /*
1084  * If sigio is on the list associated with a process or process group,
1085  * disable signalling from the device, remove sigio from the list and
1086  * free sigio.
1087  */
1088 void
1089 funsetown(struct sigio **sigiop)
1090 {
1091 	struct sigio *sigio;
1092 
1093 	/* Racy check, consumers must provide synchronization. */
1094 	if (*sigiop == NULL)
1095 		return;
1096 
1097 	SIGIO_LOCK();
1098 	sigio = funsetown_locked(*sigiop);
1099 	SIGIO_UNLOCK();
1100 	if (sigio != NULL)
1101 		sigiofree(sigio);
1102 }
1103 
1104 /*
1105  * Free a list of sigio structures.  The caller must ensure that new sigio
1106  * structures cannot be added after this point.  For process groups this is
1107  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1108  * as an interlock.
1109  */
1110 void
1111 funsetownlst(struct sigiolst *sigiolst)
1112 {
1113 	struct proc *p;
1114 	struct pgrp *pg;
1115 	struct sigio *sigio, *tmp;
1116 
1117 	/* Racy check. */
1118 	sigio = SLIST_FIRST(sigiolst);
1119 	if (sigio == NULL)
1120 		return;
1121 
1122 	p = NULL;
1123 	pg = NULL;
1124 
1125 	SIGIO_LOCK();
1126 	sigio = SLIST_FIRST(sigiolst);
1127 	if (sigio == NULL) {
1128 		SIGIO_UNLOCK();
1129 		return;
1130 	}
1131 
1132 	/*
1133 	 * Every entry of the list should belong to a single proc or pgrp.
1134 	 */
1135 	if (sigio->sio_pgid < 0) {
1136 		pg = sigio->sio_pgrp;
1137 		sx_assert(&proctree_lock, SX_XLOCKED);
1138 		PGRP_LOCK(pg);
1139 	} else /* if (sigio->sio_pgid > 0) */ {
1140 		p = sigio->sio_proc;
1141 		PROC_LOCK(p);
1142 		KASSERT((p->p_flag & P_WEXIT) != 0,
1143 		    ("%s: process %p is not exiting", __func__, p));
1144 	}
1145 
1146 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1147 		*sigio->sio_myref = NULL;
1148 		if (pg != NULL) {
1149 			KASSERT(sigio->sio_pgid < 0,
1150 			    ("Proc sigio in pgrp sigio list"));
1151 			KASSERT(sigio->sio_pgrp == pg,
1152 			    ("Bogus pgrp in sigio list"));
1153 		} else /* if (p != NULL) */ {
1154 			KASSERT(sigio->sio_pgid > 0,
1155 			    ("Pgrp sigio in proc sigio list"));
1156 			KASSERT(sigio->sio_proc == p,
1157 			    ("Bogus proc in sigio list"));
1158 		}
1159 	}
1160 
1161 	if (pg != NULL)
1162 		PGRP_UNLOCK(pg);
1163 	else
1164 		PROC_UNLOCK(p);
1165 	SIGIO_UNLOCK();
1166 
1167 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1168 		sigiofree(sigio);
1169 }
1170 
1171 /*
1172  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1173  *
1174  * After permission checking, add a sigio structure to the sigio list for
1175  * the process or process group.
1176  */
1177 int
1178 fsetown(pid_t pgid, struct sigio **sigiop)
1179 {
1180 	struct proc *proc;
1181 	struct pgrp *pgrp;
1182 	struct sigio *osigio, *sigio;
1183 	int ret;
1184 
1185 	if (pgid == 0) {
1186 		funsetown(sigiop);
1187 		return (0);
1188 	}
1189 
1190 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1191 	sigio->sio_pgid = pgid;
1192 	sigio->sio_ucred = crhold(curthread->td_ucred);
1193 	sigio->sio_myref = sigiop;
1194 
1195 	ret = 0;
1196 	if (pgid > 0) {
1197 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1198 		SIGIO_LOCK();
1199 		osigio = funsetown_locked(*sigiop);
1200 		if (ret == 0) {
1201 			PROC_LOCK(proc);
1202 			_PRELE(proc);
1203 			if ((proc->p_flag & P_WEXIT) != 0) {
1204 				ret = ESRCH;
1205 			} else if (proc->p_session !=
1206 			    curthread->td_proc->p_session) {
1207 				/*
1208 				 * Policy - Don't allow a process to FSETOWN a
1209 				 * process in another session.
1210 				 *
1211 				 * Remove this test to allow maximum flexibility
1212 				 * or restrict FSETOWN to the current process or
1213 				 * process group for maximum safety.
1214 				 */
1215 				ret = EPERM;
1216 			} else {
1217 				sigio->sio_proc = proc;
1218 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1219 				    sio_pgsigio);
1220 			}
1221 			PROC_UNLOCK(proc);
1222 		}
1223 	} else /* if (pgid < 0) */ {
1224 		sx_slock(&proctree_lock);
1225 		SIGIO_LOCK();
1226 		osigio = funsetown_locked(*sigiop);
1227 		pgrp = pgfind(-pgid);
1228 		if (pgrp == NULL) {
1229 			ret = ESRCH;
1230 		} else {
1231 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1232 				/*
1233 				 * Policy - Don't allow a process to FSETOWN a
1234 				 * process in another session.
1235 				 *
1236 				 * Remove this test to allow maximum flexibility
1237 				 * or restrict FSETOWN to the current process or
1238 				 * process group for maximum safety.
1239 				 */
1240 				ret = EPERM;
1241 			} else {
1242 				sigio->sio_pgrp = pgrp;
1243 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1244 				    sio_pgsigio);
1245 			}
1246 			PGRP_UNLOCK(pgrp);
1247 		}
1248 		sx_sunlock(&proctree_lock);
1249 	}
1250 	if (ret == 0)
1251 		*sigiop = sigio;
1252 	SIGIO_UNLOCK();
1253 	if (osigio != NULL)
1254 		sigiofree(osigio);
1255 	return (ret);
1256 }
1257 
1258 /*
1259  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1260  */
1261 pid_t
1262 fgetown(struct sigio **sigiop)
1263 {
1264 	pid_t pgid;
1265 
1266 	SIGIO_LOCK();
1267 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1268 	SIGIO_UNLOCK();
1269 	return (pgid);
1270 }
1271 
1272 static int
1273 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1274     bool audit)
1275 {
1276 	int error;
1277 
1278 	FILEDESC_XLOCK_ASSERT(fdp);
1279 
1280 	/*
1281 	 * We now hold the fp reference that used to be owned by the
1282 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1283 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1284 	 * added, and deleteing a knote for the new fd.
1285 	 */
1286 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1287 		knote_fdclose(td, fd);
1288 
1289 	/*
1290 	 * We need to notify mqueue if the object is of type mqueue.
1291 	 */
1292 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1293 		mq_fdclose(td, fd, fp);
1294 	FILEDESC_XUNLOCK(fdp);
1295 
1296 #ifdef AUDIT
1297 	if (AUDITING_TD(td) && audit)
1298 		audit_sysclose(td, fd, fp);
1299 #endif
1300 	error = closef(fp, td);
1301 
1302 	/*
1303 	 * All paths leading up to closefp() will have already removed or
1304 	 * replaced the fd in the filedesc table, so a restart would not
1305 	 * operate on the same file.
1306 	 */
1307 	if (error == ERESTART)
1308 		error = EINTR;
1309 
1310 	return (error);
1311 }
1312 
1313 static int
1314 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1315     bool holdleaders, bool audit)
1316 {
1317 	int error;
1318 
1319 	FILEDESC_XLOCK_ASSERT(fdp);
1320 
1321 	if (holdleaders) {
1322 		if (td->td_proc->p_fdtol != NULL) {
1323 			/*
1324 			 * Ask fdfree() to sleep to ensure that all relevant
1325 			 * process leaders can be traversed in closef().
1326 			 */
1327 			fdp->fd_holdleaderscount++;
1328 		} else {
1329 			holdleaders = false;
1330 		}
1331 	}
1332 
1333 	error = closefp_impl(fdp, fd, fp, td, audit);
1334 	if (holdleaders) {
1335 		FILEDESC_XLOCK(fdp);
1336 		fdp->fd_holdleaderscount--;
1337 		if (fdp->fd_holdleaderscount == 0 &&
1338 		    fdp->fd_holdleaderswakeup != 0) {
1339 			fdp->fd_holdleaderswakeup = 0;
1340 			wakeup(&fdp->fd_holdleaderscount);
1341 		}
1342 		FILEDESC_XUNLOCK(fdp);
1343 	}
1344 	return (error);
1345 }
1346 
1347 static int
1348 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1349     bool holdleaders, bool audit)
1350 {
1351 
1352 	FILEDESC_XLOCK_ASSERT(fdp);
1353 
1354 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1355 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1356 	} else {
1357 		return (closefp_impl(fdp, fd, fp, td, audit));
1358 	}
1359 }
1360 
1361 /*
1362  * Close a file descriptor.
1363  */
1364 #ifndef _SYS_SYSPROTO_H_
1365 struct close_args {
1366 	int     fd;
1367 };
1368 #endif
1369 /* ARGSUSED */
1370 int
1371 sys_close(struct thread *td, struct close_args *uap)
1372 {
1373 
1374 	return (kern_close(td, uap->fd));
1375 }
1376 
1377 int
1378 kern_close(struct thread *td, int fd)
1379 {
1380 	struct filedesc *fdp;
1381 	struct file *fp;
1382 
1383 	fdp = td->td_proc->p_fd;
1384 
1385 	FILEDESC_XLOCK(fdp);
1386 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1387 		FILEDESC_XUNLOCK(fdp);
1388 		return (EBADF);
1389 	}
1390 	fdfree(fdp, fd);
1391 
1392 	/* closefp() drops the FILEDESC lock for us. */
1393 	return (closefp(fdp, fd, fp, td, true, true));
1394 }
1395 
1396 int
1397 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1398 {
1399 	struct filedesc *fdp;
1400 	const struct fdescenttbl *fdt;
1401 	struct file *fp;
1402 	int fd;
1403 
1404 	/*
1405 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1406 	 * open should not be a usage error.  From a close_range() perspective,
1407 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1408 	 * be a usage error as all fd above 3 are in-fact already closed.
1409 	 */
1410 	if (highfd < lowfd) {
1411 		return (EINVAL);
1412 	}
1413 
1414 	fdp = td->td_proc->p_fd;
1415 	FILEDESC_XLOCK(fdp);
1416 	fdt = atomic_load_ptr(&fdp->fd_files);
1417 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1418 	fd = lowfd;
1419 	if (__predict_false(fd > highfd)) {
1420 		goto out_locked;
1421 	}
1422 	for (;;) {
1423 		fp = fdt->fdt_ofiles[fd].fde_file;
1424 		if (fp == NULL) {
1425 			if (fd == highfd)
1426 				goto out_locked;
1427 		} else {
1428 			fdfree(fdp, fd);
1429 			(void) closefp(fdp, fd, fp, td, true, true);
1430 			if (fd == highfd)
1431 				goto out_unlocked;
1432 			FILEDESC_XLOCK(fdp);
1433 			fdt = atomic_load_ptr(&fdp->fd_files);
1434 		}
1435 		fd++;
1436 	}
1437 out_locked:
1438 	FILEDESC_XUNLOCK(fdp);
1439 out_unlocked:
1440 	return (0);
1441 }
1442 
1443 #ifndef _SYS_SYSPROTO_H_
1444 struct close_range_args {
1445 	u_int	lowfd;
1446 	u_int	highfd;
1447 	int	flags;
1448 };
1449 #endif
1450 int
1451 sys_close_range(struct thread *td, struct close_range_args *uap)
1452 {
1453 
1454 	AUDIT_ARG_FD(uap->lowfd);
1455 	AUDIT_ARG_CMD(uap->highfd);
1456 	AUDIT_ARG_FFLAGS(uap->flags);
1457 
1458 	/* No flags currently defined */
1459 	if (uap->flags != 0)
1460 		return (EINVAL);
1461 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1462 }
1463 
1464 #ifdef COMPAT_FREEBSD12
1465 /*
1466  * Close open file descriptors.
1467  */
1468 #ifndef _SYS_SYSPROTO_H_
1469 struct freebsd12_closefrom_args {
1470 	int	lowfd;
1471 };
1472 #endif
1473 /* ARGSUSED */
1474 int
1475 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1476 {
1477 	u_int lowfd;
1478 
1479 	AUDIT_ARG_FD(uap->lowfd);
1480 
1481 	/*
1482 	 * Treat negative starting file descriptor values identical to
1483 	 * closefrom(0) which closes all files.
1484 	 */
1485 	lowfd = MAX(0, uap->lowfd);
1486 	return (kern_close_range(td, lowfd, ~0U));
1487 }
1488 #endif	/* COMPAT_FREEBSD12 */
1489 
1490 #if defined(COMPAT_43)
1491 /*
1492  * Return status information about a file descriptor.
1493  */
1494 #ifndef _SYS_SYSPROTO_H_
1495 struct ofstat_args {
1496 	int	fd;
1497 	struct	ostat *sb;
1498 };
1499 #endif
1500 /* ARGSUSED */
1501 int
1502 ofstat(struct thread *td, struct ofstat_args *uap)
1503 {
1504 	struct ostat oub;
1505 	struct stat ub;
1506 	int error;
1507 
1508 	error = kern_fstat(td, uap->fd, &ub);
1509 	if (error == 0) {
1510 		cvtstat(&ub, &oub);
1511 		error = copyout(&oub, uap->sb, sizeof(oub));
1512 	}
1513 	return (error);
1514 }
1515 #endif /* COMPAT_43 */
1516 
1517 #if defined(COMPAT_FREEBSD11)
1518 int
1519 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1520 {
1521 	struct stat sb;
1522 	struct freebsd11_stat osb;
1523 	int error;
1524 
1525 	error = kern_fstat(td, uap->fd, &sb);
1526 	if (error != 0)
1527 		return (error);
1528 	error = freebsd11_cvtstat(&sb, &osb);
1529 	if (error == 0)
1530 		error = copyout(&osb, uap->sb, sizeof(osb));
1531 	return (error);
1532 }
1533 #endif	/* COMPAT_FREEBSD11 */
1534 
1535 /*
1536  * Return status information about a file descriptor.
1537  */
1538 #ifndef _SYS_SYSPROTO_H_
1539 struct fstat_args {
1540 	int	fd;
1541 	struct	stat *sb;
1542 };
1543 #endif
1544 /* ARGSUSED */
1545 int
1546 sys_fstat(struct thread *td, struct fstat_args *uap)
1547 {
1548 	struct stat ub;
1549 	int error;
1550 
1551 	error = kern_fstat(td, uap->fd, &ub);
1552 	if (error == 0)
1553 		error = copyout(&ub, uap->sb, sizeof(ub));
1554 	return (error);
1555 }
1556 
1557 int
1558 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1559 {
1560 	struct file *fp;
1561 	int error;
1562 
1563 	AUDIT_ARG_FD(fd);
1564 
1565 	error = fget(td, fd, &cap_fstat_rights, &fp);
1566 	if (__predict_false(error != 0))
1567 		return (error);
1568 
1569 	AUDIT_ARG_FILE(td->td_proc, fp);
1570 
1571 	error = fo_stat(fp, sbp, td->td_ucred);
1572 	fdrop(fp, td);
1573 #ifdef __STAT_TIME_T_EXT
1574 	sbp->st_atim_ext = 0;
1575 	sbp->st_mtim_ext = 0;
1576 	sbp->st_ctim_ext = 0;
1577 	sbp->st_btim_ext = 0;
1578 #endif
1579 #ifdef KTRACE
1580 	if (KTRPOINT(td, KTR_STRUCT))
1581 		ktrstat_error(sbp, error);
1582 #endif
1583 	return (error);
1584 }
1585 
1586 #if defined(COMPAT_FREEBSD11)
1587 /*
1588  * Return status information about a file descriptor.
1589  */
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct freebsd11_nfstat_args {
1592 	int	fd;
1593 	struct	nstat *sb;
1594 };
1595 #endif
1596 /* ARGSUSED */
1597 int
1598 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1599 {
1600 	struct nstat nub;
1601 	struct stat ub;
1602 	int error;
1603 
1604 	error = kern_fstat(td, uap->fd, &ub);
1605 	if (error != 0)
1606 		return (error);
1607 	error = freebsd11_cvtnstat(&ub, &nub);
1608 	if (error != 0)
1609 		error = copyout(&nub, uap->sb, sizeof(nub));
1610 	return (error);
1611 }
1612 #endif /* COMPAT_FREEBSD11 */
1613 
1614 /*
1615  * Return pathconf information about a file descriptor.
1616  */
1617 #ifndef _SYS_SYSPROTO_H_
1618 struct fpathconf_args {
1619 	int	fd;
1620 	int	name;
1621 };
1622 #endif
1623 /* ARGSUSED */
1624 int
1625 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1626 {
1627 	long value;
1628 	int error;
1629 
1630 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1631 	if (error == 0)
1632 		td->td_retval[0] = value;
1633 	return (error);
1634 }
1635 
1636 int
1637 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1638 {
1639 	struct file *fp;
1640 	struct vnode *vp;
1641 	int error;
1642 
1643 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1644 	if (error != 0)
1645 		return (error);
1646 
1647 	if (name == _PC_ASYNC_IO) {
1648 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1649 		goto out;
1650 	}
1651 	vp = fp->f_vnode;
1652 	if (vp != NULL) {
1653 		vn_lock(vp, LK_SHARED | LK_RETRY);
1654 		error = VOP_PATHCONF(vp, name, valuep);
1655 		VOP_UNLOCK(vp);
1656 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1657 		if (name != _PC_PIPE_BUF) {
1658 			error = EINVAL;
1659 		} else {
1660 			*valuep = PIPE_BUF;
1661 			error = 0;
1662 		}
1663 	} else {
1664 		error = EOPNOTSUPP;
1665 	}
1666 out:
1667 	fdrop(fp, td);
1668 	return (error);
1669 }
1670 
1671 /*
1672  * Copy filecaps structure allocating memory for ioctls array if needed.
1673  *
1674  * The last parameter indicates whether the fdtable is locked. If it is not and
1675  * ioctls are encountered, copying fails and the caller must lock the table.
1676  *
1677  * Note that if the table was not locked, the caller has to check the relevant
1678  * sequence counter to determine whether the operation was successful.
1679  */
1680 bool
1681 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1682 {
1683 	size_t size;
1684 
1685 	if (src->fc_ioctls != NULL && !locked)
1686 		return (false);
1687 	memcpy(dst, src, sizeof(*src));
1688 	if (src->fc_ioctls == NULL)
1689 		return (true);
1690 
1691 	KASSERT(src->fc_nioctls > 0,
1692 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1693 
1694 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1695 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1696 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1697 	return (true);
1698 }
1699 
1700 static u_long *
1701 filecaps_copy_prep(const struct filecaps *src)
1702 {
1703 	u_long *ioctls;
1704 	size_t size;
1705 
1706 	if (__predict_true(src->fc_ioctls == NULL))
1707 		return (NULL);
1708 
1709 	KASSERT(src->fc_nioctls > 0,
1710 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1711 
1712 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1713 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1714 	return (ioctls);
1715 }
1716 
1717 static void
1718 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1719     u_long *ioctls)
1720 {
1721 	size_t size;
1722 
1723 	*dst = *src;
1724 	if (__predict_true(src->fc_ioctls == NULL)) {
1725 		MPASS(ioctls == NULL);
1726 		return;
1727 	}
1728 
1729 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1730 	dst->fc_ioctls = ioctls;
1731 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1732 }
1733 
1734 /*
1735  * Move filecaps structure to the new place and clear the old place.
1736  */
1737 void
1738 filecaps_move(struct filecaps *src, struct filecaps *dst)
1739 {
1740 
1741 	*dst = *src;
1742 	bzero(src, sizeof(*src));
1743 }
1744 
1745 /*
1746  * Fill the given filecaps structure with full rights.
1747  */
1748 static void
1749 filecaps_fill(struct filecaps *fcaps)
1750 {
1751 
1752 	CAP_ALL(&fcaps->fc_rights);
1753 	fcaps->fc_ioctls = NULL;
1754 	fcaps->fc_nioctls = -1;
1755 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1756 }
1757 
1758 /*
1759  * Free memory allocated within filecaps structure.
1760  */
1761 void
1762 filecaps_free(struct filecaps *fcaps)
1763 {
1764 
1765 	free(fcaps->fc_ioctls, M_FILECAPS);
1766 	bzero(fcaps, sizeof(*fcaps));
1767 }
1768 
1769 static u_long *
1770 filecaps_free_prep(struct filecaps *fcaps)
1771 {
1772 	u_long *ioctls;
1773 
1774 	ioctls = fcaps->fc_ioctls;
1775 	bzero(fcaps, sizeof(*fcaps));
1776 	return (ioctls);
1777 }
1778 
1779 static void
1780 filecaps_free_finish(u_long *ioctls)
1781 {
1782 
1783 	free(ioctls, M_FILECAPS);
1784 }
1785 
1786 /*
1787  * Validate the given filecaps structure.
1788  */
1789 static void
1790 filecaps_validate(const struct filecaps *fcaps, const char *func)
1791 {
1792 
1793 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1794 	    ("%s: invalid rights", func));
1795 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1796 	    ("%s: invalid fcntls", func));
1797 	KASSERT(fcaps->fc_fcntls == 0 ||
1798 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1799 	    ("%s: fcntls without CAP_FCNTL", func));
1800 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1801 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1802 	    ("%s: invalid ioctls", func));
1803 	KASSERT(fcaps->fc_nioctls == 0 ||
1804 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1805 	    ("%s: ioctls without CAP_IOCTL", func));
1806 }
1807 
1808 static void
1809 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1810 {
1811 	int nfd1;
1812 
1813 	FILEDESC_XLOCK_ASSERT(fdp);
1814 
1815 	nfd1 = fdp->fd_nfiles * 2;
1816 	if (nfd1 < nfd)
1817 		nfd1 = nfd;
1818 	fdgrowtable(fdp, nfd1);
1819 }
1820 
1821 /*
1822  * Grow the file table to accommodate (at least) nfd descriptors.
1823  */
1824 static void
1825 fdgrowtable(struct filedesc *fdp, int nfd)
1826 {
1827 	struct filedesc0 *fdp0;
1828 	struct freetable *ft;
1829 	struct fdescenttbl *ntable;
1830 	struct fdescenttbl *otable;
1831 	int nnfiles, onfiles;
1832 	NDSLOTTYPE *nmap, *omap;
1833 
1834 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1835 
1836 	/* save old values */
1837 	onfiles = fdp->fd_nfiles;
1838 	otable = fdp->fd_files;
1839 	omap = fdp->fd_map;
1840 
1841 	/* compute the size of the new table */
1842 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1843 	if (nnfiles <= onfiles)
1844 		/* the table is already large enough */
1845 		return;
1846 
1847 	/*
1848 	 * Allocate a new table.  We need enough space for the number of
1849 	 * entries, file entries themselves and the struct freetable we will use
1850 	 * when we decommission the table and place it on the freelist.
1851 	 * We place the struct freetable in the middle so we don't have
1852 	 * to worry about padding.
1853 	 */
1854 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1855 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1856 	    sizeof(struct freetable),
1857 	    M_FILEDESC, M_ZERO | M_WAITOK);
1858 	/* copy the old data */
1859 	ntable->fdt_nfiles = nnfiles;
1860 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1861 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1862 
1863 	/*
1864 	 * Allocate a new map only if the old is not large enough.  It will
1865 	 * grow at a slower rate than the table as it can map more
1866 	 * entries than the table can hold.
1867 	 */
1868 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1869 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1870 		    M_ZERO | M_WAITOK);
1871 		/* copy over the old data and update the pointer */
1872 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1873 		fdp->fd_map = nmap;
1874 	}
1875 
1876 	/*
1877 	 * Make sure that ntable is correctly initialized before we replace
1878 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1879 	 * data.
1880 	 */
1881 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1882 
1883 	/*
1884 	 * Free the old file table when not shared by other threads or processes.
1885 	 * The old file table is considered to be shared when either are true:
1886 	 * - The process has more than one thread.
1887 	 * - The file descriptor table has been shared via fdshare().
1888 	 *
1889 	 * When shared, the old file table will be placed on a freelist
1890 	 * which will be processed when the struct filedesc is released.
1891 	 *
1892 	 * Note that if onfiles == NDFILE, we're dealing with the original
1893 	 * static allocation contained within (struct filedesc0 *)fdp,
1894 	 * which must not be freed.
1895 	 */
1896 	if (onfiles > NDFILE) {
1897 		/*
1898 		 * Note we may be called here from fdinit while allocating a
1899 		 * table for a new process in which case ->p_fd points
1900 		 * elsewhere.
1901 		 */
1902 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1903 			free(otable, M_FILEDESC);
1904 		} else {
1905 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1906 			fdp0 = (struct filedesc0 *)fdp;
1907 			ft->ft_table = otable;
1908 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1909 		}
1910 	}
1911 	/*
1912 	 * The map does not have the same possibility of threads still
1913 	 * holding references to it.  So always free it as long as it
1914 	 * does not reference the original static allocation.
1915 	 */
1916 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1917 		free(omap, M_FILEDESC);
1918 }
1919 
1920 /*
1921  * Allocate a file descriptor for the process.
1922  */
1923 int
1924 fdalloc(struct thread *td, int minfd, int *result)
1925 {
1926 	struct proc *p = td->td_proc;
1927 	struct filedesc *fdp = p->p_fd;
1928 	int fd, maxfd, allocfd;
1929 #ifdef RACCT
1930 	int error;
1931 #endif
1932 
1933 	FILEDESC_XLOCK_ASSERT(fdp);
1934 
1935 	if (fdp->fd_freefile > minfd)
1936 		minfd = fdp->fd_freefile;
1937 
1938 	maxfd = getmaxfd(td);
1939 
1940 	/*
1941 	 * Search the bitmap for a free descriptor starting at minfd.
1942 	 * If none is found, grow the file table.
1943 	 */
1944 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1945 	if (__predict_false(fd >= maxfd))
1946 		return (EMFILE);
1947 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1948 		allocfd = min(fd * 2, maxfd);
1949 #ifdef RACCT
1950 		if (RACCT_ENABLED()) {
1951 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1952 			if (error != 0)
1953 				return (EMFILE);
1954 		}
1955 #endif
1956 		/*
1957 		 * fd is already equal to first free descriptor >= minfd, so
1958 		 * we only need to grow the table and we are done.
1959 		 */
1960 		fdgrowtable_exp(fdp, allocfd);
1961 	}
1962 
1963 	/*
1964 	 * Perform some sanity checks, then mark the file descriptor as
1965 	 * used and return it to the caller.
1966 	 */
1967 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1968 	    ("invalid descriptor %d", fd));
1969 	KASSERT(!fdisused(fdp, fd),
1970 	    ("fd_first_free() returned non-free descriptor"));
1971 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1972 	    ("file descriptor isn't free"));
1973 	fdused(fdp, fd);
1974 	*result = fd;
1975 	return (0);
1976 }
1977 
1978 /*
1979  * Allocate n file descriptors for the process.
1980  */
1981 int
1982 fdallocn(struct thread *td, int minfd, int *fds, int n)
1983 {
1984 	struct proc *p = td->td_proc;
1985 	struct filedesc *fdp = p->p_fd;
1986 	int i;
1987 
1988 	FILEDESC_XLOCK_ASSERT(fdp);
1989 
1990 	for (i = 0; i < n; i++)
1991 		if (fdalloc(td, 0, &fds[i]) != 0)
1992 			break;
1993 
1994 	if (i < n) {
1995 		for (i--; i >= 0; i--)
1996 			fdunused(fdp, fds[i]);
1997 		return (EMFILE);
1998 	}
1999 
2000 	return (0);
2001 }
2002 
2003 /*
2004  * Create a new open file structure and allocate a file descriptor for the
2005  * process that refers to it.  We add one reference to the file for the
2006  * descriptor table and one reference for resultfp. This is to prevent us
2007  * being preempted and the entry in the descriptor table closed after we
2008  * release the FILEDESC lock.
2009  */
2010 int
2011 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2012     struct filecaps *fcaps)
2013 {
2014 	struct file *fp;
2015 	int error, fd;
2016 
2017 	MPASS(resultfp != NULL);
2018 	MPASS(resultfd != NULL);
2019 
2020 	error = _falloc_noinstall(td, &fp, 2);
2021 	if (__predict_false(error != 0)) {
2022 		return (error);
2023 	}
2024 
2025 	error = finstall_refed(td, fp, &fd, flags, fcaps);
2026 	if (__predict_false(error != 0)) {
2027 		falloc_abort(td, fp);
2028 		return (error);
2029 	}
2030 
2031 	*resultfp = fp;
2032 	*resultfd = fd;
2033 
2034 	return (0);
2035 }
2036 
2037 /*
2038  * Create a new open file structure without allocating a file descriptor.
2039  */
2040 int
2041 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2042 {
2043 	struct file *fp;
2044 	int maxuserfiles = maxfiles - (maxfiles / 20);
2045 	int openfiles_new;
2046 	static struct timeval lastfail;
2047 	static int curfail;
2048 
2049 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2050 	MPASS(n > 0);
2051 
2052 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2053 	if ((openfiles_new >= maxuserfiles &&
2054 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2055 	    openfiles_new >= maxfiles) {
2056 		atomic_subtract_int(&openfiles, 1);
2057 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2058 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2059 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2060 		}
2061 		return (ENFILE);
2062 	}
2063 	fp = uma_zalloc(file_zone, M_WAITOK);
2064 	bzero(fp, sizeof(*fp));
2065 	refcount_init(&fp->f_count, n);
2066 	fp->f_cred = crhold(td->td_ucred);
2067 	fp->f_ops = &badfileops;
2068 	*resultfp = fp;
2069 	return (0);
2070 }
2071 
2072 void
2073 falloc_abort(struct thread *td, struct file *fp)
2074 {
2075 
2076 	/*
2077 	 * For assertion purposes.
2078 	 */
2079 	refcount_init(&fp->f_count, 0);
2080 	_fdrop(fp, td);
2081 }
2082 
2083 /*
2084  * Install a file in a file descriptor table.
2085  */
2086 void
2087 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2088     struct filecaps *fcaps)
2089 {
2090 	struct filedescent *fde;
2091 
2092 	MPASS(fp != NULL);
2093 	if (fcaps != NULL)
2094 		filecaps_validate(fcaps, __func__);
2095 	FILEDESC_XLOCK_ASSERT(fdp);
2096 
2097 	fde = &fdp->fd_ofiles[fd];
2098 #ifdef CAPABILITIES
2099 	seqc_write_begin(&fde->fde_seqc);
2100 #endif
2101 	fde->fde_file = fp;
2102 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2103 	if (fcaps != NULL)
2104 		filecaps_move(fcaps, &fde->fde_caps);
2105 	else
2106 		filecaps_fill(&fde->fde_caps);
2107 #ifdef CAPABILITIES
2108 	seqc_write_end(&fde->fde_seqc);
2109 #endif
2110 }
2111 
2112 int
2113 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2114     struct filecaps *fcaps)
2115 {
2116 	struct filedesc *fdp = td->td_proc->p_fd;
2117 	int error;
2118 
2119 	MPASS(fd != NULL);
2120 
2121 	FILEDESC_XLOCK(fdp);
2122 	error = fdalloc(td, 0, fd);
2123 	if (__predict_true(error == 0)) {
2124 		_finstall(fdp, fp, *fd, flags, fcaps);
2125 	}
2126 	FILEDESC_XUNLOCK(fdp);
2127 	return (error);
2128 }
2129 
2130 int
2131 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2132     struct filecaps *fcaps)
2133 {
2134 	int error;
2135 
2136 	MPASS(fd != NULL);
2137 
2138 	if (!fhold(fp))
2139 		return (EBADF);
2140 	error = finstall_refed(td, fp, fd, flags, fcaps);
2141 	if (__predict_false(error != 0)) {
2142 		fdrop(fp, td);
2143 	}
2144 	return (error);
2145 }
2146 
2147 /*
2148  * Build a new filedesc structure from another.
2149  *
2150  * If fdp is not NULL, return with it shared locked.
2151  */
2152 struct filedesc *
2153 fdinit(void)
2154 {
2155 	struct filedesc0 *newfdp0;
2156 	struct filedesc *newfdp;
2157 
2158 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2159 	newfdp = &newfdp0->fd_fd;
2160 
2161 	/* Create the file descriptor table. */
2162 	FILEDESC_LOCK_INIT(newfdp);
2163 	refcount_init(&newfdp->fd_refcnt, 1);
2164 	refcount_init(&newfdp->fd_holdcnt, 1);
2165 	newfdp->fd_map = newfdp0->fd_dmap;
2166 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2167 	newfdp->fd_files->fdt_nfiles = NDFILE;
2168 
2169 	return (newfdp);
2170 }
2171 
2172 /*
2173  * Build a pwddesc structure from another.
2174  * Copy the current, root, and jail root vnode references.
2175  *
2176  * If pdp is not NULL, return with it shared locked.
2177  */
2178 struct pwddesc *
2179 pdinit(struct pwddesc *pdp, bool keeplock)
2180 {
2181 	struct pwddesc *newpdp;
2182 	struct pwd *newpwd;
2183 
2184 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2185 
2186 	PWDDESC_LOCK_INIT(newpdp);
2187 	refcount_init(&newpdp->pd_refcount, 1);
2188 	newpdp->pd_cmask = CMASK;
2189 
2190 	if (pdp == NULL) {
2191 		newpwd = pwd_alloc();
2192 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2193 		return (newpdp);
2194 	}
2195 
2196 	PWDDESC_XLOCK(pdp);
2197 	newpwd = pwd_hold_pwddesc(pdp);
2198 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2199 	if (!keeplock)
2200 		PWDDESC_XUNLOCK(pdp);
2201 	return (newpdp);
2202 }
2203 
2204 /*
2205  * Hold either filedesc or pwddesc of the passed process.
2206  *
2207  * The process lock is used to synchronize against the target exiting and
2208  * freeing the data.
2209  *
2210  * Clearing can be ilustrated in 3 steps:
2211  * 1. set the pointer to NULL. Either routine can race against it, hence
2212  *   atomic_load_ptr.
2213  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2214  *   race to either still see the pointer or find NULL. It is still safe to
2215  *   grab a reference as clearing is stalled.
2216  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2217  *   guaranteed to see NULL, making it safe to finish clearing
2218  */
2219 static struct filedesc *
2220 fdhold(struct proc *p)
2221 {
2222 	struct filedesc *fdp;
2223 
2224 	PROC_LOCK_ASSERT(p, MA_OWNED);
2225 	fdp = atomic_load_ptr(&p->p_fd);
2226 	if (fdp != NULL)
2227 		refcount_acquire(&fdp->fd_holdcnt);
2228 	return (fdp);
2229 }
2230 
2231 static struct pwddesc *
2232 pdhold(struct proc *p)
2233 {
2234 	struct pwddesc *pdp;
2235 
2236 	PROC_LOCK_ASSERT(p, MA_OWNED);
2237 	pdp = atomic_load_ptr(&p->p_pd);
2238 	if (pdp != NULL)
2239 		refcount_acquire(&pdp->pd_refcount);
2240 	return (pdp);
2241 }
2242 
2243 static void
2244 fddrop(struct filedesc *fdp)
2245 {
2246 
2247 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2248 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2249 			return;
2250 	}
2251 
2252 	FILEDESC_LOCK_DESTROY(fdp);
2253 	uma_zfree(filedesc0_zone, fdp);
2254 }
2255 
2256 static void
2257 pddrop(struct pwddesc *pdp)
2258 {
2259 	struct pwd *pwd;
2260 
2261 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2262 		return;
2263 
2264 	PWDDESC_XLOCK(pdp);
2265 	if (refcount_release(&pdp->pd_refcount) == 0) {
2266 		PWDDESC_XUNLOCK(pdp);
2267 		return;
2268 	}
2269 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2270 	pwd_set(pdp, NULL);
2271 	PWDDESC_XUNLOCK(pdp);
2272 	pwd_drop(pwd);
2273 
2274 	PWDDESC_LOCK_DESTROY(pdp);
2275 	free(pdp, M_PWDDESC);
2276 }
2277 
2278 /*
2279  * Share a filedesc structure.
2280  */
2281 struct filedesc *
2282 fdshare(struct filedesc *fdp)
2283 {
2284 
2285 	refcount_acquire(&fdp->fd_refcnt);
2286 	return (fdp);
2287 }
2288 
2289 /*
2290  * Share a pwddesc structure.
2291  */
2292 struct pwddesc *
2293 pdshare(struct pwddesc *pdp)
2294 {
2295 	refcount_acquire(&pdp->pd_refcount);
2296 	return (pdp);
2297 }
2298 
2299 /*
2300  * Unshare a filedesc structure, if necessary by making a copy
2301  */
2302 void
2303 fdunshare(struct thread *td)
2304 {
2305 	struct filedesc *tmp;
2306 	struct proc *p = td->td_proc;
2307 
2308 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2309 		return;
2310 
2311 	tmp = fdcopy(p->p_fd);
2312 	fdescfree(td);
2313 	p->p_fd = tmp;
2314 }
2315 
2316 /*
2317  * Unshare a pwddesc structure.
2318  */
2319 void
2320 pdunshare(struct thread *td)
2321 {
2322 	struct pwddesc *pdp;
2323 	struct proc *p;
2324 
2325 	p = td->td_proc;
2326 	/* Not shared. */
2327 	if (refcount_load(&p->p_pd->pd_refcount) == 1)
2328 		return;
2329 
2330 	pdp = pdcopy(p->p_pd);
2331 	pdescfree(td);
2332 	p->p_pd = pdp;
2333 }
2334 
2335 /*
2336  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2337  * this is to ease callers, not catch errors.
2338  */
2339 struct filedesc *
2340 fdcopy(struct filedesc *fdp)
2341 {
2342 	struct filedesc *newfdp;
2343 	struct filedescent *nfde, *ofde;
2344 	int i, lastfile;
2345 
2346 	MPASS(fdp != NULL);
2347 
2348 	newfdp = fdinit();
2349 	FILEDESC_SLOCK(fdp);
2350 	for (;;) {
2351 		lastfile = fdlastfile(fdp);
2352 		if (lastfile < newfdp->fd_nfiles)
2353 			break;
2354 		FILEDESC_SUNLOCK(fdp);
2355 		fdgrowtable(newfdp, lastfile + 1);
2356 		FILEDESC_SLOCK(fdp);
2357 	}
2358 	/* copy all passable descriptors (i.e. not kqueue) */
2359 	newfdp->fd_freefile = fdp->fd_freefile;
2360 	FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2361 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2362 		    !fhold(ofde->fde_file)) {
2363 			if (newfdp->fd_freefile == fdp->fd_freefile)
2364 				newfdp->fd_freefile = i;
2365 			continue;
2366 		}
2367 		nfde = &newfdp->fd_ofiles[i];
2368 		*nfde = *ofde;
2369 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2370 		fdused_init(newfdp, i);
2371 	}
2372 	MPASS(newfdp->fd_freefile != -1);
2373 	FILEDESC_SUNLOCK(fdp);
2374 	return (newfdp);
2375 }
2376 
2377 /*
2378  * Copy a pwddesc structure.
2379  */
2380 struct pwddesc *
2381 pdcopy(struct pwddesc *pdp)
2382 {
2383 	struct pwddesc *newpdp;
2384 
2385 	MPASS(pdp != NULL);
2386 
2387 	newpdp = pdinit(pdp, true);
2388 	newpdp->pd_cmask = pdp->pd_cmask;
2389 	PWDDESC_XUNLOCK(pdp);
2390 	return (newpdp);
2391 }
2392 
2393 /*
2394  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2395  * one of processes using it exits) and the table used to be shared.
2396  */
2397 static void
2398 fdclearlocks(struct thread *td)
2399 {
2400 	struct filedesc *fdp;
2401 	struct filedesc_to_leader *fdtol;
2402 	struct flock lf;
2403 	struct file *fp;
2404 	struct proc *p;
2405 	struct vnode *vp;
2406 	int i;
2407 
2408 	p = td->td_proc;
2409 	fdp = p->p_fd;
2410 	fdtol = p->p_fdtol;
2411 	MPASS(fdtol != NULL);
2412 
2413 	FILEDESC_XLOCK(fdp);
2414 	KASSERT(fdtol->fdl_refcount > 0,
2415 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2416 	    fdtol->fdl_refcount));
2417 	if (fdtol->fdl_refcount == 1 &&
2418 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2419 		FILEDESC_FOREACH_FP(fdp, i, fp) {
2420 			if (fp->f_type != DTYPE_VNODE ||
2421 			    !fhold(fp))
2422 				continue;
2423 			FILEDESC_XUNLOCK(fdp);
2424 			lf.l_whence = SEEK_SET;
2425 			lf.l_start = 0;
2426 			lf.l_len = 0;
2427 			lf.l_type = F_UNLCK;
2428 			vp = fp->f_vnode;
2429 			(void) VOP_ADVLOCK(vp,
2430 			    (caddr_t)p->p_leader, F_UNLCK,
2431 			    &lf, F_POSIX);
2432 			FILEDESC_XLOCK(fdp);
2433 			fdrop(fp, td);
2434 		}
2435 	}
2436 retry:
2437 	if (fdtol->fdl_refcount == 1) {
2438 		if (fdp->fd_holdleaderscount > 0 &&
2439 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2440 			/*
2441 			 * close() or kern_dup() has cleared a reference
2442 			 * in a shared file descriptor table.
2443 			 */
2444 			fdp->fd_holdleaderswakeup = 1;
2445 			sx_sleep(&fdp->fd_holdleaderscount,
2446 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2447 			goto retry;
2448 		}
2449 		if (fdtol->fdl_holdcount > 0) {
2450 			/*
2451 			 * Ensure that fdtol->fdl_leader remains
2452 			 * valid in closef().
2453 			 */
2454 			fdtol->fdl_wakeup = 1;
2455 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2456 			    "fdlhold", 0);
2457 			goto retry;
2458 		}
2459 	}
2460 	fdtol->fdl_refcount--;
2461 	if (fdtol->fdl_refcount == 0 &&
2462 	    fdtol->fdl_holdcount == 0) {
2463 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2464 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2465 	} else
2466 		fdtol = NULL;
2467 	p->p_fdtol = NULL;
2468 	FILEDESC_XUNLOCK(fdp);
2469 	if (fdtol != NULL)
2470 		free(fdtol, M_FILEDESC_TO_LEADER);
2471 }
2472 
2473 /*
2474  * Release a filedesc structure.
2475  */
2476 static void
2477 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2478 {
2479 	struct filedesc0 *fdp0;
2480 	struct freetable *ft, *tft;
2481 	struct filedescent *fde;
2482 	struct file *fp;
2483 	int i;
2484 
2485 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2486 	    ("%s: fd table %p carries references", __func__, fdp));
2487 
2488 	/*
2489 	 * Serialize with threads iterating over the table, if any.
2490 	 */
2491 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2492 		FILEDESC_XLOCK(fdp);
2493 		FILEDESC_XUNLOCK(fdp);
2494 	}
2495 
2496 	FILEDESC_FOREACH_FDE(fdp, i, fde) {
2497 		fp = fde->fde_file;
2498 		fdefree_last(fde);
2499 		(void) closef(fp, td);
2500 	}
2501 
2502 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2503 		free(fdp->fd_map, M_FILEDESC);
2504 	if (fdp->fd_nfiles > NDFILE)
2505 		free(fdp->fd_files, M_FILEDESC);
2506 
2507 	fdp0 = (struct filedesc0 *)fdp;
2508 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2509 		free(ft->ft_table, M_FILEDESC);
2510 
2511 	fddrop(fdp);
2512 }
2513 
2514 void
2515 fdescfree(struct thread *td)
2516 {
2517 	struct proc *p;
2518 	struct filedesc *fdp;
2519 
2520 	p = td->td_proc;
2521 	fdp = p->p_fd;
2522 	MPASS(fdp != NULL);
2523 
2524 #ifdef RACCT
2525 	if (RACCT_ENABLED())
2526 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2527 #endif
2528 
2529 	if (p->p_fdtol != NULL)
2530 		fdclearlocks(td);
2531 
2532 	/*
2533 	 * Check fdhold for an explanation.
2534 	 */
2535 	atomic_store_ptr(&p->p_fd, NULL);
2536 	atomic_thread_fence_seq_cst();
2537 	PROC_WAIT_UNLOCKED(p);
2538 
2539 	if (refcount_release(&fdp->fd_refcnt) == 0)
2540 		return;
2541 
2542 	fdescfree_fds(td, fdp);
2543 }
2544 
2545 void
2546 pdescfree(struct thread *td)
2547 {
2548 	struct proc *p;
2549 	struct pwddesc *pdp;
2550 
2551 	p = td->td_proc;
2552 	pdp = p->p_pd;
2553 	MPASS(pdp != NULL);
2554 
2555 	/*
2556 	 * Check pdhold for an explanation.
2557 	 */
2558 	atomic_store_ptr(&p->p_pd, NULL);
2559 	atomic_thread_fence_seq_cst();
2560 	PROC_WAIT_UNLOCKED(p);
2561 
2562 	pddrop(pdp);
2563 }
2564 
2565 /*
2566  * For setugid programs, we don't want to people to use that setugidness
2567  * to generate error messages which write to a file which otherwise would
2568  * otherwise be off-limits to the process.  We check for filesystems where
2569  * the vnode can change out from under us after execve (like [lin]procfs).
2570  *
2571  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2572  * sufficient.  We also don't check for setugidness since we know we are.
2573  */
2574 static bool
2575 is_unsafe(struct file *fp)
2576 {
2577 	struct vnode *vp;
2578 
2579 	if (fp->f_type != DTYPE_VNODE)
2580 		return (false);
2581 
2582 	vp = fp->f_vnode;
2583 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2584 }
2585 
2586 /*
2587  * Make this setguid thing safe, if at all possible.
2588  */
2589 void
2590 fdsetugidsafety(struct thread *td)
2591 {
2592 	struct filedesc *fdp;
2593 	struct file *fp;
2594 	int i;
2595 
2596 	fdp = td->td_proc->p_fd;
2597 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2598 	    ("the fdtable should not be shared"));
2599 	MPASS(fdp->fd_nfiles >= 3);
2600 	for (i = 0; i <= 2; i++) {
2601 		fp = fdp->fd_ofiles[i].fde_file;
2602 		if (fp != NULL && is_unsafe(fp)) {
2603 			FILEDESC_XLOCK(fdp);
2604 			knote_fdclose(td, i);
2605 			/*
2606 			 * NULL-out descriptor prior to close to avoid
2607 			 * a race while close blocks.
2608 			 */
2609 			fdfree(fdp, i);
2610 			FILEDESC_XUNLOCK(fdp);
2611 			(void) closef(fp, td);
2612 		}
2613 	}
2614 }
2615 
2616 /*
2617  * If a specific file object occupies a specific file descriptor, close the
2618  * file descriptor entry and drop a reference on the file object.  This is a
2619  * convenience function to handle a subsequent error in a function that calls
2620  * falloc() that handles the race that another thread might have closed the
2621  * file descriptor out from under the thread creating the file object.
2622  */
2623 void
2624 fdclose(struct thread *td, struct file *fp, int idx)
2625 {
2626 	struct filedesc *fdp = td->td_proc->p_fd;
2627 
2628 	FILEDESC_XLOCK(fdp);
2629 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2630 		fdfree(fdp, idx);
2631 		FILEDESC_XUNLOCK(fdp);
2632 		fdrop(fp, td);
2633 	} else
2634 		FILEDESC_XUNLOCK(fdp);
2635 }
2636 
2637 /*
2638  * Close any files on exec?
2639  */
2640 void
2641 fdcloseexec(struct thread *td)
2642 {
2643 	struct filedesc *fdp;
2644 	struct filedescent *fde;
2645 	struct file *fp;
2646 	int i;
2647 
2648 	fdp = td->td_proc->p_fd;
2649 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2650 	    ("the fdtable should not be shared"));
2651 	FILEDESC_FOREACH_FDE(fdp, i, fde) {
2652 		fp = fde->fde_file;
2653 		if (fp->f_type == DTYPE_MQUEUE ||
2654 		    (fde->fde_flags & UF_EXCLOSE)) {
2655 			FILEDESC_XLOCK(fdp);
2656 			fdfree(fdp, i);
2657 			(void) closefp(fdp, i, fp, td, false, false);
2658 			FILEDESC_UNLOCK_ASSERT(fdp);
2659 		}
2660 	}
2661 }
2662 
2663 /*
2664  * It is unsafe for set[ug]id processes to be started with file
2665  * descriptors 0..2 closed, as these descriptors are given implicit
2666  * significance in the Standard C library.  fdcheckstd() will create a
2667  * descriptor referencing /dev/null for each of stdin, stdout, and
2668  * stderr that is not already open.
2669  */
2670 int
2671 fdcheckstd(struct thread *td)
2672 {
2673 	struct filedesc *fdp;
2674 	register_t save;
2675 	int i, error, devnull;
2676 
2677 	fdp = td->td_proc->p_fd;
2678 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2679 	    ("the fdtable should not be shared"));
2680 	MPASS(fdp->fd_nfiles >= 3);
2681 	devnull = -1;
2682 	for (i = 0; i <= 2; i++) {
2683 		if (fdp->fd_ofiles[i].fde_file != NULL)
2684 			continue;
2685 
2686 		save = td->td_retval[0];
2687 		if (devnull != -1) {
2688 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2689 		} else {
2690 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2691 			    UIO_SYSSPACE, O_RDWR, 0);
2692 			if (error == 0) {
2693 				devnull = td->td_retval[0];
2694 				KASSERT(devnull == i, ("we didn't get our fd"));
2695 			}
2696 		}
2697 		td->td_retval[0] = save;
2698 		if (error != 0)
2699 			return (error);
2700 	}
2701 	return (0);
2702 }
2703 
2704 /*
2705  * Internal form of close.  Decrement reference count on file structure.
2706  * Note: td may be NULL when closing a file that was being passed in a
2707  * message.
2708  */
2709 int
2710 closef(struct file *fp, struct thread *td)
2711 {
2712 	struct vnode *vp;
2713 	struct flock lf;
2714 	struct filedesc_to_leader *fdtol;
2715 	struct filedesc *fdp;
2716 
2717 	MPASS(td != NULL);
2718 
2719 	/*
2720 	 * POSIX record locking dictates that any close releases ALL
2721 	 * locks owned by this process.  This is handled by setting
2722 	 * a flag in the unlock to free ONLY locks obeying POSIX
2723 	 * semantics, and not to free BSD-style file locks.
2724 	 * If the descriptor was in a message, POSIX-style locks
2725 	 * aren't passed with the descriptor, and the thread pointer
2726 	 * will be NULL.  Callers should be careful only to pass a
2727 	 * NULL thread pointer when there really is no owning
2728 	 * context that might have locks, or the locks will be
2729 	 * leaked.
2730 	 */
2731 	if (fp->f_type == DTYPE_VNODE) {
2732 		vp = fp->f_vnode;
2733 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2734 			lf.l_whence = SEEK_SET;
2735 			lf.l_start = 0;
2736 			lf.l_len = 0;
2737 			lf.l_type = F_UNLCK;
2738 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2739 			    F_UNLCK, &lf, F_POSIX);
2740 		}
2741 		fdtol = td->td_proc->p_fdtol;
2742 		if (fdtol != NULL) {
2743 			/*
2744 			 * Handle special case where file descriptor table is
2745 			 * shared between multiple process leaders.
2746 			 */
2747 			fdp = td->td_proc->p_fd;
2748 			FILEDESC_XLOCK(fdp);
2749 			for (fdtol = fdtol->fdl_next;
2750 			    fdtol != td->td_proc->p_fdtol;
2751 			    fdtol = fdtol->fdl_next) {
2752 				if ((fdtol->fdl_leader->p_flag &
2753 				    P_ADVLOCK) == 0)
2754 					continue;
2755 				fdtol->fdl_holdcount++;
2756 				FILEDESC_XUNLOCK(fdp);
2757 				lf.l_whence = SEEK_SET;
2758 				lf.l_start = 0;
2759 				lf.l_len = 0;
2760 				lf.l_type = F_UNLCK;
2761 				vp = fp->f_vnode;
2762 				(void) VOP_ADVLOCK(vp,
2763 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2764 				    F_POSIX);
2765 				FILEDESC_XLOCK(fdp);
2766 				fdtol->fdl_holdcount--;
2767 				if (fdtol->fdl_holdcount == 0 &&
2768 				    fdtol->fdl_wakeup != 0) {
2769 					fdtol->fdl_wakeup = 0;
2770 					wakeup(fdtol);
2771 				}
2772 			}
2773 			FILEDESC_XUNLOCK(fdp);
2774 		}
2775 	}
2776 	return (fdrop_close(fp, td));
2777 }
2778 
2779 /*
2780  * Hack for file descriptor passing code.
2781  */
2782 void
2783 closef_nothread(struct file *fp)
2784 {
2785 
2786 	fdrop(fp, NULL);
2787 }
2788 
2789 /*
2790  * Initialize the file pointer with the specified properties.
2791  *
2792  * The ops are set with release semantics to be certain that the flags, type,
2793  * and data are visible when ops is.  This is to prevent ops methods from being
2794  * called with bad data.
2795  */
2796 void
2797 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2798 {
2799 	fp->f_data = data;
2800 	fp->f_flag = flag;
2801 	fp->f_type = type;
2802 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2803 }
2804 
2805 void
2806 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2807 {
2808 	fp->f_seqcount[UIO_READ] = 1;
2809 	fp->f_seqcount[UIO_WRITE] = 1;
2810 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2811 	    data, ops);
2812 }
2813 
2814 int
2815 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2816     struct file **fpp, struct filecaps *havecapsp)
2817 {
2818 	struct filedescent *fde;
2819 	int error;
2820 
2821 	FILEDESC_LOCK_ASSERT(fdp);
2822 
2823 	*fpp = NULL;
2824 	fde = fdeget_locked(fdp, fd);
2825 	if (fde == NULL) {
2826 		error = EBADF;
2827 		goto out;
2828 	}
2829 
2830 #ifdef CAPABILITIES
2831 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2832 	if (error != 0)
2833 		goto out;
2834 #endif
2835 
2836 	if (havecapsp != NULL)
2837 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2838 
2839 	*fpp = fde->fde_file;
2840 
2841 	error = 0;
2842 out:
2843 	return (error);
2844 }
2845 
2846 #ifdef CAPABILITIES
2847 int
2848 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2849     struct file **fpp, struct filecaps *havecapsp)
2850 {
2851 	struct filedesc *fdp = td->td_proc->p_fd;
2852 	int error;
2853 	struct file *fp;
2854 	seqc_t seq;
2855 
2856 	*fpp = NULL;
2857 	for (;;) {
2858 		error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2859 		if (error != 0)
2860 			return (error);
2861 
2862 		if (havecapsp != NULL) {
2863 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2864 			    havecapsp, false)) {
2865 				fdrop(fp, td);
2866 				goto get_locked;
2867 			}
2868 		}
2869 
2870 		if (!fd_modified(fdp, fd, seq))
2871 			break;
2872 		fdrop(fp, td);
2873 	}
2874 
2875 	*fpp = fp;
2876 	return (0);
2877 
2878 get_locked:
2879 	FILEDESC_SLOCK(fdp);
2880 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2881 	if (error == 0 && !fhold(*fpp))
2882 		error = EBADF;
2883 	FILEDESC_SUNLOCK(fdp);
2884 	return (error);
2885 }
2886 #else
2887 int
2888 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2889     struct file **fpp, struct filecaps *havecapsp)
2890 {
2891 	int error;
2892 	error = fget_unlocked(td, fd, needrightsp, fpp);
2893 	if (havecapsp != NULL && error == 0)
2894 		filecaps_fill(havecapsp);
2895 
2896 	return (error);
2897 }
2898 #endif
2899 
2900 #ifdef CAPABILITIES
2901 int
2902 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2903 {
2904 	const struct filedescent *fde;
2905 	const struct fdescenttbl *fdt;
2906 	struct filedesc *fdp;
2907 	struct file *fp;
2908 	struct vnode *vp;
2909 	const cap_rights_t *haverights;
2910 	cap_rights_t rights;
2911 	seqc_t seq;
2912 
2913 	VFS_SMR_ASSERT_ENTERED();
2914 
2915 	rights = *ndp->ni_rightsneeded;
2916 	cap_rights_set_one(&rights, CAP_LOOKUP);
2917 
2918 	fdp = curproc->p_fd;
2919 	fdt = fdp->fd_files;
2920 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2921 		return (EBADF);
2922 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2923 	fde = &fdt->fdt_ofiles[fd];
2924 	haverights = cap_rights_fde_inline(fde);
2925 	fp = fde->fde_file;
2926 	if (__predict_false(fp == NULL))
2927 		return (EAGAIN);
2928 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2929 		return (EAGAIN);
2930 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2931 	vp = fp->f_vnode;
2932 	if (__predict_false(vp == NULL)) {
2933 		return (EAGAIN);
2934 	}
2935 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2936 		return (EAGAIN);
2937 	}
2938 	/*
2939 	 * Use an acquire barrier to force re-reading of fdt so it is
2940 	 * refreshed for verification.
2941 	 */
2942 	atomic_thread_fence_acq();
2943 	fdt = fdp->fd_files;
2944 	if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
2945 		return (EAGAIN);
2946 	/*
2947 	 * If file descriptor doesn't have all rights,
2948 	 * all lookups relative to it must also be
2949 	 * strictly relative.
2950 	 *
2951 	 * Not yet supported by fast path.
2952 	 */
2953 	CAP_ALL(&rights);
2954 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2955 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2956 	    ndp->ni_filecaps.fc_nioctls != -1) {
2957 #ifdef notyet
2958 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2959 #else
2960 		return (EAGAIN);
2961 #endif
2962 	}
2963 	*vpp = vp;
2964 	return (0);
2965 }
2966 #else
2967 int
2968 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2969 {
2970 	const struct fdescenttbl *fdt;
2971 	struct filedesc *fdp;
2972 	struct file *fp;
2973 	struct vnode *vp;
2974 
2975 	VFS_SMR_ASSERT_ENTERED();
2976 
2977 	fdp = curproc->p_fd;
2978 	fdt = fdp->fd_files;
2979 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2980 		return (EBADF);
2981 	fp = fdt->fdt_ofiles[fd].fde_file;
2982 	if (__predict_false(fp == NULL))
2983 		return (EAGAIN);
2984 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2985 	vp = fp->f_vnode;
2986 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2987 		return (EAGAIN);
2988 	}
2989 	/*
2990 	 * Use an acquire barrier to force re-reading of fdt so it is
2991 	 * refreshed for verification.
2992 	 */
2993 	atomic_thread_fence_acq();
2994 	fdt = fdp->fd_files;
2995 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2996 		return (EAGAIN);
2997 	filecaps_fill(&ndp->ni_filecaps);
2998 	*vpp = vp;
2999 	return (0);
3000 }
3001 #endif
3002 
3003 /*
3004  * Fetch the descriptor locklessly.
3005  *
3006  * We avoid fdrop() races by never raising a refcount above 0.  To accomplish
3007  * this we have to use a cmpset loop rather than an atomic_add.  The descriptor
3008  * must be re-verified once we acquire a reference to be certain that the
3009  * identity is still correct and we did not lose a race due to preemption.
3010  *
3011  * Force a reload of fdt when looping. Another thread could reallocate
3012  * the table before this fd was closed, so it is possible that there is
3013  * a stale fp pointer in cached version.
3014  */
3015 #ifdef CAPABILITIES
3016 static int
3017 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3018     struct file **fpp, seqc_t *seqp)
3019 {
3020 	struct filedesc *fdp;
3021 	const struct filedescent *fde;
3022 	const struct fdescenttbl *fdt;
3023 	struct file *fp;
3024 	seqc_t seq;
3025 	cap_rights_t haverights;
3026 	int error;
3027 
3028 	fdp = td->td_proc->p_fd;
3029 	fdt = fdp->fd_files;
3030 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3031 		return (EBADF);
3032 
3033 	for (;;) {
3034 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3035 		fde = &fdt->fdt_ofiles[fd];
3036 		haverights = *cap_rights_fde_inline(fde);
3037 		fp = fde->fde_file;
3038 		if (__predict_false(fp == NULL)) {
3039 			if (seqc_consistent(fd_seqc(fdt, fd), seq))
3040 				return (EBADF);
3041 			fdt = atomic_load_ptr(&fdp->fd_files);
3042 			continue;
3043 		}
3044 		error = cap_check_inline(&haverights, needrightsp);
3045 		if (__predict_false(error != 0)) {
3046 			if (seqc_consistent(fd_seqc(fdt, fd), seq))
3047 				return (error);
3048 			fdt = atomic_load_ptr(&fdp->fd_files);
3049 			continue;
3050 		}
3051 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3052 			fdt = atomic_load_ptr(&fdp->fd_files);
3053 			continue;
3054 		}
3055 		/*
3056 		 * Use an acquire barrier to force re-reading of fdt so it is
3057 		 * refreshed for verification.
3058 		 */
3059 		atomic_thread_fence_acq();
3060 		fdt = fdp->fd_files;
3061 		if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3062 			break;
3063 		fdrop(fp, td);
3064 	}
3065 	*fpp = fp;
3066 	if (seqp != NULL) {
3067 		*seqp = seq;
3068 	}
3069 	return (0);
3070 }
3071 #else
3072 static int
3073 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3074     struct file **fpp, seqc_t *seqp __unused)
3075 {
3076 	struct filedesc *fdp;
3077 	const struct fdescenttbl *fdt;
3078 	struct file *fp;
3079 
3080 	fdp = td->td_proc->p_fd;
3081 	fdt = fdp->fd_files;
3082 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3083 		return (EBADF);
3084 
3085 	for (;;) {
3086 		fp = fdt->fdt_ofiles[fd].fde_file;
3087 		if (__predict_false(fp == NULL))
3088 			return (EBADF);
3089 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3090 			fdt = atomic_load_ptr(&fdp->fd_files);
3091 			continue;
3092 		}
3093 		/*
3094 		 * Use an acquire barrier to force re-reading of fdt so it is
3095 		 * refreshed for verification.
3096 		 */
3097 		atomic_thread_fence_acq();
3098 		fdt = fdp->fd_files;
3099 		if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3100 			break;
3101 		fdrop(fp, td);
3102 	}
3103 	*fpp = fp;
3104 	return (0);
3105 }
3106 #endif
3107 
3108 /*
3109  * See the comments in fget_unlocked_seq for an explanation of how this works.
3110  *
3111  * This is a simplified variant which bails out to the aforementioned routine
3112  * if anything goes wrong. In practice this only happens when userspace is
3113  * racing with itself.
3114  */
3115 int
3116 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3117     struct file **fpp)
3118 {
3119 	struct filedesc *fdp;
3120 #ifdef CAPABILITIES
3121 	const struct filedescent *fde;
3122 #endif
3123 	const struct fdescenttbl *fdt;
3124 	struct file *fp;
3125 #ifdef CAPABILITIES
3126 	seqc_t seq;
3127 	const cap_rights_t *haverights;
3128 #endif
3129 
3130 	fdp = td->td_proc->p_fd;
3131 	fdt = fdp->fd_files;
3132 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3133 		*fpp = NULL;
3134 		return (EBADF);
3135 	}
3136 #ifdef CAPABILITIES
3137 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3138 	fde = &fdt->fdt_ofiles[fd];
3139 	haverights = cap_rights_fde_inline(fde);
3140 	fp = fde->fde_file;
3141 #else
3142 	fp = fdt->fdt_ofiles[fd].fde_file;
3143 #endif
3144 	if (__predict_false(fp == NULL))
3145 		goto out_fallback;
3146 #ifdef CAPABILITIES
3147 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3148 		goto out_fallback;
3149 #endif
3150 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3151 		goto out_fallback;
3152 
3153 	/*
3154 	 * Use an acquire barrier to force re-reading of fdt so it is
3155 	 * refreshed for verification.
3156 	 */
3157 	atomic_thread_fence_acq();
3158 	fdt = fdp->fd_files;
3159 #ifdef	CAPABILITIES
3160 	if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3161 #else
3162 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3163 #endif
3164 		goto out_fdrop;
3165 	*fpp = fp;
3166 	return (0);
3167 out_fdrop:
3168 	fdrop(fp, td);
3169 out_fallback:
3170 	*fpp = NULL;
3171 	return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3172 }
3173 
3174 /*
3175  * Translate fd -> file when the caller guarantees the file descriptor table
3176  * can't be changed by others.
3177  *
3178  * Note this does not mean the file object itself is only visible to the caller,
3179  * merely that it wont disappear without having to be referenced.
3180  *
3181  * Must be paired with fput_only_user.
3182  */
3183 #ifdef	CAPABILITIES
3184 int
3185 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3186     struct file **fpp)
3187 {
3188 	const struct filedescent *fde;
3189 	const struct fdescenttbl *fdt;
3190 	const cap_rights_t *haverights;
3191 	struct file *fp;
3192 	int error;
3193 
3194 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3195 
3196 	*fpp = NULL;
3197 	if (__predict_false(fd >= fdp->fd_nfiles))
3198 		return (EBADF);
3199 
3200 	fdt = fdp->fd_files;
3201 	fde = &fdt->fdt_ofiles[fd];
3202 	fp = fde->fde_file;
3203 	if (__predict_false(fp == NULL))
3204 		return (EBADF);
3205 	MPASS(refcount_load(&fp->f_count) > 0);
3206 	haverights = cap_rights_fde_inline(fde);
3207 	error = cap_check_inline(haverights, needrightsp);
3208 	if (__predict_false(error != 0))
3209 		return (error);
3210 	*fpp = fp;
3211 	return (0);
3212 }
3213 #else
3214 int
3215 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3216     struct file **fpp)
3217 {
3218 	struct file *fp;
3219 
3220 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3221 
3222 	*fpp = NULL;
3223 	if (__predict_false(fd >= fdp->fd_nfiles))
3224 		return (EBADF);
3225 
3226 	fp = fdp->fd_ofiles[fd].fde_file;
3227 	if (__predict_false(fp == NULL))
3228 		return (EBADF);
3229 
3230 	MPASS(refcount_load(&fp->f_count) > 0);
3231 	*fpp = fp;
3232 	return (0);
3233 }
3234 #endif
3235 
3236 /*
3237  * Extract the file pointer associated with the specified descriptor for the
3238  * current user process.
3239  *
3240  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3241  * returned.
3242  *
3243  * File's rights will be checked against the capability rights mask.
3244  *
3245  * If an error occurred the non-zero error is returned and *fpp is set to
3246  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3247  * responsible for fdrop().
3248  */
3249 static __inline int
3250 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3251     cap_rights_t *needrightsp)
3252 {
3253 	struct file *fp;
3254 	int error;
3255 
3256 	*fpp = NULL;
3257 	error = fget_unlocked(td, fd, needrightsp, &fp);
3258 	if (__predict_false(error != 0))
3259 		return (error);
3260 	if (__predict_false(fp->f_ops == &badfileops)) {
3261 		fdrop(fp, td);
3262 		return (EBADF);
3263 	}
3264 
3265 	/*
3266 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3267 	 */
3268 	error = 0;
3269 	switch (flags) {
3270 	case FREAD:
3271 	case FWRITE:
3272 		if ((fp->f_flag & flags) == 0)
3273 			error = EBADF;
3274 		break;
3275 	case FEXEC:
3276 		if (fp->f_ops != &path_fileops &&
3277 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3278 		    (fp->f_flag & FWRITE) != 0))
3279 			error = EBADF;
3280 		break;
3281 	case 0:
3282 		break;
3283 	default:
3284 		KASSERT(0, ("wrong flags"));
3285 	}
3286 
3287 	if (error != 0) {
3288 		fdrop(fp, td);
3289 		return (error);
3290 	}
3291 
3292 	*fpp = fp;
3293 	return (0);
3294 }
3295 
3296 int
3297 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3298 {
3299 
3300 	return (_fget(td, fd, fpp, 0, rightsp));
3301 }
3302 
3303 int
3304 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3305     struct file **fpp)
3306 {
3307 	int error;
3308 #ifndef CAPABILITIES
3309 	error = _fget(td, fd, fpp, 0, rightsp);
3310 	if (maxprotp != NULL)
3311 		*maxprotp = VM_PROT_ALL;
3312 	return (error);
3313 #else
3314 	cap_rights_t fdrights;
3315 	struct filedesc *fdp;
3316 	struct file *fp;
3317 	seqc_t seq;
3318 
3319 	*fpp = NULL;
3320 	fdp = td->td_proc->p_fd;
3321 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3322 	for (;;) {
3323 		error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3324 		if (__predict_false(error != 0))
3325 			return (error);
3326 		if (__predict_false(fp->f_ops == &badfileops)) {
3327 			fdrop(fp, td);
3328 			return (EBADF);
3329 		}
3330 		if (maxprotp != NULL)
3331 			fdrights = *cap_rights(fdp, fd);
3332 		if (!fd_modified(fdp, fd, seq))
3333 			break;
3334 		fdrop(fp, td);
3335 	}
3336 
3337 	/*
3338 	 * If requested, convert capability rights to access flags.
3339 	 */
3340 	if (maxprotp != NULL)
3341 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3342 	*fpp = fp;
3343 	return (0);
3344 #endif
3345 }
3346 
3347 int
3348 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3349 {
3350 
3351 	return (_fget(td, fd, fpp, FREAD, rightsp));
3352 }
3353 
3354 int
3355 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3356 {
3357 
3358 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3359 }
3360 
3361 int
3362 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3363     struct file **fpp)
3364 {
3365 #ifndef CAPABILITIES
3366 	return (fget_unlocked(td, fd, rightsp, fpp));
3367 #else
3368 	struct filedesc *fdp = td->td_proc->p_fd;
3369 	struct file *fp;
3370 	int error;
3371 	seqc_t seq;
3372 
3373 	*fpp = NULL;
3374 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3375 	for (;;) {
3376 		error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3377 		if (error != 0)
3378 			return (error);
3379 		error = cap_fcntl_check(fdp, fd, needfcntl);
3380 		if (!fd_modified(fdp, fd, seq))
3381 			break;
3382 		fdrop(fp, td);
3383 	}
3384 	if (error != 0) {
3385 		fdrop(fp, td);
3386 		return (error);
3387 	}
3388 	*fpp = fp;
3389 	return (0);
3390 #endif
3391 }
3392 
3393 /*
3394  * Like fget() but loads the underlying vnode, or returns an error if the
3395  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3396  * never have VM objects.  The returned vnode will be vref()'d.
3397  *
3398  * XXX: what about the unused flags ?
3399  */
3400 static __inline int
3401 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3402     struct vnode **vpp)
3403 {
3404 	struct file *fp;
3405 	int error;
3406 
3407 	*vpp = NULL;
3408 	error = _fget(td, fd, &fp, flags, needrightsp);
3409 	if (error != 0)
3410 		return (error);
3411 	if (fp->f_vnode == NULL) {
3412 		error = EINVAL;
3413 	} else {
3414 		*vpp = fp->f_vnode;
3415 		vref(*vpp);
3416 	}
3417 	fdrop(fp, td);
3418 
3419 	return (error);
3420 }
3421 
3422 int
3423 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3424 {
3425 
3426 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3427 }
3428 
3429 int
3430 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3431     struct filecaps *havecaps, struct vnode **vpp)
3432 {
3433 	struct filecaps caps;
3434 	struct file *fp;
3435 	int error;
3436 
3437 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3438 	if (error != 0)
3439 		return (error);
3440 	if (fp->f_ops == &badfileops) {
3441 		error = EBADF;
3442 		goto out;
3443 	}
3444 	if (fp->f_vnode == NULL) {
3445 		error = EINVAL;
3446 		goto out;
3447 	}
3448 
3449 	*havecaps = caps;
3450 	*vpp = fp->f_vnode;
3451 	vref(*vpp);
3452 	fdrop(fp, td);
3453 
3454 	return (0);
3455 out:
3456 	filecaps_free(&caps);
3457 	fdrop(fp, td);
3458 	return (error);
3459 }
3460 
3461 int
3462 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3463 {
3464 
3465 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3466 }
3467 
3468 int
3469 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3470 {
3471 
3472 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3473 }
3474 
3475 #ifdef notyet
3476 int
3477 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3478     struct vnode **vpp)
3479 {
3480 
3481 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3482 }
3483 #endif
3484 
3485 /*
3486  * Handle the last reference to a file being closed.
3487  *
3488  * Without the noinline attribute clang keeps inlining the func thorough this
3489  * file when fdrop is used.
3490  */
3491 int __noinline
3492 _fdrop(struct file *fp, struct thread *td)
3493 {
3494 	int error;
3495 #ifdef INVARIANTS
3496 	int count;
3497 
3498 	count = refcount_load(&fp->f_count);
3499 	if (count != 0)
3500 		panic("fdrop: fp %p count %d", fp, count);
3501 #endif
3502 	error = fo_close(fp, td);
3503 	atomic_subtract_int(&openfiles, 1);
3504 	crfree(fp->f_cred);
3505 	free(fp->f_advice, M_FADVISE);
3506 	uma_zfree(file_zone, fp);
3507 
3508 	return (error);
3509 }
3510 
3511 /*
3512  * Apply an advisory lock on a file descriptor.
3513  *
3514  * Just attempt to get a record lock of the requested type on the entire file
3515  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3516  */
3517 #ifndef _SYS_SYSPROTO_H_
3518 struct flock_args {
3519 	int	fd;
3520 	int	how;
3521 };
3522 #endif
3523 /* ARGSUSED */
3524 int
3525 sys_flock(struct thread *td, struct flock_args *uap)
3526 {
3527 	struct file *fp;
3528 	struct vnode *vp;
3529 	struct flock lf;
3530 	int error;
3531 
3532 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3533 	if (error != 0)
3534 		return (error);
3535 	error = EOPNOTSUPP;
3536 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3537 		goto done;
3538 	}
3539 	if (fp->f_ops == &path_fileops) {
3540 		goto done;
3541 	}
3542 
3543 	error = 0;
3544 	vp = fp->f_vnode;
3545 	lf.l_whence = SEEK_SET;
3546 	lf.l_start = 0;
3547 	lf.l_len = 0;
3548 	if (uap->how & LOCK_UN) {
3549 		lf.l_type = F_UNLCK;
3550 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3551 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3552 		goto done;
3553 	}
3554 	if (uap->how & LOCK_EX)
3555 		lf.l_type = F_WRLCK;
3556 	else if (uap->how & LOCK_SH)
3557 		lf.l_type = F_RDLCK;
3558 	else {
3559 		error = EBADF;
3560 		goto done;
3561 	}
3562 	atomic_set_int(&fp->f_flag, FHASLOCK);
3563 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3564 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3565 done:
3566 	fdrop(fp, td);
3567 	return (error);
3568 }
3569 /*
3570  * Duplicate the specified descriptor to a free descriptor.
3571  */
3572 int
3573 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3574     int openerror, int *indxp)
3575 {
3576 	struct filedescent *newfde, *oldfde;
3577 	struct file *fp;
3578 	u_long *ioctls;
3579 	int error, indx;
3580 
3581 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3582 	    ("unexpected error %d in %s", openerror, __func__));
3583 
3584 	/*
3585 	 * If the to-be-dup'd fd number is greater than the allowed number
3586 	 * of file descriptors, or the fd to be dup'd has already been
3587 	 * closed, then reject.
3588 	 */
3589 	FILEDESC_XLOCK(fdp);
3590 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3591 		FILEDESC_XUNLOCK(fdp);
3592 		return (EBADF);
3593 	}
3594 
3595 	error = fdalloc(td, 0, &indx);
3596 	if (error != 0) {
3597 		FILEDESC_XUNLOCK(fdp);
3598 		return (error);
3599 	}
3600 
3601 	/*
3602 	 * There are two cases of interest here.
3603 	 *
3604 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3605 	 *
3606 	 * For ENXIO steal away the file structure from (dfd) and store it in
3607 	 * (indx).  (dfd) is effectively closed by this operation.
3608 	 */
3609 	switch (openerror) {
3610 	case ENODEV:
3611 		/*
3612 		 * Check that the mode the file is being opened for is a
3613 		 * subset of the mode of the existing descriptor.
3614 		 */
3615 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3616 			fdunused(fdp, indx);
3617 			FILEDESC_XUNLOCK(fdp);
3618 			return (EACCES);
3619 		}
3620 		if (!fhold(fp)) {
3621 			fdunused(fdp, indx);
3622 			FILEDESC_XUNLOCK(fdp);
3623 			return (EBADF);
3624 		}
3625 		newfde = &fdp->fd_ofiles[indx];
3626 		oldfde = &fdp->fd_ofiles[dfd];
3627 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3628 #ifdef CAPABILITIES
3629 		seqc_write_begin(&newfde->fde_seqc);
3630 #endif
3631 		fde_copy(oldfde, newfde);
3632 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3633 		    ioctls);
3634 #ifdef CAPABILITIES
3635 		seqc_write_end(&newfde->fde_seqc);
3636 #endif
3637 		break;
3638 	case ENXIO:
3639 		/*
3640 		 * Steal away the file pointer from dfd and stuff it into indx.
3641 		 */
3642 		newfde = &fdp->fd_ofiles[indx];
3643 		oldfde = &fdp->fd_ofiles[dfd];
3644 #ifdef CAPABILITIES
3645 		seqc_write_begin(&oldfde->fde_seqc);
3646 		seqc_write_begin(&newfde->fde_seqc);
3647 #endif
3648 		fde_copy(oldfde, newfde);
3649 		oldfde->fde_file = NULL;
3650 		fdunused(fdp, dfd);
3651 #ifdef CAPABILITIES
3652 		seqc_write_end(&newfde->fde_seqc);
3653 		seqc_write_end(&oldfde->fde_seqc);
3654 #endif
3655 		break;
3656 	}
3657 	FILEDESC_XUNLOCK(fdp);
3658 	*indxp = indx;
3659 	return (0);
3660 }
3661 
3662 /*
3663  * This sysctl determines if we will allow a process to chroot(2) if it
3664  * has a directory open:
3665  *	0: disallowed for all processes.
3666  *	1: allowed for processes that were not already chroot(2)'ed.
3667  *	2: allowed for all processes.
3668  */
3669 
3670 static int chroot_allow_open_directories = 1;
3671 
3672 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3673     &chroot_allow_open_directories, 0,
3674     "Allow a process to chroot(2) if it has a directory open");
3675 
3676 /*
3677  * Helper function for raised chroot(2) security function:  Refuse if
3678  * any filedescriptors are open directories.
3679  */
3680 static int
3681 chroot_refuse_vdir_fds(struct filedesc *fdp)
3682 {
3683 	struct vnode *vp;
3684 	struct file *fp;
3685 	int i;
3686 
3687 	FILEDESC_LOCK_ASSERT(fdp);
3688 
3689 	FILEDESC_FOREACH_FP(fdp, i, fp) {
3690 		if (fp->f_type == DTYPE_VNODE) {
3691 			vp = fp->f_vnode;
3692 			if (vp->v_type == VDIR)
3693 				return (EPERM);
3694 		}
3695 	}
3696 	return (0);
3697 }
3698 
3699 static void
3700 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3701 {
3702 
3703 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3704 		vrefact(oldpwd->pwd_cdir);
3705 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3706 	}
3707 
3708 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3709 		vrefact(oldpwd->pwd_rdir);
3710 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3711 	}
3712 
3713 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3714 		vrefact(oldpwd->pwd_jdir);
3715 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3716 	}
3717 }
3718 
3719 struct pwd *
3720 pwd_hold_pwddesc(struct pwddesc *pdp)
3721 {
3722 	struct pwd *pwd;
3723 
3724 	PWDDESC_ASSERT_XLOCKED(pdp);
3725 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3726 	if (pwd != NULL)
3727 		refcount_acquire(&pwd->pwd_refcount);
3728 	return (pwd);
3729 }
3730 
3731 bool
3732 pwd_hold_smr(struct pwd *pwd)
3733 {
3734 
3735 	MPASS(pwd != NULL);
3736 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3737 		return (true);
3738 	}
3739 	return (false);
3740 }
3741 
3742 struct pwd *
3743 pwd_hold(struct thread *td)
3744 {
3745 	struct pwddesc *pdp;
3746 	struct pwd *pwd;
3747 
3748 	pdp = td->td_proc->p_pd;
3749 
3750 	vfs_smr_enter();
3751 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3752 	if (pwd_hold_smr(pwd)) {
3753 		vfs_smr_exit();
3754 		return (pwd);
3755 	}
3756 	vfs_smr_exit();
3757 	PWDDESC_XLOCK(pdp);
3758 	pwd = pwd_hold_pwddesc(pdp);
3759 	MPASS(pwd != NULL);
3760 	PWDDESC_XUNLOCK(pdp);
3761 	return (pwd);
3762 }
3763 
3764 struct pwd *
3765 pwd_hold_proc(struct proc *p)
3766 {
3767 	struct pwddesc *pdp;
3768 	struct pwd *pwd;
3769 
3770 	PROC_ASSERT_HELD(p);
3771 	PROC_LOCK(p);
3772 	pdp = pdhold(p);
3773 	MPASS(pdp != NULL);
3774 	PROC_UNLOCK(p);
3775 
3776 	PWDDESC_XLOCK(pdp);
3777 	pwd = pwd_hold_pwddesc(pdp);
3778 	MPASS(pwd != NULL);
3779 	PWDDESC_XUNLOCK(pdp);
3780 	pddrop(pdp);
3781 	return (pwd);
3782 }
3783 
3784 static struct pwd *
3785 pwd_alloc(void)
3786 {
3787 	struct pwd *pwd;
3788 
3789 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3790 	bzero(pwd, sizeof(*pwd));
3791 	refcount_init(&pwd->pwd_refcount, 1);
3792 	return (pwd);
3793 }
3794 
3795 void
3796 pwd_drop(struct pwd *pwd)
3797 {
3798 
3799 	if (!refcount_release(&pwd->pwd_refcount))
3800 		return;
3801 
3802 	if (pwd->pwd_cdir != NULL)
3803 		vrele(pwd->pwd_cdir);
3804 	if (pwd->pwd_rdir != NULL)
3805 		vrele(pwd->pwd_rdir);
3806 	if (pwd->pwd_jdir != NULL)
3807 		vrele(pwd->pwd_jdir);
3808 	uma_zfree_smr(pwd_zone, pwd);
3809 }
3810 
3811 /*
3812 * The caller is responsible for invoking priv_check() and
3813 * mac_vnode_check_chroot() to authorize this operation.
3814 */
3815 int
3816 pwd_chroot(struct thread *td, struct vnode *vp)
3817 {
3818 	struct pwddesc *pdp;
3819 	struct filedesc *fdp;
3820 	struct pwd *newpwd, *oldpwd;
3821 	int error;
3822 
3823 	fdp = td->td_proc->p_fd;
3824 	pdp = td->td_proc->p_pd;
3825 	newpwd = pwd_alloc();
3826 	FILEDESC_SLOCK(fdp);
3827 	PWDDESC_XLOCK(pdp);
3828 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3829 	if (chroot_allow_open_directories == 0 ||
3830 	    (chroot_allow_open_directories == 1 &&
3831 	    oldpwd->pwd_rdir != rootvnode)) {
3832 		error = chroot_refuse_vdir_fds(fdp);
3833 		FILEDESC_SUNLOCK(fdp);
3834 		if (error != 0) {
3835 			PWDDESC_XUNLOCK(pdp);
3836 			pwd_drop(newpwd);
3837 			return (error);
3838 		}
3839 	} else {
3840 		FILEDESC_SUNLOCK(fdp);
3841 	}
3842 
3843 	vrefact(vp);
3844 	newpwd->pwd_rdir = vp;
3845 	if (oldpwd->pwd_jdir == NULL) {
3846 		vrefact(vp);
3847 		newpwd->pwd_jdir = vp;
3848 	}
3849 	pwd_fill(oldpwd, newpwd);
3850 	pwd_set(pdp, newpwd);
3851 	PWDDESC_XUNLOCK(pdp);
3852 	pwd_drop(oldpwd);
3853 	return (0);
3854 }
3855 
3856 void
3857 pwd_chdir(struct thread *td, struct vnode *vp)
3858 {
3859 	struct pwddesc *pdp;
3860 	struct pwd *newpwd, *oldpwd;
3861 
3862 	VNPASS(vp->v_usecount > 0, vp);
3863 
3864 	newpwd = pwd_alloc();
3865 	pdp = td->td_proc->p_pd;
3866 	PWDDESC_XLOCK(pdp);
3867 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3868 	newpwd->pwd_cdir = vp;
3869 	pwd_fill(oldpwd, newpwd);
3870 	pwd_set(pdp, newpwd);
3871 	PWDDESC_XUNLOCK(pdp);
3872 	pwd_drop(oldpwd);
3873 }
3874 
3875 /*
3876  * jail_attach(2) changes both root and working directories.
3877  */
3878 int
3879 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3880 {
3881 	struct pwddesc *pdp;
3882 	struct filedesc *fdp;
3883 	struct pwd *newpwd, *oldpwd;
3884 	int error;
3885 
3886 	fdp = td->td_proc->p_fd;
3887 	pdp = td->td_proc->p_pd;
3888 	newpwd = pwd_alloc();
3889 	FILEDESC_SLOCK(fdp);
3890 	PWDDESC_XLOCK(pdp);
3891 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3892 	error = chroot_refuse_vdir_fds(fdp);
3893 	FILEDESC_SUNLOCK(fdp);
3894 	if (error != 0) {
3895 		PWDDESC_XUNLOCK(pdp);
3896 		pwd_drop(newpwd);
3897 		return (error);
3898 	}
3899 
3900 	vrefact(vp);
3901 	newpwd->pwd_rdir = vp;
3902 	vrefact(vp);
3903 	newpwd->pwd_cdir = vp;
3904 	if (oldpwd->pwd_jdir == NULL) {
3905 		vrefact(vp);
3906 		newpwd->pwd_jdir = vp;
3907 	}
3908 	pwd_fill(oldpwd, newpwd);
3909 	pwd_set(pdp, newpwd);
3910 	PWDDESC_XUNLOCK(pdp);
3911 	pwd_drop(oldpwd);
3912 	return (0);
3913 }
3914 
3915 void
3916 pwd_ensure_dirs(void)
3917 {
3918 	struct pwddesc *pdp;
3919 	struct pwd *oldpwd, *newpwd;
3920 
3921 	pdp = curproc->p_pd;
3922 	PWDDESC_XLOCK(pdp);
3923 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3924 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3925 		PWDDESC_XUNLOCK(pdp);
3926 		return;
3927 	}
3928 	PWDDESC_XUNLOCK(pdp);
3929 
3930 	newpwd = pwd_alloc();
3931 	PWDDESC_XLOCK(pdp);
3932 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3933 	pwd_fill(oldpwd, newpwd);
3934 	if (newpwd->pwd_cdir == NULL) {
3935 		vrefact(rootvnode);
3936 		newpwd->pwd_cdir = rootvnode;
3937 	}
3938 	if (newpwd->pwd_rdir == NULL) {
3939 		vrefact(rootvnode);
3940 		newpwd->pwd_rdir = rootvnode;
3941 	}
3942 	pwd_set(pdp, newpwd);
3943 	PWDDESC_XUNLOCK(pdp);
3944 	pwd_drop(oldpwd);
3945 }
3946 
3947 void
3948 pwd_set_rootvnode(void)
3949 {
3950 	struct pwddesc *pdp;
3951 	struct pwd *oldpwd, *newpwd;
3952 
3953 	pdp = curproc->p_pd;
3954 
3955 	newpwd = pwd_alloc();
3956 	PWDDESC_XLOCK(pdp);
3957 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3958 	vrefact(rootvnode);
3959 	newpwd->pwd_cdir = rootvnode;
3960 	vrefact(rootvnode);
3961 	newpwd->pwd_rdir = rootvnode;
3962 	pwd_fill(oldpwd, newpwd);
3963 	pwd_set(pdp, newpwd);
3964 	PWDDESC_XUNLOCK(pdp);
3965 	pwd_drop(oldpwd);
3966 }
3967 
3968 /*
3969  * Scan all active processes and prisons to see if any of them have a current
3970  * or root directory of `olddp'. If so, replace them with the new mount point.
3971  */
3972 void
3973 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3974 {
3975 	struct pwddesc *pdp;
3976 	struct pwd *newpwd, *oldpwd;
3977 	struct prison *pr;
3978 	struct proc *p;
3979 	int nrele;
3980 
3981 	if (vrefcnt(olddp) == 1)
3982 		return;
3983 	nrele = 0;
3984 	newpwd = pwd_alloc();
3985 	sx_slock(&allproc_lock);
3986 	FOREACH_PROC_IN_SYSTEM(p) {
3987 		PROC_LOCK(p);
3988 		pdp = pdhold(p);
3989 		PROC_UNLOCK(p);
3990 		if (pdp == NULL)
3991 			continue;
3992 		PWDDESC_XLOCK(pdp);
3993 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3994 		if (oldpwd == NULL ||
3995 		    (oldpwd->pwd_cdir != olddp &&
3996 		    oldpwd->pwd_rdir != olddp &&
3997 		    oldpwd->pwd_jdir != olddp)) {
3998 			PWDDESC_XUNLOCK(pdp);
3999 			pddrop(pdp);
4000 			continue;
4001 		}
4002 		if (oldpwd->pwd_cdir == olddp) {
4003 			vrefact(newdp);
4004 			newpwd->pwd_cdir = newdp;
4005 		}
4006 		if (oldpwd->pwd_rdir == olddp) {
4007 			vrefact(newdp);
4008 			newpwd->pwd_rdir = newdp;
4009 		}
4010 		if (oldpwd->pwd_jdir == olddp) {
4011 			vrefact(newdp);
4012 			newpwd->pwd_jdir = newdp;
4013 		}
4014 		pwd_fill(oldpwd, newpwd);
4015 		pwd_set(pdp, newpwd);
4016 		PWDDESC_XUNLOCK(pdp);
4017 		pwd_drop(oldpwd);
4018 		pddrop(pdp);
4019 		newpwd = pwd_alloc();
4020 	}
4021 	sx_sunlock(&allproc_lock);
4022 	pwd_drop(newpwd);
4023 	if (rootvnode == olddp) {
4024 		vrefact(newdp);
4025 		rootvnode = newdp;
4026 		nrele++;
4027 	}
4028 	mtx_lock(&prison0.pr_mtx);
4029 	if (prison0.pr_root == olddp) {
4030 		vrefact(newdp);
4031 		prison0.pr_root = newdp;
4032 		nrele++;
4033 	}
4034 	mtx_unlock(&prison0.pr_mtx);
4035 	sx_slock(&allprison_lock);
4036 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4037 		mtx_lock(&pr->pr_mtx);
4038 		if (pr->pr_root == olddp) {
4039 			vrefact(newdp);
4040 			pr->pr_root = newdp;
4041 			nrele++;
4042 		}
4043 		mtx_unlock(&pr->pr_mtx);
4044 	}
4045 	sx_sunlock(&allprison_lock);
4046 	while (nrele--)
4047 		vrele(olddp);
4048 }
4049 
4050 struct filedesc_to_leader *
4051 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4052 {
4053 	struct filedesc_to_leader *fdtol;
4054 
4055 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4056 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4057 	fdtol->fdl_refcount = 1;
4058 	fdtol->fdl_holdcount = 0;
4059 	fdtol->fdl_wakeup = 0;
4060 	fdtol->fdl_leader = leader;
4061 	if (old != NULL) {
4062 		FILEDESC_XLOCK(fdp);
4063 		fdtol->fdl_next = old->fdl_next;
4064 		fdtol->fdl_prev = old;
4065 		old->fdl_next = fdtol;
4066 		fdtol->fdl_next->fdl_prev = fdtol;
4067 		FILEDESC_XUNLOCK(fdp);
4068 	} else {
4069 		fdtol->fdl_next = fdtol;
4070 		fdtol->fdl_prev = fdtol;
4071 	}
4072 	return (fdtol);
4073 }
4074 
4075 static int
4076 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4077 {
4078 	NDSLOTTYPE *map;
4079 	struct filedesc *fdp;
4080 	u_int namelen;
4081 	int count, off, minoff;
4082 
4083 	namelen = arg2;
4084 	if (namelen != 1)
4085 		return (EINVAL);
4086 
4087 	if (*(int *)arg1 != 0)
4088 		return (EINVAL);
4089 
4090 	fdp = curproc->p_fd;
4091 	count = 0;
4092 	FILEDESC_SLOCK(fdp);
4093 	map = fdp->fd_map;
4094 	off = NDSLOT(fdp->fd_nfiles - 1);
4095 	for (minoff = NDSLOT(0); off >= minoff; --off)
4096 		count += bitcountl(map[off]);
4097 	FILEDESC_SUNLOCK(fdp);
4098 
4099 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4100 }
4101 
4102 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4103     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4104     "Number of open file descriptors");
4105 
4106 /*
4107  * Get file structures globally.
4108  */
4109 static int
4110 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4111 {
4112 	struct xfile xf;
4113 	struct filedesc *fdp;
4114 	struct file *fp;
4115 	struct proc *p;
4116 	int error, n;
4117 
4118 	error = sysctl_wire_old_buffer(req, 0);
4119 	if (error != 0)
4120 		return (error);
4121 	if (req->oldptr == NULL) {
4122 		n = 0;
4123 		sx_slock(&allproc_lock);
4124 		FOREACH_PROC_IN_SYSTEM(p) {
4125 			PROC_LOCK(p);
4126 			if (p->p_state == PRS_NEW) {
4127 				PROC_UNLOCK(p);
4128 				continue;
4129 			}
4130 			fdp = fdhold(p);
4131 			PROC_UNLOCK(p);
4132 			if (fdp == NULL)
4133 				continue;
4134 			/* overestimates sparse tables. */
4135 			n += fdp->fd_nfiles;
4136 			fddrop(fdp);
4137 		}
4138 		sx_sunlock(&allproc_lock);
4139 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4140 	}
4141 	error = 0;
4142 	bzero(&xf, sizeof(xf));
4143 	xf.xf_size = sizeof(xf);
4144 	sx_slock(&allproc_lock);
4145 	FOREACH_PROC_IN_SYSTEM(p) {
4146 		PROC_LOCK(p);
4147 		if (p->p_state == PRS_NEW) {
4148 			PROC_UNLOCK(p);
4149 			continue;
4150 		}
4151 		if (p_cansee(req->td, p) != 0) {
4152 			PROC_UNLOCK(p);
4153 			continue;
4154 		}
4155 		xf.xf_pid = p->p_pid;
4156 		xf.xf_uid = p->p_ucred->cr_uid;
4157 		fdp = fdhold(p);
4158 		PROC_UNLOCK(p);
4159 		if (fdp == NULL)
4160 			continue;
4161 		FILEDESC_SLOCK(fdp);
4162 		FILEDESC_FOREACH_FP(fdp, n, fp) {
4163 			if (refcount_load(&fdp->fd_refcnt) == 0)
4164 				break;
4165 			xf.xf_fd = n;
4166 			xf.xf_file = (uintptr_t)fp;
4167 			xf.xf_data = (uintptr_t)fp->f_data;
4168 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4169 			xf.xf_type = (uintptr_t)fp->f_type;
4170 			xf.xf_count = refcount_load(&fp->f_count);
4171 			xf.xf_msgcount = 0;
4172 			xf.xf_offset = foffset_get(fp);
4173 			xf.xf_flag = fp->f_flag;
4174 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4175 			if (error)
4176 				break;
4177 		}
4178 		FILEDESC_SUNLOCK(fdp);
4179 		fddrop(fdp);
4180 		if (error)
4181 			break;
4182 	}
4183 	sx_sunlock(&allproc_lock);
4184 	return (error);
4185 }
4186 
4187 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4188     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4189 
4190 #ifdef KINFO_FILE_SIZE
4191 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4192 #endif
4193 
4194 static int
4195 xlate_fflags(int fflags)
4196 {
4197 	static const struct {
4198 		int	fflag;
4199 		int	kf_fflag;
4200 	} fflags_table[] = {
4201 		{ FAPPEND, KF_FLAG_APPEND },
4202 		{ FASYNC, KF_FLAG_ASYNC },
4203 		{ FFSYNC, KF_FLAG_FSYNC },
4204 		{ FHASLOCK, KF_FLAG_HASLOCK },
4205 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4206 		{ FREAD, KF_FLAG_READ },
4207 		{ FWRITE, KF_FLAG_WRITE },
4208 		{ O_CREAT, KF_FLAG_CREAT },
4209 		{ O_DIRECT, KF_FLAG_DIRECT },
4210 		{ O_EXCL, KF_FLAG_EXCL },
4211 		{ O_EXEC, KF_FLAG_EXEC },
4212 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4213 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4214 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4215 		{ O_TRUNC, KF_FLAG_TRUNC }
4216 	};
4217 	unsigned int i;
4218 	int kflags;
4219 
4220 	kflags = 0;
4221 	for (i = 0; i < nitems(fflags_table); i++)
4222 		if (fflags & fflags_table[i].fflag)
4223 			kflags |=  fflags_table[i].kf_fflag;
4224 	return (kflags);
4225 }
4226 
4227 /* Trim unused data from kf_path by truncating the structure size. */
4228 void
4229 pack_kinfo(struct kinfo_file *kif)
4230 {
4231 
4232 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4233 	    strlen(kif->kf_path) + 1;
4234 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4235 }
4236 
4237 static void
4238 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4239     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4240 {
4241 	int error;
4242 
4243 	bzero(kif, sizeof(*kif));
4244 
4245 	/* Set a default type to allow for empty fill_kinfo() methods. */
4246 	kif->kf_type = KF_TYPE_UNKNOWN;
4247 	kif->kf_flags = xlate_fflags(fp->f_flag);
4248 	if (rightsp != NULL)
4249 		kif->kf_cap_rights = *rightsp;
4250 	else
4251 		cap_rights_init_zero(&kif->kf_cap_rights);
4252 	kif->kf_fd = fd;
4253 	kif->kf_ref_count = refcount_load(&fp->f_count);
4254 	kif->kf_offset = foffset_get(fp);
4255 
4256 	/*
4257 	 * This may drop the filedesc lock, so the 'fp' cannot be
4258 	 * accessed after this call.
4259 	 */
4260 	error = fo_fill_kinfo(fp, kif, fdp);
4261 	if (error == 0)
4262 		kif->kf_status |= KF_ATTR_VALID;
4263 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4264 		pack_kinfo(kif);
4265 	else
4266 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4267 }
4268 
4269 static void
4270 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4271     struct kinfo_file *kif, int flags)
4272 {
4273 	int error;
4274 
4275 	bzero(kif, sizeof(*kif));
4276 
4277 	kif->kf_type = KF_TYPE_VNODE;
4278 	error = vn_fill_kinfo_vnode(vp, kif);
4279 	if (error == 0)
4280 		kif->kf_status |= KF_ATTR_VALID;
4281 	kif->kf_flags = xlate_fflags(fflags);
4282 	cap_rights_init_zero(&kif->kf_cap_rights);
4283 	kif->kf_fd = fd;
4284 	kif->kf_ref_count = -1;
4285 	kif->kf_offset = -1;
4286 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4287 		pack_kinfo(kif);
4288 	else
4289 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4290 	vrele(vp);
4291 }
4292 
4293 struct export_fd_buf {
4294 	struct filedesc		*fdp;
4295 	struct pwddesc	*pdp;
4296 	struct sbuf 		*sb;
4297 	ssize_t			remainder;
4298 	struct kinfo_file	kif;
4299 	int			flags;
4300 };
4301 
4302 static int
4303 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4304 {
4305 	struct kinfo_file *kif;
4306 
4307 	kif = &efbuf->kif;
4308 	if (efbuf->remainder != -1) {
4309 		if (efbuf->remainder < kif->kf_structsize)
4310 			return (ENOMEM);
4311 		efbuf->remainder -= kif->kf_structsize;
4312 	}
4313 	if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4314 		return (sbuf_error(efbuf->sb));
4315 	return (0);
4316 }
4317 
4318 static int
4319 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4320     struct export_fd_buf *efbuf)
4321 {
4322 	int error;
4323 
4324 	if (efbuf->remainder == 0)
4325 		return (ENOMEM);
4326 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4327 	    efbuf->flags);
4328 	FILEDESC_SUNLOCK(efbuf->fdp);
4329 	error = export_kinfo_to_sb(efbuf);
4330 	FILEDESC_SLOCK(efbuf->fdp);
4331 	return (error);
4332 }
4333 
4334 static int
4335 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4336     struct export_fd_buf *efbuf)
4337 {
4338 	int error;
4339 
4340 	if (efbuf->remainder == 0)
4341 		return (ENOMEM);
4342 	if (efbuf->pdp != NULL)
4343 		PWDDESC_XUNLOCK(efbuf->pdp);
4344 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4345 	error = export_kinfo_to_sb(efbuf);
4346 	if (efbuf->pdp != NULL)
4347 		PWDDESC_XLOCK(efbuf->pdp);
4348 	return (error);
4349 }
4350 
4351 /*
4352  * Store a process file descriptor information to sbuf.
4353  *
4354  * Takes a locked proc as argument, and returns with the proc unlocked.
4355  */
4356 int
4357 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4358     int flags)
4359 {
4360 	struct file *fp;
4361 	struct filedesc *fdp;
4362 	struct pwddesc *pdp;
4363 	struct export_fd_buf *efbuf;
4364 	struct vnode *cttyvp, *textvp, *tracevp;
4365 	struct pwd *pwd;
4366 	int error, i;
4367 	cap_rights_t rights;
4368 
4369 	PROC_LOCK_ASSERT(p, MA_OWNED);
4370 
4371 	/* ktrace vnode */
4372 	tracevp = ktr_get_tracevp(p, true);
4373 	/* text vnode */
4374 	textvp = p->p_textvp;
4375 	if (textvp != NULL)
4376 		vrefact(textvp);
4377 	/* Controlling tty. */
4378 	cttyvp = NULL;
4379 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4380 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4381 		if (cttyvp != NULL)
4382 			vrefact(cttyvp);
4383 	}
4384 	fdp = fdhold(p);
4385 	pdp = pdhold(p);
4386 	PROC_UNLOCK(p);
4387 
4388 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4389 	efbuf->fdp = NULL;
4390 	efbuf->pdp = NULL;
4391 	efbuf->sb = sb;
4392 	efbuf->remainder = maxlen;
4393 	efbuf->flags = flags;
4394 
4395 	error = 0;
4396 	if (tracevp != NULL)
4397 		error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4398 		    FREAD | FWRITE, efbuf);
4399 	if (error == 0 && textvp != NULL)
4400 		error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4401 		    efbuf);
4402 	if (error == 0 && cttyvp != NULL)
4403 		error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4404 		    FREAD | FWRITE, efbuf);
4405 	if (error != 0 || pdp == NULL || fdp == NULL)
4406 		goto fail;
4407 	efbuf->fdp = fdp;
4408 	efbuf->pdp = pdp;
4409 	PWDDESC_XLOCK(pdp);
4410 	pwd = pwd_hold_pwddesc(pdp);
4411 	if (pwd != NULL) {
4412 		/* working directory */
4413 		if (pwd->pwd_cdir != NULL) {
4414 			vrefact(pwd->pwd_cdir);
4415 			error = export_vnode_to_sb(pwd->pwd_cdir,
4416 			    KF_FD_TYPE_CWD, FREAD, efbuf);
4417 		}
4418 		/* root directory */
4419 		if (error == 0 && pwd->pwd_rdir != NULL) {
4420 			vrefact(pwd->pwd_rdir);
4421 			error = export_vnode_to_sb(pwd->pwd_rdir,
4422 			    KF_FD_TYPE_ROOT, FREAD, efbuf);
4423 		}
4424 		/* jail directory */
4425 		if (error == 0 && pwd->pwd_jdir != NULL) {
4426 			vrefact(pwd->pwd_jdir);
4427 			error = export_vnode_to_sb(pwd->pwd_jdir,
4428 			    KF_FD_TYPE_JAIL, FREAD, efbuf);
4429 		}
4430 	}
4431 	PWDDESC_XUNLOCK(pdp);
4432 	if (error != 0)
4433 		goto fail;
4434 	if (pwd != NULL)
4435 		pwd_drop(pwd);
4436 	FILEDESC_SLOCK(fdp);
4437 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4438 		if (refcount_load(&fdp->fd_refcnt) == 0)
4439 			break;
4440 #ifdef CAPABILITIES
4441 		rights = *cap_rights(fdp, i);
4442 #else /* !CAPABILITIES */
4443 		rights = cap_no_rights;
4444 #endif
4445 		/*
4446 		 * Create sysctl entry.  It is OK to drop the filedesc
4447 		 * lock inside of export_file_to_sb() as we will
4448 		 * re-validate and re-evaluate its properties when the
4449 		 * loop continues.
4450 		 */
4451 		error = export_file_to_sb(fp, i, &rights, efbuf);
4452 		if (error != 0)
4453 			break;
4454 	}
4455 	FILEDESC_SUNLOCK(fdp);
4456 fail:
4457 	if (fdp != NULL)
4458 		fddrop(fdp);
4459 	if (pdp != NULL)
4460 		pddrop(pdp);
4461 	free(efbuf, M_TEMP);
4462 	return (error);
4463 }
4464 
4465 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4466 
4467 /*
4468  * Get per-process file descriptors for use by procstat(1), et al.
4469  */
4470 static int
4471 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4472 {
4473 	struct sbuf sb;
4474 	struct proc *p;
4475 	ssize_t maxlen;
4476 	u_int namelen;
4477 	int error, error2, *name;
4478 
4479 	namelen = arg2;
4480 	if (namelen != 1)
4481 		return (EINVAL);
4482 
4483 	name = (int *)arg1;
4484 
4485 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4486 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4487 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4488 	if (error != 0) {
4489 		sbuf_delete(&sb);
4490 		return (error);
4491 	}
4492 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4493 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4494 	    KERN_FILEDESC_PACK_KINFO);
4495 	error2 = sbuf_finish(&sb);
4496 	sbuf_delete(&sb);
4497 	return (error != 0 ? error : error2);
4498 }
4499 
4500 #ifdef COMPAT_FREEBSD7
4501 #ifdef KINFO_OFILE_SIZE
4502 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4503 #endif
4504 
4505 static void
4506 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4507 {
4508 
4509 	okif->kf_structsize = sizeof(*okif);
4510 	okif->kf_type = kif->kf_type;
4511 	okif->kf_fd = kif->kf_fd;
4512 	okif->kf_ref_count = kif->kf_ref_count;
4513 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4514 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4515 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4516 	okif->kf_offset = kif->kf_offset;
4517 	if (kif->kf_type == KF_TYPE_VNODE)
4518 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4519 	else
4520 		okif->kf_vnode_type = KF_VTYPE_VNON;
4521 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4522 	if (kif->kf_type == KF_TYPE_SOCKET) {
4523 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4524 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4525 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4526 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4527 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4528 	} else {
4529 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4530 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4531 	}
4532 }
4533 
4534 static int
4535 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4536     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4537 {
4538 	int error;
4539 
4540 	vrefact(vp);
4541 	PWDDESC_XUNLOCK(pdp);
4542 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4543 	kinfo_to_okinfo(kif, okif);
4544 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4545 	PWDDESC_XLOCK(pdp);
4546 	return (error);
4547 }
4548 
4549 /*
4550  * Get per-process file descriptors for use by procstat(1), et al.
4551  */
4552 static int
4553 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4554 {
4555 	struct kinfo_ofile *okif;
4556 	struct kinfo_file *kif;
4557 	struct filedesc *fdp;
4558 	struct pwddesc *pdp;
4559 	struct pwd *pwd;
4560 	u_int namelen;
4561 	int error, i, *name;
4562 	struct file *fp;
4563 	struct proc *p;
4564 
4565 	namelen = arg2;
4566 	if (namelen != 1)
4567 		return (EINVAL);
4568 
4569 	name = (int *)arg1;
4570 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4571 	if (error != 0)
4572 		return (error);
4573 	fdp = fdhold(p);
4574 	if (fdp != NULL)
4575 		pdp = pdhold(p);
4576 	PROC_UNLOCK(p);
4577 	if (fdp == NULL || pdp == NULL) {
4578 		if (fdp != NULL)
4579 			fddrop(fdp);
4580 		return (ENOENT);
4581 	}
4582 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4583 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4584 	PWDDESC_XLOCK(pdp);
4585 	pwd = pwd_hold_pwddesc(pdp);
4586 	if (pwd != NULL) {
4587 		if (pwd->pwd_cdir != NULL)
4588 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4589 			    okif, pdp, req);
4590 		if (pwd->pwd_rdir != NULL)
4591 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4592 			    okif, pdp, req);
4593 		if (pwd->pwd_jdir != NULL)
4594 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4595 			    okif, pdp, req);
4596 	}
4597 	PWDDESC_XUNLOCK(pdp);
4598 	if (pwd != NULL)
4599 		pwd_drop(pwd);
4600 	FILEDESC_SLOCK(fdp);
4601 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4602 		if (refcount_load(&fdp->fd_refcnt) == 0)
4603 			break;
4604 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4605 		    KERN_FILEDESC_PACK_KINFO);
4606 		FILEDESC_SUNLOCK(fdp);
4607 		kinfo_to_okinfo(kif, okif);
4608 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4609 		FILEDESC_SLOCK(fdp);
4610 		if (error)
4611 			break;
4612 	}
4613 	FILEDESC_SUNLOCK(fdp);
4614 	fddrop(fdp);
4615 	pddrop(pdp);
4616 	free(kif, M_TEMP);
4617 	free(okif, M_TEMP);
4618 	return (0);
4619 }
4620 
4621 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4622     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4623     "Process ofiledesc entries");
4624 #endif	/* COMPAT_FREEBSD7 */
4625 
4626 int
4627 vntype_to_kinfo(int vtype)
4628 {
4629 	struct {
4630 		int	vtype;
4631 		int	kf_vtype;
4632 	} vtypes_table[] = {
4633 		{ VBAD, KF_VTYPE_VBAD },
4634 		{ VBLK, KF_VTYPE_VBLK },
4635 		{ VCHR, KF_VTYPE_VCHR },
4636 		{ VDIR, KF_VTYPE_VDIR },
4637 		{ VFIFO, KF_VTYPE_VFIFO },
4638 		{ VLNK, KF_VTYPE_VLNK },
4639 		{ VNON, KF_VTYPE_VNON },
4640 		{ VREG, KF_VTYPE_VREG },
4641 		{ VSOCK, KF_VTYPE_VSOCK }
4642 	};
4643 	unsigned int i;
4644 
4645 	/*
4646 	 * Perform vtype translation.
4647 	 */
4648 	for (i = 0; i < nitems(vtypes_table); i++)
4649 		if (vtypes_table[i].vtype == vtype)
4650 			return (vtypes_table[i].kf_vtype);
4651 
4652 	return (KF_VTYPE_UNKNOWN);
4653 }
4654 
4655 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4656     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4657     "Process filedesc entries");
4658 
4659 /*
4660  * Store a process current working directory information to sbuf.
4661  *
4662  * Takes a locked proc as argument, and returns with the proc unlocked.
4663  */
4664 int
4665 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4666 {
4667 	struct pwddesc *pdp;
4668 	struct pwd *pwd;
4669 	struct export_fd_buf *efbuf;
4670 	struct vnode *cdir;
4671 	int error;
4672 
4673 	PROC_LOCK_ASSERT(p, MA_OWNED);
4674 
4675 	pdp = pdhold(p);
4676 	PROC_UNLOCK(p);
4677 	if (pdp == NULL)
4678 		return (EINVAL);
4679 
4680 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4681 	efbuf->fdp = NULL;
4682 	efbuf->pdp = pdp;
4683 	efbuf->sb = sb;
4684 	efbuf->remainder = maxlen;
4685 	efbuf->flags = 0;
4686 
4687 	PWDDESC_XLOCK(pdp);
4688 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4689 	cdir = pwd->pwd_cdir;
4690 	if (cdir == NULL) {
4691 		error = EINVAL;
4692 	} else {
4693 		vrefact(cdir);
4694 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4695 	}
4696 	PWDDESC_XUNLOCK(pdp);
4697 	pddrop(pdp);
4698 	free(efbuf, M_TEMP);
4699 	return (error);
4700 }
4701 
4702 /*
4703  * Get per-process current working directory.
4704  */
4705 static int
4706 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4707 {
4708 	struct sbuf sb;
4709 	struct proc *p;
4710 	ssize_t maxlen;
4711 	u_int namelen;
4712 	int error, error2, *name;
4713 
4714 	namelen = arg2;
4715 	if (namelen != 1)
4716 		return (EINVAL);
4717 
4718 	name = (int *)arg1;
4719 
4720 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4721 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4722 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4723 	if (error != 0) {
4724 		sbuf_delete(&sb);
4725 		return (error);
4726 	}
4727 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4728 	error = kern_proc_cwd_out(p, &sb, maxlen);
4729 	error2 = sbuf_finish(&sb);
4730 	sbuf_delete(&sb);
4731 	return (error != 0 ? error : error2);
4732 }
4733 
4734 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4735     sysctl_kern_proc_cwd, "Process current working directory");
4736 
4737 #ifdef DDB
4738 /*
4739  * For the purposes of debugging, generate a human-readable string for the
4740  * file type.
4741  */
4742 static const char *
4743 file_type_to_name(short type)
4744 {
4745 
4746 	switch (type) {
4747 	case 0:
4748 		return ("zero");
4749 	case DTYPE_VNODE:
4750 		return ("vnode");
4751 	case DTYPE_SOCKET:
4752 		return ("socket");
4753 	case DTYPE_PIPE:
4754 		return ("pipe");
4755 	case DTYPE_FIFO:
4756 		return ("fifo");
4757 	case DTYPE_KQUEUE:
4758 		return ("kqueue");
4759 	case DTYPE_CRYPTO:
4760 		return ("crypto");
4761 	case DTYPE_MQUEUE:
4762 		return ("mqueue");
4763 	case DTYPE_SHM:
4764 		return ("shm");
4765 	case DTYPE_SEM:
4766 		return ("ksem");
4767 	case DTYPE_PTS:
4768 		return ("pts");
4769 	case DTYPE_DEV:
4770 		return ("dev");
4771 	case DTYPE_PROCDESC:
4772 		return ("proc");
4773 	case DTYPE_EVENTFD:
4774 		return ("eventfd");
4775 	case DTYPE_LINUXTFD:
4776 		return ("ltimer");
4777 	default:
4778 		return ("unkn");
4779 	}
4780 }
4781 
4782 /*
4783  * For the purposes of debugging, identify a process (if any, perhaps one of
4784  * many) that references the passed file in its file descriptor array. Return
4785  * NULL if none.
4786  */
4787 static struct proc *
4788 file_to_first_proc(struct file *fp)
4789 {
4790 	struct filedesc *fdp;
4791 	struct proc *p;
4792 	int n;
4793 
4794 	FOREACH_PROC_IN_SYSTEM(p) {
4795 		if (p->p_state == PRS_NEW)
4796 			continue;
4797 		fdp = p->p_fd;
4798 		if (fdp == NULL)
4799 			continue;
4800 		for (n = 0; n < fdp->fd_nfiles; n++) {
4801 			if (fp == fdp->fd_ofiles[n].fde_file)
4802 				return (p);
4803 		}
4804 	}
4805 	return (NULL);
4806 }
4807 
4808 static void
4809 db_print_file(struct file *fp, int header)
4810 {
4811 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4812 	struct proc *p;
4813 
4814 	if (header)
4815 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4816 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4817 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4818 		    "FCmd");
4819 	p = file_to_first_proc(fp);
4820 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4821 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4822 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4823 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4824 
4825 #undef XPTRWIDTH
4826 }
4827 
4828 DB_SHOW_COMMAND(file, db_show_file)
4829 {
4830 	struct file *fp;
4831 
4832 	if (!have_addr) {
4833 		db_printf("usage: show file <addr>\n");
4834 		return;
4835 	}
4836 	fp = (struct file *)addr;
4837 	db_print_file(fp, 1);
4838 }
4839 
4840 DB_SHOW_COMMAND(files, db_show_files)
4841 {
4842 	struct filedesc *fdp;
4843 	struct file *fp;
4844 	struct proc *p;
4845 	int header;
4846 	int n;
4847 
4848 	header = 1;
4849 	FOREACH_PROC_IN_SYSTEM(p) {
4850 		if (p->p_state == PRS_NEW)
4851 			continue;
4852 		if ((fdp = p->p_fd) == NULL)
4853 			continue;
4854 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4855 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4856 				continue;
4857 			db_print_file(fp, header);
4858 			header = 0;
4859 		}
4860 	}
4861 }
4862 #endif
4863 
4864 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4865     &maxfilesperproc, 0, "Maximum files allowed open per process");
4866 
4867 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4868     &maxfiles, 0, "Maximum number of files");
4869 
4870 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4871     &openfiles, 0, "System-wide number of open files");
4872 
4873 /* ARGSUSED*/
4874 static void
4875 filelistinit(void *dummy)
4876 {
4877 
4878 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4879 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4880 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4881 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4882 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4883 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4884 	/*
4885 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4886 	 * the vnode zone.
4887 	 */
4888 	vfs_smr = uma_zone_get_smr(pwd_zone);
4889 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4890 }
4891 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4892 
4893 /*-------------------------------------------------------------------*/
4894 
4895 static int
4896 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4897     int flags, struct thread *td)
4898 {
4899 
4900 	return (EBADF);
4901 }
4902 
4903 static int
4904 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4905     struct thread *td)
4906 {
4907 
4908 	return (EINVAL);
4909 }
4910 
4911 static int
4912 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4913     struct thread *td)
4914 {
4915 
4916 	return (EBADF);
4917 }
4918 
4919 static int
4920 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4921     struct thread *td)
4922 {
4923 
4924 	return (0);
4925 }
4926 
4927 static int
4928 badfo_kqfilter(struct file *fp, struct knote *kn)
4929 {
4930 
4931 	return (EBADF);
4932 }
4933 
4934 static int
4935 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4936 {
4937 
4938 	return (EBADF);
4939 }
4940 
4941 static int
4942 badfo_close(struct file *fp, struct thread *td)
4943 {
4944 
4945 	return (0);
4946 }
4947 
4948 static int
4949 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4950     struct thread *td)
4951 {
4952 
4953 	return (EBADF);
4954 }
4955 
4956 static int
4957 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4958     struct thread *td)
4959 {
4960 
4961 	return (EBADF);
4962 }
4963 
4964 static int
4965 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4966     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4967     struct thread *td)
4968 {
4969 
4970 	return (EBADF);
4971 }
4972 
4973 static int
4974 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4975 {
4976 
4977 	return (0);
4978 }
4979 
4980 struct fileops badfileops = {
4981 	.fo_read = badfo_readwrite,
4982 	.fo_write = badfo_readwrite,
4983 	.fo_truncate = badfo_truncate,
4984 	.fo_ioctl = badfo_ioctl,
4985 	.fo_poll = badfo_poll,
4986 	.fo_kqfilter = badfo_kqfilter,
4987 	.fo_stat = badfo_stat,
4988 	.fo_close = badfo_close,
4989 	.fo_chmod = badfo_chmod,
4990 	.fo_chown = badfo_chown,
4991 	.fo_sendfile = badfo_sendfile,
4992 	.fo_fill_kinfo = badfo_fill_kinfo,
4993 };
4994 
4995 static int
4996 path_poll(struct file *fp, int events, struct ucred *active_cred,
4997     struct thread *td)
4998 {
4999 	return (POLLNVAL);
5000 }
5001 
5002 static int
5003 path_close(struct file *fp, struct thread *td)
5004 {
5005 	MPASS(fp->f_type == DTYPE_VNODE);
5006 	fp->f_ops = &badfileops;
5007 	vdrop(fp->f_vnode);
5008 	return (0);
5009 }
5010 
5011 struct fileops path_fileops = {
5012 	.fo_read = badfo_readwrite,
5013 	.fo_write = badfo_readwrite,
5014 	.fo_truncate = badfo_truncate,
5015 	.fo_ioctl = badfo_ioctl,
5016 	.fo_poll = path_poll,
5017 	.fo_kqfilter = vn_kqfilter_opath,
5018 	.fo_stat = vn_statfile,
5019 	.fo_close = path_close,
5020 	.fo_chmod = badfo_chmod,
5021 	.fo_chown = badfo_chown,
5022 	.fo_sendfile = badfo_sendfile,
5023 	.fo_fill_kinfo = vn_fill_kinfo,
5024 	.fo_flags = DFLAG_PASSABLE,
5025 };
5026 
5027 int
5028 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5029     int flags, struct thread *td)
5030 {
5031 
5032 	return (EOPNOTSUPP);
5033 }
5034 
5035 int
5036 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5037     struct thread *td)
5038 {
5039 
5040 	return (EINVAL);
5041 }
5042 
5043 int
5044 invfo_ioctl(struct file *fp, u_long com, void *data,
5045     struct ucred *active_cred, struct thread *td)
5046 {
5047 
5048 	return (ENOTTY);
5049 }
5050 
5051 int
5052 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5053     struct thread *td)
5054 {
5055 
5056 	return (poll_no_poll(events));
5057 }
5058 
5059 int
5060 invfo_kqfilter(struct file *fp, struct knote *kn)
5061 {
5062 
5063 	return (EINVAL);
5064 }
5065 
5066 int
5067 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5068     struct thread *td)
5069 {
5070 
5071 	return (EINVAL);
5072 }
5073 
5074 int
5075 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5076     struct thread *td)
5077 {
5078 
5079 	return (EINVAL);
5080 }
5081 
5082 int
5083 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5084     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5085     struct thread *td)
5086 {
5087 
5088 	return (EINVAL);
5089 }
5090 
5091 /*-------------------------------------------------------------------*/
5092 
5093 /*
5094  * File Descriptor pseudo-device driver (/dev/fd/).
5095  *
5096  * Opening minor device N dup()s the file (if any) connected to file
5097  * descriptor N belonging to the calling process.  Note that this driver
5098  * consists of only the ``open()'' routine, because all subsequent
5099  * references to this file will be direct to the other driver.
5100  *
5101  * XXX: we could give this one a cloning event handler if necessary.
5102  */
5103 
5104 /* ARGSUSED */
5105 static int
5106 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5107 {
5108 
5109 	/*
5110 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5111 	 * the file descriptor being sought for duplication. The error
5112 	 * return ensures that the vnode for this device will be released
5113 	 * by vn_open. Open will detect this special error and take the
5114 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5115 	 * will simply report the error.
5116 	 */
5117 	td->td_dupfd = dev2unit(dev);
5118 	return (ENODEV);
5119 }
5120 
5121 static struct cdevsw fildesc_cdevsw = {
5122 	.d_version =	D_VERSION,
5123 	.d_open =	fdopen,
5124 	.d_name =	"FD",
5125 };
5126 
5127 static void
5128 fildesc_drvinit(void *unused)
5129 {
5130 	struct cdev *dev;
5131 
5132 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5133 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5134 	make_dev_alias(dev, "stdin");
5135 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5136 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5137 	make_dev_alias(dev, "stdout");
5138 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5139 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5140 	make_dev_alias(dev, "stderr");
5141 }
5142 
5143 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5144